fs/btrfs/inode.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/kernel.h>
  20 #include <linux/bio.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/file.h>
  23 #include <linux/fs.h>
  24 #include <linux/pagemap.h>
  25 #include <linux/highmem.h>
  26 #include <linux/time.h>
  27 #include <linux/init.h>
  28 #include <linux/string.h>
  29 #include <linux/backing-dev.h>
  30 #include <linux/mpage.h>
  31 #include <linux/swap.h>
  32 #include <linux/writeback.h>
  33 #include <linux/statfs.h>
  34 #include <linux/compat.h>
  35 #include <linux/bit_spinlock.h>
  36 #include <linux/xattr.h>
  37 #include <linux/posix_acl.h>
  38 #include <linux/falloc.h>
  39 #include <linux/slab.h>
  40 #include <linux/ratelimit.h>
  41 #include <linux/mount.h>
  42 #include "compat.h"
  43 #include "ctree.h"
  44 #include "disk-io.h"
  45 #include "transaction.h"
  46 #include "btrfs_inode.h"
  47 #include "ioctl.h"
  48 #include "print-tree.h"
  49 #include "ordered-data.h"
  50 #include "xattr.h"
  51 #include "tree-log.h"
  52 #include "volumes.h"
  53 #include "compression.h"
  54 #include "locking.h"
  55 #include "free-space-cache.h"
  56 #include "inode-map.h"
  57
  58 struct btrfs_iget_args {
  59         u64 ino;
  60         struct btrfs_root *root;
  61 };
  62
  63 static const struct inode_operations btrfs_dir_inode_operations;
  64 static const struct inode_operations btrfs_symlink_inode_operations;
  65 static const struct inode_operations btrfs_dir_ro_inode_operations;
  66 static const struct inode_operations btrfs_special_inode_operations;
  67 static const struct inode_operations btrfs_file_inode_operations;
  68 static const struct address_space_operations btrfs_aops;
  69 static const struct address_space_operations btrfs_symlink_aops;
  70 static const struct file_operations btrfs_dir_file_operations;
  71 static struct extent_io_ops btrfs_extent_io_ops;
  72
  73 static struct kmem_cache *btrfs_inode_cachep;
  74 static struct kmem_cache *btrfs_delalloc_work_cachep;
  75 struct kmem_cache *btrfs_trans_handle_cachep;
  76 struct kmem_cache *btrfs_transaction_cachep;
  77 struct kmem_cache *btrfs_path_cachep;
  78 struct kmem_cache *btrfs_free_space_cachep;
  79
  80 #define S_SHIFT 12
  81 static unsigned char btrfs_type_by_mode[S_IFMT >> S_SHIFT] = {
  82         [S_IFREG >> S_SHIFT]    = BTRFS_FT_REG_FILE,
  83         [S_IFDIR >> S_SHIFT]    = BTRFS_FT_DIR,
  84         [S_IFCHR >> S_SHIFT]    = BTRFS_FT_CHRDEV,
  85         [S_IFBLK >> S_SHIFT]    = BTRFS_FT_BLKDEV,
  86         [S_IFIFO >> S_SHIFT]    = BTRFS_FT_FIFO,
  87         [S_IFSOCK >> S_SHIFT]   = BTRFS_FT_SOCK,
  88         [S_IFLNK >> S_SHIFT]    = BTRFS_FT_SYMLINK,
  89 };
  90
  91 static int btrfs_setsize(struct inode *inode, loff_t newsize);
  92 static int btrfs_truncate(struct inode *inode);
  93 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent);
  94 static noinline int cow_file_range(struct inode *inode,
  95                                    struct page *locked_page,
  96                                    u64 start, u64 end, int *page_started,
  97                                    unsigned long *nr_written, int unlock);
  98
  99 static int btrfs_init_inode_security(struct btrfs_trans_handle *trans,
 100                                      struct inode *inode,  struct inode *dir,
 101                                      const struct qstr *qstr)
 102 {
 103         int err;
 104
 105         err = btrfs_init_acl(trans, inode, dir);
 106         if (!err)
 107                 err = btrfs_xattr_security_init(trans, inode, dir, qstr);
 108         return err;
 109 }
 110
 111 /*
 112  * this does all the hard work for inserting an inline extent into
 113  * the btree.  The caller should have done a btrfs_drop_extents so that
 114  * no overlapping inline items exist in the btree
 115  */
 116 static noinline int insert_inline_extent(struct btrfs_trans_handle *trans,
 117                                 struct btrfs_root *root, struct inode *inode,
 118                                 u64 start, size_t size, size_t compressed_size,
 119                                 int compress_type,
 120                                 struct page **compressed_pages)
 121 {
 122         struct btrfs_key key;
 123         struct btrfs_path *path;
 124         struct extent_buffer *leaf;
 125         struct page *page = NULL;
 126         char *kaddr;
 127         unsigned long ptr;
 128         struct btrfs_file_extent_item *ei;
 129         int err = 0;
 130         int ret;
 131         size_t cur_size = size;
 132         size_t datasize;
 133         unsigned long offset;
 134
 135         if (compressed_size && compressed_pages)
 136                 cur_size = compressed_size;
 137
 138         path = btrfs_alloc_path();
 139         if (!path)
 140                 return -ENOMEM;
 141
 142         path->leave_spinning = 1;
 143
 144         key.objectid = btrfs_ino(inode);
 145         key.offset = start;
 146         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
 147         datasize = btrfs_file_extent_calc_inline_size(cur_size);
 148
 149         inode_add_bytes(inode, size);
 150         ret = btrfs_insert_empty_item(trans, root, path, &key,
 151                                       datasize);
 152         if (ret) {
 153                 err = ret;
 154                 goto fail;
 155         }
 156         leaf = path->nodes[0];
 157         ei = btrfs_item_ptr(leaf, path->slots[0],
 158                             struct btrfs_file_extent_item);
 159         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
 160         btrfs_set_file_extent_type(leaf, ei, BTRFS_FILE_EXTENT_INLINE);
 161         btrfs_set_file_extent_encryption(leaf, ei, 0);
 162         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
 163         btrfs_set_file_extent_ram_bytes(leaf, ei, size);
 164         ptr = btrfs_file_extent_inline_start(ei);
 165
 166         if (compress_type != BTRFS_COMPRESS_NONE) {
 167                 struct page *cpage;
 168                 int i = 0;
 169                 while (compressed_size > 0) {
 170                         cpage = compressed_pages[i];
 171                         cur_size = min_t(unsigned long, compressed_size,
 172                                        PAGE_CACHE_SIZE);
 173
 174                         kaddr = kmap_atomic(cpage);
 175                         write_extent_buffer(leaf, kaddr, ptr, cur_size);
 176                         kunmap_atomic(kaddr);
 177
 178                         i++;
 179                         ptr += cur_size;
 180                         compressed_size -= cur_size;
 181                 }
 182                 btrfs_set_file_extent_compression(leaf, ei,
 183                                                   compress_type);
 184         } else {
 185                 page = find_get_page(inode->i_mapping,
 186                                      start >> PAGE_CACHE_SHIFT);
 187                 btrfs_set_file_extent_compression(leaf, ei, 0);
 188                 kaddr = kmap_atomic(page);
 189                 offset = start & (PAGE_CACHE_SIZE - 1);
 190                 write_extent_buffer(leaf, kaddr + offset, ptr, size);
 191                 kunmap_atomic(kaddr);
 192                 page_cache_release(page);
 193         }
 194         btrfs_mark_buffer_dirty(leaf);
 195         btrfs_free_path(path);
 196
 197         /*
 198          * we're an inline extent, so nobody can
 199          * extend the file past i_size without locking
 200          * a page we already have locked.
 201          *
 202          * We must do any isize and inode updates
 203          * before we unlock the pages.  Otherwise we
 204          * could end up racing with unlink.
 205          */
 206         BTRFS_I(inode)->disk_i_size = inode->i_size;
 207         ret = btrfs_update_inode(trans, root, inode);
 208
 209         return ret;
 210 fail:
 211         btrfs_free_path(path);
 212         return err;
 213 }
 214
 215
 216 /*
 217  * conditionally insert an inline extent into the file.  This
 218  * does the checks required to make sure the data is small enough
 219  * to fit as an inline extent.
 220  */
 221 static noinline int cow_file_range_inline(struct btrfs_trans_handle *trans,
 222                                  struct btrfs_root *root,
 223                                  struct inode *inode, u64 start, u64 end,
 224                                  size_t compressed_size, int compress_type,
 225                                  struct page **compressed_pages)
 226 {
 227         u64 isize = i_size_read(inode);
 228         u64 actual_end = min(end + 1, isize);
 229         u64 inline_len = actual_end - start;
 230         u64 aligned_end = (end + root->sectorsize - 1) &
 231                         ~((u64)root->sectorsize - 1);
 232         u64 data_len = inline_len;
 233         int ret;
 234
 235         if (compressed_size)
 236                 data_len = compressed_size;
 237
 238         if (start > 0 ||
 239             actual_end >= PAGE_CACHE_SIZE ||
 240             data_len >= BTRFS_MAX_INLINE_DATA_SIZE(root) ||
 241             (!compressed_size &&
 242             (actual_end & (root->sectorsize - 1)) == 0) ||
 243             end + 1 < isize ||
 244             data_len > root->fs_info->max_inline) {
 245                 return 1;
 246         }
 247
 248         ret = btrfs_drop_extents(trans, root, inode, start, aligned_end, 1);
 249         if (ret)
 250                 return ret;
 251
 252         if (isize > actual_end)
 253                 inline_len = min_t(u64, isize, actual_end);
 254         ret = insert_inline_extent(trans, root, inode, start,
 255                                    inline_len, compressed_size,
 256                                    compress_type, compressed_pages);
 257         if (ret && ret != -ENOSPC) {
 258                 btrfs_abort_transaction(trans, root, ret);
 259                 return ret;
 260         } else if (ret == -ENOSPC) {
 261                 return 1;
 262         }
 263
 264         btrfs_delalloc_release_metadata(inode, end + 1 - start);
 265         btrfs_drop_extent_cache(inode, start, aligned_end - 1, 0);
 266         return 0;
 267 }
 268
 269 struct async_extent {
 270         u64 start;
 271         u64 ram_size;
 272         u64 compressed_size;
 273         struct page **pages;
 274         unsigned long nr_pages;
 275         int compress_type;
 276         struct list_head list;
 277 };
 278
 279 struct async_cow {
 280         struct inode *inode;
 281         struct btrfs_root *root;
 282         struct page *locked_page;
 283         u64 start;
 284         u64 end;
 285         struct list_head extents;
 286         struct btrfs_work work;
 287 };
 288
 289 static noinline int add_async_extent(struct async_cow *cow,
 290                                      u64 start, u64 ram_size,
 291                                      u64 compressed_size,
 292                                      struct page **pages,
 293                                      unsigned long nr_pages,
 294                                      int compress_type)
 295 {
 296         struct async_extent *async_extent;
 297
 298         async_extent = kmalloc(sizeof(*async_extent), GFP_NOFS);
 299         BUG_ON(!async_extent); /* -ENOMEM */
 300         async_extent->start = start;
 301         async_extent->ram_size = ram_size;
 302         async_extent->compressed_size = compressed_size;
 303         async_extent->pages = pages;
 304         async_extent->nr_pages = nr_pages;
 305         async_extent->compress_type = compress_type;
 306         list_add_tail(&async_extent->list, &cow->extents);
 307         return 0;
 308 }
 309
 310 /*
 311  * we create compressed extents in two phases.  The first
 312  * phase compresses a range of pages that have already been
 313  * locked (both pages and state bits are locked).
 314  *
 315  * This is done inside an ordered work queue, and the compression
 316  * is spread across many cpus.  The actual IO submission is step
 317  * two, and the ordered work queue takes care of making sure that
 318  * happens in the same order things were put onto the queue by
 319  * writepages and friends.
 320  *
 321  * If this code finds it can't get good compression, it puts an
 322  * entry onto the work queue to write the uncompressed bytes.  This
 323  * makes sure that both compressed inodes and uncompressed inodes
 324  * are written in the same order that the flusher thread sent them
 325  * down.
 326  */
 327 static noinline int compress_file_range(struct inode *inode,
 328                                         struct page *locked_page,
 329                                         u64 start, u64 end,
 330                                         struct async_cow *async_cow,
 331                                         int *num_added)
 332 {
 333         struct btrfs_root *root = BTRFS_I(inode)->root;
 334         struct btrfs_trans_handle *trans;
 335         u64 num_bytes;
 336         u64 blocksize = root->sectorsize;
 337         u64 actual_end;
 338         u64 isize = i_size_read(inode);
 339         int ret = 0;
 340         struct page **pages = NULL;
 341         unsigned long nr_pages;
 342         unsigned long nr_pages_ret = 0;
 343         unsigned long total_compressed = 0;
 344         unsigned long total_in = 0;
 345         unsigned long max_compressed = 128 * 1024;
 346         unsigned long max_uncompressed = 128 * 1024;
 347         int i;
 348         int will_compress;
 349         int compress_type = root->fs_info->compress_type;
 350
 351         /* if this is a small write inside eof, kick off a defrag */
 352         if ((end - start + 1) < 16 * 1024 &&
 353             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 354                 btrfs_add_inode_defrag(NULL, inode);
 355
 356         actual_end = min_t(u64, isize, end + 1);
 357 again:
 358         will_compress = 0;
 359         nr_pages = (end >> PAGE_CACHE_SHIFT) - (start >> PAGE_CACHE_SHIFT) + 1;
 360         nr_pages = min(nr_pages, (128 * 1024UL) / PAGE_CACHE_SIZE);
 361
 362         /*
 363          * we don't want to send crud past the end of i_size through
 364          * compression, that's just a waste of CPU time.  So, if the
 365          * end of the file is before the start of our current
 366          * requested range of bytes, we bail out to the uncompressed
 367          * cleanup code that can deal with all of this.
 368          *
 369          * It isn't really the fastest way to fix things, but this is a
 370          * very uncommon corner.
 371          */
 372         if (actual_end <= start)
 373                 goto cleanup_and_bail_uncompressed;
 374
 375         total_compressed = actual_end - start;
 376
 377         /* we want to make sure that amount of ram required to uncompress
 378          * an extent is reasonable, so we limit the total size in ram
 379          * of a compressed extent to 128k.  This is a crucial number
 380          * because it also controls how easily we can spread reads across
 381          * cpus for decompression.
 382          *
 383          * We also want to make sure the amount of IO required to do
 384          * a random read is reasonably small, so we limit the size of
 385          * a compressed extent to 128k.
 386          */
 387         total_compressed = min(total_compressed, max_uncompressed);
 388         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 389         num_bytes = max(blocksize,  num_bytes);
 390         total_in = 0;
 391         ret = 0;
 392
 393         /*
 394          * we do compression for mount -o compress and when the
 395          * inode has not been flagged as nocompress.  This flag can
 396          * change at any time if we discover bad compression ratios.
 397          */
 398         if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS) &&
 399             (btrfs_test_opt(root, COMPRESS) ||
 400              (BTRFS_I(inode)->force_compress) ||
 401              (BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS))) {
 402                 WARN_ON(pages);
 403                 pages = kzalloc(sizeof(struct page *) * nr_pages, GFP_NOFS);
 404                 if (!pages) {
 405                         /* just bail out to the uncompressed code */
 406                         goto cont;
 407                 }
 408
 409                 if (BTRFS_I(inode)->force_compress)
 410                         compress_type = BTRFS_I(inode)->force_compress;
 411
 412                 ret = btrfs_compress_pages(compress_type,
 413                                            inode->i_mapping, start,
 414                                            total_compressed, pages,
 415                                            nr_pages, &nr_pages_ret,
 416                                            &total_in,
 417                                            &total_compressed,
 418                                            max_compressed);
 419
 420                 if (!ret) {
 421                         unsigned long offset = total_compressed &
 422                                 (PAGE_CACHE_SIZE - 1);
 423                         struct page *page = pages[nr_pages_ret - 1];
 424                         char *kaddr;
 425
 426                         /* zero the tail end of the last page, we might be
 427                          * sending it down to disk
 428                          */
 429                         if (offset) {
 430                                 kaddr = kmap_atomic(page);
 431                                 memset(kaddr + offset, 0,
 432                                        PAGE_CACHE_SIZE - offset);
 433                                 kunmap_atomic(kaddr);
 434                         }
 435                         will_compress = 1;
 436                 }
 437         }
 438 cont:
 439         if (start == 0) {
 440                 trans = btrfs_join_transaction(root);
 441                 if (IS_ERR(trans)) {
 442                         ret = PTR_ERR(trans);
 443                         trans = NULL;
 444                         goto cleanup_and_out;
 445                 }
 446                 trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 447
 448                 /* lets try to make an inline extent */
 449                 if (ret || total_in < (actual_end - start)) {
 450                         /* we didn't compress the entire range, try
 451                          * to make an uncompressed inline extent.
 452                          */
 453                         ret = cow_file_range_inline(trans, root, inode,
 454                                                     start, end, 0, 0, NULL);
 455                 } else {
 456                         /* try making a compressed inline extent */
 457                         ret = cow_file_range_inline(trans, root, inode,
 458                                                     start, end,
 459                                                     total_compressed,
 460                                                     compress_type, pages);
 461                 }
 462                 if (ret <= 0) {
 463                         /*
 464                          * inline extent creation worked or returned error,
 465                          * we don't need to create any more async work items.
 466                          * Unlock and free up our temp pages.
 467                          */
 468                         extent_clear_unlock_delalloc(inode,
 469                              &BTRFS_I(inode)->io_tree,
 470                              start, end, NULL,
 471                              EXTENT_CLEAR_UNLOCK_PAGE | EXTENT_CLEAR_DIRTY |
 472                              EXTENT_CLEAR_DELALLOC |
 473                              EXTENT_SET_WRITEBACK | EXTENT_END_WRITEBACK);
 474
 475                         btrfs_end_transaction(trans, root);
 476                         goto free_pages_out;
 477                 }
 478                 btrfs_end_transaction(trans, root);
 479         }
 480
 481         if (will_compress) {
 482                 /*
 483                  * we aren't doing an inline extent round the compressed size
 484                  * up to a block size boundary so the allocator does sane
 485                  * things
 486                  */
 487                 total_compressed = (total_compressed + blocksize - 1) &
 488                         ~(blocksize - 1);
 489
 490                 /*
 491                  * one last check to make sure the compression is really a
 492                  * win, compare the page count read with the blocks on disk
 493                  */
 494                 total_in = (total_in + PAGE_CACHE_SIZE - 1) &
 495                         ~(PAGE_CACHE_SIZE - 1);
 496                 if (total_compressed >= total_in) {
 497                         will_compress = 0;
 498                 } else {
 499                         num_bytes = total_in;
 500                 }
 501         }
 502         if (!will_compress && pages) {
 503                 /*
 504                  * the compression code ran but failed to make things smaller,
 505                  * free any pages it allocated and our page pointer array
 506                  */
 507                 for (i = 0; i < nr_pages_ret; i++) {
 508                         WARN_ON(pages[i]->mapping);
 509                         page_cache_release(pages[i]);
 510                 }
 511                 kfree(pages);
 512                 pages = NULL;
 513                 total_compressed = 0;
 514                 nr_pages_ret = 0;
 515
 516                 /* flag the file so we don't compress in the future */
 517                 if (!btrfs_test_opt(root, FORCE_COMPRESS) &&
 518                     !(BTRFS_I(inode)->force_compress)) {
 519                         BTRFS_I(inode)->flags |= BTRFS_INODE_NOCOMPRESS;
 520                 }
 521         }
 522         if (will_compress) {
 523                 *num_added += 1;
 524
 525                 /* the async work queues will take care of doing actual
 526                  * allocation on disk for these compressed pages,
 527                  * and will submit them to the elevator.
 528                  */
 529                 add_async_extent(async_cow, start, num_bytes,
 530                                  total_compressed, pages, nr_pages_ret,
 531                                  compress_type);
 532
 533                 if (start + num_bytes < end) {
 534                         start += num_bytes;
 535                         pages = NULL;
 536                         cond_resched();
 537                         goto again;
 538                 }
 539         } else {
 540 cleanup_and_bail_uncompressed:
 541                 /*
 542                  * No compression, but we still need to write the pages in
 543                  * the file we've been given so far.  redirty the locked
 544                  * page if it corresponds to our extent and set things up
 545                  * for the async work queue to run cow_file_range to do
 546                  * the normal delalloc dance
 547                  */
 548                 if (page_offset(locked_page) >= start &&
 549                     page_offset(locked_page) <= end) {
 550                         __set_page_dirty_nobuffers(locked_page);
 551                         /* unlocked later on in the async handlers */
 552                 }
 553                 add_async_extent(async_cow, start, end - start + 1,
 554                                  0, NULL, 0, BTRFS_COMPRESS_NONE);
 555                 *num_added += 1;
 556         }
 557
 558 out:
 559         return ret;
 560
 561 free_pages_out:
 562         for (i = 0; i < nr_pages_ret; i++) {
 563                 WARN_ON(pages[i]->mapping);
 564                 page_cache_release(pages[i]);
 565         }
 566         kfree(pages);
 567
 568         goto out;
 569
 570 cleanup_and_out:
 571         extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 572                                      start, end, NULL,
 573                                      EXTENT_CLEAR_UNLOCK_PAGE |
 574                                      EXTENT_CLEAR_DIRTY |
 575                                      EXTENT_CLEAR_DELALLOC |
 576                                      EXTENT_SET_WRITEBACK |
 577                                      EXTENT_END_WRITEBACK);
 578         if (!trans || IS_ERR(trans))
 579                 btrfs_error(root->fs_info, ret, "Failed to join transaction");
 580         else
 581                 btrfs_abort_transaction(trans, root, ret);
 582         goto free_pages_out;
 583 }
 584
 585 /*
 586  * phase two of compressed writeback.  This is the ordered portion
 587  * of the code, which only gets called in the order the work was
 588  * queued.  We walk all the async extents created by compress_file_range
 589  * and send them down to the disk.
 590  */
 591 static noinline int submit_compressed_extents(struct inode *inode,
 592                                               struct async_cow *async_cow)
 593 {
 594         struct async_extent *async_extent;
 595         u64 alloc_hint = 0;
 596         struct btrfs_trans_handle *trans;
 597         struct btrfs_key ins;
 598         struct extent_map *em;
 599         struct btrfs_root *root = BTRFS_I(inode)->root;
 600         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 601         struct extent_io_tree *io_tree;
 602         int ret = 0;
 603
 604         if (list_empty(&async_cow->extents))
 605                 return 0;
 606
 607
 608         while (!list_empty(&async_cow->extents)) {
 609                 async_extent = list_entry(async_cow->extents.next,
 610                                           struct async_extent, list);
 611                 list_del(&async_extent->list);
 612
 613                 io_tree = &BTRFS_I(inode)->io_tree;
 614
 615 retry:
 616                 /* did the compression code fall back to uncompressed IO? */
 617                 if (!async_extent->pages) {
 618                         int page_started = 0;
 619                         unsigned long nr_written = 0;
 620
 621                         lock_extent(io_tree, async_extent->start,
 622                                          async_extent->start +
 623                                          async_extent->ram_size - 1);
 624
 625                         /* allocate blocks */
 626                         ret = cow_file_range(inode, async_cow->locked_page,
 627                                              async_extent->start,
 628                                              async_extent->start +
 629                                              async_extent->ram_size - 1,
 630                                              &page_started, &nr_written, 0);
 631
 632                         /* JDM XXX */
 633
 634                         /*
 635                          * if page_started, cow_file_range inserted an
 636                          * inline extent and took care of all the unlocking
 637                          * and IO for us.  Otherwise, we need to submit
 638                          * all those pages down to the drive.
 639                          */
 640                         if (!page_started && !ret)
 641                                 extent_write_locked_range(io_tree,
 642                                                   inode, async_extent->start,
 643                                                   async_extent->start +
 644                                                   async_extent->ram_size - 1,
 645                                                   btrfs_get_extent,
 646                                                   WB_SYNC_ALL);
 647                         kfree(async_extent);
 648                         cond_resched();
 649                         continue;
 650                 }
 651
 652                 lock_extent(io_tree, async_extent->start,
 653                             async_extent->start + async_extent->ram_size - 1);
 654
 655                 trans = btrfs_join_transaction(root);
 656                 if (IS_ERR(trans)) {
 657                         ret = PTR_ERR(trans);
 658                 } else {
 659                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 660                         ret = btrfs_reserve_extent(trans, root,
 661                                            async_extent->compressed_size,
 662                                            async_extent->compressed_size,
 663                                            0, alloc_hint, &ins, 1);
 664                         if (ret && ret != -ENOSPC)
 665                                 btrfs_abort_transaction(trans, root, ret);
 666                         btrfs_end_transaction(trans, root);
 667                 }
 668
 669                 if (ret) {
 670                         int i;
 671                         for (i = 0; i < async_extent->nr_pages; i++) {
 672                                 WARN_ON(async_extent->pages[i]->mapping);
 673                                 page_cache_release(async_extent->pages[i]);
 674                         }
 675                         kfree(async_extent->pages);
 676                         async_extent->nr_pages = 0;
 677                         async_extent->pages = NULL;
 678                         unlock_extent(io_tree, async_extent->start,
 679                                       async_extent->start +
 680                                       async_extent->ram_size - 1);
 681                         if (ret == -ENOSPC)
 682                                 goto retry;
 683                         goto out_free; /* JDM: Requeue? */
 684                 }
 685
 686                 /*
 687                  * here we're doing allocation and writeback of the
 688                  * compressed pages
 689                  */
 690                 btrfs_drop_extent_cache(inode, async_extent->start,
 691                                         async_extent->start +
 692                                         async_extent->ram_size - 1, 0);
 693
 694                 em = alloc_extent_map();
 695                 BUG_ON(!em); /* -ENOMEM */
 696                 em->start = async_extent->start;
 697                 em->len = async_extent->ram_size;
 698                 em->orig_start = em->start;
 699
 700                 em->block_start = ins.objectid;
 701                 em->block_len = ins.offset;
 702                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 703                 em->compress_type = async_extent->compress_type;
 704                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 705                 set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 706
 707                 while (1) {
 708                         write_lock(&em_tree->lock);
 709                         ret = add_extent_mapping(em_tree, em);
 710                         write_unlock(&em_tree->lock);
 711                         if (ret != -EEXIST) {
 712                                 free_extent_map(em);
 713                                 break;
 714                         }
 715                         btrfs_drop_extent_cache(inode, async_extent->start,
 716                                                 async_extent->start +
 717                                                 async_extent->ram_size - 1, 0);
 718                 }
 719
 720                 ret = btrfs_add_ordered_extent_compress(inode,
 721                                                 async_extent->start,
 722                                                 ins.objectid,
 723                                                 async_extent->ram_size,
 724                                                 ins.offset,
 725                                                 BTRFS_ORDERED_COMPRESSED,
 726                                                 async_extent->compress_type);
 727                 BUG_ON(ret); /* -ENOMEM */
 728
 729                 /*
 730                  * clear dirty, set writeback and unlock the pages.
 731                  */
 732                 extent_clear_unlock_delalloc(inode,
 733                                 &BTRFS_I(inode)->io_tree,
 734                                 async_extent->start,
 735                                 async_extent->start +
 736                                 async_extent->ram_size - 1,
 737                                 NULL, EXTENT_CLEAR_UNLOCK_PAGE |
 738                                 EXTENT_CLEAR_UNLOCK |
 739                                 EXTENT_CLEAR_DELALLOC |
 740                                 EXTENT_CLEAR_DIRTY | EXTENT_SET_WRITEBACK);
 741
 742                 ret = btrfs_submit_compressed_write(inode,
 743                                     async_extent->start,
 744                                     async_extent->ram_size,
 745                                     ins.objectid,
 746                                     ins.offset, async_extent->pages,
 747                                     async_extent->nr_pages);
 748
 749                 BUG_ON(ret); /* -ENOMEM */
 750                 alloc_hint = ins.objectid + ins.offset;
 751                 kfree(async_extent);
 752                 cond_resched();
 753         }
 754         ret = 0;
 755 out:
 756         return ret;
 757 out_free:
 758         kfree(async_extent);
 759         goto out;
 760 }
 761
 762 static u64 get_extent_allocation_hint(struct inode *inode, u64 start,
 763                                       u64 num_bytes)
 764 {
 765         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 766         struct extent_map *em;
 767         u64 alloc_hint = 0;
 768
 769         read_lock(&em_tree->lock);
 770         em = search_extent_mapping(em_tree, start, num_bytes);
 771         if (em) {
 772                 /*
 773                  * if block start isn't an actual block number then find the
 774                  * first block in this inode and use that as a hint.  If that
 775                  * block is also bogus then just don't worry about it.
 776                  */
 777                 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
 778                         free_extent_map(em);
 779                         em = search_extent_mapping(em_tree, 0, 0);
 780                         if (em && em->block_start < EXTENT_MAP_LAST_BYTE)
 781                                 alloc_hint = em->block_start;
 782                         if (em)
 783                                 free_extent_map(em);
 784                 } else {
 785                         alloc_hint = em->block_start;
 786                         free_extent_map(em);
 787                 }
 788         }
 789         read_unlock(&em_tree->lock);
 790
 791         return alloc_hint;
 792 }
 793
 794 /*
 795  * when extent_io.c finds a delayed allocation range in the file,
 796  * the call backs end up in this code.  The basic idea is to
 797  * allocate extents on disk for the range, and create ordered data structs
 798  * in ram to track those extents.
 799  *
 800  * locked_page is the page that writepage had locked already.  We use
 801  * it to make sure we don't do extra locks or unlocks.
 802  *
 803  * *page_started is set to one if we unlock locked_page and do everything
 804  * required to start IO on it.  It may be clean and already done with
 805  * IO when we return.
 806  */
 807 static noinline int __cow_file_range(struct btrfs_trans_handle *trans,
 808                                      struct inode *inode,
 809                                      struct btrfs_root *root,
 810                                      struct page *locked_page,
 811                                      u64 start, u64 end, int *page_started,
 812                                      unsigned long *nr_written,
 813                                      int unlock)
 814 {
 815         u64 alloc_hint = 0;
 816         u64 num_bytes;
 817         unsigned long ram_size;
 818         u64 disk_num_bytes;
 819         u64 cur_alloc_size;
 820         u64 blocksize = root->sectorsize;
 821         struct btrfs_key ins;
 822         struct extent_map *em;
 823         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 824         int ret = 0;
 825
 826         BUG_ON(btrfs_is_free_space_inode(inode));
 827
 828         num_bytes = (end - start + blocksize) & ~(blocksize - 1);
 829         num_bytes = max(blocksize,  num_bytes);
 830         disk_num_bytes = num_bytes;
 831
 832         /* if this is a small write inside eof, kick off defrag */
 833         if (num_bytes < 64 * 1024 &&
 834             (start > 0 || end + 1 < BTRFS_I(inode)->disk_i_size))
 835                 btrfs_add_inode_defrag(trans, inode);
 836
 837         if (start == 0) {
 838                 /* lets try to make an inline extent */
 839                 ret = cow_file_range_inline(trans, root, inode,
 840                                             start, end, 0, 0, NULL);
 841                 if (ret == 0) {
 842                         extent_clear_unlock_delalloc(inode,
 843                                      &BTRFS_I(inode)->io_tree,
 844                                      start, end, NULL,
 845                                      EXTENT_CLEAR_UNLOCK_PAGE |
 846                                      EXTENT_CLEAR_UNLOCK |
 847                                      EXTENT_CLEAR_DELALLOC |
 848                                      EXTENT_CLEAR_DIRTY |
 849                                      EXTENT_SET_WRITEBACK |
 850                                      EXTENT_END_WRITEBACK);
 851
 852                         *nr_written = *nr_written +
 853                              (end - start + PAGE_CACHE_SIZE) / PAGE_CACHE_SIZE;
 854                         *page_started = 1;
 855                         goto out;
 856                 } else if (ret < 0) {
 857                         btrfs_abort_transaction(trans, root, ret);
 858                         goto out_unlock;
 859                 }
 860         }
 861
 862         BUG_ON(disk_num_bytes >
 863                btrfs_super_total_bytes(root->fs_info->super_copy));
 864
 865         alloc_hint = get_extent_allocation_hint(inode, start, num_bytes);
 866         btrfs_drop_extent_cache(inode, start, start + num_bytes - 1, 0);
 867
 868         while (disk_num_bytes > 0) {
 869                 unsigned long op;
 870
 871                 cur_alloc_size = disk_num_bytes;
 872                 ret = btrfs_reserve_extent(trans, root, cur_alloc_size,
 873                                            root->sectorsize, 0, alloc_hint,
 874                                            &ins, 1);
 875                 if (ret < 0) {
 876                         btrfs_abort_transaction(trans, root, ret);
 877                         goto out_unlock;
 878                 }
 879
 880                 em = alloc_extent_map();
 881                 BUG_ON(!em); /* -ENOMEM */
 882                 em->start = start;
 883                 em->orig_start = em->start;
 884                 ram_size = ins.offset;
 885                 em->len = ins.offset;
 886
 887                 em->block_start = ins.objectid;
 888                 em->block_len = ins.offset;
 889                 em->bdev = root->fs_info->fs_devices->latest_bdev;
 890                 set_bit(EXTENT_FLAG_PINNED, &em->flags);
 891
 892                 while (1) {
 893                         write_lock(&em_tree->lock);
 894                         ret = add_extent_mapping(em_tree, em);
 895                         write_unlock(&em_tree->lock);
 896                         if (ret != -EEXIST) {
 897                                 free_extent_map(em);
 898                                 break;
 899                         }
 900                         btrfs_drop_extent_cache(inode, start,
 901                                                 start + ram_size - 1, 0);
 902                 }
 903
 904                 cur_alloc_size = ins.offset;
 905                 ret = btrfs_add_ordered_extent(inode, start, ins.objectid,
 906                                                ram_size, cur_alloc_size, 0);
 907                 BUG_ON(ret); /* -ENOMEM */
 908
 909                 if (root->root_key.objectid ==
 910                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
 911                         ret = btrfs_reloc_clone_csums(inode, start,
 912                                                       cur_alloc_size);
 913                         if (ret) {
 914                                 btrfs_abort_transaction(trans, root, ret);
 915                                 goto out_unlock;
 916                         }
 917                 }
 918
 919                 if (disk_num_bytes < cur_alloc_size)
 920                         break;
 921
 922                 /* we're not doing compressed IO, don't unlock the first
 923                  * page (which the caller expects to stay locked), don't
 924                  * clear any dirty bits and don't set any writeback bits
 925                  *
 926                  * Do set the Private2 bit so we know this page was properly
 927                  * setup for writepage
 928                  */
 929                 op = unlock ? EXTENT_CLEAR_UNLOCK_PAGE : 0;
 930                 op |= EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
 931                         EXTENT_SET_PRIVATE2;
 932
 933                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
 934                                              start, start + ram_size - 1,
 935                                              locked_page, op);
 936                 disk_num_bytes -= cur_alloc_size;
 937                 num_bytes -= cur_alloc_size;
 938                 alloc_hint = ins.objectid + ins.offset;
 939                 start += cur_alloc_size;
 940         }
 941 out:
 942         return ret;
 943
 944 out_unlock:
 945         extent_clear_unlock_delalloc(inode,
 946                      &BTRFS_I(inode)->io_tree,
 947                      start, end, locked_page,
 948                      EXTENT_CLEAR_UNLOCK_PAGE |
 949                      EXTENT_CLEAR_UNLOCK |
 950                      EXTENT_CLEAR_DELALLOC |
 951                      EXTENT_CLEAR_DIRTY |
 952                      EXTENT_SET_WRITEBACK |
 953                      EXTENT_END_WRITEBACK);
 954
 955         goto out;
 956 }
 957
 958 static noinline int cow_file_range(struct inode *inode,
 959                                    struct page *locked_page,
 960                                    u64 start, u64 end, int *page_started,
 961                                    unsigned long *nr_written,
 962                                    int unlock)
 963 {
 964         struct btrfs_trans_handle *trans;
 965         struct btrfs_root *root = BTRFS_I(inode)->root;
 966         int ret;
 967
 968         trans = btrfs_join_transaction(root);
 969         if (IS_ERR(trans)) {
 970                 extent_clear_unlock_delalloc(inode,
 971                              &BTRFS_I(inode)->io_tree,
 972                              start, end, locked_page,
 973                              EXTENT_CLEAR_UNLOCK_PAGE |
 974                              EXTENT_CLEAR_UNLOCK |
 975                              EXTENT_CLEAR_DELALLOC |
 976                              EXTENT_CLEAR_DIRTY |
 977                              EXTENT_SET_WRITEBACK |
 978                              EXTENT_END_WRITEBACK);
 979                 return PTR_ERR(trans);
 980         }
 981         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
 982
 983         ret = __cow_file_range(trans, inode, root, locked_page, start, end,
 984                                page_started, nr_written, unlock);
 985
 986         btrfs_end_transaction(trans, root);
 987
 988         return ret;
 989 }
 990
 991 /*
 992  * work queue call back to started compression on a file and pages
 993  */
 994 static noinline void async_cow_start(struct btrfs_work *work)
 995 {
 996         struct async_cow *async_cow;
 997         int num_added = 0;
 998         async_cow = container_of(work, struct async_cow, work);
 999
1000         compress_file_range(async_cow->inode, async_cow->locked_page,
1001                             async_cow->start, async_cow->end, async_cow,
1002                             &num_added);
1003         if (num_added == 0) {
1004                 btrfs_add_delayed_iput(async_cow->inode);
1005                 async_cow->inode = NULL;
1006         }
1007 }
1008
1009 /*
1010  * work queue call back to submit previously compressed pages
1011  */
1012 static noinline void async_cow_submit(struct btrfs_work *work)
1013 {
1014         struct async_cow *async_cow;
1015         struct btrfs_root *root;
1016         unsigned long nr_pages;
1017
1018         async_cow = container_of(work, struct async_cow, work);
1019
1020         root = async_cow->root;
1021         nr_pages = (async_cow->end - async_cow->start + PAGE_CACHE_SIZE) >>
1022                 PAGE_CACHE_SHIFT;
1023
1024         if (atomic_sub_return(nr_pages, &root->fs_info->async_delalloc_pages) <
1025             5 * 1024 * 1024 &&
1026             waitqueue_active(&root->fs_info->async_submit_wait))
1027                 wake_up(&root->fs_info->async_submit_wait);
1028
1029         if (async_cow->inode)
1030                 submit_compressed_extents(async_cow->inode, async_cow);
1031 }
1032
1033 static noinline void async_cow_free(struct btrfs_work *work)
1034 {
1035         struct async_cow *async_cow;
1036         async_cow = container_of(work, struct async_cow, work);
1037         if (async_cow->inode)
1038                 btrfs_add_delayed_iput(async_cow->inode);
1039         kfree(async_cow);
1040 }
1041
1042 static int cow_file_range_async(struct inode *inode, struct page *locked_page,
1043                                 u64 start, u64 end, int *page_started,
1044                                 unsigned long *nr_written)
1045 {
1046         struct async_cow *async_cow;
1047         struct btrfs_root *root = BTRFS_I(inode)->root;
1048         unsigned long nr_pages;
1049         u64 cur_end;
1050         int limit = 10 * 1024 * 1024;
1051
1052         clear_extent_bit(&BTRFS_I(inode)->io_tree, start, end, EXTENT_LOCKED,
1053                          1, 0, NULL, GFP_NOFS);
1054         while (start < end) {
1055                 async_cow = kmalloc(sizeof(*async_cow), GFP_NOFS);
1056                 BUG_ON(!async_cow); /* -ENOMEM */
1057                 async_cow->inode = igrab(inode);
1058                 async_cow->root = root;
1059                 async_cow->locked_page = locked_page;
1060                 async_cow->start = start;
1061
1062                 if (BTRFS_I(inode)->flags & BTRFS_INODE_NOCOMPRESS)
1063                         cur_end = end;
1064                 else
1065                         cur_end = min(end, start + 512 * 1024 - 1);
1066
1067                 async_cow->end = cur_end;
1068                 INIT_LIST_HEAD(&async_cow->extents);
1069
1070                 async_cow->work.func = async_cow_start;
1071                 async_cow->work.ordered_func = async_cow_submit;
1072                 async_cow->work.ordered_free = async_cow_free;
1073                 async_cow->work.flags = 0;
1074
1075                 nr_pages = (cur_end - start + PAGE_CACHE_SIZE) >>
1076                         PAGE_CACHE_SHIFT;
1077                 atomic_add(nr_pages, &root->fs_info->async_delalloc_pages);
1078
1079                 btrfs_queue_worker(&root->fs_info->delalloc_workers,
1080                                    &async_cow->work);
1081
1082                 if (atomic_read(&root->fs_info->async_delalloc_pages) > limit) {
1083                         wait_event(root->fs_info->async_submit_wait,
1084                            (atomic_read(&root->fs_info->async_delalloc_pages) <
1085                             limit));
1086                 }
1087
1088                 while (atomic_read(&root->fs_info->async_submit_draining) &&
1089                       atomic_read(&root->fs_info->async_delalloc_pages)) {
1090                         wait_event(root->fs_info->async_submit_wait,
1091                           (atomic_read(&root->fs_info->async_delalloc_pages) ==
1092                            0));
1093                 }
1094
1095                 *nr_written += nr_pages;
1096                 start = cur_end + 1;
1097         }
1098         *page_started = 1;
1099         return 0;
1100 }
1101
1102 static noinline int csum_exist_in_range(struct btrfs_root *root,
1103                                         u64 bytenr, u64 num_bytes)
1104 {
1105         int ret;
1106         struct btrfs_ordered_sum *sums;
1107         LIST_HEAD(list);
1108
1109         ret = btrfs_lookup_csums_range(root->fs_info->csum_root, bytenr,
1110                                        bytenr + num_bytes - 1, &list, 0);
1111         if (ret == 0 && list_empty(&list))
1112                 return 0;
1113
1114         while (!list_empty(&list)) {
1115                 sums = list_entry(list.next, struct btrfs_ordered_sum, list);
1116                 list_del(&sums->list);
1117                 kfree(sums);
1118         }
1119         return 1;
1120 }
1121
1122 /*
1123  * when nowcow writeback call back.  This checks for snapshots or COW copies
1124  * of the extents that exist in the file, and COWs the file as required.
1125  *
1126  * If no cow copies or snapshots exist, we write directly to the existing
1127  * blocks on disk
1128  */
1129 static noinline int run_delalloc_nocow(struct inode *inode,
1130                                        struct page *locked_page,
1131                               u64 start, u64 end, int *page_started, int force,
1132                               unsigned long *nr_written)
1133 {
1134         struct btrfs_root *root = BTRFS_I(inode)->root;
1135         struct btrfs_trans_handle *trans;
1136         struct extent_buffer *leaf;
1137         struct btrfs_path *path;
1138         struct btrfs_file_extent_item *fi;
1139         struct btrfs_key found_key;
1140         u64 cow_start;
1141         u64 cur_offset;
1142         u64 extent_end;
1143         u64 extent_offset;
1144         u64 disk_bytenr;
1145         u64 num_bytes;
1146         int extent_type;
1147         int ret, err;
1148         int type;
1149         int nocow;
1150         int check_prev = 1;
1151         bool nolock;
1152         u64 ino = btrfs_ino(inode);
1153
1154         path = btrfs_alloc_path();
1155         if (!path) {
1156                 extent_clear_unlock_delalloc(inode,
1157                              &BTRFS_I(inode)->io_tree,
1158                              start, end, locked_page,
1159                              EXTENT_CLEAR_UNLOCK_PAGE |
1160                              EXTENT_CLEAR_UNLOCK |
1161                              EXTENT_CLEAR_DELALLOC |
1162                              EXTENT_CLEAR_DIRTY |
1163                              EXTENT_SET_WRITEBACK |
1164                              EXTENT_END_WRITEBACK);
1165                 return -ENOMEM;
1166         }
1167
1168         nolock = btrfs_is_free_space_inode(inode);
1169
1170         if (nolock)
1171                 trans = btrfs_join_transaction_nolock(root);
1172         else
1173                 trans = btrfs_join_transaction(root);
1174
1175         if (IS_ERR(trans)) {
1176                 extent_clear_unlock_delalloc(inode,
1177                              &BTRFS_I(inode)->io_tree,
1178                              start, end, locked_page,
1179                              EXTENT_CLEAR_UNLOCK_PAGE |
1180                              EXTENT_CLEAR_UNLOCK |
1181                              EXTENT_CLEAR_DELALLOC |
1182                              EXTENT_CLEAR_DIRTY |
1183                              EXTENT_SET_WRITEBACK |
1184                              EXTENT_END_WRITEBACK);
1185                 btrfs_free_path(path);
1186                 return PTR_ERR(trans);
1187         }
1188
1189         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1190
1191         cow_start = (u64)-1;
1192         cur_offset = start;
1193         while (1) {
1194                 ret = btrfs_lookup_file_extent(trans, root, path, ino,
1195                                                cur_offset, 0);
1196                 if (ret < 0) {
1197                         btrfs_abort_transaction(trans, root, ret);
1198                         goto error;
1199                 }
1200                 if (ret > 0 && path->slots[0] > 0 && check_prev) {
1201                         leaf = path->nodes[0];
1202                         btrfs_item_key_to_cpu(leaf, &found_key,
1203                                               path->slots[0] - 1);
1204                         if (found_key.objectid == ino &&
1205                             found_key.type == BTRFS_EXTENT_DATA_KEY)
1206                                 path->slots[0]--;
1207                 }
1208                 check_prev = 0;
1209 next_slot:
1210                 leaf = path->nodes[0];
1211                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1212                         ret = btrfs_next_leaf(root, path);
1213                         if (ret < 0) {
1214                                 btrfs_abort_transaction(trans, root, ret);
1215                                 goto error;
1216                         }
1217                         if (ret > 0)
1218                                 break;
1219                         leaf = path->nodes[0];
1220                 }
1221
1222                 nocow = 0;
1223                 disk_bytenr = 0;
1224                 num_bytes = 0;
1225                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1226
1227                 if (found_key.objectid > ino ||
1228                     found_key.type > BTRFS_EXTENT_DATA_KEY ||
1229                     found_key.offset > end)
1230                         break;
1231
1232                 if (found_key.offset > cur_offset) {
1233                         extent_end = found_key.offset;
1234                         extent_type = 0;
1235                         goto out_check;
1236                 }
1237
1238                 fi = btrfs_item_ptr(leaf, path->slots[0],
1239                                     struct btrfs_file_extent_item);
1240                 extent_type = btrfs_file_extent_type(leaf, fi);
1241
1242                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
1243                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1244                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
1245                         extent_offset = btrfs_file_extent_offset(leaf, fi);
1246                         extent_end = found_key.offset +
1247                                 btrfs_file_extent_num_bytes(leaf, fi);
1248                         if (extent_end <= start) {
1249                                 path->slots[0]++;
1250                                 goto next_slot;
1251                         }
1252                         if (disk_bytenr == 0)
1253                                 goto out_check;
1254                         if (btrfs_file_extent_compression(leaf, fi) ||
1255                             btrfs_file_extent_encryption(leaf, fi) ||
1256                             btrfs_file_extent_other_encoding(leaf, fi))
1257                                 goto out_check;
1258                         if (extent_type == BTRFS_FILE_EXTENT_REG && !force)
1259                                 goto out_check;
1260                         if (btrfs_extent_readonly(root, disk_bytenr))
1261                                 goto out_check;
1262                         if (btrfs_cross_ref_exist(trans, root, ino,
1263                                                   found_key.offset -
1264                                                   extent_offset, disk_bytenr))
1265                                 goto out_check;
1266                         disk_bytenr += extent_offset;
1267                         disk_bytenr += cur_offset - found_key.offset;
1268                         num_bytes = min(end + 1, extent_end) - cur_offset;
1269                         /*
1270                          * force cow if csum exists in the range.
1271                          * this ensure that csum for a given extent are
1272                          * either valid or do not exist.
1273                          */
1274                         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
1275                                 goto out_check;
1276                         nocow = 1;
1277                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
1278                         extent_end = found_key.offset +
1279                                 btrfs_file_extent_inline_len(leaf, fi);
1280                         extent_end = ALIGN(extent_end, root->sectorsize);
1281                 } else {
1282                         BUG_ON(1);
1283                 }
1284 out_check:
1285                 if (extent_end <= start) {
1286                         path->slots[0]++;
1287                         goto next_slot;
1288                 }
1289                 if (!nocow) {
1290                         if (cow_start == (u64)-1)
1291                                 cow_start = cur_offset;
1292                         cur_offset = extent_end;
1293                         if (cur_offset > end)
1294                                 break;
1295                         path->slots[0]++;
1296                         goto next_slot;
1297                 }
1298
1299                 btrfs_release_path(path);
1300                 if (cow_start != (u64)-1) {
1301                         ret = __cow_file_range(trans, inode, root, locked_page,
1302                                                cow_start, found_key.offset - 1,
1303                                                page_started, nr_written, 1);
1304                         if (ret) {
1305                                 btrfs_abort_transaction(trans, root, ret);
1306                                 goto error;
1307                         }
1308                         cow_start = (u64)-1;
1309                 }
1310
1311                 if (extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
1312                         struct extent_map *em;
1313                         struct extent_map_tree *em_tree;
1314                         em_tree = &BTRFS_I(inode)->extent_tree;
1315                         em = alloc_extent_map();
1316                         BUG_ON(!em); /* -ENOMEM */
1317                         em->start = cur_offset;
1318                         em->orig_start = em->start;
1319                         em->len = num_bytes;
1320                         em->block_len = num_bytes;
1321                         em->block_start = disk_bytenr;
1322                         em->bdev = root->fs_info->fs_devices->latest_bdev;
1323                         set_bit(EXTENT_FLAG_PINNED, &em->flags);
1324                         set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
1325                         while (1) {
1326                                 write_lock(&em_tree->lock);
1327                                 ret = add_extent_mapping(em_tree, em);
1328                                 write_unlock(&em_tree->lock);
1329                                 if (ret != -EEXIST) {
1330                                         free_extent_map(em);
1331                                         break;
1332                                 }
1333                                 btrfs_drop_extent_cache(inode, em->start,
1334                                                 em->start + em->len - 1, 0);
1335                         }
1336                         type = BTRFS_ORDERED_PREALLOC;
1337                 } else {
1338                         type = BTRFS_ORDERED_NOCOW;
1339                 }
1340
1341                 ret = btrfs_add_ordered_extent(inode, cur_offset, disk_bytenr,
1342                                                num_bytes, num_bytes, type);
1343                 BUG_ON(ret); /* -ENOMEM */
1344
1345                 if (root->root_key.objectid ==
1346                     BTRFS_DATA_RELOC_TREE_OBJECTID) {
1347                         ret = btrfs_reloc_clone_csums(inode, cur_offset,
1348                                                       num_bytes);
1349                         if (ret) {
1350                                 btrfs_abort_transaction(trans, root, ret);
1351                                 goto error;
1352                         }
1353                 }
1354
1355                 extent_clear_unlock_delalloc(inode, &BTRFS_I(inode)->io_tree,
1356                                 cur_offset, cur_offset + num_bytes - 1,
1357                                 locked_page, EXTENT_CLEAR_UNLOCK_PAGE |
1358                                 EXTENT_CLEAR_UNLOCK | EXTENT_CLEAR_DELALLOC |
1359                                 EXTENT_SET_PRIVATE2);
1360                 cur_offset = extent_end;
1361                 if (cur_offset > end)
1362                         break;
1363         }
1364         btrfs_release_path(path);
1365
1366         if (cur_offset <= end && cow_start == (u64)-1) {
1367                 cow_start = cur_offset;
1368                 cur_offset = end;
1369         }
1370
1371         if (cow_start != (u64)-1) {
1372                 ret = __cow_file_range(trans, inode, root, locked_page,
1373                                        cow_start, end,
1374                                        page_started, nr_written, 1);
1375                 if (ret) {
1376                         btrfs_abort_transaction(trans, root, ret);
1377                         goto error;
1378                 }
1379         }
1380
1381 error:
1382         err = btrfs_end_transaction(trans, root);
1383         if (!ret)
1384                 ret = err;
1385
1386         if (ret && cur_offset < end)
1387                 extent_clear_unlock_delalloc(inode,
1388                              &BTRFS_I(inode)->io_tree,
1389                              cur_offset, end, locked_page,
1390                              EXTENT_CLEAR_UNLOCK_PAGE |
1391                              EXTENT_CLEAR_UNLOCK |
1392                              EXTENT_CLEAR_DELALLOC |
1393                              EXTENT_CLEAR_DIRTY |
1394                              EXTENT_SET_WRITEBACK |
1395                              EXTENT_END_WRITEBACK);
1396
1397         btrfs_free_path(path);
1398         return ret;
1399 }
1400
1401 /*
1402  * extent_io.c call back to do delayed allocation processing
1403  */
1404 static int run_delalloc_range(struct inode *inode, struct page *locked_page,
1405                               u64 start, u64 end, int *page_started,
1406                               unsigned long *nr_written)
1407 {
1408         int ret;
1409         struct btrfs_root *root = BTRFS_I(inode)->root;
1410
1411         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) {
1412                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1413                                          page_started, 1, nr_written);
1414         } else if (BTRFS_I(inode)->flags & BTRFS_INODE_PREALLOC) {
1415                 ret = run_delalloc_nocow(inode, locked_page, start, end,
1416                                          page_started, 0, nr_written);
1417         } else if (!btrfs_test_opt(root, COMPRESS) &&
1418                    !(BTRFS_I(inode)->force_compress) &&
1419                    !(BTRFS_I(inode)->flags & BTRFS_INODE_COMPRESS)) {
1420                 ret = cow_file_range(inode, locked_page, start, end,
1421                                       page_started, nr_written, 1);
1422         } else {
1423                 set_bit(BTRFS_INODE_HAS_ASYNC_EXTENT,
1424                         &BTRFS_I(inode)->runtime_flags);
1425                 ret = cow_file_range_async(inode, locked_page, start, end,
1426                                            page_started, nr_written);
1427         }
1428         return ret;
1429 }
1430
1431 static void btrfs_split_extent_hook(struct inode *inode,
1432                                     struct extent_state *orig, u64 split)
1433 {
1434         /* not delalloc, ignore it */
1435         if (!(orig->state & EXTENT_DELALLOC))
1436                 return;
1437
1438         spin_lock(&BTRFS_I(inode)->lock);
1439         BTRFS_I(inode)->outstanding_extents++;
1440         spin_unlock(&BTRFS_I(inode)->lock);
1441 }
1442
1443 /*
1444  * extent_io.c merge_extent_hook, used to track merged delayed allocation
1445  * extents so we can keep track of new extents that are just merged onto old
1446  * extents, such as when we are doing sequential writes, so we can properly
1447  * account for the metadata space we'll need.
1448  */
1449 static void btrfs_merge_extent_hook(struct inode *inode,
1450                                     struct extent_state *new,
1451                                     struct extent_state *other)
1452 {
1453         /* not delalloc, ignore it */
1454         if (!(other->state & EXTENT_DELALLOC))
1455                 return;
1456
1457         spin_lock(&BTRFS_I(inode)->lock);
1458         BTRFS_I(inode)->outstanding_extents--;
1459         spin_unlock(&BTRFS_I(inode)->lock);
1460 }
1461
1462 /*
1463  * extent_io.c set_bit_hook, used to track delayed allocation
1464  * bytes in this file, and to maintain the list of inodes that
1465  * have pending delalloc work to be done.
1466  */
1467 static void btrfs_set_bit_hook(struct inode *inode,
1468                                struct extent_state *state, int *bits)
1469 {
1470
1471         /*
1472          * set_bit and clear bit hooks normally require _irqsave/restore
1473          * but in this case, we are only testing for the DELALLOC
1474          * bit, which is only set or cleared with irqs on
1475          */
1476         if (!(state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1477                 struct btrfs_root *root = BTRFS_I(inode)->root;
1478                 u64 len = state->end + 1 - state->start;
1479                 bool do_list = !btrfs_is_free_space_inode(inode);
1480
1481                 if (*bits & EXTENT_FIRST_DELALLOC) {
1482                         *bits &= ~EXTENT_FIRST_DELALLOC;
1483                 } else {
1484                         spin_lock(&BTRFS_I(inode)->lock);
1485                         BTRFS_I(inode)->outstanding_extents++;
1486                         spin_unlock(&BTRFS_I(inode)->lock);
1487                 }
1488
1489                 spin_lock(&root->fs_info->delalloc_lock);
1490                 BTRFS_I(inode)->delalloc_bytes += len;
1491                 root->fs_info->delalloc_bytes += len;
1492                 if (do_list && list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1493                         list_add_tail(&BTRFS_I(inode)->delalloc_inodes,
1494                                       &root->fs_info->delalloc_inodes);
1495                 }
1496                 spin_unlock(&root->fs_info->delalloc_lock);
1497         }
1498 }
1499
1500 /*
1501  * extent_io.c clear_bit_hook, see set_bit_hook for why
1502  */
1503 static void btrfs_clear_bit_hook(struct inode *inode,
1504                                  struct extent_state *state, int *bits)
1505 {
1506         /*
1507          * set_bit and clear bit hooks normally require _irqsave/restore
1508          * but in this case, we are only testing for the DELALLOC
1509          * bit, which is only set or cleared with irqs on
1510          */
1511         if ((state->state & EXTENT_DELALLOC) && (*bits & EXTENT_DELALLOC)) {
1512                 struct btrfs_root *root = BTRFS_I(inode)->root;
1513                 u64 len = state->end + 1 - state->start;
1514                 bool do_list = !btrfs_is_free_space_inode(inode);
1515
1516                 if (*bits & EXTENT_FIRST_DELALLOC) {
1517                         *bits &= ~EXTENT_FIRST_DELALLOC;
1518                 } else if (!(*bits & EXTENT_DO_ACCOUNTING)) {
1519                         spin_lock(&BTRFS_I(inode)->lock);
1520                         BTRFS_I(inode)->outstanding_extents--;
1521                         spin_unlock(&BTRFS_I(inode)->lock);
1522                 }
1523
1524                 if (*bits & EXTENT_DO_ACCOUNTING)
1525                         btrfs_delalloc_release_metadata(inode, len);
1526
1527                 if (root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID
1528                     && do_list)
1529                         btrfs_free_reserved_data_space(inode, len);
1530
1531                 spin_lock(&root->fs_info->delalloc_lock);
1532                 root->fs_info->delalloc_bytes -= len;
1533                 BTRFS_I(inode)->delalloc_bytes -= len;
1534
1535                 if (do_list && BTRFS_I(inode)->delalloc_bytes == 0 &&
1536                     !list_empty(&BTRFS_I(inode)->delalloc_inodes)) {
1537                         list_del_init(&BTRFS_I(inode)->delalloc_inodes);
1538                 }
1539                 spin_unlock(&root->fs_info->delalloc_lock);
1540         }
1541 }
1542
1543 /*
1544  * extent_io.c merge_bio_hook, this must check the chunk tree to make sure
1545  * we don't create bios that span stripes or chunks
1546  */
1547 int btrfs_merge_bio_hook(struct page *page, unsigned long offset,
1548                          size_t size, struct bio *bio,
1549                          unsigned long bio_flags)
1550 {
1551         struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
1552         u64 logical = (u64)bio->bi_sector << 9;
1553         u64 length = 0;
1554         u64 map_length;
1555         int ret;
1556
1557         if (bio_flags & EXTENT_BIO_COMPRESSED)
1558                 return 0;
1559
1560         length = bio->bi_size;
1561         map_length = length;
1562         ret = btrfs_map_block(root->fs_info, READ, logical,
1563                               &map_length, NULL, 0);
1564         /* Will always return 0 with map_multi == NULL */
1565         BUG_ON(ret < 0);
1566         if (map_length < length + size)
1567                 return 1;
1568         return 0;
1569 }
1570
1571 /*
1572  * in order to insert checksums into the metadata in large chunks,
1573  * we wait until bio submission time.   All the pages in the bio are
1574  * checksummed and sums are attached onto the ordered extent record.
1575  *
1576  * At IO completion time the cums attached on the ordered extent record
1577  * are inserted into the btree
1578  */
1579 static int __btrfs_submit_bio_start(struct inode *inode, int rw,
1580                                     struct bio *bio, int mirror_num,
1581                                     unsigned long bio_flags,
1582                                     u64 bio_offset)
1583 {
1584         struct btrfs_root *root = BTRFS_I(inode)->root;
1585         int ret = 0;
1586
1587         ret = btrfs_csum_one_bio(root, inode, bio, 0, 0);
1588         BUG_ON(ret); /* -ENOMEM */
1589         return 0;
1590 }
1591
1592 /*
1593  * in order to insert checksums into the metadata in large chunks,
1594  * we wait until bio submission time.   All the pages in the bio are
1595  * checksummed and sums are attached onto the ordered extent record.
1596  *
1597  * At IO completion time the cums attached on the ordered extent record
1598  * are inserted into the btree
1599  */
1600 static int __btrfs_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
1601                           int mirror_num, unsigned long bio_flags,
1602                           u64 bio_offset)
1603 {
1604         struct btrfs_root *root = BTRFS_I(inode)->root;
1605         int ret;
1606
1607         ret = btrfs_map_bio(root, rw, bio, mirror_num, 1);
1608         if (ret)
1609                 bio_endio(bio, ret);
1610         return ret;
1611 }
1612
1613 /*
1614  * extent_io.c submission hook. This does the right thing for csum calculation
1615  * on write, or reading the csums from the tree before a read
1616  */
1617 static int btrfs_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
1618                           int mirror_num, unsigned long bio_flags,
1619                           u64 bio_offset)
1620 {
1621         struct btrfs_root *root = BTRFS_I(inode)->root;
1622         int ret = 0;
1623         int skip_sum;
1624         int metadata = 0;
1625
1626         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
1627
1628         if (btrfs_is_free_space_inode(inode))
1629                 metadata = 2;
1630
1631         if (!(rw & REQ_WRITE)) {
1632                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, metadata);
1633                 if (ret)
1634                         goto out;
1635
1636                 if (bio_flags & EXTENT_BIO_COMPRESSED) {
1637                         ret = btrfs_submit_compressed_read(inode, bio,
1638                                                            mirror_num,
1639                                                            bio_flags);
1640                         goto out;
1641                 } else if (!skip_sum) {
1642                         ret = btrfs_lookup_bio_sums(root, inode, bio, NULL);
1643                         if (ret)
1644                                 goto out;
1645                 }
1646                 goto mapit;
1647         } else if (!skip_sum) {
1648                 /* csum items have already been cloned */
1649                 if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID)
1650                         goto mapit;
1651                 /* we're doing a write, do the async checksumming */
1652                 ret = btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
1653                                    inode, rw, bio, mirror_num,
1654                                    bio_flags, bio_offset,
1655                                    __btrfs_submit_bio_start,
1656                                    __btrfs_submit_bio_done);
1657                 goto out;
1658         }
1659
1660 mapit:
1661         ret = btrfs_map_bio(root, rw, bio, mirror_num, 0);
1662
1663 out:
1664         if (ret < 0)
1665                 bio_endio(bio, ret);
1666         return ret;
1667 }
1668
1669 /*
1670  * given a list of ordered sums record them in the inode.  This happens
1671  * at IO completion time based on sums calculated at bio submission time.
1672  */
1673 static noinline int add_pending_csums(struct btrfs_trans_handle *trans,
1674                              struct inode *inode, u64 file_offset,
1675                              struct list_head *list)
1676 {
1677         struct btrfs_ordered_sum *sum;
1678
1679         list_for_each_entry(sum, list, list) {
1680                 btrfs_csum_file_blocks(trans,
1681                        BTRFS_I(inode)->root->fs_info->csum_root, sum);
1682         }
1683         return 0;
1684 }
1685
1686 int btrfs_set_extent_delalloc(struct inode *inode, u64 start, u64 end,
1687                               struct extent_state **cached_state)
1688 {
1689         WARN_ON((end & (PAGE_CACHE_SIZE - 1)) == 0);
1690         return set_extent_delalloc(&BTRFS_I(inode)->io_tree, start, end,
1691                                    cached_state, GFP_NOFS);
1692 }
1693
1694 /* see btrfs_writepage_start_hook for details on why this is required */
1695 struct btrfs_writepage_fixup {
1696         struct page *page;
1697         struct btrfs_work work;
1698 };
1699
1700 static void btrfs_writepage_fixup_worker(struct btrfs_work *work)
1701 {
1702         struct btrfs_writepage_fixup *fixup;
1703         struct btrfs_ordered_extent *ordered;
1704         struct extent_state *cached_state = NULL;
1705         struct page *page;
1706         struct inode *inode;
1707         u64 page_start;
1708         u64 page_end;
1709         int ret;
1710
1711         fixup = container_of(work, struct btrfs_writepage_fixup, work);
1712         page = fixup->page;
1713 again:
1714         lock_page(page);
1715         if (!page->mapping || !PageDirty(page) || !PageChecked(page)) {
1716                 ClearPageChecked(page);
1717                 goto out_page;
1718         }
1719
1720         inode = page->mapping->host;
1721         page_start = page_offset(page);
1722         page_end = page_offset(page) + PAGE_CACHE_SIZE - 1;
1723
1724         lock_extent_bits(&BTRFS_I(inode)->io_tree, page_start, page_end, 0,
1725                          &cached_state);
1726
1727         /* already ordered? We're done */
1728         if (PagePrivate2(page))
1729                 goto out;
1730
1731         ordered = btrfs_lookup_ordered_extent(inode, page_start);
1732         if (ordered) {
1733                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start,
1734                                      page_end, &cached_state, GFP_NOFS);
1735                 unlock_page(page);
1736                 btrfs_start_ordered_extent(inode, ordered, 1);
1737                 btrfs_put_ordered_extent(ordered);
1738                 goto again;
1739         }
1740
1741         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
1742         if (ret) {
1743                 mapping_set_error(page->mapping, ret);
1744                 end_extent_writepage(page, ret, page_start, page_end);
1745                 ClearPageChecked(page);
1746                 goto out;
1747          }
1748
1749         btrfs_set_extent_delalloc(inode, page_start, page_end, &cached_state);
1750         ClearPageChecked(page);
1751         set_page_dirty(page);
1752 out:
1753         unlock_extent_cached(&BTRFS_I(inode)->io_tree, page_start, page_end,
1754                              &cached_state, GFP_NOFS);
1755 out_page:
1756         unlock_page(page);
1757         page_cache_release(page);
1758         kfree(fixup);
1759 }
1760
1761 /*
1762  * There are a few paths in the higher layers of the kernel that directly
1763  * set the page dirty bit without asking the filesystem if it is a
1764  * good idea.  This causes problems because we want to make sure COW
1765  * properly happens and the data=ordered rules are followed.
1766  *
1767  * In our case any range that doesn't have the ORDERED bit set
1768  * hasn't been properly setup for IO.  We kick off an async process
1769  * to fix it up.  The async helper will wait for ordered extents, set
1770  * the delalloc bit and make it safe to write the page.
1771  */
1772 static int btrfs_writepage_start_hook(struct page *page, u64 start, u64 end)
1773 {
1774         struct inode *inode = page->mapping->host;
1775         struct btrfs_writepage_fixup *fixup;
1776         struct btrfs_root *root = BTRFS_I(inode)->root;
1777
1778         /* this page is properly in the ordered list */
1779         if (TestClearPagePrivate2(page))
1780                 return 0;
1781
1782         if (PageChecked(page))
1783                 return -EAGAIN;
1784
1785         fixup = kzalloc(sizeof(*fixup), GFP_NOFS);
1786         if (!fixup)
1787                 return -EAGAIN;
1788
1789         SetPageChecked(page);
1790         page_cache_get(page);
1791         fixup->work.func = btrfs_writepage_fixup_worker;
1792         fixup->page = page;
1793         btrfs_queue_worker(&root->fs_info->fixup_workers, &fixup->work);
1794         return -EBUSY;
1795 }
1796
1797 static int insert_reserved_file_extent(struct btrfs_trans_handle *trans,
1798                                        struct inode *inode, u64 file_pos,
1799                                        u64 disk_bytenr, u64 disk_num_bytes,
1800                                        u64 num_bytes, u64 ram_bytes,
1801                                        u8 compression, u8 encryption,
1802                                        u16 other_encoding, int extent_type)
1803 {
1804         struct btrfs_root *root = BTRFS_I(inode)->root;
1805         struct btrfs_file_extent_item *fi;
1806         struct btrfs_path *path;
1807         struct extent_buffer *leaf;
1808         struct btrfs_key ins;
1809         int ret;
1810
1811         path = btrfs_alloc_path();
1812         if (!path)
1813                 return -ENOMEM;
1814
1815         path->leave_spinning = 1;
1816
1817         /*
1818          * we may be replacing one extent in the tree with another.
1819          * The new extent is pinned in the extent map, and we don't want
1820          * to drop it from the cache until it is completely in the btree.
1821          *
1822          * So, tell btrfs_drop_extents to leave this extent in the cache.
1823          * the caller is expected to unpin it and allow it to be merged
1824          * with the others.
1825          */
1826         ret = btrfs_drop_extents(trans, root, inode, file_pos,
1827                                  file_pos + num_bytes, 0);
1828         if (ret)
1829                 goto out;
1830
1831         ins.objectid = btrfs_ino(inode);
1832         ins.offset = file_pos;
1833         ins.type = BTRFS_EXTENT_DATA_KEY;
1834         ret = btrfs_insert_empty_item(trans, root, path, &ins, sizeof(*fi));
1835         if (ret)
1836                 goto out;
1837         leaf = path->nodes[0];
1838         fi = btrfs_item_ptr(leaf, path->slots[0],
1839                             struct btrfs_file_extent_item);
1840         btrfs_set_file_extent_generation(leaf, fi, trans->transid);
1841         btrfs_set_file_extent_type(leaf, fi, extent_type);
1842         btrfs_set_file_extent_disk_bytenr(leaf, fi, disk_bytenr);
1843         btrfs_set_file_extent_disk_num_bytes(leaf, fi, disk_num_bytes);
1844         btrfs_set_file_extent_offset(leaf, fi, 0);
1845         btrfs_set_file_extent_num_bytes(leaf, fi, num_bytes);
1846         btrfs_set_file_extent_ram_bytes(leaf, fi, ram_bytes);
1847         btrfs_set_file_extent_compression(leaf, fi, compression);
1848         btrfs_set_file_extent_encryption(leaf, fi, encryption);
1849         btrfs_set_file_extent_other_encoding(leaf, fi, other_encoding);
1850
1851         btrfs_mark_buffer_dirty(leaf);
1852         btrfs_release_path(path);
1853
1854         inode_add_bytes(inode, num_bytes);
1855
1856         ins.objectid = disk_bytenr;
1857         ins.offset = disk_num_bytes;
1858         ins.type = BTRFS_EXTENT_ITEM_KEY;
1859         ret = btrfs_alloc_reserved_file_extent(trans, root,
1860                                         root->root_key.objectid,
1861                                         btrfs_ino(inode), file_pos, &ins);
1862 out:
1863         btrfs_free_path(path);
1864
1865         return ret;
1866 }
1867
1868 /*
1869  * helper function for btrfs_finish_ordered_io, this
1870  * just reads in some of the csum leaves to prime them into ram
1871  * before we start the transaction.  It limits the amount of btree
1872  * reads required while inside the transaction.
1873  */
1874 /* as ordered data IO finishes, this gets called so we can finish
1875  * an ordered extent if the range of bytes in the file it covers are
1876  * fully written.
1877  */
1878 static int btrfs_finish_ordered_io(struct btrfs_ordered_extent *ordered_extent)
1879 {
1880         struct inode *inode = ordered_extent->inode;
1881         struct btrfs_root *root = BTRFS_I(inode)->root;
1882         struct btrfs_trans_handle *trans = NULL;
1883         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
1884         struct extent_state *cached_state = NULL;
1885         int compress_type = 0;
1886         int ret;
1887         bool nolock;
1888
1889         nolock = btrfs_is_free_space_inode(inode);
1890
1891         if (test_bit(BTRFS_ORDERED_IOERR, &ordered_extent->flags)) {
1892                 ret = -EIO;
1893                 goto out;
1894         }
1895
1896         if (test_bit(BTRFS_ORDERED_NOCOW, &ordered_extent->flags)) {
1897                 BUG_ON(!list_empty(&ordered_extent->list)); /* Logic error */
1898                 ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1899                 if (!ret) {
1900                         if (nolock)
1901                                 trans = btrfs_join_transaction_nolock(root);
1902                         else
1903                                 trans = btrfs_join_transaction(root);
1904                         if (IS_ERR(trans)) {
1905                                 ret = PTR_ERR(trans);
1906                                 trans = NULL;
1907                                 goto out;
1908                         }
1909                         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1910                         ret = btrfs_update_inode_fallback(trans, root, inode);
1911                         if (ret) /* -ENOMEM or corruption */
1912                                 btrfs_abort_transaction(trans, root, ret);
1913                 }
1914                 goto out;
1915         }
1916
1917         lock_extent_bits(io_tree, ordered_extent->file_offset,
1918                          ordered_extent->file_offset + ordered_extent->len - 1,
1919                          0, &cached_state);
1920
1921         if (nolock)
1922                 trans = btrfs_join_transaction_nolock(root);
1923         else
1924                 trans = btrfs_join_transaction(root);
1925         if (IS_ERR(trans)) {
1926                 ret = PTR_ERR(trans);
1927                 trans = NULL;
1928                 goto out_unlock;
1929         }
1930         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
1931
1932         if (test_bit(BTRFS_ORDERED_COMPRESSED, &ordered_extent->flags))
1933                 compress_type = ordered_extent->compress_type;
1934         if (test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1935                 BUG_ON(compress_type);
1936                 ret = btrfs_mark_extent_written(trans, inode,
1937                                                 ordered_extent->file_offset,
1938                                                 ordered_extent->file_offset +
1939                                                 ordered_extent->len);
1940         } else {
1941                 BUG_ON(root == root->fs_info->tree_root);
1942                 ret = insert_reserved_file_extent(trans, inode,
1943                                                 ordered_extent->file_offset,
1944                                                 ordered_extent->start,
1945                                                 ordered_extent->disk_len,
1946                                                 ordered_extent->len,
1947                                                 ordered_extent->len,
1948                                                 compress_type, 0, 0,
1949                                                 BTRFS_FILE_EXTENT_REG);
1950         }
1951         unpin_extent_cache(&BTRFS_I(inode)->extent_tree,
1952                            ordered_extent->file_offset, ordered_extent->len,
1953                            trans->transid);
1954         if (ret < 0) {
1955                 btrfs_abort_transaction(trans, root, ret);
1956                 goto out_unlock;
1957         }
1958
1959         add_pending_csums(trans, inode, ordered_extent->file_offset,
1960                           &ordered_extent->list);
1961
1962         ret = btrfs_ordered_update_i_size(inode, 0, ordered_extent);
1963         if (!ret || !test_bit(BTRFS_ORDERED_PREALLOC, &ordered_extent->flags)) {
1964                 ret = btrfs_update_inode_fallback(trans, root, inode);
1965                 if (ret) { /* -ENOMEM or corruption */
1966                         btrfs_abort_transaction(trans, root, ret);
1967                         goto out_unlock;
1968                 }
1969         } else {
1970                 btrfs_set_inode_last_trans(trans, inode);
1971         }
1972         ret = 0;
1973 out_unlock:
1974         unlock_extent_cached(io_tree, ordered_extent->file_offset,
1975                              ordered_extent->file_offset +
1976                              ordered_extent->len - 1, &cached_state, GFP_NOFS);
1977 out:
1978         if (root != root->fs_info->tree_root)
1979                 btrfs_delalloc_release_metadata(inode, ordered_extent->len);
1980         if (trans)
1981                 btrfs_end_transaction(trans, root);
1982
1983         if (ret)
1984                 clear_extent_uptodate(io_tree, ordered_extent->file_offset,
1985                                       ordered_extent->file_offset +
1986                                       ordered_extent->len - 1, NULL, GFP_NOFS);
1987
1988         /*
1989          * This needs to be done to make sure anybody waiting knows we are done
1990          * updating everything for this ordered extent.
1991          */
1992         btrfs_remove_ordered_extent(inode, ordered_extent);
1993
1994         /* once for us */
1995         btrfs_put_ordered_extent(ordered_extent);
1996         /* once for the tree */
1997         btrfs_put_ordered_extent(ordered_extent);
1998
1999         return ret;
2000 }
2001
2002 static void finish_ordered_fn(struct btrfs_work *work)
2003 {
2004         struct btrfs_ordered_extent *ordered_extent;
2005         ordered_extent = container_of(work, struct btrfs_ordered_extent, work);
2006         btrfs_finish_ordered_io(ordered_extent);
2007 }
2008
2009 static int btrfs_writepage_end_io_hook(struct page *page, u64 start, u64 end,
2010                                 struct extent_state *state, int uptodate)
2011 {
2012         struct inode *inode = page->mapping->host;
2013         struct btrfs_root *root = BTRFS_I(inode)->root;
2014         struct btrfs_ordered_extent *ordered_extent = NULL;
2015         struct btrfs_workers *workers;
2016
2017         trace_btrfs_writepage_end_io_hook(page, start, end, uptodate);
2018
2019         ClearPagePrivate2(page);
2020         if (!btrfs_dec_test_ordered_pending(inode, &ordered_extent, start,
2021                                             end - start + 1, uptodate))
2022                 return 0;
2023
2024         ordered_extent->work.func = finish_ordered_fn;
2025         ordered_extent->work.flags = 0;
2026
2027         if (btrfs_is_free_space_inode(inode))
2028                 workers = &root->fs_info->endio_freespace_worker;
2029         else
2030                 workers = &root->fs_info->endio_write_workers;
2031         btrfs_queue_worker(workers, &ordered_extent->work);
2032
2033         return 0;
2034 }
2035
2036 /*
2037  * when reads are done, we need to check csums to verify the data is correct
2038  * if there's a match, we allow the bio to finish.  If not, the code in
2039  * extent_io.c will try to find good copies for us.
2040  */
2041 static int btrfs_readpage_end_io_hook(struct page *page, u64 start, u64 end,
2042                                struct extent_state *state, int mirror)
2043 {
2044         size_t offset = start - ((u64)page->index << PAGE_CACHE_SHIFT);
2045         struct inode *inode = page->mapping->host;
2046         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2047         char *kaddr;
2048         u64 private = ~(u32)0;
2049         int ret;
2050         struct btrfs_root *root = BTRFS_I(inode)->root;
2051         u32 csum = ~(u32)0;
2052
2053         if (PageChecked(page)) {
2054                 ClearPageChecked(page);
2055                 goto good;
2056         }
2057
2058         if (BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)
2059                 goto good;
2060
2061         if (root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID &&
2062             test_range_bit(io_tree, start, end, EXTENT_NODATASUM, 1, NULL)) {
2063                 clear_extent_bits(io_tree, start, end, EXTENT_NODATASUM,
2064                                   GFP_NOFS);
2065                 return 0;
2066         }
2067
2068         if (state && state->start == start) {
2069                 private = state->private;
2070                 ret = 0;
2071         } else {
2072                 ret = get_state_private(io_tree, start, &private);
2073         }
2074         kaddr = kmap_atomic(page);
2075         if (ret)
2076                 goto zeroit;
2077
2078         csum = btrfs_csum_data(root, kaddr + offset, csum,  end - start + 1);
2079         btrfs_csum_final(csum, (char *)&csum);
2080         if (csum != private)
2081                 goto zeroit;
2082
2083         kunmap_atomic(kaddr);
2084 good:
2085         return 0;
2086
2087 zeroit:
2088         printk_ratelimited(KERN_INFO "btrfs csum failed ino %llu off %llu csum %u "
2089                        "private %llu\n",
2090                        (unsigned long long)btrfs_ino(page->mapping->host),
2091                        (unsigned long long)start, csum,
2092                        (unsigned long long)private);
2093         memset(kaddr + offset, 1, end - start + 1);
2094         flush_dcache_page(page);
2095         kunmap_atomic(kaddr);
2096         if (private == 0)
2097                 return 0;
2098         return -EIO;
2099 }
2100
2101 struct delayed_iput {
2102         struct list_head list;
2103         struct inode *inode;
2104 };
2105
2106 /* JDM: If this is fs-wide, why can't we add a pointer to
2107  * btrfs_inode instead and avoid the allocation? */
2108 void btrfs_add_delayed_iput(struct inode *inode)
2109 {
2110         struct btrfs_fs_info *fs_info = BTRFS_I(inode)->root->fs_info;
2111         struct delayed_iput *delayed;
2112
2113         if (atomic_add_unless(&inode->i_count, -1, 1))
2114                 return;
2115
2116         delayed = kmalloc(sizeof(*delayed), GFP_NOFS | __GFP_NOFAIL);
2117         delayed->inode = inode;
2118
2119         spin_lock(&fs_info->delayed_iput_lock);
2120         list_add_tail(&delayed->list, &fs_info->delayed_iputs);
2121         spin_unlock(&fs_info->delayed_iput_lock);
2122 }
2123
2124 void btrfs_run_delayed_iputs(struct btrfs_root *root)
2125 {
2126         LIST_HEAD(list);
2127         struct btrfs_fs_info *fs_info = root->fs_info;
2128         struct delayed_iput *delayed;
2129         int empty;
2130
2131         spin_lock(&fs_info->delayed_iput_lock);
2132         empty = list_empty(&fs_info->delayed_iputs);
2133         spin_unlock(&fs_info->delayed_iput_lock);
2134         if (empty)
2135                 return;
2136
2137         spin_lock(&fs_info->delayed_iput_lock);
2138         list_splice_init(&fs_info->delayed_iputs, &list);
2139         spin_unlock(&fs_info->delayed_iput_lock);
2140
2141         while (!list_empty(&list)) {
2142                 delayed = list_entry(list.next, struct delayed_iput, list);
2143                 list_del(&delayed->list);
2144                 iput(delayed->inode);
2145                 kfree(delayed);
2146         }
2147 }
2148
2149 enum btrfs_orphan_cleanup_state {
2150         ORPHAN_CLEANUP_STARTED  = 1,
2151         ORPHAN_CLEANUP_DONE     = 2,
2152 };
2153
2154 /*
2155  * This is called in transaction commit time. If there are no orphan
2156  * files in the subvolume, it removes orphan item and frees block_rsv
2157  * structure.
2158  */
2159 void btrfs_orphan_commit_root(struct btrfs_trans_handle *trans,
2160                               struct btrfs_root *root)
2161 {
2162         struct btrfs_block_rsv *block_rsv;
2163         int ret;
2164
2165         if (atomic_read(&root->orphan_inodes) ||
2166             root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE)
2167                 return;
2168
2169         spin_lock(&root->orphan_lock);
2170         if (atomic_read(&root->orphan_inodes)) {
2171                 spin_unlock(&root->orphan_lock);
2172                 return;
2173         }
2174
2175         if (root->orphan_cleanup_state != ORPHAN_CLEANUP_DONE) {
2176                 spin_unlock(&root->orphan_lock);
2177                 return;
2178         }
2179
2180         block_rsv = root->orphan_block_rsv;
2181         root->orphan_block_rsv = NULL;
2182         spin_unlock(&root->orphan_lock);
2183
2184         if (root->orphan_item_inserted &&
2185             btrfs_root_refs(&root->root_item) > 0) {
2186                 ret = btrfs_del_orphan_item(trans, root->fs_info->tree_root,
2187                                             root->root_key.objectid);
2188                 BUG_ON(ret);
2189                 root->orphan_item_inserted = 0;
2190         }
2191
2192         if (block_rsv) {
2193                 WARN_ON(block_rsv->size > 0);
2194                 btrfs_free_block_rsv(root, block_rsv);
2195         }
2196 }
2197
2198 /*
2199  * This creates an orphan entry for the given inode in case something goes
2200  * wrong in the middle of an unlink/truncate.
2201  *
2202  * NOTE: caller of this function should reserve 5 units of metadata for
2203  *       this function.
2204  */
2205 int btrfs_orphan_add(struct btrfs_trans_handle *trans, struct inode *inode)
2206 {
2207         struct btrfs_root *root = BTRFS_I(inode)->root;
2208         struct btrfs_block_rsv *block_rsv = NULL;
2209         int reserve = 0;
2210         int insert = 0;
2211         int ret;
2212
2213         if (!root->orphan_block_rsv) {
2214                 block_rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
2215                 if (!block_rsv)
2216                         return -ENOMEM;
2217         }
2218
2219         spin_lock(&root->orphan_lock);
2220         if (!root->orphan_block_rsv) {
2221                 root->orphan_block_rsv = block_rsv;
2222         } else if (block_rsv) {
2223                 btrfs_free_block_rsv(root, block_rsv);
2224                 block_rsv = NULL;
2225         }
2226
2227         if (!test_and_set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2228                               &BTRFS_I(inode)->runtime_flags)) {
2229 #if 0
2230                 /*
2231                  * For proper ENOSPC handling, we should do orphan
2232                  * cleanup when mounting. But this introduces backward
2233                  * compatibility issue.
2234                  */
2235                 if (!xchg(&root->orphan_item_inserted, 1))
2236                         insert = 2;
2237                 else
2238                         insert = 1;
2239 #endif
2240                 insert = 1;
2241                 atomic_inc(&root->orphan_inodes);
2242         }
2243
2244         if (!test_and_set_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2245                               &BTRFS_I(inode)->runtime_flags))
2246                 reserve = 1;
2247         spin_unlock(&root->orphan_lock);
2248
2249         /* grab metadata reservation from transaction handle */
2250         if (reserve) {
2251                 ret = btrfs_orphan_reserve_metadata(trans, inode);
2252                 BUG_ON(ret); /* -ENOSPC in reservation; Logic error? JDM */
2253         }
2254
2255         /* insert an orphan item to track this unlinked/truncated file */
2256         if (insert >= 1) {
2257                 ret = btrfs_insert_orphan_item(trans, root, btrfs_ino(inode));
2258                 if (ret && ret != -EEXIST) {
2259                         clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2260                                   &BTRFS_I(inode)->runtime_flags);
2261                         btrfs_abort_transaction(trans, root, ret);
2262                         return ret;
2263                 }
2264                 ret = 0;
2265         }
2266
2267         /* insert an orphan item to track subvolume contains orphan files */
2268         if (insert >= 2) {
2269                 ret = btrfs_insert_orphan_item(trans, root->fs_info->tree_root,
2270                                                root->root_key.objectid);
2271                 if (ret && ret != -EEXIST) {
2272                         btrfs_abort_transaction(trans, root, ret);
2273                         return ret;
2274                 }
2275         }
2276         return 0;
2277 }
2278
2279 /*
2280  * We have done the truncate/delete so we can go ahead and remove the orphan
2281  * item for this particular inode.
2282  */
2283 int btrfs_orphan_del(struct btrfs_trans_handle *trans, struct inode *inode)
2284 {
2285         struct btrfs_root *root = BTRFS_I(inode)->root;
2286         int delete_item = 0;
2287         int release_rsv = 0;
2288         int ret = 0;
2289
2290         spin_lock(&root->orphan_lock);
2291         if (test_and_clear_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2292                                &BTRFS_I(inode)->runtime_flags))
2293                 delete_item = 1;
2294
2295         if (test_and_clear_bit(BTRFS_INODE_ORPHAN_META_RESERVED,
2296                                &BTRFS_I(inode)->runtime_flags))
2297                 release_rsv = 1;
2298         spin_unlock(&root->orphan_lock);
2299
2300         if (trans && delete_item) {
2301                 ret = btrfs_del_orphan_item(trans, root, btrfs_ino(inode));
2302                 BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
2303         }
2304
2305         if (release_rsv) {
2306                 btrfs_orphan_release_metadata(inode);
2307                 atomic_dec(&root->orphan_inodes);
2308         }
2309
2310         return 0;
2311 }
2312
2313 /*
2314  * this cleans up any orphans that may be left on the list from the last use
2315  * of this root.
2316  */
2317 int btrfs_orphan_cleanup(struct btrfs_root *root)
2318 {
2319         struct btrfs_path *path;
2320         struct extent_buffer *leaf;
2321         struct btrfs_key key, found_key;
2322         struct btrfs_trans_handle *trans;
2323         struct inode *inode;
2324         u64 last_objectid = 0;
2325         int ret = 0, nr_unlink = 0, nr_truncate = 0;
2326
2327         if (cmpxchg(&root->orphan_cleanup_state, 0, ORPHAN_CLEANUP_STARTED))
2328                 return 0;
2329
2330         path = btrfs_alloc_path();
2331         if (!path) {
2332                 ret = -ENOMEM;
2333                 goto out;
2334         }
2335         path->reada = -1;
2336
2337         key.objectid = BTRFS_ORPHAN_OBJECTID;
2338         btrfs_set_key_type(&key, BTRFS_ORPHAN_ITEM_KEY);
2339         key.offset = (u64)-1;
2340
2341         while (1) {
2342                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2343                 if (ret < 0)
2344                         goto out;
2345
2346                 /*
2347                  * if ret == 0 means we found what we were searching for, which
2348                  * is weird, but possible, so only screw with path if we didn't
2349                  * find the key and see if we have stuff that matches
2350                  */
2351                 if (ret > 0) {
2352                         ret = 0;
2353                         if (path->slots[0] == 0)
2354                                 break;
2355                         path->slots[0]--;
2356                 }
2357
2358                 /* pull out the item */
2359                 leaf = path->nodes[0];
2360                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2361
2362                 /* make sure the item matches what we want */
2363                 if (found_key.objectid != BTRFS_ORPHAN_OBJECTID)
2364                         break;
2365                 if (btrfs_key_type(&found_key) != BTRFS_ORPHAN_ITEM_KEY)
2366                         break;
2367
2368                 /* release the path since we're done with it */
2369                 btrfs_release_path(path);
2370
2371                 /*
2372                  * this is where we are basically btrfs_lookup, without the
2373                  * crossing root thing.  we store the inode number in the
2374                  * offset of the orphan item.
2375                  */
2376
2377                 if (found_key.offset == last_objectid) {
2378                         printk(KERN_ERR "btrfs: Error removing orphan entry, "
2379                                "stopping orphan cleanup\n");
2380                         ret = -EINVAL;
2381                         goto out;
2382                 }
2383
2384                 last_objectid = found_key.offset;
2385
2386                 found_key.objectid = found_key.offset;
2387                 found_key.type = BTRFS_INODE_ITEM_KEY;
2388                 found_key.offset = 0;
2389                 inode = btrfs_iget(root->fs_info->sb, &found_key, root, NULL);
2390                 ret = PTR_RET(inode);
2391                 if (ret && ret != -ESTALE)
2392                         goto out;
2393
2394                 if (ret == -ESTALE && root == root->fs_info->tree_root) {
2395                         struct btrfs_root *dead_root;
2396                         struct btrfs_fs_info *fs_info = root->fs_info;
2397                         int is_dead_root = 0;
2398
2399                         /*
2400                          * this is an orphan in the tree root. Currently these
2401                          * could come from 2 sources:
2402                          *  a) a snapshot deletion in progress
2403                          *  b) a free space cache inode
2404                          * We need to distinguish those two, as the snapshot
2405                          * orphan must not get deleted.
2406                          * find_dead_roots already ran before us, so if this
2407                          * is a snapshot deletion, we should find the root
2408                          * in the dead_roots list
2409                          */
2410                         spin_lock(&fs_info->trans_lock);
2411                         list_for_each_entry(dead_root, &fs_info->dead_roots,
2412                                             root_list) {
2413                                 if (dead_root->root_key.objectid ==
2414                                     found_key.objectid) {
2415                                         is_dead_root = 1;
2416                                         break;
2417                                 }
2418                         }
2419                         spin_unlock(&fs_info->trans_lock);
2420                         if (is_dead_root) {
2421                                 /* prevent this orphan from being found again */
2422                                 key.offset = found_key.objectid - 1;
2423                                 continue;
2424                         }
2425                 }
2426                 /*
2427                  * Inode is already gone but the orphan item is still there,
2428                  * kill the orphan item.
2429                  */
2430                 if (ret == -ESTALE) {
2431                         trans = btrfs_start_transaction(root, 1);
2432                         if (IS_ERR(trans)) {
2433                                 ret = PTR_ERR(trans);
2434                                 goto out;
2435                         }
2436                         printk(KERN_ERR "auto deleting %Lu\n",
2437                                found_key.objectid);
2438                         ret = btrfs_del_orphan_item(trans, root,
2439                                                     found_key.objectid);
2440                         BUG_ON(ret); /* -ENOMEM or corruption (JDM: Recheck) */
2441                         btrfs_end_transaction(trans, root);
2442                         continue;
2443                 }
2444
2445                 /*
2446                  * add this inode to the orphan list so btrfs_orphan_del does
2447                  * the proper thing when we hit it
2448                  */
2449                 set_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
2450                         &BTRFS_I(inode)->runtime_flags);
2451
2452                 /* if we have links, this was a truncate, lets do that */
2453                 if (inode->i_nlink) {
2454                         if (!S_ISREG(inode->i_mode)) {
2455                                 WARN_ON(1);
2456                                 iput(inode);
2457                                 continue;
2458                         }
2459                         nr_truncate++;
2460                         ret = btrfs_truncate(inode);
2461                 } else {
2462                         nr_unlink++;
2463                 }
2464
2465                 /* this will do delete_inode and everything for us */
2466                 iput(inode);
2467                 if (ret)
2468                         goto out;
2469         }
2470         /* release the path since we're done with it */
2471         btrfs_release_path(path);
2472
2473         root->orphan_cleanup_state = ORPHAN_CLEANUP_DONE;
2474
2475         if (root->orphan_block_rsv)
2476                 btrfs_block_rsv_release(root, root->orphan_block_rsv,
2477                                         (u64)-1);
2478
2479         if (root->orphan_block_rsv || root->orphan_item_inserted) {
2480                 trans = btrfs_join_transaction(root);
2481                 if (!IS_ERR(trans))
2482                         btrfs_end_transaction(trans, root);
2483         }
2484
2485         if (nr_unlink)
2486                 printk(KERN_INFO "btrfs: unlinked %d orphans\n", nr_unlink);
2487         if (nr_truncate)
2488                 printk(KERN_INFO "btrfs: truncated %d orphans\n", nr_truncate);
2489
2490 out:
2491         if (ret)
2492                 printk(KERN_CRIT "btrfs: could not do orphan cleanup %d\n", ret);
2493         btrfs_free_path(path);
2494         return ret;
2495 }
2496
2497 /*
2498  * very simple check to peek ahead in the leaf looking for xattrs.  If we
2499  * don't find any xattrs, we know there can't be any acls.
2500  *
2501  * slot is the slot the inode is in, objectid is the objectid of the inode
2502  */
2503 static noinline int acls_after_inode_item(struct extent_buffer *leaf,
2504                                           int slot, u64 objectid)
2505 {
2506         u32 nritems = btrfs_header_nritems(leaf);
2507         struct btrfs_key found_key;
2508         int scanned = 0;
2509
2510         slot++;
2511         while (slot < nritems) {
2512                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2513
2514                 /* we found a different objectid, there must not be acls */
2515                 if (found_key.objectid != objectid)
2516                         return 0;
2517
2518                 /* we found an xattr, assume we've got an acl */
2519                 if (found_key.type == BTRFS_XATTR_ITEM_KEY)
2520                         return 1;
2521
2522                 /*
2523                  * we found a key greater than an xattr key, there can't
2524                  * be any acls later on
2525                  */
2526                 if (found_key.type > BTRFS_XATTR_ITEM_KEY)
2527                         return 0;
2528
2529                 slot++;
2530                 scanned++;
2531
2532                 /*
2533                  * it goes inode, inode backrefs, xattrs, extents,
2534                  * so if there are a ton of hard links to an inode there can
2535                  * be a lot of backrefs.  Don't waste time searching too hard,
2536                  * this is just an optimization
2537                  */
2538                 if (scanned >= 8)
2539                         break;
2540         }
2541         /* we hit the end of the leaf before we found an xattr or
2542          * something larger than an xattr.  We have to assume the inode
2543          * has acls
2544          */
2545         return 1;
2546 }
2547
2548 /*
2549  * read an inode from the btree into the in-memory inode
2550  */
2551 static void btrfs_read_locked_inode(struct inode *inode)
2552 {
2553         struct btrfs_path *path;
2554         struct extent_buffer *leaf;
2555         struct btrfs_inode_item *inode_item;
2556         struct btrfs_timespec *tspec;
2557         struct btrfs_root *root = BTRFS_I(inode)->root;
2558         struct btrfs_key location;
2559         int maybe_acls;
2560         u32 rdev;
2561         int ret;
2562         bool filled = false;
2563
2564         ret = btrfs_fill_inode(inode, &rdev);
2565         if (!ret)
2566                 filled = true;
2567
2568         path = btrfs_alloc_path();
2569         if (!path)
2570                 goto make_bad;
2571
2572         path->leave_spinning = 1;
2573         memcpy(&location, &BTRFS_I(inode)->location, sizeof(location));
2574
2575         ret = btrfs_lookup_inode(NULL, root, path, &location, 0);
2576         if (ret)
2577                 goto make_bad;
2578
2579         leaf = path->nodes[0];
2580
2581         if (filled)
2582                 goto cache_acl;
2583
2584         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2585                                     struct btrfs_inode_item);
2586         inode->i_mode = btrfs_inode_mode(leaf, inode_item);
2587         set_nlink(inode, btrfs_inode_nlink(leaf, inode_item));
2588         i_uid_write(inode, btrfs_inode_uid(leaf, inode_item));
2589         i_gid_write(inode, btrfs_inode_gid(leaf, inode_item));
2590         btrfs_i_size_write(inode, btrfs_inode_size(leaf, inode_item));
2591
2592         tspec = btrfs_inode_atime(inode_item);
2593         inode->i_atime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2594         inode->i_atime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2595
2596         tspec = btrfs_inode_mtime(inode_item);
2597         inode->i_mtime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2598         inode->i_mtime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2599
2600         tspec = btrfs_inode_ctime(inode_item);
2601         inode->i_ctime.tv_sec = btrfs_timespec_sec(leaf, tspec);
2602         inode->i_ctime.tv_nsec = btrfs_timespec_nsec(leaf, tspec);
2603
2604         inode_set_bytes(inode, btrfs_inode_nbytes(leaf, inode_item));
2605         BTRFS_I(inode)->generation = btrfs_inode_generation(leaf, inode_item);
2606         BTRFS_I(inode)->last_trans = btrfs_inode_transid(leaf, inode_item);
2607
2608         /*
2609          * If we were modified in the current generation and evicted from memory
2610          * and then re-read we need to do a full sync since we don't have any
2611          * idea about which extents were modified before we were evicted from
2612          * cache.
2613          */
2614         if (BTRFS_I(inode)->last_trans == root->fs_info->generation)
2615                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
2616                         &BTRFS_I(inode)->runtime_flags);
2617
2618         inode->i_version = btrfs_inode_sequence(leaf, inode_item);
2619         inode->i_generation = BTRFS_I(inode)->generation;
2620         inode->i_rdev = 0;
2621         rdev = btrfs_inode_rdev(leaf, inode_item);
2622
2623         BTRFS_I(inode)->index_cnt = (u64)-1;
2624         BTRFS_I(inode)->flags = btrfs_inode_flags(leaf, inode_item);
2625 cache_acl:
2626         /*
2627          * try to precache a NULL acl entry for files that don't have
2628          * any xattrs or acls
2629          */
2630         maybe_acls = acls_after_inode_item(leaf, path->slots[0],
2631                                            btrfs_ino(inode));
2632         if (!maybe_acls)
2633                 cache_no_acl(inode);
2634
2635         btrfs_free_path(path);
2636
2637         switch (inode->i_mode & S_IFMT) {
2638         case S_IFREG:
2639                 inode->i_mapping->a_ops = &btrfs_aops;
2640                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2641                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
2642                 inode->i_fop = &btrfs_file_operations;
2643                 inode->i_op = &btrfs_file_inode_operations;
2644                 break;
2645         case S_IFDIR:
2646                 inode->i_fop = &btrfs_dir_file_operations;
2647                 if (root == root->fs_info->tree_root)
2648                         inode->i_op = &btrfs_dir_ro_inode_operations;
2649                 else
2650                         inode->i_op = &btrfs_dir_inode_operations;
2651                 break;
2652         case S_IFLNK:
2653                 inode->i_op = &btrfs_symlink_inode_operations;
2654                 inode->i_mapping->a_ops = &btrfs_symlink_aops;
2655                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
2656                 break;
2657         default:
2658                 inode->i_op = &btrfs_special_inode_operations;
2659                 init_special_inode(inode, inode->i_mode, rdev);
2660                 break;
2661         }
2662
2663         btrfs_update_iflags(inode);
2664         return;
2665
2666 make_bad:
2667         btrfs_free_path(path);
2668         make_bad_inode(inode);
2669 }
2670
2671 /*
2672  * given a leaf and an inode, copy the inode fields into the leaf
2673  */
2674 static void fill_inode_item(struct btrfs_trans_handle *trans,
2675                             struct extent_buffer *leaf,
2676                             struct btrfs_inode_item *item,
2677                             struct inode *inode)
2678 {
2679         btrfs_set_inode_uid(leaf, item, i_uid_read(inode));
2680         btrfs_set_inode_gid(leaf, item, i_gid_read(inode));
2681         btrfs_set_inode_size(leaf, item, BTRFS_I(inode)->disk_i_size);
2682         btrfs_set_inode_mode(leaf, item, inode->i_mode);
2683         btrfs_set_inode_nlink(leaf, item, inode->i_nlink);
2684
2685         btrfs_set_timespec_sec(leaf, btrfs_inode_atime(item),
2686                                inode->i_atime.tv_sec);
2687         btrfs_set_timespec_nsec(leaf, btrfs_inode_atime(item),
2688                                 inode->i_atime.tv_nsec);
2689
2690         btrfs_set_timespec_sec(leaf, btrfs_inode_mtime(item),
2691                                inode->i_mtime.tv_sec);
2692         btrfs_set_timespec_nsec(leaf, btrfs_inode_mtime(item),
2693                                 inode->i_mtime.tv_nsec);
2694
2695         btrfs_set_timespec_sec(leaf, btrfs_inode_ctime(item),
2696                                inode->i_ctime.tv_sec);
2697         btrfs_set_timespec_nsec(leaf, btrfs_inode_ctime(item),
2698                                 inode->i_ctime.tv_nsec);
2699
2700         btrfs_set_inode_nbytes(leaf, item, inode_get_bytes(inode));
2701         btrfs_set_inode_generation(leaf, item, BTRFS_I(inode)->generation);
2702         btrfs_set_inode_sequence(leaf, item, inode->i_version);
2703         btrfs_set_inode_transid(leaf, item, trans->transid);
2704         btrfs_set_inode_rdev(leaf, item, inode->i_rdev);
2705         btrfs_set_inode_flags(leaf, item, BTRFS_I(inode)->flags);
2706         btrfs_set_inode_block_group(leaf, item, 0);
2707 }
2708
2709 /*
2710  * copy everything in the in-memory inode into the btree.
2711  */
2712 static noinline int btrfs_update_inode_item(struct btrfs_trans_handle *trans,
2713                                 struct btrfs_root *root, struct inode *inode)
2714 {
2715         struct btrfs_inode_item *inode_item;
2716         struct btrfs_path *path;
2717         struct extent_buffer *leaf;
2718         int ret;
2719
2720         path = btrfs_alloc_path();
2721         if (!path)
2722                 return -ENOMEM;
2723
2724         path->leave_spinning = 1;
2725         ret = btrfs_lookup_inode(trans, root, path, &BTRFS_I(inode)->location,
2726                                  1);
2727         if (ret) {
2728                 if (ret > 0)
2729                         ret = -ENOENT;
2730                 goto failed;
2731         }
2732
2733         btrfs_unlock_up_safe(path, 1);
2734         leaf = path->nodes[0];
2735         inode_item = btrfs_item_ptr(leaf, path->slots[0],
2736                                     struct btrfs_inode_item);
2737
2738         fill_inode_item(trans, leaf, inode_item, inode);
2739         btrfs_mark_buffer_dirty(leaf);
2740         btrfs_set_inode_last_trans(trans, inode);
2741         ret = 0;
2742 failed:
2743         btrfs_free_path(path);
2744         return ret;
2745 }
2746
2747 /*
2748  * copy everything in the in-memory inode into the btree.
2749  */
2750 noinline int btrfs_update_inode(struct btrfs_trans_handle *trans,
2751                                 struct btrfs_root *root, struct inode *inode)
2752 {
2753         int ret;
2754
2755         /*
2756          * If the inode is a free space inode, we can deadlock during commit
2757          * if we put it into the delayed code.
2758          *
2759          * The data relocation inode should also be directly updated
2760          * without delay
2761          */
2762         if (!btrfs_is_free_space_inode(inode)
2763             && root->root_key.objectid != BTRFS_DATA_RELOC_TREE_OBJECTID) {
2764                 btrfs_update_root_times(trans, root);
2765
2766                 ret = btrfs_delayed_update_inode(trans, root, inode);
2767                 if (!ret)
2768                         btrfs_set_inode_last_trans(trans, inode);
2769                 return ret;
2770         }
2771
2772         return btrfs_update_inode_item(trans, root, inode);
2773 }
2774
2775 noinline int btrfs_update_inode_fallback(struct btrfs_trans_handle *trans,
2776                                          struct btrfs_root *root,
2777                                          struct inode *inode)
2778 {
2779         int ret;
2780
2781         ret = btrfs_update_inode(trans, root, inode);
2782         if (ret == -ENOSPC)
2783                 return btrfs_update_inode_item(trans, root, inode);
2784         return ret;
2785 }
2786
2787 /*
2788  * unlink helper that gets used here in inode.c and in the tree logging
2789  * recovery code.  It remove a link in a directory with a given name, and
2790  * also drops the back refs in the inode to the directory
2791  */
2792 static int __btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2793                                 struct btrfs_root *root,
2794                                 struct inode *dir, struct inode *inode,
2795                                 const char *name, int name_len)
2796 {
2797         struct btrfs_path *path;
2798         int ret = 0;
2799         struct extent_buffer *leaf;
2800         struct btrfs_dir_item *di;
2801         struct btrfs_key key;
2802         u64 index;
2803         u64 ino = btrfs_ino(inode);
2804         u64 dir_ino = btrfs_ino(dir);
2805
2806         path = btrfs_alloc_path();
2807         if (!path) {
2808                 ret = -ENOMEM;
2809                 goto out;
2810         }
2811
2812         path->leave_spinning = 1;
2813         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
2814                                     name, name_len, -1);
2815         if (IS_ERR(di)) {
2816                 ret = PTR_ERR(di);
2817                 goto err;
2818         }
2819         if (!di) {
2820                 ret = -ENOENT;
2821                 goto err;
2822         }
2823         leaf = path->nodes[0];
2824         btrfs_dir_item_key_to_cpu(leaf, di, &key);
2825         ret = btrfs_delete_one_dir_name(trans, root, path, di);
2826         if (ret)
2827                 goto err;
2828         btrfs_release_path(path);
2829
2830         ret = btrfs_del_inode_ref(trans, root, name, name_len, ino,
2831                                   dir_ino, &index);
2832         if (ret) {
2833                 printk(KERN_INFO "btrfs failed to delete reference to %.*s, "
2834                        "inode %llu parent %llu\n", name_len, name,
2835                        (unsigned long long)ino, (unsigned long long)dir_ino);
2836                 btrfs_abort_transaction(trans, root, ret);
2837                 goto err;
2838         }
2839
2840         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
2841         if (ret) {
2842                 btrfs_abort_transaction(trans, root, ret);
2843                 goto err;
2844         }
2845
2846         ret = btrfs_del_inode_ref_in_log(trans, root, name, name_len,
2847                                          inode, dir_ino);
2848         if (ret != 0 && ret != -ENOENT) {
2849                 btrfs_abort_transaction(trans, root, ret);
2850                 goto err;
2851         }
2852
2853         ret = btrfs_del_dir_entries_in_log(trans, root, name, name_len,
2854                                            dir, index);
2855         if (ret == -ENOENT)
2856                 ret = 0;
2857 err:
2858         btrfs_free_path(path);
2859         if (ret)
2860                 goto out;
2861
2862         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
2863         inode_inc_iversion(inode);
2864         inode_inc_iversion(dir);
2865         inode->i_ctime = dir->i_mtime = dir->i_ctime = CURRENT_TIME;
2866         ret = btrfs_update_inode(trans, root, dir);
2867 out:
2868         return ret;
2869 }
2870
2871 int btrfs_unlink_inode(struct btrfs_trans_handle *trans,
2872                        struct btrfs_root *root,
2873                        struct inode *dir, struct inode *inode,
2874                        const char *name, int name_len)
2875 {
2876         int ret;
2877         ret = __btrfs_unlink_inode(trans, root, dir, inode, name, name_len);
2878         if (!ret) {
2879                 btrfs_drop_nlink(inode);
2880                 ret = btrfs_update_inode(trans, root, inode);
2881         }
2882         return ret;
2883 }
2884
2885
2886 /* helper to check if there is any shared block in the path */
2887 static int check_path_shared(struct btrfs_root *root,
2888                              struct btrfs_path *path)
2889 {
2890         struct extent_buffer *eb;
2891         int level;
2892         u64 refs = 1;
2893
2894         for (level = 0; level < BTRFS_MAX_LEVEL; level++) {
2895                 int ret;
2896
2897                 if (!path->nodes[level])
2898                         break;
2899                 eb = path->nodes[level];
2900                 if (!btrfs_block_can_be_shared(root, eb))
2901                         continue;
2902                 ret = btrfs_lookup_extent_info(NULL, root, eb->start, eb->len,
2903                                                &refs, NULL);
2904                 if (refs > 1)
2905                         return 1;
2906         }
2907         return 0;
2908 }
2909
2910 /*
2911  * helper to start transaction for unlink and rmdir.
2912  *
2913  * unlink and rmdir are special in btrfs, they do not always free space.
2914  * so in enospc case, we should make sure they will free space before
2915  * allowing them to use the global metadata reservation.
2916  */
2917 static struct btrfs_trans_handle *__unlink_start_trans(struct inode *dir,
2918                                                        struct dentry *dentry)
2919 {
2920         struct btrfs_trans_handle *trans;
2921         struct btrfs_root *root = BTRFS_I(dir)->root;
2922         struct btrfs_path *path;
2923         struct btrfs_dir_item *di;
2924         struct inode *inode = dentry->d_inode;
2925         u64 index;
2926         int check_link = 1;
2927         int err = -ENOSPC;
2928         int ret;
2929         u64 ino = btrfs_ino(inode);
2930         u64 dir_ino = btrfs_ino(dir);
2931
2932         /*
2933          * 1 for the possible orphan item
2934          * 1 for the dir item
2935          * 1 for the dir index
2936          * 1 for the inode ref
2937          * 1 for the inode ref in the tree log
2938          * 2 for the dir entries in the log
2939          * 1 for the inode
2940          */
2941         trans = btrfs_start_transaction(root, 8);
2942         if (!IS_ERR(trans) || PTR_ERR(trans) != -ENOSPC)
2943                 return trans;
2944
2945         if (ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
2946                 return ERR_PTR(-ENOSPC);
2947
2948         /* check if there is someone else holds reference */
2949         if (S_ISDIR(inode->i_mode) && atomic_read(&inode->i_count) > 1)
2950                 return ERR_PTR(-ENOSPC);
2951
2952         if (atomic_read(&inode->i_count) > 2)
2953                 return ERR_PTR(-ENOSPC);
2954
2955         if (xchg(&root->fs_info->enospc_unlink, 1))
2956                 return ERR_PTR(-ENOSPC);
2957
2958         path = btrfs_alloc_path();
2959         if (!path) {
2960                 root->fs_info->enospc_unlink = 0;
2961                 return ERR_PTR(-ENOMEM);
2962         }
2963
2964         /* 1 for the orphan item */
2965         trans = btrfs_start_transaction(root, 1);
2966         if (IS_ERR(trans)) {
2967                 btrfs_free_path(path);
2968                 root->fs_info->enospc_unlink = 0;
2969                 return trans;
2970         }
2971
2972         path->skip_locking = 1;
2973         path->search_commit_root = 1;
2974
2975         ret = btrfs_lookup_inode(trans, root, path,
2976                                 &BTRFS_I(dir)->location, 0);
2977         if (ret < 0) {
2978                 err = ret;
2979                 goto out;
2980         }
2981         if (ret == 0) {
2982                 if (check_path_shared(root, path))
2983                         goto out;
2984         } else {
2985                 check_link = 0;
2986         }
2987         btrfs_release_path(path);
2988
2989         ret = btrfs_lookup_inode(trans, root, path,
2990                                 &BTRFS_I(inode)->location, 0);
2991         if (ret < 0) {
2992                 err = ret;
2993                 goto out;
2994         }
2995         if (ret == 0) {
2996                 if (check_path_shared(root, path))
2997                         goto out;
2998         } else {
2999                 check_link = 0;
3000         }
3001         btrfs_release_path(path);
3002
3003         if (ret == 0 && S_ISREG(inode->i_mode)) {
3004                 ret = btrfs_lookup_file_extent(trans, root, path,
3005                                                ino, (u64)-1, 0);
3006                 if (ret < 0) {
3007                         err = ret;
3008                         goto out;
3009                 }
3010                 BUG_ON(ret == 0); /* Corruption */
3011                 if (check_path_shared(root, path))
3012                         goto out;
3013                 btrfs_release_path(path);
3014         }
3015
3016         if (!check_link) {
3017                 err = 0;
3018                 goto out;
3019         }
3020
3021         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3022                                 dentry->d_name.name, dentry->d_name.len, 0);
3023         if (IS_ERR(di)) {
3024                 err = PTR_ERR(di);
3025                 goto out;
3026         }
3027         if (di) {
3028                 if (check_path_shared(root, path))
3029                         goto out;
3030         } else {
3031                 err = 0;
3032                 goto out;
3033         }
3034         btrfs_release_path(path);
3035
3036         ret = btrfs_get_inode_ref_index(trans, root, path, dentry->d_name.name,
3037                                         dentry->d_name.len, ino, dir_ino, 0,
3038                                         &index);
3039         if (ret) {
3040                 err = ret;
3041                 goto out;
3042         }
3043
3044         if (check_path_shared(root, path))
3045                 goto out;
3046
3047         btrfs_release_path(path);
3048
3049         /*
3050          * This is a commit root search, if we can lookup inode item and other
3051          * relative items in the commit root, it means the transaction of
3052          * dir/file creation has been committed, and the dir index item that we
3053          * delay to insert has also been inserted into the commit root. So
3054          * we needn't worry about the delayed insertion of the dir index item
3055          * here.
3056          */
3057         di = btrfs_lookup_dir_index_item(trans, root, path, dir_ino, index,
3058                                 dentry->d_name.name, dentry->d_name.len, 0);
3059         if (IS_ERR(di)) {
3060                 err = PTR_ERR(di);
3061                 goto out;
3062         }
3063         BUG_ON(ret == -ENOENT);
3064         if (check_path_shared(root, path))
3065                 goto out;
3066
3067         err = 0;
3068 out:
3069         btrfs_free_path(path);
3070         /* Migrate the orphan reservation over */
3071         if (!err)
3072                 err = btrfs_block_rsv_migrate(trans->block_rsv,
3073                                 &root->fs_info->global_block_rsv,
3074                                 trans->bytes_reserved);
3075
3076         if (err) {
3077                 btrfs_end_transaction(trans, root);
3078                 root->fs_info->enospc_unlink = 0;
3079                 return ERR_PTR(err);
3080         }
3081
3082         trans->block_rsv = &root->fs_info->global_block_rsv;
3083         return trans;
3084 }
3085
3086 static void __unlink_end_trans(struct btrfs_trans_handle *trans,
3087                                struct btrfs_root *root)
3088 {
3089         if (trans->block_rsv->type == BTRFS_BLOCK_RSV_GLOBAL) {
3090                 btrfs_block_rsv_release(root, trans->block_rsv,
3091                                         trans->bytes_reserved);
3092                 trans->block_rsv = &root->fs_info->trans_block_rsv;
3093                 BUG_ON(!root->fs_info->enospc_unlink);
3094                 root->fs_info->enospc_unlink = 0;
3095         }
3096         btrfs_end_transaction(trans, root);
3097 }
3098
3099 static int btrfs_unlink(struct inode *dir, struct dentry *dentry)
3100 {
3101         struct btrfs_root *root = BTRFS_I(dir)->root;
3102         struct btrfs_trans_handle *trans;
3103         struct inode *inode = dentry->d_inode;
3104         int ret;
3105
3106         trans = __unlink_start_trans(dir, dentry);
3107         if (IS_ERR(trans))
3108                 return PTR_ERR(trans);
3109
3110         btrfs_record_unlink_dir(trans, dir, dentry->d_inode, 0);
3111
3112         ret = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3113                                  dentry->d_name.name, dentry->d_name.len);
3114         if (ret)
3115                 goto out;
3116
3117         if (inode->i_nlink == 0) {
3118                 ret = btrfs_orphan_add(trans, inode);
3119                 if (ret)
3120                         goto out;
3121         }
3122
3123 out:
3124         __unlink_end_trans(trans, root);
3125         btrfs_btree_balance_dirty(root);
3126         return ret;
3127 }
3128
3129 int btrfs_unlink_subvol(struct btrfs_trans_handle *trans,
3130                         struct btrfs_root *root,
3131                         struct inode *dir, u64 objectid,
3132                         const char *name, int name_len)
3133 {
3134         struct btrfs_path *path;
3135         struct extent_buffer *leaf;
3136         struct btrfs_dir_item *di;
3137         struct btrfs_key key;
3138         u64 index;
3139         int ret;
3140         u64 dir_ino = btrfs_ino(dir);
3141
3142         path = btrfs_alloc_path();
3143         if (!path)
3144                 return -ENOMEM;
3145
3146         di = btrfs_lookup_dir_item(trans, root, path, dir_ino,
3147                                    name, name_len, -1);
3148         if (IS_ERR_OR_NULL(di)) {
3149                 if (!di)
3150                         ret = -ENOENT;
3151                 else
3152                         ret = PTR_ERR(di);
3153                 goto out;
3154         }
3155
3156         leaf = path->nodes[0];
3157         btrfs_dir_item_key_to_cpu(leaf, di, &key);
3158         WARN_ON(key.type != BTRFS_ROOT_ITEM_KEY || key.objectid != objectid);
3159         ret = btrfs_delete_one_dir_name(trans, root, path, di);
3160         if (ret) {
3161                 btrfs_abort_transaction(trans, root, ret);
3162                 goto out;
3163         }
3164         btrfs_release_path(path);
3165
3166         ret = btrfs_del_root_ref(trans, root->fs_info->tree_root,
3167                                  objectid, root->root_key.objectid,
3168                                  dir_ino, &index, name, name_len);
3169         if (ret < 0) {
3170                 if (ret != -ENOENT) {
3171                         btrfs_abort_transaction(trans, root, ret);
3172                         goto out;
3173                 }
3174                 di = btrfs_search_dir_index_item(root, path, dir_ino,
3175                                                  name, name_len);
3176                 if (IS_ERR_OR_NULL(di)) {
3177                         if (!di)
3178                                 ret = -ENOENT;
3179                         else
3180                                 ret = PTR_ERR(di);
3181                         btrfs_abort_transaction(trans, root, ret);
3182                         goto out;
3183                 }
3184
3185                 leaf = path->nodes[0];
3186                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
3187                 btrfs_release_path(path);
3188                 index = key.offset;
3189         }
3190         btrfs_release_path(path);
3191
3192         ret = btrfs_delete_delayed_dir_index(trans, root, dir, index);
3193         if (ret) {
3194                 btrfs_abort_transaction(trans, root, ret);
3195                 goto out;
3196         }
3197
3198         btrfs_i_size_write(dir, dir->i_size - name_len * 2);
3199         inode_inc_iversion(dir);
3200         dir->i_mtime = dir->i_ctime = CURRENT_TIME;
3201         ret = btrfs_update_inode_fallback(trans, root, dir);
3202         if (ret)
3203                 btrfs_abort_transaction(trans, root, ret);
3204 out:
3205         btrfs_free_path(path);
3206         return ret;
3207 }
3208
3209 static int btrfs_rmdir(struct inode *dir, struct dentry *dentry)
3210 {
3211         struct inode *inode = dentry->d_inode;
3212         int err = 0;
3213         struct btrfs_root *root = BTRFS_I(dir)->root;
3214         struct btrfs_trans_handle *trans;
3215
3216         if (inode->i_size > BTRFS_EMPTY_DIR_SIZE)
3217                 return -ENOTEMPTY;
3218         if (btrfs_ino(inode) == BTRFS_FIRST_FREE_OBJECTID)
3219                 return -EPERM;
3220
3221         trans = __unlink_start_trans(dir, dentry);
3222         if (IS_ERR(trans))
3223                 return PTR_ERR(trans);
3224
3225         if (unlikely(btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
3226                 err = btrfs_unlink_subvol(trans, root, dir,
3227                                           BTRFS_I(inode)->location.objectid,
3228                                           dentry->d_name.name,
3229                                           dentry->d_name.len);
3230                 goto out;
3231         }
3232
3233         err = btrfs_orphan_add(trans, inode);
3234         if (err)
3235                 goto out;
3236
3237         /* now the directory is empty */
3238         err = btrfs_unlink_inode(trans, root, dir, dentry->d_inode,
3239                                  dentry->d_name.name, dentry->d_name.len);
3240         if (!err)
3241                 btrfs_i_size_write(inode, 0);
3242 out:
3243         __unlink_end_trans(trans, root);
3244         btrfs_btree_balance_dirty(root);
3245
3246         return err;
3247 }
3248
3249 /*
3250  * this can truncate away extent items, csum items and directory items.
3251  * It starts at a high offset and removes keys until it can't find
3252  * any higher than new_size
3253  *
3254  * csum items that cross the new i_size are truncated to the new size
3255  * as well.
3256  *
3257  * min_type is the minimum key type to truncate down to.  If set to 0, this
3258  * will kill all the items on this inode, including the INODE_ITEM_KEY.
3259  */
3260 int btrfs_truncate_inode_items(struct btrfs_trans_handle *trans,
3261                                struct btrfs_root *root,
3262                                struct inode *inode,
3263                                u64 new_size, u32 min_type)
3264 {
3265         struct btrfs_path *path;
3266         struct extent_buffer *leaf;
3267         struct btrfs_file_extent_item *fi;
3268         struct btrfs_key key;
3269         struct btrfs_key found_key;
3270         u64 extent_start = 0;
3271         u64 extent_num_bytes = 0;
3272         u64 extent_offset = 0;
3273         u64 item_end = 0;
3274         u64 mask = root->sectorsize - 1;
3275         u32 found_type = (u8)-1;
3276         int found_extent;
3277         int del_item;
3278         int pending_del_nr = 0;
3279         int pending_del_slot = 0;
3280         int extent_type = -1;
3281         int ret;
3282         int err = 0;
3283         u64 ino = btrfs_ino(inode);
3284
3285         BUG_ON(new_size > 0 && min_type != BTRFS_EXTENT_DATA_KEY);
3286
3287         path = btrfs_alloc_path();
3288         if (!path)
3289                 return -ENOMEM;
3290         path->reada = -1;
3291
3292         /*
3293          * We want to drop from the next block forward in case this new size is
3294          * not block aligned since we will be keeping the last block of the
3295          * extent just the way it is.
3296          */
3297         if (root->ref_cows || root == root->fs_info->tree_root)
3298                 btrfs_drop_extent_cache(inode, (new_size + mask) & (~mask), (u64)-1, 0);
3299
3300         /*
3301          * This function is also used to drop the items in the log tree before
3302          * we relog the inode, so if root != BTRFS_I(inode)->root, it means
3303          * it is used to drop the loged items. So we shouldn't kill the delayed
3304          * items.
3305          */
3306         if (min_type == 0 && root == BTRFS_I(inode)->root)
3307                 btrfs_kill_delayed_inode_items(inode);
3308
3309         key.objectid = ino;
3310         key.offset = (u64)-1;
3311         key.type = (u8)-1;
3312
3313 search_again:
3314         path->leave_spinning = 1;
3315         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
3316         if (ret < 0) {
3317                 err = ret;
3318                 goto out;
3319         }
3320
3321         if (ret > 0) {
3322                 /* there are no items in the tree for us to truncate, we're
3323                  * done
3324                  */
3325                 if (path->slots[0] == 0)
3326                         goto out;
3327                 path->slots[0]--;
3328         }
3329
3330         while (1) {
3331                 fi = NULL;
3332                 leaf = path->nodes[0];
3333                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
3334                 found_type = btrfs_key_type(&found_key);
3335
3336                 if (found_key.objectid != ino)
3337                         break;
3338
3339                 if (found_type < min_type)
3340                         break;
3341
3342                 item_end = found_key.offset;
3343                 if (found_type == BTRFS_EXTENT_DATA_KEY) {
3344                         fi = btrfs_item_ptr(leaf, path->slots[0],
3345                                             struct btrfs_file_extent_item);
3346                         extent_type = btrfs_file_extent_type(leaf, fi);
3347                         if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3348                                 item_end +=
3349                                     btrfs_file_extent_num_bytes(leaf, fi);
3350                         } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3351                                 item_end += btrfs_file_extent_inline_len(leaf,
3352                                                                          fi);
3353                         }
3354                         item_end--;
3355                 }
3356                 if (found_type > min_type) {
3357                         del_item = 1;
3358                 } else {
3359                         if (item_end < new_size)
3360                                 break;
3361                         if (found_key.offset >= new_size)
3362                                 del_item = 1;
3363                         else
3364                                 del_item = 0;
3365                 }
3366                 found_extent = 0;
3367                 /* FIXME, shrink the extent if the ref count is only 1 */
3368                 if (found_type != BTRFS_EXTENT_DATA_KEY)
3369                         goto delete;
3370
3371                 if (extent_type != BTRFS_FILE_EXTENT_INLINE) {
3372                         u64 num_dec;
3373                         extent_start = btrfs_file_extent_disk_bytenr(leaf, fi);
3374                         if (!del_item) {
3375                                 u64 orig_num_bytes =
3376                                         btrfs_file_extent_num_bytes(leaf, fi);
3377                                 extent_num_bytes = new_size -
3378                                         found_key.offset + root->sectorsize - 1;
3379                                 extent_num_bytes = extent_num_bytes &
3380                                         ~((u64)root->sectorsize - 1);
3381                                 btrfs_set_file_extent_num_bytes(leaf, fi,
3382                                                          extent_num_bytes);
3383                                 num_dec = (orig_num_bytes -
3384                                            extent_num_bytes);
3385                                 if (root->ref_cows && extent_start != 0)
3386                                         inode_sub_bytes(inode, num_dec);
3387                                 btrfs_mark_buffer_dirty(leaf);
3388                         } else {
3389                                 extent_num_bytes =
3390                                         btrfs_file_extent_disk_num_bytes(leaf,
3391                                                                          fi);
3392                                 extent_offset = found_key.offset -
3393                                         btrfs_file_extent_offset(leaf, fi);
3394
3395                                 /* FIXME blocksize != 4096 */
3396                                 num_dec = btrfs_file_extent_num_bytes(leaf, fi);
3397                                 if (extent_start != 0) {
3398                                         found_extent = 1;
3399                                         if (root->ref_cows)
3400                                                 inode_sub_bytes(inode, num_dec);
3401                                 }
3402                         }
3403                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
3404                         /*
3405                          * we can't truncate inline items that have had
3406                          * special encodings
3407                          */
3408                         if (!del_item &&
3409                             btrfs_file_extent_compression(leaf, fi) == 0 &&
3410                             btrfs_file_extent_encryption(leaf, fi) == 0 &&
3411                             btrfs_file_extent_other_encoding(leaf, fi) == 0) {
3412                                 u32 size = new_size - found_key.offset;
3413
3414                                 if (root->ref_cows) {
3415                                         inode_sub_bytes(inode, item_end + 1 -
3416                                                         new_size);
3417                                 }
3418                                 size =
3419                                     btrfs_file_extent_calc_inline_size(size);
3420                                 btrfs_truncate_item(trans, root, path,
3421                                                     size, 1);
3422                         } else if (root->ref_cows) {
3423                                 inode_sub_bytes(inode, item_end + 1 -
3424                                                 found_key.offset);
3425                         }
3426                 }
3427 delete:
3428                 if (del_item) {
3429                         if (!pending_del_nr) {
3430                                 /* no pending yet, add ourselves */
3431                                 pending_del_slot = path->slots[0];
3432                                 pending_del_nr = 1;
3433                         } else if (pending_del_nr &&
3434                                    path->slots[0] + 1 == pending_del_slot) {
3435                                 /* hop on the pending chunk */
3436                                 pending_del_nr++;
3437                                 pending_del_slot = path->slots[0];
3438                         } else {
3439                                 BUG();
3440                         }
3441                 } else {
3442                         break;
3443                 }
3444                 if (found_extent && (root->ref_cows ||
3445                                      root == root->fs_info->tree_root)) {
3446                         btrfs_set_path_blocking(path);
3447                         ret = btrfs_free_extent(trans, root, extent_start,
3448                                                 extent_num_bytes, 0,
3449                                                 btrfs_header_owner(leaf),
3450                                                 ino, extent_offset, 0);
3451                         BUG_ON(ret);
3452                 }
3453
3454                 if (found_type == BTRFS_INODE_ITEM_KEY)
3455                         break;
3456
3457                 if (path->slots[0] == 0 ||
3458                     path->slots[0] != pending_del_slot) {
3459                         if (pending_del_nr) {
3460                                 ret = btrfs_del_items(trans, root, path,
3461                                                 pending_del_slot,
3462                                                 pending_del_nr);
3463                                 if (ret) {
3464                                         btrfs_abort_transaction(trans,
3465                                                                 root, ret);
3466                                         goto error;
3467                                 }
3468                                 pending_del_nr = 0;
3469                         }
3470                         btrfs_release_path(path);
3471                         goto search_again;
3472                 } else {
3473                         path->slots[0]--;
3474                 }
3475         }
3476 out:
3477         if (pending_del_nr) {
3478                 ret = btrfs_del_items(trans, root, path, pending_del_slot,
3479                                       pending_del_nr);
3480                 if (ret)
3481                         btrfs_abort_transaction(trans, root, ret);
3482         }
3483 error:
3484         btrfs_free_path(path);
3485         return err;
3486 }
3487
3488 /*
3489  * btrfs_truncate_page - read, zero a chunk and write a page
3490  * @inode - inode that we're zeroing
3491  * @from - the offset to start zeroing
3492  * @len - the length to zero, 0 to zero the entire range respective to the
3493  *      offset
3494  * @front - zero up to the offset instead of from the offset on
3495  *
3496  * This will find the page for the "from" offset and cow the page and zero the
3497  * part we want to zero.  This is used with truncate and hole punching.
3498  */
3499 int btrfs_truncate_page(struct inode *inode, loff_t from, loff_t len,
3500                         int front)
3501 {
3502         struct address_space *mapping = inode->i_mapping;
3503         struct btrfs_root *root = BTRFS_I(inode)->root;
3504         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3505         struct btrfs_ordered_extent *ordered;
3506         struct extent_state *cached_state = NULL;
3507         char *kaddr;
3508         u32 blocksize = root->sectorsize;
3509         pgoff_t index = from >> PAGE_CACHE_SHIFT;
3510         unsigned offset = from & (PAGE_CACHE_SIZE-1);
3511         struct page *page;
3512         gfp_t mask = btrfs_alloc_write_mask(mapping);
3513         int ret = 0;
3514         u64 page_start;
3515         u64 page_end;
3516
3517         if ((offset & (blocksize - 1)) == 0 &&
3518             (!len || ((len & (blocksize - 1)) == 0)))
3519                 goto out;
3520         ret = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
3521         if (ret)
3522                 goto out;
3523
3524         ret = -ENOMEM;
3525 again:
3526         page = find_or_create_page(mapping, index, mask);
3527         if (!page) {
3528                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3529                 goto out;
3530         }
3531
3532         page_start = page_offset(page);
3533         page_end = page_start + PAGE_CACHE_SIZE - 1;
3534
3535         if (!PageUptodate(page)) {
3536                 ret = btrfs_readpage(NULL, page);
3537                 lock_page(page);
3538                 if (page->mapping != mapping) {
3539                         unlock_page(page);
3540                         page_cache_release(page);
3541                         goto again;
3542                 }
3543                 if (!PageUptodate(page)) {
3544                         ret = -EIO;
3545                         goto out_unlock;
3546                 }
3547         }
3548         wait_on_page_writeback(page);
3549
3550         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
3551         set_page_extent_mapped(page);
3552
3553         ordered = btrfs_lookup_ordered_extent(inode, page_start);
3554         if (ordered) {
3555                 unlock_extent_cached(io_tree, page_start, page_end,
3556                                      &cached_state, GFP_NOFS);
3557                 unlock_page(page);
3558                 page_cache_release(page);
3559                 btrfs_start_ordered_extent(inode, ordered, 1);
3560                 btrfs_put_ordered_extent(ordered);
3561                 goto again;
3562         }
3563
3564         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
3565                           EXTENT_DIRTY | EXTENT_DELALLOC |
3566                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
3567                           0, 0, &cached_state, GFP_NOFS);
3568
3569         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
3570                                         &cached_state);
3571         if (ret) {
3572                 unlock_extent_cached(io_tree, page_start, page_end,
3573                                      &cached_state, GFP_NOFS);
3574                 goto out_unlock;
3575         }
3576
3577         ret = 0;
3578         if (offset != PAGE_CACHE_SIZE) {
3579                 if (!len)
3580                         len = PAGE_CACHE_SIZE - offset;
3581                 kaddr = kmap(page);
3582                 if (front)
3583                         memset(kaddr, 0, offset);
3584                 else
3585                         memset(kaddr + offset, 0, len);
3586                 flush_dcache_page(page);
3587                 kunmap(page);
3588         }
3589         ClearPageChecked(page);
3590         set_page_dirty(page);
3591         unlock_extent_cached(io_tree, page_start, page_end, &cached_state,
3592                              GFP_NOFS);
3593
3594 out_unlock:
3595         if (ret)
3596                 btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
3597         unlock_page(page);
3598         page_cache_release(page);
3599 out:
3600         return ret;
3601 }
3602
3603 /*
3604  * This function puts in dummy file extents for the area we're creating a hole
3605  * for.  So if we are truncating this file to a larger size we need to insert
3606  * these file extents so that btrfs_get_extent will return a EXTENT_MAP_HOLE for
3607  * the range between oldsize and size
3608  */
3609 int btrfs_cont_expand(struct inode *inode, loff_t oldsize, loff_t size)
3610 {
3611         struct btrfs_trans_handle *trans;
3612         struct btrfs_root *root = BTRFS_I(inode)->root;
3613         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
3614         struct extent_map *em = NULL;
3615         struct extent_state *cached_state = NULL;
3616         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
3617         u64 mask = root->sectorsize - 1;
3618         u64 hole_start = (oldsize + mask) & ~mask;
3619         u64 block_end = (size + mask) & ~mask;
3620         u64 last_byte;
3621         u64 cur_offset;
3622         u64 hole_size;
3623         int err = 0;
3624
3625         if (size <= hole_start)
3626                 return 0;
3627
3628         while (1) {
3629                 struct btrfs_ordered_extent *ordered;
3630                 btrfs_wait_ordered_range(inode, hole_start,
3631                                          block_end - hole_start);
3632                 lock_extent_bits(io_tree, hole_start, block_end - 1, 0,
3633                                  &cached_state);
3634                 ordered = btrfs_lookup_ordered_extent(inode, hole_start);
3635                 if (!ordered)
3636                         break;
3637                 unlock_extent_cached(io_tree, hole_start, block_end - 1,
3638                                      &cached_state, GFP_NOFS);
3639                 btrfs_put_ordered_extent(ordered);
3640         }
3641
3642         cur_offset = hole_start;
3643         while (1) {
3644                 em = btrfs_get_extent(inode, NULL, 0, cur_offset,
3645                                 block_end - cur_offset, 0);
3646                 if (IS_ERR(em)) {
3647                         err = PTR_ERR(em);
3648                         break;
3649                 }
3650                 last_byte = min(extent_map_end(em), block_end);
3651                 last_byte = (last_byte + mask) & ~mask;
3652                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags)) {
3653                         struct extent_map *hole_em;
3654                         hole_size = last_byte - cur_offset;
3655
3656                         trans = btrfs_start_transaction(root, 3);
3657                         if (IS_ERR(trans)) {
3658                                 err = PTR_ERR(trans);
3659                                 break;
3660                         }
3661
3662                         err = btrfs_drop_extents(trans, root, inode,
3663                                                  cur_offset,
3664                                                  cur_offset + hole_size, 1);
3665                         if (err) {
3666                                 btrfs_abort_transaction(trans, root, err);
3667                                 btrfs_end_transaction(trans, root);
3668                                 break;
3669                         }
3670
3671                         err = btrfs_insert_file_extent(trans, root,
3672                                         btrfs_ino(inode), cur_offset, 0,
3673                                         0, hole_size, 0, hole_size,
3674                                         0, 0, 0);
3675                         if (err) {
3676                                 btrfs_abort_transaction(trans, root, err);
3677                                 btrfs_end_transaction(trans, root);
3678                                 break;
3679                         }
3680
3681                         btrfs_drop_extent_cache(inode, cur_offset,
3682                                                 cur_offset + hole_size - 1, 0);
3683                         hole_em = alloc_extent_map();
3684                         if (!hole_em) {
3685                                 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
3686                                         &BTRFS_I(inode)->runtime_flags);
3687                                 goto next;
3688                         }
3689                         hole_em->start = cur_offset;
3690                         hole_em->len = hole_size;
3691                         hole_em->orig_start = cur_offset;
3692
3693                         hole_em->block_start = EXTENT_MAP_HOLE;
3694                         hole_em->block_len = 0;
3695                         hole_em->bdev = root->fs_info->fs_devices->latest_bdev;
3696                         hole_em->compress_type = BTRFS_COMPRESS_NONE;
3697                         hole_em->generation = trans->transid;
3698
3699                         while (1) {
3700                                 write_lock(&em_tree->lock);
3701                                 err = add_extent_mapping(em_tree, hole_em);
3702                                 if (!err)
3703                                         list_move(&hole_em->list,
3704                                                   &em_tree->modified_extents);
3705                                 write_unlock(&em_tree->lock);
3706                                 if (err != -EEXIST)
3707                                         break;
3708                                 btrfs_drop_extent_cache(inode, cur_offset,
3709                                                         cur_offset +
3710                                                         hole_size - 1, 0);
3711                         }
3712                         free_extent_map(hole_em);
3713 next:
3714                         btrfs_update_inode(trans, root, inode);
3715                         btrfs_end_transaction(trans, root);
3716                 }
3717                 free_extent_map(em);
3718                 em = NULL;
3719                 cur_offset = last_byte;
3720                 if (cur_offset >= block_end)
3721                         break;
3722         }
3723
3724         free_extent_map(em);
3725         unlock_extent_cached(io_tree, hole_start, block_end - 1, &cached_state,
3726                              GFP_NOFS);
3727         return err;
3728 }
3729
3730 static int btrfs_setsize(struct inode *inode, loff_t newsize)
3731 {
3732         struct btrfs_root *root = BTRFS_I(inode)->root;
3733         struct btrfs_trans_handle *trans;
3734         loff_t oldsize = i_size_read(inode);
3735         int ret;
3736
3737         if (newsize == oldsize)
3738                 return 0;
3739
3740         if (newsize > oldsize) {
3741                 truncate_pagecache(inode, oldsize, newsize);
3742                 ret = btrfs_cont_expand(inode, oldsize, newsize);
3743                 if (ret)
3744                         return ret;
3745
3746                 trans = btrfs_start_transaction(root, 1);
3747                 if (IS_ERR(trans))
3748                         return PTR_ERR(trans);
3749
3750                 i_size_write(inode, newsize);
3751                 btrfs_ordered_update_i_size(inode, i_size_read(inode), NULL);
3752                 ret = btrfs_update_inode(trans, root, inode);
3753                 btrfs_end_transaction(trans, root);
3754         } else {
3755
3756                 /*
3757                  * We're truncating a file that used to have good data down to
3758                  * zero. Make sure it gets into the ordered flush list so that
3759                  * any new writes get down to disk quickly.
3760                  */
3761                 if (newsize == 0)
3762                         set_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
3763                                 &BTRFS_I(inode)->runtime_flags);
3764
3765                 /* we don't support swapfiles, so vmtruncate shouldn't fail */
3766                 truncate_setsize(inode, newsize);
3767                 ret = btrfs_truncate(inode);
3768         }
3769
3770         return ret;
3771 }
3772
3773 static int btrfs_setattr(struct dentry *dentry, struct iattr *attr)
3774 {
3775         struct inode *inode = dentry->d_inode;
3776         struct btrfs_root *root = BTRFS_I(inode)->root;
3777         int err;
3778
3779         if (btrfs_root_readonly(root))
3780                 return -EROFS;
3781
3782         err = inode_change_ok(inode, attr);
3783         if (err)
3784                 return err;
3785
3786         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
3787                 err = btrfs_setsize(inode, attr->ia_size);
3788                 if (err)
3789                         return err;
3790         }
3791
3792         if (attr->ia_valid) {
3793                 setattr_copy(inode, attr);
3794                 inode_inc_iversion(inode);
3795                 err = btrfs_dirty_inode(inode);
3796
3797                 if (!err && attr->ia_valid & ATTR_MODE)
3798                         err = btrfs_acl_chmod(inode);
3799         }
3800
3801         return err;
3802 }
3803
3804 void btrfs_evict_inode(struct inode *inode)
3805 {
3806         struct btrfs_trans_handle *trans;
3807         struct btrfs_root *root = BTRFS_I(inode)->root;
3808         struct btrfs_block_rsv *rsv, *global_rsv;
3809         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
3810         int ret;
3811
3812         trace_btrfs_inode_evict(inode);
3813
3814         truncate_inode_pages(&inode->i_data, 0);
3815         if (inode->i_nlink && (btrfs_root_refs(&root->root_item) != 0 ||
3816                                btrfs_is_free_space_inode(inode)))
3817                 goto no_delete;
3818
3819         if (is_bad_inode(inode)) {
3820                 btrfs_orphan_del(NULL, inode);
3821                 goto no_delete;
3822         }
3823         /* do we really want it for ->i_nlink > 0 and zero btrfs_root_refs? */
3824         btrfs_wait_ordered_range(inode, 0, (u64)-1);
3825
3826         if (root->fs_info->log_root_recovering) {
3827                 BUG_ON(test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
3828                                  &BTRFS_I(inode)->runtime_flags));
3829                 goto no_delete;
3830         }
3831
3832         if (inode->i_nlink > 0) {
3833                 BUG_ON(btrfs_root_refs(&root->root_item) != 0);
3834                 goto no_delete;
3835         }
3836
3837         rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
3838         if (!rsv) {
3839                 btrfs_orphan_del(NULL, inode);
3840                 goto no_delete;
3841         }
3842         rsv->size = min_size;
3843         rsv->failfast = 1;
3844         global_rsv = &root->fs_info->global_block_rsv;
3845
3846         btrfs_i_size_write(inode, 0);
3847
3848         /*
3849          * This is a bit simpler than btrfs_truncate since we've already
3850          * reserved our space for our orphan item in the unlink, so we just
3851          * need to reserve some slack space in case we add bytes and update
3852          * inode item when doing the truncate.
3853          */
3854         while (1) {
3855                 ret = btrfs_block_rsv_refill(root, rsv, min_size,
3856                                              BTRFS_RESERVE_FLUSH_LIMIT);
3857
3858                 /*
3859                  * Try and steal from the global reserve since we will
3860                  * likely not use this space anyway, we want to try as
3861                  * hard as possible to get this to work.
3862                  */
3863                 if (ret)
3864                         ret = btrfs_block_rsv_migrate(global_rsv, rsv, min_size);
3865
3866                 if (ret) {
3867                         printk(KERN_WARNING "Could not get space for a "
3868                                "delete, will truncate on mount %d\n", ret);
3869                         btrfs_orphan_del(NULL, inode);
3870                         btrfs_free_block_rsv(root, rsv);
3871                         goto no_delete;
3872                 }
3873
3874                 trans = btrfs_start_transaction_lflush(root, 1);
3875                 if (IS_ERR(trans)) {
3876                         btrfs_orphan_del(NULL, inode);
3877                         btrfs_free_block_rsv(root, rsv);
3878                         goto no_delete;
3879                 }
3880
3881                 trans->block_rsv = rsv;
3882
3883                 ret = btrfs_truncate_inode_items(trans, root, inode, 0, 0);
3884                 if (ret != -ENOSPC)
3885                         break;
3886
3887                 trans->block_rsv = &root->fs_info->trans_block_rsv;
3888                 ret = btrfs_update_inode(trans, root, inode);
3889                 BUG_ON(ret);
3890
3891                 btrfs_end_transaction(trans, root);
3892                 trans = NULL;
3893                 btrfs_btree_balance_dirty(root);
3894         }
3895
3896         btrfs_free_block_rsv(root, rsv);
3897
3898         if (ret == 0) {
3899                 trans->block_rsv = root->orphan_block_rsv;
3900                 ret = btrfs_orphan_del(trans, inode);
3901                 BUG_ON(ret);
3902         }
3903
3904         trans->block_rsv = &root->fs_info->trans_block_rsv;
3905         if (!(root == root->fs_info->tree_root ||
3906               root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID))
3907                 btrfs_return_ino(root, btrfs_ino(inode));
3908
3909         btrfs_end_transaction(trans, root);
3910         btrfs_btree_balance_dirty(root);
3911 no_delete:
3912         clear_inode(inode);
3913         return;
3914 }
3915
3916 /*
3917  * this returns the key found in the dir entry in the location pointer.
3918  * If no dir entries were found, location->objectid is 0.
3919  */
3920 static int btrfs_inode_by_name(struct inode *dir, struct dentry *dentry,
3921                                struct btrfs_key *location)
3922 {
3923         const char *name = dentry->d_name.name;
3924         int namelen = dentry->d_name.len;
3925         struct btrfs_dir_item *di;
3926         struct btrfs_path *path;
3927         struct btrfs_root *root = BTRFS_I(dir)->root;
3928         int ret = 0;
3929
3930         path = btrfs_alloc_path();
3931         if (!path)
3932                 return -ENOMEM;
3933
3934         di = btrfs_lookup_dir_item(NULL, root, path, btrfs_ino(dir), name,
3935                                     namelen, 0);
3936         if (IS_ERR(di))
3937                 ret = PTR_ERR(di);
3938
3939         if (IS_ERR_OR_NULL(di))
3940                 goto out_err;
3941
3942         btrfs_dir_item_key_to_cpu(path->nodes[0], di, location);
3943 out:
3944         btrfs_free_path(path);
3945         return ret;
3946 out_err:
3947         location->objectid = 0;
3948         goto out;
3949 }
3950
3951 /*
3952  * when we hit a tree root in a directory, the btrfs part of the inode
3953  * needs to be changed to reflect the root directory of the tree root.  This
3954  * is kind of like crossing a mount point.
3955  */
3956 static int fixup_tree_root_location(struct btrfs_root *root,
3957                                     struct inode *dir,
3958                                     struct dentry *dentry,
3959                                     struct btrfs_key *location,
3960                                     struct btrfs_root **sub_root)
3961 {
3962         struct btrfs_path *path;
3963         struct btrfs_root *new_root;
3964         struct btrfs_root_ref *ref;
3965         struct extent_buffer *leaf;
3966         int ret;
3967         int err = 0;
3968
3969         path = btrfs_alloc_path();
3970         if (!path) {
3971                 err = -ENOMEM;
3972                 goto out;
3973         }
3974
3975         err = -ENOENT;
3976         ret = btrfs_find_root_ref(root->fs_info->tree_root, path,
3977                                   BTRFS_I(dir)->root->root_key.objectid,
3978                                   location->objectid);
3979         if (ret) {
3980                 if (ret < 0)
3981                         err = ret;
3982                 goto out;
3983         }
3984
3985         leaf = path->nodes[0];
3986         ref = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_root_ref);
3987         if (btrfs_root_ref_dirid(leaf, ref) != btrfs_ino(dir) ||
3988             btrfs_root_ref_name_len(leaf, ref) != dentry->d_name.len)
3989                 goto out;
3990
3991         ret = memcmp_extent_buffer(leaf, dentry->d_name.name,
3992                                    (unsigned long)(ref + 1),
3993                                    dentry->d_name.len);
3994         if (ret)
3995                 goto out;
3996
3997         btrfs_release_path(path);
3998
3999         new_root = btrfs_read_fs_root_no_name(root->fs_info, location);
4000         if (IS_ERR(new_root)) {
4001                 err = PTR_ERR(new_root);
4002                 goto out;
4003         }
4004
4005         if (btrfs_root_refs(&new_root->root_item) == 0) {
4006                 err = -ENOENT;
4007                 goto out;
4008         }
4009
4010         *sub_root = new_root;
4011         location->objectid = btrfs_root_dirid(&new_root->root_item);
4012         location->type = BTRFS_INODE_ITEM_KEY;
4013         location->offset = 0;
4014         err = 0;
4015 out:
4016         btrfs_free_path(path);
4017         return err;
4018 }
4019
4020 static void inode_tree_add(struct inode *inode)
4021 {
4022         struct btrfs_root *root = BTRFS_I(inode)->root;
4023         struct btrfs_inode *entry;
4024         struct rb_node **p;
4025         struct rb_node *parent;
4026         u64 ino = btrfs_ino(inode);
4027 again:
4028         p = &root->inode_tree.rb_node;
4029         parent = NULL;
4030
4031         if (inode_unhashed(inode))
4032                 return;
4033
4034         spin_lock(&root->inode_lock);
4035         while (*p) {
4036                 parent = *p;
4037                 entry = rb_entry(parent, struct btrfs_inode, rb_node);
4038
4039                 if (ino < btrfs_ino(&entry->vfs_inode))
4040                         p = &parent->rb_left;
4041                 else if (ino > btrfs_ino(&entry->vfs_inode))
4042                         p = &parent->rb_right;
4043                 else {
4044                         WARN_ON(!(entry->vfs_inode.i_state &
4045                                   (I_WILL_FREE | I_FREEING)));
4046                         rb_erase(parent, &root->inode_tree);
4047                         RB_CLEAR_NODE(parent);
4048                         spin_unlock(&root->inode_lock);
4049                         goto again;
4050                 }
4051         }
4052         rb_link_node(&BTRFS_I(inode)->rb_node, parent, p);
4053         rb_insert_color(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4054         spin_unlock(&root->inode_lock);
4055 }
4056
4057 static void inode_tree_del(struct inode *inode)
4058 {
4059         struct btrfs_root *root = BTRFS_I(inode)->root;
4060         int empty = 0;
4061
4062         spin_lock(&root->inode_lock);
4063         if (!RB_EMPTY_NODE(&BTRFS_I(inode)->rb_node)) {
4064                 rb_erase(&BTRFS_I(inode)->rb_node, &root->inode_tree);
4065                 RB_CLEAR_NODE(&BTRFS_I(inode)->rb_node);
4066                 empty = RB_EMPTY_ROOT(&root->inode_tree);
4067         }
4068         spin_unlock(&root->inode_lock);
4069
4070         /*
4071          * Free space cache has inodes in the tree root, but the tree root has a
4072          * root_refs of 0, so this could end up dropping the tree root as a
4073          * snapshot, so we need the extra !root->fs_info->tree_root check to
4074          * make sure we don't drop it.
4075          */
4076         if (empty && btrfs_root_refs(&root->root_item) == 0 &&
4077             root != root->fs_info->tree_root) {
4078                 synchronize_srcu(&root->fs_info->subvol_srcu);
4079                 spin_lock(&root->inode_lock);
4080                 empty = RB_EMPTY_ROOT(&root->inode_tree);
4081                 spin_unlock(&root->inode_lock);
4082                 if (empty)
4083                         btrfs_add_dead_root(root);
4084         }
4085 }
4086
4087 void btrfs_invalidate_inodes(struct btrfs_root *root)
4088 {
4089         struct rb_node *node;
4090         struct rb_node *prev;
4091         struct btrfs_inode *entry;
4092         struct inode *inode;
4093         u64 objectid = 0;
4094
4095         WARN_ON(btrfs_root_refs(&root->root_item) != 0);
4096
4097         spin_lock(&root->inode_lock);
4098 again:
4099         node = root->inode_tree.rb_node;
4100         prev = NULL;
4101         while (node) {
4102                 prev = node;
4103                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4104
4105                 if (objectid < btrfs_ino(&entry->vfs_inode))
4106                         node = node->rb_left;
4107                 else if (objectid > btrfs_ino(&entry->vfs_inode))
4108                         node = node->rb_right;
4109                 else
4110                         break;
4111         }
4112         if (!node) {
4113                 while (prev) {
4114                         entry = rb_entry(prev, struct btrfs_inode, rb_node);
4115                         if (objectid <= btrfs_ino(&entry->vfs_inode)) {
4116                                 node = prev;
4117                                 break;
4118                         }
4119                         prev = rb_next(prev);
4120                 }
4121         }
4122         while (node) {
4123                 entry = rb_entry(node, struct btrfs_inode, rb_node);
4124                 objectid = btrfs_ino(&entry->vfs_inode) + 1;
4125                 inode = igrab(&entry->vfs_inode);
4126                 if (inode) {
4127                         spin_unlock(&root->inode_lock);
4128                         if (atomic_read(&inode->i_count) > 1)
4129                                 d_prune_aliases(inode);
4130                         /*
4131                          * btrfs_drop_inode will have it removed from
4132                          * the inode cache when its usage count
4133                          * hits zero.
4134                          */
4135                         iput(inode);
4136                         cond_resched();
4137                         spin_lock(&root->inode_lock);
4138                         goto again;
4139                 }
4140
4141                 if (cond_resched_lock(&root->inode_lock))
4142                         goto again;
4143
4144                 node = rb_next(node);
4145         }
4146         spin_unlock(&root->inode_lock);
4147 }
4148
4149 static int btrfs_init_locked_inode(struct inode *inode, void *p)
4150 {
4151         struct btrfs_iget_args *args = p;
4152         inode->i_ino = args->ino;
4153         BTRFS_I(inode)->root = args->root;
4154         return 0;
4155 }
4156
4157 static int btrfs_find_actor(struct inode *inode, void *opaque)
4158 {
4159         struct btrfs_iget_args *args = opaque;
4160         return args->ino == btrfs_ino(inode) &&
4161                 args->root == BTRFS_I(inode)->root;
4162 }
4163
4164 static struct inode *btrfs_iget_locked(struct super_block *s,
4165                                        u64 objectid,
4166                                        struct btrfs_root *root)
4167 {
4168         struct inode *inode;
4169         struct btrfs_iget_args args;
4170         args.ino = objectid;
4171         args.root = root;
4172
4173         inode = iget5_locked(s, objectid, btrfs_find_actor,
4174                              btrfs_init_locked_inode,
4175                              (void *)&args);
4176         return inode;
4177 }
4178
4179 /* Get an inode object given its location and corresponding root.
4180  * Returns in *is_new if the inode was read from disk
4181  */
4182 struct inode *btrfs_iget(struct super_block *s, struct btrfs_key *location,
4183                          struct btrfs_root *root, int *new)
4184 {
4185         struct inode *inode;
4186
4187         inode = btrfs_iget_locked(s, location->objectid, root);
4188         if (!inode)
4189                 return ERR_PTR(-ENOMEM);
4190
4191         if (inode->i_state & I_NEW) {
4192                 BTRFS_I(inode)->root = root;
4193                 memcpy(&BTRFS_I(inode)->location, location, sizeof(*location));
4194                 btrfs_read_locked_inode(inode);
4195                 if (!is_bad_inode(inode)) {
4196                         inode_tree_add(inode);
4197                         unlock_new_inode(inode);
4198                         if (new)
4199                                 *new = 1;
4200                 } else {
4201                         unlock_new_inode(inode);
4202                         iput(inode);
4203                         inode = ERR_PTR(-ESTALE);
4204                 }
4205         }
4206
4207         return inode;
4208 }
4209
4210 static struct inode *new_simple_dir(struct super_block *s,
4211                                     struct btrfs_key *key,
4212                                     struct btrfs_root *root)
4213 {
4214         struct inode *inode = new_inode(s);
4215
4216         if (!inode)
4217                 return ERR_PTR(-ENOMEM);
4218
4219         BTRFS_I(inode)->root = root;
4220         memcpy(&BTRFS_I(inode)->location, key, sizeof(*key));
4221         set_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags);
4222
4223         inode->i_ino = BTRFS_EMPTY_SUBVOL_DIR_OBJECTID;
4224         inode->i_op = &btrfs_dir_ro_inode_operations;
4225         inode->i_fop = &simple_dir_operations;
4226         inode->i_mode = S_IFDIR | S_IRUGO | S_IWUSR | S_IXUGO;
4227         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4228
4229         return inode;
4230 }
4231
4232 struct inode *btrfs_lookup_dentry(struct inode *dir, struct dentry *dentry)
4233 {
4234         struct inode *inode;
4235         struct btrfs_root *root = BTRFS_I(dir)->root;
4236         struct btrfs_root *sub_root = root;
4237         struct btrfs_key location;
4238         int index;
4239         int ret = 0;
4240
4241         if (dentry->d_name.len > BTRFS_NAME_LEN)
4242                 return ERR_PTR(-ENAMETOOLONG);
4243
4244         if (unlikely(d_need_lookup(dentry))) {
4245                 memcpy(&location, dentry->d_fsdata, sizeof(struct btrfs_key));
4246                 kfree(dentry->d_fsdata);
4247                 dentry->d_fsdata = NULL;
4248                 /* This thing is hashed, drop it for now */
4249                 d_drop(dentry);
4250         } else {
4251                 ret = btrfs_inode_by_name(dir, dentry, &location);
4252         }
4253
4254         if (ret < 0)
4255                 return ERR_PTR(ret);
4256
4257         if (location.objectid == 0)
4258                 return NULL;
4259
4260         if (location.type == BTRFS_INODE_ITEM_KEY) {
4261                 inode = btrfs_iget(dir->i_sb, &location, root, NULL);
4262                 return inode;
4263         }
4264
4265         BUG_ON(location.type != BTRFS_ROOT_ITEM_KEY);
4266
4267         index = srcu_read_lock(&root->fs_info->subvol_srcu);
4268         ret = fixup_tree_root_location(root, dir, dentry,
4269                                        &location, &sub_root);
4270         if (ret < 0) {
4271                 if (ret != -ENOENT)
4272                         inode = ERR_PTR(ret);
4273                 else
4274                         inode = new_simple_dir(dir->i_sb, &location, sub_root);
4275         } else {
4276                 inode = btrfs_iget(dir->i_sb, &location, sub_root, NULL);
4277         }
4278         srcu_read_unlock(&root->fs_info->subvol_srcu, index);
4279
4280         if (!IS_ERR(inode) && root != sub_root) {
4281                 down_read(&root->fs_info->cleanup_work_sem);
4282                 if (!(inode->i_sb->s_flags & MS_RDONLY))
4283                         ret = btrfs_orphan_cleanup(sub_root);
4284                 up_read(&root->fs_info->cleanup_work_sem);
4285                 if (ret)
4286                         inode = ERR_PTR(ret);
4287         }
4288
4289         return inode;
4290 }
4291
4292 static int btrfs_dentry_delete(const struct dentry *dentry)
4293 {
4294         struct btrfs_root *root;
4295         struct inode *inode = dentry->d_inode;
4296
4297         if (!inode && !IS_ROOT(dentry))
4298                 inode = dentry->d_parent->d_inode;
4299
4300         if (inode) {
4301                 root = BTRFS_I(inode)->root;
4302                 if (btrfs_root_refs(&root->root_item) == 0)
4303                         return 1;
4304
4305                 if (btrfs_ino(inode) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
4306                         return 1;
4307         }
4308         return 0;
4309 }
4310
4311 static void btrfs_dentry_release(struct dentry *dentry)
4312 {
4313         if (dentry->d_fsdata)
4314                 kfree(dentry->d_fsdata);
4315 }
4316
4317 static struct dentry *btrfs_lookup(struct inode *dir, struct dentry *dentry,
4318                                    unsigned int flags)
4319 {
4320         struct dentry *ret;
4321
4322         ret = d_splice_alias(btrfs_lookup_dentry(dir, dentry), dentry);
4323         if (unlikely(d_need_lookup(dentry))) {
4324                 spin_lock(&dentry->d_lock);
4325                 dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
4326                 spin_unlock(&dentry->d_lock);
4327         }
4328         return ret;
4329 }
4330
4331 unsigned char btrfs_filetype_table[] = {
4332         DT_UNKNOWN, DT_REG, DT_DIR, DT_CHR, DT_BLK, DT_FIFO, DT_SOCK, DT_LNK
4333 };
4334
4335 static int btrfs_real_readdir(struct file *filp, void *dirent,
4336                               filldir_t filldir)
4337 {
4338         struct inode *inode = filp->f_dentry->d_inode;
4339         struct btrfs_root *root = BTRFS_I(inode)->root;
4340         struct btrfs_item *item;
4341         struct btrfs_dir_item *di;
4342         struct btrfs_key key;
4343         struct btrfs_key found_key;
4344         struct btrfs_path *path;
4345         struct list_head ins_list;
4346         struct list_head del_list;
4347         int ret;
4348         struct extent_buffer *leaf;
4349         int slot;
4350         unsigned char d_type;
4351         int over = 0;
4352         u32 di_cur;
4353         u32 di_total;
4354         u32 di_len;
4355         int key_type = BTRFS_DIR_INDEX_KEY;
4356         char tmp_name[32];
4357         char *name_ptr;
4358         int name_len;
4359         int is_curr = 0;        /* filp->f_pos points to the current index? */
4360
4361         /* FIXME, use a real flag for deciding about the key type */
4362         if (root->fs_info->tree_root == root)
4363                 key_type = BTRFS_DIR_ITEM_KEY;
4364
4365         /* special case for "." */
4366         if (filp->f_pos == 0) {
4367                 over = filldir(dirent, ".", 1,
4368                                filp->f_pos, btrfs_ino(inode), DT_DIR);
4369                 if (over)
4370                         return 0;
4371                 filp->f_pos = 1;
4372         }
4373         /* special case for .., just use the back ref */
4374         if (filp->f_pos == 1) {
4375                 u64 pino = parent_ino(filp->f_path.dentry);
4376                 over = filldir(dirent, "..", 2,
4377                                filp->f_pos, pino, DT_DIR);
4378                 if (over)
4379                         return 0;
4380                 filp->f_pos = 2;
4381         }
4382         path = btrfs_alloc_path();
4383         if (!path)
4384                 return -ENOMEM;
4385
4386         path->reada = 1;
4387
4388         if (key_type == BTRFS_DIR_INDEX_KEY) {
4389                 INIT_LIST_HEAD(&ins_list);
4390                 INIT_LIST_HEAD(&del_list);
4391                 btrfs_get_delayed_items(inode, &ins_list, &del_list);
4392         }
4393
4394         btrfs_set_key_type(&key, key_type);
4395         key.offset = filp->f_pos;
4396         key.objectid = btrfs_ino(inode);
4397
4398         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4399         if (ret < 0)
4400                 goto err;
4401
4402         while (1) {
4403                 leaf = path->nodes[0];
4404                 slot = path->slots[0];
4405                 if (slot >= btrfs_header_nritems(leaf)) {
4406                         ret = btrfs_next_leaf(root, path);
4407                         if (ret < 0)
4408                                 goto err;
4409                         else if (ret > 0)
4410                                 break;
4411                         continue;
4412                 }
4413
4414                 item = btrfs_item_nr(leaf, slot);
4415                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
4416
4417                 if (found_key.objectid != key.objectid)
4418                         break;
4419                 if (btrfs_key_type(&found_key) != key_type)
4420                         break;
4421                 if (found_key.offset < filp->f_pos)
4422                         goto next;
4423                 if (key_type == BTRFS_DIR_INDEX_KEY &&
4424                     btrfs_should_delete_dir_index(&del_list,
4425                                                   found_key.offset))
4426                         goto next;
4427
4428                 filp->f_pos = found_key.offset;
4429                 is_curr = 1;
4430
4431                 di = btrfs_item_ptr(leaf, slot, struct btrfs_dir_item);
4432                 di_cur = 0;
4433                 di_total = btrfs_item_size(leaf, item);
4434
4435                 while (di_cur < di_total) {
4436                         struct btrfs_key location;
4437
4438                         if (verify_dir_item(root, leaf, di))
4439                                 break;
4440
4441                         name_len = btrfs_dir_name_len(leaf, di);
4442                         if (name_len <= sizeof(tmp_name)) {
4443                                 name_ptr = tmp_name;
4444                         } else {
4445                                 name_ptr = kmalloc(name_len, GFP_NOFS);
4446                                 if (!name_ptr) {
4447                                         ret = -ENOMEM;
4448                                         goto err;
4449                                 }
4450                         }
4451                         read_extent_buffer(leaf, name_ptr,
4452                                            (unsigned long)(di + 1), name_len);
4453
4454                         d_type = btrfs_filetype_table[btrfs_dir_type(leaf, di)];
4455                         btrfs_dir_item_key_to_cpu(leaf, di, &location);
4456
4457
4458                         /* is this a reference to our own snapshot? If so
4459                          * skip it.
4460                          *
4461                          * In contrast to old kernels, we insert the snapshot's
4462                          * dir item and dir index after it has been created, so
4463                          * we won't find a reference to our own snapshot. We
4464                          * still keep the following code for backward
4465                          * compatibility.
4466                          */
4467                         if (location.type == BTRFS_ROOT_ITEM_KEY &&
4468                             location.objectid == root->root_key.objectid) {
4469                                 over = 0;
4470                                 goto skip;
4471                         }
4472                         over = filldir(dirent, name_ptr, name_len,
4473                                        found_key.offset, location.objectid,
4474                                        d_type);
4475
4476 skip:
4477                         if (name_ptr != tmp_name)
4478                                 kfree(name_ptr);
4479
4480                         if (over)
4481                                 goto nopos;
4482                         di_len = btrfs_dir_name_len(leaf, di) +
4483                                  btrfs_dir_data_len(leaf, di) + sizeof(*di);
4484                         di_cur += di_len;
4485                         di = (struct btrfs_dir_item *)((char *)di + di_len);
4486                 }
4487 next:
4488                 path->slots[0]++;
4489         }
4490
4491         if (key_type == BTRFS_DIR_INDEX_KEY) {
4492                 if (is_curr)
4493                         filp->f_pos++;
4494                 ret = btrfs_readdir_delayed_dir_index(filp, dirent, filldir,
4495                                                       &ins_list);
4496                 if (ret)
4497                         goto nopos;
4498         }
4499
4500         /* Reached end of directory/root. Bump pos past the last item. */
4501         if (key_type == BTRFS_DIR_INDEX_KEY)
4502                 /*
4503                  * 32-bit glibc will use getdents64, but then strtol -
4504                  * so the last number we can serve is this.
4505                  */
4506                 filp->f_pos = 0x7fffffff;
4507         else
4508                 filp->f_pos++;
4509 nopos:
4510         ret = 0;
4511 err:
4512         if (key_type == BTRFS_DIR_INDEX_KEY)
4513                 btrfs_put_delayed_items(&ins_list, &del_list);
4514         btrfs_free_path(path);
4515         return ret;
4516 }
4517
4518 int btrfs_write_inode(struct inode *inode, struct writeback_control *wbc)
4519 {
4520         struct btrfs_root *root = BTRFS_I(inode)->root;
4521         struct btrfs_trans_handle *trans;
4522         int ret = 0;
4523         bool nolock = false;
4524
4525         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4526                 return 0;
4527
4528         if (btrfs_fs_closing(root->fs_info) && btrfs_is_free_space_inode(inode))
4529                 nolock = true;
4530
4531         if (wbc->sync_mode == WB_SYNC_ALL) {
4532                 if (nolock)
4533                         trans = btrfs_join_transaction_nolock(root);
4534                 else
4535                         trans = btrfs_join_transaction(root);
4536                 if (IS_ERR(trans))
4537                         return PTR_ERR(trans);
4538                 ret = btrfs_commit_transaction(trans, root);
4539         }
4540         return ret;
4541 }
4542
4543 /*
4544  * This is somewhat expensive, updating the tree every time the
4545  * inode changes.  But, it is most likely to find the inode in cache.
4546  * FIXME, needs more benchmarking...there are no reasons other than performance
4547  * to keep or drop this code.
4548  */
4549 int btrfs_dirty_inode(struct inode *inode)
4550 {
4551         struct btrfs_root *root = BTRFS_I(inode)->root;
4552         struct btrfs_trans_handle *trans;
4553         int ret;
4554
4555         if (test_bit(BTRFS_INODE_DUMMY, &BTRFS_I(inode)->runtime_flags))
4556                 return 0;
4557
4558         trans = btrfs_join_transaction(root);
4559         if (IS_ERR(trans))
4560                 return PTR_ERR(trans);
4561
4562         ret = btrfs_update_inode(trans, root, inode);
4563         if (ret && ret == -ENOSPC) {
4564                 /* whoops, lets try again with the full transaction */
4565                 btrfs_end_transaction(trans, root);
4566                 trans = btrfs_start_transaction(root, 1);
4567                 if (IS_ERR(trans))
4568                         return PTR_ERR(trans);
4569
4570                 ret = btrfs_update_inode(trans, root, inode);
4571         }
4572         btrfs_end_transaction(trans, root);
4573         if (BTRFS_I(inode)->delayed_node)
4574                 btrfs_balance_delayed_items(root);
4575
4576         return ret;
4577 }
4578
4579 /*
4580  * This is a copy of file_update_time.  We need this so we can return error on
4581  * ENOSPC for updating the inode in the case of file write and mmap writes.
4582  */
4583 static int btrfs_update_time(struct inode *inode, struct timespec *now,
4584                              int flags)
4585 {
4586         struct btrfs_root *root = BTRFS_I(inode)->root;
4587
4588         if (btrfs_root_readonly(root))
4589                 return -EROFS;
4590
4591         if (flags & S_VERSION)
4592                 inode_inc_iversion(inode);
4593         if (flags & S_CTIME)
4594                 inode->i_ctime = *now;
4595         if (flags & S_MTIME)
4596                 inode->i_mtime = *now;
4597         if (flags & S_ATIME)
4598                 inode->i_atime = *now;
4599         return btrfs_dirty_inode(inode);
4600 }
4601
4602 /*
4603  * find the highest existing sequence number in a directory
4604  * and then set the in-memory index_cnt variable to reflect
4605  * free sequence numbers
4606  */
4607 static int btrfs_set_inode_index_count(struct inode *inode)
4608 {
4609         struct btrfs_root *root = BTRFS_I(inode)->root;
4610         struct btrfs_key key, found_key;
4611         struct btrfs_path *path;
4612         struct extent_buffer *leaf;
4613         int ret;
4614
4615         key.objectid = btrfs_ino(inode);
4616         btrfs_set_key_type(&key, BTRFS_DIR_INDEX_KEY);
4617         key.offset = (u64)-1;
4618
4619         path = btrfs_alloc_path();
4620         if (!path)
4621                 return -ENOMEM;
4622
4623         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4624         if (ret < 0)
4625                 goto out;
4626         /* FIXME: we should be able to handle this */
4627         if (ret == 0)
4628                 goto out;
4629         ret = 0;
4630
4631         /*
4632          * MAGIC NUMBER EXPLANATION:
4633          * since we search a directory based on f_pos we have to start at 2
4634          * since '.' and '..' have f_pos of 0 and 1 respectively, so everybody
4635          * else has to start at 2
4636          */
4637         if (path->slots[0] == 0) {
4638                 BTRFS_I(inode)->index_cnt = 2;
4639                 goto out;
4640         }
4641
4642         path->slots[0]--;
4643
4644         leaf = path->nodes[0];
4645         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4646
4647         if (found_key.objectid != btrfs_ino(inode) ||
4648             btrfs_key_type(&found_key) != BTRFS_DIR_INDEX_KEY) {
4649                 BTRFS_I(inode)->index_cnt = 2;
4650                 goto out;
4651         }
4652
4653         BTRFS_I(inode)->index_cnt = found_key.offset + 1;
4654 out:
4655         btrfs_free_path(path);
4656         return ret;
4657 }
4658
4659 /*
4660  * helper to find a free sequence number in a given directory.  This current
4661  * code is very simple, later versions will do smarter things in the btree
4662  */
4663 int btrfs_set_inode_index(struct inode *dir, u64 *index)
4664 {
4665         int ret = 0;
4666
4667         if (BTRFS_I(dir)->index_cnt == (u64)-1) {
4668                 ret = btrfs_inode_delayed_dir_index_count(dir);
4669                 if (ret) {
4670                         ret = btrfs_set_inode_index_count(dir);
4671                         if (ret)
4672                                 return ret;
4673                 }
4674         }
4675
4676         *index = BTRFS_I(dir)->index_cnt;
4677         BTRFS_I(dir)->index_cnt++;
4678
4679         return ret;
4680 }
4681
4682 static struct inode *btrfs_new_inode(struct btrfs_trans_handle *trans,
4683                                      struct btrfs_root *root,
4684                                      struct inode *dir,
4685                                      const char *name, int name_len,
4686                                      u64 ref_objectid, u64 objectid,
4687                                      umode_t mode, u64 *index)
4688 {
4689         struct inode *inode;
4690         struct btrfs_inode_item *inode_item;
4691         struct btrfs_key *location;
4692         struct btrfs_path *path;
4693         struct btrfs_inode_ref *ref;
4694         struct btrfs_key key[2];
4695         u32 sizes[2];
4696         unsigned long ptr;
4697         int ret;
4698         int owner;
4699
4700         path = btrfs_alloc_path();
4701         if (!path)
4702                 return ERR_PTR(-ENOMEM);
4703
4704         inode = new_inode(root->fs_info->sb);
4705         if (!inode) {
4706                 btrfs_free_path(path);
4707                 return ERR_PTR(-ENOMEM);
4708         }
4709
4710         /*
4711          * we have to initialize this early, so we can reclaim the inode
4712          * number if we fail afterwards in this function.
4713          */
4714         inode->i_ino = objectid;
4715
4716         if (dir) {
4717                 trace_btrfs_inode_request(dir);
4718
4719                 ret = btrfs_set_inode_index(dir, index);
4720                 if (ret) {
4721                         btrfs_free_path(path);
4722                         iput(inode);
4723                         return ERR_PTR(ret);
4724                 }
4725         }
4726         /*
4727          * index_cnt is ignored for everything but a dir,
4728          * btrfs_get_inode_index_count has an explanation for the magic
4729          * number
4730          */
4731         BTRFS_I(inode)->index_cnt = 2;
4732         BTRFS_I(inode)->root = root;
4733         BTRFS_I(inode)->generation = trans->transid;
4734         inode->i_generation = BTRFS_I(inode)->generation;
4735
4736         /*
4737          * We could have gotten an inode number from somebody who was fsynced
4738          * and then removed in this same transaction, so let's just set full
4739          * sync since it will be a full sync anyway and this will blow away the
4740          * old info in the log.
4741          */
4742         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
4743
4744         if (S_ISDIR(mode))
4745                 owner = 0;
4746         else
4747                 owner = 1;
4748
4749         key[0].objectid = objectid;
4750         btrfs_set_key_type(&key[0], BTRFS_INODE_ITEM_KEY);
4751         key[0].offset = 0;
4752
4753         /*
4754          * Start new inodes with an inode_ref. This is slightly more
4755          * efficient for small numbers of hard links since they will
4756          * be packed into one item. Extended refs will kick in if we
4757          * add more hard links than can fit in the ref item.
4758          */
4759         key[1].objectid = objectid;
4760         btrfs_set_key_type(&key[1], BTRFS_INODE_REF_KEY);
4761         key[1].offset = ref_objectid;
4762
4763         sizes[0] = sizeof(struct btrfs_inode_item);
4764         sizes[1] = name_len + sizeof(*ref);
4765
4766         path->leave_spinning = 1;
4767         ret = btrfs_insert_empty_items(trans, root, path, key, sizes, 2);
4768         if (ret != 0)
4769                 goto fail;
4770
4771         inode_init_owner(inode, dir, mode);
4772         inode_set_bytes(inode, 0);
4773         inode->i_mtime = inode->i_atime = inode->i_ctime = CURRENT_TIME;
4774         inode_item = btrfs_item_ptr(path->nodes[0], path->slots[0],
4775                                   struct btrfs_inode_item);
4776         memset_extent_buffer(path->nodes[0], 0, (unsigned long)inode_item,
4777                              sizeof(*inode_item));
4778         fill_inode_item(trans, path->nodes[0], inode_item, inode);
4779
4780         ref = btrfs_item_ptr(path->nodes[0], path->slots[0] + 1,
4781                              struct btrfs_inode_ref);
4782         btrfs_set_inode_ref_name_len(path->nodes[0], ref, name_len);
4783         btrfs_set_inode_ref_index(path->nodes[0], ref, *index);
4784         ptr = (unsigned long)(ref + 1);
4785         write_extent_buffer(path->nodes[0], name, ptr, name_len);
4786
4787         btrfs_mark_buffer_dirty(path->nodes[0]);
4788         btrfs_free_path(path);
4789
4790         location = &BTRFS_I(inode)->location;
4791         location->objectid = objectid;
4792         location->offset = 0;
4793         btrfs_set_key_type(location, BTRFS_INODE_ITEM_KEY);
4794
4795         btrfs_inherit_iflags(inode, dir);
4796
4797         if (S_ISREG(mode)) {
4798                 if (btrfs_test_opt(root, NODATASUM))
4799                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATASUM;
4800                 if (btrfs_test_opt(root, NODATACOW) ||
4801                     (BTRFS_I(dir)->flags & BTRFS_INODE_NODATACOW))
4802                         BTRFS_I(inode)->flags |= BTRFS_INODE_NODATACOW;
4803         }
4804
4805         insert_inode_hash(inode);
4806         inode_tree_add(inode);
4807
4808         trace_btrfs_inode_new(inode);
4809         btrfs_set_inode_last_trans(trans, inode);
4810
4811         btrfs_update_root_times(trans, root);
4812
4813         return inode;
4814 fail:
4815         if (dir)
4816                 BTRFS_I(dir)->index_cnt--;
4817         btrfs_free_path(path);
4818         iput(inode);
4819         return ERR_PTR(ret);
4820 }
4821
4822 static inline u8 btrfs_inode_type(struct inode *inode)
4823 {
4824         return btrfs_type_by_mode[(inode->i_mode & S_IFMT) >> S_SHIFT];
4825 }
4826
4827 /*
4828  * utility function to add 'inode' into 'parent_inode' with
4829  * a give name and a given sequence number.
4830  * if 'add_backref' is true, also insert a backref from the
4831  * inode to the parent directory.
4832  */
4833 int btrfs_add_link(struct btrfs_trans_handle *trans,
4834                    struct inode *parent_inode, struct inode *inode,
4835                    const char *name, int name_len, int add_backref, u64 index)
4836 {
4837         int ret = 0;
4838         struct btrfs_key key;
4839         struct btrfs_root *root = BTRFS_I(parent_inode)->root;
4840         u64 ino = btrfs_ino(inode);
4841         u64 parent_ino = btrfs_ino(parent_inode);
4842
4843         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4844                 memcpy(&key, &BTRFS_I(inode)->root->root_key, sizeof(key));
4845         } else {
4846                 key.objectid = ino;
4847                 btrfs_set_key_type(&key, BTRFS_INODE_ITEM_KEY);
4848                 key.offset = 0;
4849         }
4850
4851         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4852                 ret = btrfs_add_root_ref(trans, root->fs_info->tree_root,
4853                                          key.objectid, root->root_key.objectid,
4854                                          parent_ino, index, name, name_len);
4855         } else if (add_backref) {
4856                 ret = btrfs_insert_inode_ref(trans, root, name, name_len, ino,
4857                                              parent_ino, index);
4858         }
4859
4860         /* Nothing to clean up yet */
4861         if (ret)
4862                 return ret;
4863
4864         ret = btrfs_insert_dir_item(trans, root, name, name_len,
4865                                     parent_inode, &key,
4866                                     btrfs_inode_type(inode), index);
4867         if (ret == -EEXIST)
4868                 goto fail_dir_item;
4869         else if (ret) {
4870                 btrfs_abort_transaction(trans, root, ret);
4871                 return ret;
4872         }
4873
4874         btrfs_i_size_write(parent_inode, parent_inode->i_size +
4875                            name_len * 2);
4876         inode_inc_iversion(parent_inode);
4877         parent_inode->i_mtime = parent_inode->i_ctime = CURRENT_TIME;
4878         ret = btrfs_update_inode(trans, root, parent_inode);
4879         if (ret)
4880                 btrfs_abort_transaction(trans, root, ret);
4881         return ret;
4882
4883 fail_dir_item:
4884         if (unlikely(ino == BTRFS_FIRST_FREE_OBJECTID)) {
4885                 u64 local_index;
4886                 int err;
4887                 err = btrfs_del_root_ref(trans, root->fs_info->tree_root,
4888                                  key.objectid, root->root_key.objectid,
4889                                  parent_ino, &local_index, name, name_len);
4890
4891         } else if (add_backref) {
4892                 u64 local_index;
4893                 int err;
4894
4895                 err = btrfs_del_inode_ref(trans, root, name, name_len,
4896                                           ino, parent_ino, &local_index);
4897         }
4898         return ret;
4899 }
4900
4901 static int btrfs_add_nondir(struct btrfs_trans_handle *trans,
4902                             struct inode *dir, struct dentry *dentry,
4903                             struct inode *inode, int backref, u64 index)
4904 {
4905         int err = btrfs_add_link(trans, dir, inode,
4906                                  dentry->d_name.name, dentry->d_name.len,
4907                                  backref, index);
4908         if (err > 0)
4909                 err = -EEXIST;
4910         return err;
4911 }
4912
4913 static int btrfs_mknod(struct inode *dir, struct dentry *dentry,
4914                         umode_t mode, dev_t rdev)
4915 {
4916         struct btrfs_trans_handle *trans;
4917         struct btrfs_root *root = BTRFS_I(dir)->root;
4918         struct inode *inode = NULL;
4919         int err;
4920         int drop_inode = 0;
4921         u64 objectid;
4922         u64 index = 0;
4923
4924         if (!new_valid_dev(rdev))
4925                 return -EINVAL;
4926
4927         /*
4928          * 2 for inode item and ref
4929          * 2 for dir items
4930          * 1 for xattr if selinux is on
4931          */
4932         trans = btrfs_start_transaction(root, 5);
4933         if (IS_ERR(trans))
4934                 return PTR_ERR(trans);
4935
4936         err = btrfs_find_free_ino(root, &objectid);
4937         if (err)
4938                 goto out_unlock;
4939
4940         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
4941                                 dentry->d_name.len, btrfs_ino(dir), objectid,
4942                                 mode, &index);
4943         if (IS_ERR(inode)) {
4944                 err = PTR_ERR(inode);
4945                 goto out_unlock;
4946         }
4947
4948         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
4949         if (err) {
4950                 drop_inode = 1;
4951                 goto out_unlock;
4952         }
4953
4954         /*
4955         * If the active LSM wants to access the inode during
4956         * d_instantiate it needs these. Smack checks to see
4957         * if the filesystem supports xattrs by looking at the
4958         * ops vector.
4959         */
4960
4961         inode->i_op = &btrfs_special_inode_operations;
4962         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
4963         if (err)
4964                 drop_inode = 1;
4965         else {
4966                 init_special_inode(inode, inode->i_mode, rdev);
4967                 btrfs_update_inode(trans, root, inode);
4968                 d_instantiate(dentry, inode);
4969         }
4970 out_unlock:
4971         btrfs_end_transaction(trans, root);
4972         btrfs_btree_balance_dirty(root);
4973         if (drop_inode) {
4974                 inode_dec_link_count(inode);
4975                 iput(inode);
4976         }
4977         return err;
4978 }
4979
4980 static int btrfs_create(struct inode *dir, struct dentry *dentry,
4981                         umode_t mode, bool excl)
4982 {
4983         struct btrfs_trans_handle *trans;
4984         struct btrfs_root *root = BTRFS_I(dir)->root;
4985         struct inode *inode = NULL;
4986         int drop_inode = 0;
4987         int err;
4988         u64 objectid;
4989         u64 index = 0;
4990
4991         /*
4992          * 2 for inode item and ref
4993          * 2 for dir items
4994          * 1 for xattr if selinux is on
4995          */
4996         trans = btrfs_start_transaction(root, 5);
4997         if (IS_ERR(trans))
4998                 return PTR_ERR(trans);
4999
5000         err = btrfs_find_free_ino(root, &objectid);
5001         if (err)
5002                 goto out_unlock;
5003
5004         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5005                                 dentry->d_name.len, btrfs_ino(dir), objectid,
5006                                 mode, &index);
5007         if (IS_ERR(inode)) {
5008                 err = PTR_ERR(inode);
5009                 goto out_unlock;
5010         }
5011
5012         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5013         if (err) {
5014                 drop_inode = 1;
5015                 goto out_unlock;
5016         }
5017
5018         /*
5019         * If the active LSM wants to access the inode during
5020         * d_instantiate it needs these. Smack checks to see
5021         * if the filesystem supports xattrs by looking at the
5022         * ops vector.
5023         */
5024         inode->i_fop = &btrfs_file_operations;
5025         inode->i_op = &btrfs_file_inode_operations;
5026
5027         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
5028         if (err)
5029                 drop_inode = 1;
5030         else {
5031                 inode->i_mapping->a_ops = &btrfs_aops;
5032                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
5033                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
5034                 d_instantiate(dentry, inode);
5035         }
5036 out_unlock:
5037         btrfs_end_transaction(trans, root);
5038         if (drop_inode) {
5039                 inode_dec_link_count(inode);
5040                 iput(inode);
5041         }
5042         btrfs_btree_balance_dirty(root);
5043         return err;
5044 }
5045
5046 static int btrfs_link(struct dentry *old_dentry, struct inode *dir,
5047                       struct dentry *dentry)
5048 {
5049         struct btrfs_trans_handle *trans;
5050         struct btrfs_root *root = BTRFS_I(dir)->root;
5051         struct inode *inode = old_dentry->d_inode;
5052         u64 index;
5053         int err;
5054         int drop_inode = 0;
5055
5056         /* do not allow sys_link's with other subvols of the same device */
5057         if (root->objectid != BTRFS_I(inode)->root->objectid)
5058                 return -EXDEV;
5059
5060         if (inode->i_nlink >= BTRFS_LINK_MAX)
5061                 return -EMLINK;
5062
5063         err = btrfs_set_inode_index(dir, &index);
5064         if (err)
5065                 goto fail;
5066
5067         /*
5068          * 2 items for inode and inode ref
5069          * 2 items for dir items
5070          * 1 item for parent inode
5071          */
5072         trans = btrfs_start_transaction(root, 5);
5073         if (IS_ERR(trans)) {
5074                 err = PTR_ERR(trans);
5075                 goto fail;
5076         }
5077
5078         btrfs_inc_nlink(inode);
5079         inode_inc_iversion(inode);
5080         inode->i_ctime = CURRENT_TIME;
5081         ihold(inode);
5082
5083         err = btrfs_add_nondir(trans, dir, dentry, inode, 1, index);
5084
5085         if (err) {
5086                 drop_inode = 1;
5087         } else {
5088                 struct dentry *parent = dentry->d_parent;
5089                 err = btrfs_update_inode(trans, root, inode);
5090                 if (err)
5091                         goto fail;
5092                 d_instantiate(dentry, inode);
5093                 btrfs_log_new_name(trans, inode, NULL, parent);
5094         }
5095
5096         btrfs_end_transaction(trans, root);
5097 fail:
5098         if (drop_inode) {
5099                 inode_dec_link_count(inode);
5100                 iput(inode);
5101         }
5102         btrfs_btree_balance_dirty(root);
5103         return err;
5104 }
5105
5106 static int btrfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
5107 {
5108         struct inode *inode = NULL;
5109         struct btrfs_trans_handle *trans;
5110         struct btrfs_root *root = BTRFS_I(dir)->root;
5111         int err = 0;
5112         int drop_on_err = 0;
5113         u64 objectid = 0;
5114         u64 index = 0;
5115
5116         /*
5117          * 2 items for inode and ref
5118          * 2 items for dir items
5119          * 1 for xattr if selinux is on
5120          */
5121         trans = btrfs_start_transaction(root, 5);
5122         if (IS_ERR(trans))
5123                 return PTR_ERR(trans);
5124
5125         err = btrfs_find_free_ino(root, &objectid);
5126         if (err)
5127                 goto out_fail;
5128
5129         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
5130                                 dentry->d_name.len, btrfs_ino(dir), objectid,
5131                                 S_IFDIR | mode, &index);
5132         if (IS_ERR(inode)) {
5133                 err = PTR_ERR(inode);
5134                 goto out_fail;
5135         }
5136
5137         drop_on_err = 1;
5138
5139         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
5140         if (err)
5141                 goto out_fail;
5142
5143         inode->i_op = &btrfs_dir_inode_operations;
5144         inode->i_fop = &btrfs_dir_file_operations;
5145
5146         btrfs_i_size_write(inode, 0);
5147         err = btrfs_update_inode(trans, root, inode);
5148         if (err)
5149                 goto out_fail;
5150
5151         err = btrfs_add_link(trans, dir, inode, dentry->d_name.name,
5152                              dentry->d_name.len, 0, index);
5153         if (err)
5154                 goto out_fail;
5155
5156         d_instantiate(dentry, inode);
5157         drop_on_err = 0;
5158
5159 out_fail:
5160         btrfs_end_transaction(trans, root);
5161         if (drop_on_err)
5162                 iput(inode);
5163         btrfs_btree_balance_dirty(root);
5164         return err;
5165 }
5166
5167 /* helper for btfs_get_extent.  Given an existing extent in the tree,
5168  * and an extent that you want to insert, deal with overlap and insert
5169  * the new extent into the tree.
5170  */
5171 static int merge_extent_mapping(struct extent_map_tree *em_tree,
5172                                 struct extent_map *existing,
5173                                 struct extent_map *em,
5174                                 u64 map_start, u64 map_len)
5175 {
5176         u64 start_diff;
5177
5178         BUG_ON(map_start < em->start || map_start >= extent_map_end(em));
5179         start_diff = map_start - em->start;
5180         em->start = map_start;
5181         em->len = map_len;
5182         if (em->block_start < EXTENT_MAP_LAST_BYTE &&
5183             !test_bit(EXTENT_FLAG_COMPRESSED, &em->flags)) {
5184                 em->block_start += start_diff;
5185                 em->block_len -= start_diff;
5186         }
5187         return add_extent_mapping(em_tree, em);
5188 }
5189
5190 static noinline int uncompress_inline(struct btrfs_path *path,
5191                                       struct inode *inode, struct page *page,
5192                                       size_t pg_offset, u64 extent_offset,
5193                                       struct btrfs_file_extent_item *item)
5194 {
5195         int ret;
5196         struct extent_buffer *leaf = path->nodes[0];
5197         char *tmp;
5198         size_t max_size;
5199         unsigned long inline_size;
5200         unsigned long ptr;
5201         int compress_type;
5202
5203         WARN_ON(pg_offset != 0);
5204         compress_type = btrfs_file_extent_compression(leaf, item);
5205         max_size = btrfs_file_extent_ram_bytes(leaf, item);
5206         inline_size = btrfs_file_extent_inline_item_len(leaf,
5207                                         btrfs_item_nr(leaf, path->slots[0]));
5208         tmp = kmalloc(inline_size, GFP_NOFS);
5209         if (!tmp)
5210                 return -ENOMEM;
5211         ptr = btrfs_file_extent_inline_start(item);
5212
5213         read_extent_buffer(leaf, tmp, ptr, inline_size);
5214
5215         max_size = min_t(unsigned long, PAGE_CACHE_SIZE, max_size);
5216         ret = btrfs_decompress(compress_type, tmp, page,
5217                                extent_offset, inline_size, max_size);
5218         if (ret) {
5219                 char *kaddr = kmap_atomic(page);
5220                 unsigned long copy_size = min_t(u64,
5221                                   PAGE_CACHE_SIZE - pg_offset,
5222                                   max_size - extent_offset);
5223                 memset(kaddr + pg_offset, 0, copy_size);
5224                 kunmap_atomic(kaddr);
5225         }
5226         kfree(tmp);
5227         return 0;
5228 }
5229
5230 /*
5231  * a bit scary, this does extent mapping from logical file offset to the disk.
5232  * the ugly parts come from merging extents from the disk with the in-ram
5233  * representation.  This gets more complex because of the data=ordered code,
5234  * where the in-ram extents might be locked pending data=ordered completion.
5235  *
5236  * This also copies inline extents directly into the page.
5237  */
5238
5239 struct extent_map *btrfs_get_extent(struct inode *inode, struct page *page,
5240                                     size_t pg_offset, u64 start, u64 len,
5241                                     int create)
5242 {
5243         int ret;
5244         int err = 0;
5245         u64 bytenr;
5246         u64 extent_start = 0;
5247         u64 extent_end = 0;
5248         u64 objectid = btrfs_ino(inode);
5249         u32 found_type;
5250         struct btrfs_path *path = NULL;
5251         struct btrfs_root *root = BTRFS_I(inode)->root;
5252         struct btrfs_file_extent_item *item;
5253         struct extent_buffer *leaf;
5254         struct btrfs_key found_key;
5255         struct extent_map *em = NULL;
5256         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5257         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
5258         struct btrfs_trans_handle *trans = NULL;
5259         int compress_type;
5260
5261 again:
5262         read_lock(&em_tree->lock);
5263         em = lookup_extent_mapping(em_tree, start, len);
5264         if (em)
5265                 em->bdev = root->fs_info->fs_devices->latest_bdev;
5266         read_unlock(&em_tree->lock);
5267
5268         if (em) {
5269                 if (em->start > start || em->start + em->len <= start)
5270                         free_extent_map(em);
5271                 else if (em->block_start == EXTENT_MAP_INLINE && page)
5272                         free_extent_map(em);
5273                 else
5274                         goto out;
5275         }
5276         em = alloc_extent_map();
5277         if (!em) {
5278                 err = -ENOMEM;
5279                 goto out;
5280         }
5281         em->bdev = root->fs_info->fs_devices->latest_bdev;
5282         em->start = EXTENT_MAP_HOLE;
5283         em->orig_start = EXTENT_MAP_HOLE;
5284         em->len = (u64)-1;
5285         em->block_len = (u64)-1;
5286
5287         if (!path) {
5288                 path = btrfs_alloc_path();
5289                 if (!path) {
5290                         err = -ENOMEM;
5291                         goto out;
5292                 }
5293                 /*
5294                  * Chances are we'll be called again, so go ahead and do
5295                  * readahead
5296                  */
5297                 path->reada = 1;
5298         }
5299
5300         ret = btrfs_lookup_file_extent(trans, root, path,
5301                                        objectid, start, trans != NULL);
5302         if (ret < 0) {
5303                 err = ret;
5304                 goto out;
5305         }
5306
5307         if (ret != 0) {
5308                 if (path->slots[0] == 0)
5309                         goto not_found;
5310                 path->slots[0]--;
5311         }
5312
5313         leaf = path->nodes[0];
5314         item = btrfs_item_ptr(leaf, path->slots[0],
5315                               struct btrfs_file_extent_item);
5316         /* are we inside the extent that was found? */
5317         btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5318         found_type = btrfs_key_type(&found_key);
5319         if (found_key.objectid != objectid ||
5320             found_type != BTRFS_EXTENT_DATA_KEY) {
5321                 goto not_found;
5322         }
5323
5324         found_type = btrfs_file_extent_type(leaf, item);
5325         extent_start = found_key.offset;
5326         compress_type = btrfs_file_extent_compression(leaf, item);
5327         if (found_type == BTRFS_FILE_EXTENT_REG ||
5328             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5329                 extent_end = extent_start +
5330                        btrfs_file_extent_num_bytes(leaf, item);
5331         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5332                 size_t size;
5333                 size = btrfs_file_extent_inline_len(leaf, item);
5334                 extent_end = (extent_start + size + root->sectorsize - 1) &
5335                         ~((u64)root->sectorsize - 1);
5336         }
5337
5338         if (start >= extent_end) {
5339                 path->slots[0]++;
5340                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
5341                         ret = btrfs_next_leaf(root, path);
5342                         if (ret < 0) {
5343                                 err = ret;
5344                                 goto out;
5345                         }
5346                         if (ret > 0)
5347                                 goto not_found;
5348                         leaf = path->nodes[0];
5349                 }
5350                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5351                 if (found_key.objectid != objectid ||
5352                     found_key.type != BTRFS_EXTENT_DATA_KEY)
5353                         goto not_found;
5354                 if (start + len <= found_key.offset)
5355                         goto not_found;
5356                 em->start = start;
5357                 em->len = found_key.offset - start;
5358                 goto not_found_em;
5359         }
5360
5361         if (found_type == BTRFS_FILE_EXTENT_REG ||
5362             found_type == BTRFS_FILE_EXTENT_PREALLOC) {
5363                 em->start = extent_start;
5364                 em->len = extent_end - extent_start;
5365                 em->orig_start = extent_start -
5366                                  btrfs_file_extent_offset(leaf, item);
5367                 bytenr = btrfs_file_extent_disk_bytenr(leaf, item);
5368                 if (bytenr == 0) {
5369                         em->block_start = EXTENT_MAP_HOLE;
5370                         goto insert;
5371                 }
5372                 if (compress_type != BTRFS_COMPRESS_NONE) {
5373                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5374                         em->compress_type = compress_type;
5375                         em->block_start = bytenr;
5376                         em->block_len = btrfs_file_extent_disk_num_bytes(leaf,
5377                                                                          item);
5378                 } else {
5379                         bytenr += btrfs_file_extent_offset(leaf, item);
5380                         em->block_start = bytenr;
5381                         em->block_len = em->len;
5382                         if (found_type == BTRFS_FILE_EXTENT_PREALLOC)
5383                                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5384                 }
5385                 goto insert;
5386         } else if (found_type == BTRFS_FILE_EXTENT_INLINE) {
5387                 unsigned long ptr;
5388                 char *map;
5389                 size_t size;
5390                 size_t extent_offset;
5391                 size_t copy_size;
5392
5393                 em->block_start = EXTENT_MAP_INLINE;
5394                 if (!page || create) {
5395                         em->start = extent_start;
5396                         em->len = extent_end - extent_start;
5397                         goto out;
5398                 }
5399
5400                 size = btrfs_file_extent_inline_len(leaf, item);
5401                 extent_offset = page_offset(page) + pg_offset - extent_start;
5402                 copy_size = min_t(u64, PAGE_CACHE_SIZE - pg_offset,
5403                                 size - extent_offset);
5404                 em->start = extent_start + extent_offset;
5405                 em->len = (copy_size + root->sectorsize - 1) &
5406                         ~((u64)root->sectorsize - 1);
5407                 em->orig_start = EXTENT_MAP_INLINE;
5408                 if (compress_type) {
5409                         set_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
5410                         em->compress_type = compress_type;
5411                 }
5412                 ptr = btrfs_file_extent_inline_start(item) + extent_offset;
5413                 if (create == 0 && !PageUptodate(page)) {
5414                         if (btrfs_file_extent_compression(leaf, item) !=
5415                             BTRFS_COMPRESS_NONE) {
5416                                 ret = uncompress_inline(path, inode, page,
5417                                                         pg_offset,
5418                                                         extent_offset, item);
5419                                 BUG_ON(ret); /* -ENOMEM */
5420                         } else {
5421                                 map = kmap(page);
5422                                 read_extent_buffer(leaf, map + pg_offset, ptr,
5423                                                    copy_size);
5424                                 if (pg_offset + copy_size < PAGE_CACHE_SIZE) {
5425                                         memset(map + pg_offset + copy_size, 0,
5426                                                PAGE_CACHE_SIZE - pg_offset -
5427                                                copy_size);
5428                                 }
5429                                 kunmap(page);
5430                         }
5431                         flush_dcache_page(page);
5432                 } else if (create && PageUptodate(page)) {
5433                         BUG();
5434                         if (!trans) {
5435                                 kunmap(page);
5436                                 free_extent_map(em);
5437                                 em = NULL;
5438
5439                                 btrfs_release_path(path);
5440                                 trans = btrfs_join_transaction(root);
5441
5442                                 if (IS_ERR(trans))
5443                                         return ERR_CAST(trans);
5444                                 goto again;
5445                         }
5446                         map = kmap(page);
5447                         write_extent_buffer(leaf, map + pg_offset, ptr,
5448                                             copy_size);
5449                         kunmap(page);
5450                         btrfs_mark_buffer_dirty(leaf);
5451                 }
5452                 set_extent_uptodate(io_tree, em->start,
5453                                     extent_map_end(em) - 1, NULL, GFP_NOFS);
5454                 goto insert;
5455         } else {
5456                 WARN(1, KERN_ERR "btrfs unknown found_type %d\n", found_type);
5457         }
5458 not_found:
5459         em->start = start;
5460         em->len = len;
5461 not_found_em:
5462         em->block_start = EXTENT_MAP_HOLE;
5463         set_bit(EXTENT_FLAG_VACANCY, &em->flags);
5464 insert:
5465         btrfs_release_path(path);
5466         if (em->start > start || extent_map_end(em) <= start) {
5467                 printk(KERN_ERR "Btrfs: bad extent! em: [%llu %llu] passed "
5468                        "[%llu %llu]\n", (unsigned long long)em->start,
5469                        (unsigned long long)em->len,
5470                        (unsigned long long)start,
5471                        (unsigned long long)len);
5472                 err = -EIO;
5473                 goto out;
5474         }
5475
5476         err = 0;
5477         write_lock(&em_tree->lock);
5478         ret = add_extent_mapping(em_tree, em);
5479         /* it is possible that someone inserted the extent into the tree
5480          * while we had the lock dropped.  It is also possible that
5481          * an overlapping map exists in the tree
5482          */
5483         if (ret == -EEXIST) {
5484                 struct extent_map *existing;
5485
5486                 ret = 0;
5487
5488                 existing = lookup_extent_mapping(em_tree, start, len);
5489                 if (existing && (existing->start > start ||
5490                     existing->start + existing->len <= start)) {
5491                         free_extent_map(existing);
5492                         existing = NULL;
5493                 }
5494                 if (!existing) {
5495                         existing = lookup_extent_mapping(em_tree, em->start,
5496                                                          em->len);
5497                         if (existing) {
5498                                 err = merge_extent_mapping(em_tree, existing,
5499                                                            em, start,
5500                                                            root->sectorsize);
5501                                 free_extent_map(existing);
5502                                 if (err) {
5503                                         free_extent_map(em);
5504                                         em = NULL;
5505                                 }
5506                         } else {
5507                                 err = -EIO;
5508                                 free_extent_map(em);
5509                                 em = NULL;
5510                         }
5511                 } else {
5512                         free_extent_map(em);
5513                         em = existing;
5514                         err = 0;
5515                 }
5516         }
5517         write_unlock(&em_tree->lock);
5518 out:
5519
5520         if (em)
5521                 trace_btrfs_get_extent(root, em);
5522
5523         if (path)
5524                 btrfs_free_path(path);
5525         if (trans) {
5526                 ret = btrfs_end_transaction(trans, root);
5527                 if (!err)
5528                         err = ret;
5529         }
5530         if (err) {
5531                 free_extent_map(em);
5532                 return ERR_PTR(err);
5533         }
5534         BUG_ON(!em); /* Error is always set */
5535         return em;
5536 }
5537
5538 struct extent_map *btrfs_get_extent_fiemap(struct inode *inode, struct page *page,
5539                                            size_t pg_offset, u64 start, u64 len,
5540                                            int create)
5541 {
5542         struct extent_map *em;
5543         struct extent_map *hole_em = NULL;
5544         u64 range_start = start;
5545         u64 end;
5546         u64 found;
5547         u64 found_end;
5548         int err = 0;
5549
5550         em = btrfs_get_extent(inode, page, pg_offset, start, len, create);
5551         if (IS_ERR(em))
5552                 return em;
5553         if (em) {
5554                 /*
5555                  * if our em maps to a hole, there might
5556                  * actually be delalloc bytes behind it
5557                  */
5558                 if (em->block_start != EXTENT_MAP_HOLE)
5559                         return em;
5560                 else
5561                         hole_em = em;
5562         }
5563
5564         /* check to see if we've wrapped (len == -1 or similar) */
5565         end = start + len;
5566         if (end < start)
5567                 end = (u64)-1;
5568         else
5569                 end -= 1;
5570
5571         em = NULL;
5572
5573         /* ok, we didn't find anything, lets look for delalloc */
5574         found = count_range_bits(&BTRFS_I(inode)->io_tree, &range_start,
5575                                  end, len, EXTENT_DELALLOC, 1);
5576         found_end = range_start + found;
5577         if (found_end < range_start)
5578                 found_end = (u64)-1;
5579
5580         /*
5581          * we didn't find anything useful, return
5582          * the original results from get_extent()
5583          */
5584         if (range_start > end || found_end <= start) {
5585                 em = hole_em;
5586                 hole_em = NULL;
5587                 goto out;
5588         }
5589
5590         /* adjust the range_start to make sure it doesn't
5591          * go backwards from the start they passed in
5592          */
5593         range_start = max(start,range_start);
5594         found = found_end - range_start;
5595
5596         if (found > 0) {
5597                 u64 hole_start = start;
5598                 u64 hole_len = len;
5599
5600                 em = alloc_extent_map();
5601                 if (!em) {
5602                         err = -ENOMEM;
5603                         goto out;
5604                 }
5605                 /*
5606                  * when btrfs_get_extent can't find anything it
5607                  * returns one huge hole
5608                  *
5609                  * make sure what it found really fits our range, and
5610                  * adjust to make sure it is based on the start from
5611                  * the caller
5612                  */
5613                 if (hole_em) {
5614                         u64 calc_end = extent_map_end(hole_em);
5615
5616                         if (calc_end <= start || (hole_em->start > end)) {
5617                                 free_extent_map(hole_em);
5618                                 hole_em = NULL;
5619                         } else {
5620                                 hole_start = max(hole_em->start, start);
5621                                 hole_len = calc_end - hole_start;
5622                         }
5623                 }
5624                 em->bdev = NULL;
5625                 if (hole_em && range_start > hole_start) {
5626                         /* our hole starts before our delalloc, so we
5627                          * have to return just the parts of the hole
5628                          * that go until  the delalloc starts
5629                          */
5630                         em->len = min(hole_len,
5631                                       range_start - hole_start);
5632                         em->start = hole_start;
5633                         em->orig_start = hole_start;
5634                         /*
5635                          * don't adjust block start at all,
5636                          * it is fixed at EXTENT_MAP_HOLE
5637                          */
5638                         em->block_start = hole_em->block_start;
5639                         em->block_len = hole_len;
5640                 } else {
5641                         em->start = range_start;
5642                         em->len = found;
5643                         em->orig_start = range_start;
5644                         em->block_start = EXTENT_MAP_DELALLOC;
5645                         em->block_len = found;
5646                 }
5647         } else if (hole_em) {
5648                 return hole_em;
5649         }
5650 out:
5651
5652         free_extent_map(hole_em);
5653         if (err) {
5654                 free_extent_map(em);
5655                 return ERR_PTR(err);
5656         }
5657         return em;
5658 }
5659
5660 static struct extent_map *btrfs_new_extent_direct(struct inode *inode,
5661                                                   struct extent_map *em,
5662                                                   u64 start, u64 len)
5663 {
5664         struct btrfs_root *root = BTRFS_I(inode)->root;
5665         struct btrfs_trans_handle *trans;
5666         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
5667         struct btrfs_key ins;
5668         u64 alloc_hint;
5669         int ret;
5670         bool insert = false;
5671
5672         /*
5673          * Ok if the extent map we looked up is a hole and is for the exact
5674          * range we want, there is no reason to allocate a new one, however if
5675          * it is not right then we need to free this one and drop the cache for
5676          * our range.
5677          */
5678         if (em->block_start != EXTENT_MAP_HOLE || em->start != start ||
5679             em->len != len) {
5680                 free_extent_map(em);
5681                 em = NULL;
5682                 insert = true;
5683                 btrfs_drop_extent_cache(inode, start, start + len - 1, 0);
5684         }
5685
5686         trans = btrfs_join_transaction(root);
5687         if (IS_ERR(trans))
5688                 return ERR_CAST(trans);
5689
5690         if (start <= BTRFS_I(inode)->disk_i_size && len < 64 * 1024)
5691                 btrfs_add_inode_defrag(trans, inode);
5692
5693         trans->block_rsv = &root->fs_info->delalloc_block_rsv;
5694
5695         alloc_hint = get_extent_allocation_hint(inode, start, len);
5696         ret = btrfs_reserve_extent(trans, root, len, root->sectorsize, 0,
5697                                    alloc_hint, &ins, 1);
5698         if (ret) {
5699                 em = ERR_PTR(ret);
5700                 goto out;
5701         }
5702
5703         if (!em) {
5704                 em = alloc_extent_map();
5705                 if (!em) {
5706                         em = ERR_PTR(-ENOMEM);
5707                         goto out;
5708                 }
5709         }
5710
5711         em->start = start;
5712         em->orig_start = em->start;
5713         em->len = ins.offset;
5714
5715         em->block_start = ins.objectid;
5716         em->block_len = ins.offset;
5717         em->bdev = root->fs_info->fs_devices->latest_bdev;
5718
5719         /*
5720          * We need to do this because if we're using the original em we searched
5721          * for, we could have EXTENT_FLAG_VACANCY set, and we don't want that.
5722          */
5723         em->flags = 0;
5724         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5725
5726         while (insert) {
5727                 write_lock(&em_tree->lock);
5728                 ret = add_extent_mapping(em_tree, em);
5729                 write_unlock(&em_tree->lock);
5730                 if (ret != -EEXIST)
5731                         break;
5732                 btrfs_drop_extent_cache(inode, start, start + em->len - 1, 0);
5733         }
5734
5735         ret = btrfs_add_ordered_extent_dio(inode, start, ins.objectid,
5736                                            ins.offset, ins.offset, 0);
5737         if (ret) {
5738                 btrfs_free_reserved_extent(root, ins.objectid, ins.offset);
5739                 em = ERR_PTR(ret);
5740         }
5741 out:
5742         btrfs_end_transaction(trans, root);
5743         return em;
5744 }
5745
5746 /*
5747  * returns 1 when the nocow is safe, < 1 on error, 0 if the
5748  * block must be cow'd
5749  */
5750 static noinline int can_nocow_odirect(struct btrfs_trans_handle *trans,
5751                                       struct inode *inode, u64 offset, u64 len)
5752 {
5753         struct btrfs_path *path;
5754         int ret;
5755         struct extent_buffer *leaf;
5756         struct btrfs_root *root = BTRFS_I(inode)->root;
5757         struct btrfs_file_extent_item *fi;
5758         struct btrfs_key key;
5759         u64 disk_bytenr;
5760         u64 backref_offset;
5761         u64 extent_end;
5762         u64 num_bytes;
5763         int slot;
5764         int found_type;
5765
5766         path = btrfs_alloc_path();
5767         if (!path)
5768                 return -ENOMEM;
5769
5770         ret = btrfs_lookup_file_extent(trans, root, path, btrfs_ino(inode),
5771                                        offset, 0);
5772         if (ret < 0)
5773                 goto out;
5774
5775         slot = path->slots[0];
5776         if (ret == 1) {
5777                 if (slot == 0) {
5778                         /* can't find the item, must cow */
5779                         ret = 0;
5780                         goto out;
5781                 }
5782                 slot--;
5783         }
5784         ret = 0;
5785         leaf = path->nodes[0];
5786         btrfs_item_key_to_cpu(leaf, &key, slot);
5787         if (key.objectid != btrfs_ino(inode) ||
5788             key.type != BTRFS_EXTENT_DATA_KEY) {
5789                 /* not our file or wrong item type, must cow */
5790                 goto out;
5791         }
5792
5793         if (key.offset > offset) {
5794                 /* Wrong offset, must cow */
5795                 goto out;
5796         }
5797
5798         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
5799         found_type = btrfs_file_extent_type(leaf, fi);
5800         if (found_type != BTRFS_FILE_EXTENT_REG &&
5801             found_type != BTRFS_FILE_EXTENT_PREALLOC) {
5802                 /* not a regular extent, must cow */
5803                 goto out;
5804         }
5805         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
5806         backref_offset = btrfs_file_extent_offset(leaf, fi);
5807
5808         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
5809         if (extent_end < offset + len) {
5810                 /* extent doesn't include our full range, must cow */
5811                 goto out;
5812         }
5813
5814         if (btrfs_extent_readonly(root, disk_bytenr))
5815                 goto out;
5816
5817         /*
5818          * look for other files referencing this extent, if we
5819          * find any we must cow
5820          */
5821         if (btrfs_cross_ref_exist(trans, root, btrfs_ino(inode),
5822                                   key.offset - backref_offset, disk_bytenr))
5823                 goto out;
5824
5825         /*
5826          * adjust disk_bytenr and num_bytes to cover just the bytes
5827          * in this extent we are about to write.  If there
5828          * are any csums in that range we have to cow in order
5829          * to keep the csums correct
5830          */
5831         disk_bytenr += backref_offset;
5832         disk_bytenr += offset - key.offset;
5833         num_bytes = min(offset + len, extent_end) - offset;
5834         if (csum_exist_in_range(root, disk_bytenr, num_bytes))
5835                                 goto out;
5836         /*
5837          * all of the above have passed, it is safe to overwrite this extent
5838          * without cow
5839          */
5840         ret = 1;
5841 out:
5842         btrfs_free_path(path);
5843         return ret;
5844 }
5845
5846 static int lock_extent_direct(struct inode *inode, u64 lockstart, u64 lockend,
5847                               struct extent_state **cached_state, int writing)
5848 {
5849         struct btrfs_ordered_extent *ordered;
5850         int ret = 0;
5851
5852         while (1) {
5853                 lock_extent_bits(&BTRFS_I(inode)->io_tree, lockstart, lockend,
5854                                  0, cached_state);
5855                 /*
5856                  * We're concerned with the entire range that we're going to be
5857                  * doing DIO to, so we need to make sure theres no ordered
5858                  * extents in this range.
5859                  */
5860                 ordered = btrfs_lookup_ordered_range(inode, lockstart,
5861                                                      lockend - lockstart + 1);
5862
5863                 /*
5864                  * We need to make sure there are no buffered pages in this
5865                  * range either, we could have raced between the invalidate in
5866                  * generic_file_direct_write and locking the extent.  The
5867                  * invalidate needs to happen so that reads after a write do not
5868                  * get stale data.
5869                  */
5870                 if (!ordered && (!writing ||
5871                     !test_range_bit(&BTRFS_I(inode)->io_tree,
5872                                     lockstart, lockend, EXTENT_UPTODATE, 0,
5873                                     *cached_state)))
5874                         break;
5875
5876                 unlock_extent_cached(&BTRFS_I(inode)->io_tree, lockstart, lockend,
5877                                      cached_state, GFP_NOFS);
5878
5879                 if (ordered) {
5880                         btrfs_start_ordered_extent(inode, ordered, 1);
5881                         btrfs_put_ordered_extent(ordered);
5882                 } else {
5883                         /* Screw you mmap */
5884                         ret = filemap_write_and_wait_range(inode->i_mapping,
5885                                                            lockstart,
5886                                                            lockend);
5887                         if (ret)
5888                                 break;
5889
5890                         /*
5891                          * If we found a page that couldn't be invalidated just
5892                          * fall back to buffered.
5893                          */
5894                         ret = invalidate_inode_pages2_range(inode->i_mapping,
5895                                         lockstart >> PAGE_CACHE_SHIFT,
5896                                         lockend >> PAGE_CACHE_SHIFT);
5897                         if (ret)
5898                                 break;
5899                 }
5900
5901                 cond_resched();
5902         }
5903
5904         return ret;
5905 }
5906
5907 static struct extent_map *create_pinned_em(struct inode *inode, u64 start,
5908                                            u64 len, u64 orig_start,
5909                                            u64 block_start, u64 block_len,
5910                                            int type)
5911 {
5912         struct extent_map_tree *em_tree;
5913         struct extent_map *em;
5914         struct btrfs_root *root = BTRFS_I(inode)->root;
5915         int ret;
5916
5917         em_tree = &BTRFS_I(inode)->extent_tree;
5918         em = alloc_extent_map();
5919         if (!em)
5920                 return ERR_PTR(-ENOMEM);
5921
5922         em->start = start;
5923         em->orig_start = orig_start;
5924         em->len = len;
5925         em->block_len = block_len;
5926         em->block_start = block_start;
5927         em->bdev = root->fs_info->fs_devices->latest_bdev;
5928         set_bit(EXTENT_FLAG_PINNED, &em->flags);
5929         if (type == BTRFS_ORDERED_PREALLOC)
5930                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
5931
5932         do {
5933                 btrfs_drop_extent_cache(inode, em->start,
5934                                 em->start + em->len - 1, 0);
5935                 write_lock(&em_tree->lock);
5936                 ret = add_extent_mapping(em_tree, em);
5937                 write_unlock(&em_tree->lock);
5938         } while (ret == -EEXIST);
5939
5940         if (ret) {
5941                 free_extent_map(em);
5942                 return ERR_PTR(ret);
5943         }
5944
5945         return em;
5946 }
5947
5948
5949 static int btrfs_get_blocks_direct(struct inode *inode, sector_t iblock,
5950                                    struct buffer_head *bh_result, int create)
5951 {
5952         struct extent_map *em;
5953         struct btrfs_root *root = BTRFS_I(inode)->root;
5954         struct extent_state *cached_state = NULL;
5955         u64 start = iblock << inode->i_blkbits;
5956         u64 lockstart, lockend;
5957         u64 len = bh_result->b_size;
5958         struct btrfs_trans_handle *trans;
5959         int unlock_bits = EXTENT_LOCKED;
5960         int ret;
5961
5962         if (create) {
5963                 ret = btrfs_delalloc_reserve_space(inode, len);
5964                 if (ret)
5965                         return ret;
5966                 unlock_bits |= EXTENT_DELALLOC | EXTENT_DIRTY;
5967         } else {
5968                 len = min_t(u64, len, root->sectorsize);
5969         }
5970
5971         lockstart = start;
5972         lockend = start + len - 1;
5973
5974         /*
5975          * If this errors out it's because we couldn't invalidate pagecache for
5976          * this range and we need to fallback to buffered.
5977          */
5978         if (lock_extent_direct(inode, lockstart, lockend, &cached_state, create))
5979                 return -ENOTBLK;
5980
5981         if (create) {
5982                 ret = set_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
5983                                      lockend, EXTENT_DELALLOC, NULL,
5984                                      &cached_state, GFP_NOFS);
5985                 if (ret)
5986                         goto unlock_err;
5987         }
5988
5989         em = btrfs_get_extent(inode, NULL, 0, start, len, 0);
5990         if (IS_ERR(em)) {
5991                 ret = PTR_ERR(em);
5992                 goto unlock_err;
5993         }
5994
5995         /*
5996          * Ok for INLINE and COMPRESSED extents we need to fallback on buffered
5997          * io.  INLINE is special, and we could probably kludge it in here, but
5998          * it's still buffered so for safety lets just fall back to the generic
5999          * buffered path.
6000          *
6001          * For COMPRESSED we _have_ to read the entire extent in so we can
6002          * decompress it, so there will be buffering required no matter what we
6003          * do, so go ahead and fallback to buffered.
6004          *
6005          * We return -ENOTBLK because thats what makes DIO go ahead and go back
6006          * to buffered IO.  Don't blame me, this is the price we pay for using
6007          * the generic code.
6008          */
6009         if (test_bit(EXTENT_FLAG_COMPRESSED, &em->flags) ||
6010             em->block_start == EXTENT_MAP_INLINE) {
6011                 free_extent_map(em);
6012                 ret = -ENOTBLK;
6013                 goto unlock_err;
6014         }
6015
6016         /* Just a good old fashioned hole, return */
6017         if (!create && (em->block_start == EXTENT_MAP_HOLE ||
6018                         test_bit(EXTENT_FLAG_PREALLOC, &em->flags))) {
6019                 free_extent_map(em);
6020                 ret = 0;
6021                 goto unlock_err;
6022         }
6023
6024         /*
6025          * We don't allocate a new extent in the following cases
6026          *
6027          * 1) The inode is marked as NODATACOW.  In this case we'll just use the
6028          * existing extent.
6029          * 2) The extent is marked as PREALLOC.  We're good to go here and can
6030          * just use the extent.
6031          *
6032          */
6033         if (!create) {
6034                 len = min(len, em->len - (start - em->start));
6035                 lockstart = start + len;
6036                 goto unlock;
6037         }
6038
6039         if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags) ||
6040             ((BTRFS_I(inode)->flags & BTRFS_INODE_NODATACOW) &&
6041              em->block_start != EXTENT_MAP_HOLE)) {
6042                 int type;
6043                 int ret;
6044                 u64 block_start;
6045
6046                 if (test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6047                         type = BTRFS_ORDERED_PREALLOC;
6048                 else
6049                         type = BTRFS_ORDERED_NOCOW;
6050                 len = min(len, em->len - (start - em->start));
6051                 block_start = em->block_start + (start - em->start);
6052
6053                 /*
6054                  * we're not going to log anything, but we do need
6055                  * to make sure the current transaction stays open
6056                  * while we look for nocow cross refs
6057                  */
6058                 trans = btrfs_join_transaction(root);
6059                 if (IS_ERR(trans))
6060                         goto must_cow;
6061
6062                 if (can_nocow_odirect(trans, inode, start, len) == 1) {
6063                         u64 orig_start = em->start;
6064
6065                         if (type == BTRFS_ORDERED_PREALLOC) {
6066                                 free_extent_map(em);
6067                                 em = create_pinned_em(inode, start, len,
6068                                                        orig_start,
6069                                                        block_start, len, type);
6070                                 if (IS_ERR(em)) {
6071                                         btrfs_end_transaction(trans, root);
6072                                         goto unlock_err;
6073                                 }
6074                         }
6075
6076                         ret = btrfs_add_ordered_extent_dio(inode, start,
6077                                            block_start, len, len, type);
6078                         btrfs_end_transaction(trans, root);
6079                         if (ret) {
6080                                 free_extent_map(em);
6081                                 goto unlock_err;
6082                         }
6083                         goto unlock;
6084                 }
6085                 btrfs_end_transaction(trans, root);
6086         }
6087 must_cow:
6088         /*
6089          * this will cow the extent, reset the len in case we changed
6090          * it above
6091          */
6092         len = bh_result->b_size;
6093         em = btrfs_new_extent_direct(inode, em, start, len);
6094         if (IS_ERR(em)) {
6095                 ret = PTR_ERR(em);
6096                 goto unlock_err;
6097         }
6098         len = min(len, em->len - (start - em->start));
6099 unlock:
6100         bh_result->b_blocknr = (em->block_start + (start - em->start)) >>
6101                 inode->i_blkbits;
6102         bh_result->b_size = len;
6103         bh_result->b_bdev = em->bdev;
6104         set_buffer_mapped(bh_result);
6105         if (create) {
6106                 if (!test_bit(EXTENT_FLAG_PREALLOC, &em->flags))
6107                         set_buffer_new(bh_result);
6108
6109                 /*
6110                  * Need to update the i_size under the extent lock so buffered
6111                  * readers will get the updated i_size when we unlock.
6112                  */
6113                 if (start + len > i_size_read(inode))
6114                         i_size_write(inode, start + len);
6115         }
6116
6117         /*
6118          * In the case of write we need to clear and unlock the entire range,
6119          * in the case of read we need to unlock only the end area that we
6120          * aren't using if there is any left over space.
6121          */
6122         if (lockstart < lockend) {
6123                 if (create && len < lockend - lockstart) {
6124                         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6125                                          lockstart + len - 1,
6126                                          unlock_bits | EXTENT_DEFRAG, 1, 0,
6127                                          &cached_state, GFP_NOFS);
6128                         /*
6129                          * Beside unlock, we also need to cleanup reserved space
6130                          * for the left range by attaching EXTENT_DO_ACCOUNTING.
6131                          */
6132                         clear_extent_bit(&BTRFS_I(inode)->io_tree,
6133                                          lockstart + len, lockend,
6134                                          unlock_bits | EXTENT_DO_ACCOUNTING |
6135                                          EXTENT_DEFRAG, 1, 0, NULL, GFP_NOFS);
6136                 } else {
6137                         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart,
6138                                          lockend, unlock_bits, 1, 0,
6139                                          &cached_state, GFP_NOFS);
6140                 }
6141         } else {
6142                 free_extent_state(cached_state);
6143         }
6144
6145         free_extent_map(em);
6146
6147         return 0;
6148
6149 unlock_err:
6150         if (create)
6151                 unlock_bits |= EXTENT_DO_ACCOUNTING;
6152
6153         clear_extent_bit(&BTRFS_I(inode)->io_tree, lockstart, lockend,
6154                          unlock_bits, 1, 0, &cached_state, GFP_NOFS);
6155         return ret;
6156 }
6157
6158 struct btrfs_dio_private {
6159         struct inode *inode;
6160         u64 logical_offset;
6161         u64 disk_bytenr;
6162         u64 bytes;
6163         void *private;
6164
6165         /* number of bios pending for this dio */
6166         atomic_t pending_bios;
6167
6168         /* IO errors */
6169         int errors;
6170
6171         struct bio *orig_bio;
6172 };
6173
6174 static void btrfs_endio_direct_read(struct bio *bio, int err)
6175 {
6176         struct btrfs_dio_private *dip = bio->bi_private;
6177         struct bio_vec *bvec_end = bio->bi_io_vec + bio->bi_vcnt - 1;
6178         struct bio_vec *bvec = bio->bi_io_vec;
6179         struct inode *inode = dip->inode;
6180         struct btrfs_root *root = BTRFS_I(inode)->root;
6181         u64 start;
6182
6183         start = dip->logical_offset;
6184         do {
6185                 if (!(BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM)) {
6186                         struct page *page = bvec->bv_page;
6187                         char *kaddr;
6188                         u32 csum = ~(u32)0;
6189                         u64 private = ~(u32)0;
6190                         unsigned long flags;
6191
6192                         if (get_state_private(&BTRFS_I(inode)->io_tree,
6193                                               start, &private))
6194                                 goto failed;
6195                         local_irq_save(flags);
6196                         kaddr = kmap_atomic(page);
6197                         csum = btrfs_csum_data(root, kaddr + bvec->bv_offset,
6198                                                csum, bvec->bv_len);
6199                         btrfs_csum_final(csum, (char *)&csum);
6200                         kunmap_atomic(kaddr);
6201                         local_irq_restore(flags);
6202
6203                         flush_dcache_page(bvec->bv_page);
6204                         if (csum != private) {
6205 failed:
6206                                 printk(KERN_ERR "btrfs csum failed ino %llu off"
6207                                       " %llu csum %u private %u\n",
6208                                       (unsigned long long)btrfs_ino(inode),
6209                                       (unsigned long long)start,
6210                                       csum, (unsigned)private);
6211                                 err = -EIO;
6212                         }
6213                 }
6214
6215                 start += bvec->bv_len;
6216                 bvec++;
6217         } while (bvec <= bvec_end);
6218
6219         unlock_extent(&BTRFS_I(inode)->io_tree, dip->logical_offset,
6220                       dip->logical_offset + dip->bytes - 1);
6221         bio->bi_private = dip->private;
6222
6223         kfree(dip);
6224
6225         /* If we had a csum failure make sure to clear the uptodate flag */
6226         if (err)
6227                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
6228         dio_end_io(bio, err);
6229 }
6230
6231 static void btrfs_endio_direct_write(struct bio *bio, int err)
6232 {
6233         struct btrfs_dio_private *dip = bio->bi_private;
6234         struct inode *inode = dip->inode;
6235         struct btrfs_root *root = BTRFS_I(inode)->root;
6236         struct btrfs_ordered_extent *ordered = NULL;
6237         u64 ordered_offset = dip->logical_offset;
6238         u64 ordered_bytes = dip->bytes;
6239         int ret;
6240
6241         if (err)
6242                 goto out_done;
6243 again:
6244         ret = btrfs_dec_test_first_ordered_pending(inode, &ordered,
6245                                                    &ordered_offset,
6246                                                    ordered_bytes, !err);
6247         if (!ret)
6248                 goto out_test;
6249
6250         ordered->work.func = finish_ordered_fn;
6251         ordered->work.flags = 0;
6252         btrfs_queue_worker(&root->fs_info->endio_write_workers,
6253                            &ordered->work);
6254 out_test:
6255         /*
6256          * our bio might span multiple ordered extents.  If we haven't
6257          * completed the accounting for the whole dio, go back and try again
6258          */
6259         if (ordered_offset < dip->logical_offset + dip->bytes) {
6260                 ordered_bytes = dip->logical_offset + dip->bytes -
6261                         ordered_offset;
6262                 ordered = NULL;
6263                 goto again;
6264         }
6265 out_done:
6266         bio->bi_private = dip->private;
6267
6268         kfree(dip);
6269
6270         /* If we had an error make sure to clear the uptodate flag */
6271         if (err)
6272                 clear_bit(BIO_UPTODATE, &bio->bi_flags);
6273         dio_end_io(bio, err);
6274 }
6275
6276 static int __btrfs_submit_bio_start_direct_io(struct inode *inode, int rw,
6277                                     struct bio *bio, int mirror_num,
6278                                     unsigned long bio_flags, u64 offset)
6279 {
6280         int ret;
6281         struct btrfs_root *root = BTRFS_I(inode)->root;
6282         ret = btrfs_csum_one_bio(root, inode, bio, offset, 1);
6283         BUG_ON(ret); /* -ENOMEM */
6284         return 0;
6285 }
6286
6287 static void btrfs_end_dio_bio(struct bio *bio, int err)
6288 {
6289         struct btrfs_dio_private *dip = bio->bi_private;
6290
6291         if (err) {
6292                 printk(KERN_ERR "btrfs direct IO failed ino %llu rw %lu "
6293                       "sector %#Lx len %u err no %d\n",
6294                       (unsigned long long)btrfs_ino(dip->inode), bio->bi_rw,
6295                       (unsigned long long)bio->bi_sector, bio->bi_size, err);
6296                 dip->errors = 1;
6297
6298                 /*
6299                  * before atomic variable goto zero, we must make sure
6300                  * dip->errors is perceived to be set.
6301                  */
6302                 smp_mb__before_atomic_dec();
6303         }
6304
6305         /* if there are more bios still pending for this dio, just exit */
6306         if (!atomic_dec_and_test(&dip->pending_bios))
6307                 goto out;
6308
6309         if (dip->errors)
6310                 bio_io_error(dip->orig_bio);
6311         else {
6312                 set_bit(BIO_UPTODATE, &dip->orig_bio->bi_flags);
6313                 bio_endio(dip->orig_bio, 0);
6314         }
6315 out:
6316         bio_put(bio);
6317 }
6318
6319 static struct bio *btrfs_dio_bio_alloc(struct block_device *bdev,
6320                                        u64 first_sector, gfp_t gfp_flags)
6321 {
6322         int nr_vecs = bio_get_nr_vecs(bdev);
6323         return btrfs_bio_alloc(bdev, first_sector, nr_vecs, gfp_flags);
6324 }
6325
6326 static inline int __btrfs_submit_dio_bio(struct bio *bio, struct inode *inode,
6327                                          int rw, u64 file_offset, int skip_sum,
6328                                          int async_submit)
6329 {
6330         int write = rw & REQ_WRITE;
6331         struct btrfs_root *root = BTRFS_I(inode)->root;
6332         int ret;
6333
6334         bio_get(bio);
6335
6336         if (!write) {
6337                 ret = btrfs_bio_wq_end_io(root->fs_info, bio, 0);
6338                 if (ret)
6339                         goto err;
6340         }
6341
6342         if (skip_sum)
6343                 goto map;
6344
6345         if (write && async_submit) {
6346                 ret = btrfs_wq_submit_bio(root->fs_info,
6347                                    inode, rw, bio, 0, 0,
6348                                    file_offset,
6349                                    __btrfs_submit_bio_start_direct_io,
6350                                    __btrfs_submit_bio_done);
6351                 goto err;
6352         } else if (write) {
6353                 /*
6354                  * If we aren't doing async submit, calculate the csum of the
6355                  * bio now.
6356                  */
6357                 ret = btrfs_csum_one_bio(root, inode, bio, file_offset, 1);
6358                 if (ret)
6359                         goto err;
6360         } else if (!skip_sum) {
6361                 ret = btrfs_lookup_bio_sums_dio(root, inode, bio, file_offset);
6362                 if (ret)
6363                         goto err;
6364         }
6365
6366 map:
6367         ret = btrfs_map_bio(root, rw, bio, 0, async_submit);
6368 err:
6369         bio_put(bio);
6370         return ret;
6371 }
6372
6373 static int btrfs_submit_direct_hook(int rw, struct btrfs_dio_private *dip,
6374                                     int skip_sum)
6375 {
6376         struct inode *inode = dip->inode;
6377         struct btrfs_root *root = BTRFS_I(inode)->root;
6378         struct bio *bio;
6379         struct bio *orig_bio = dip->orig_bio;
6380         struct bio_vec *bvec = orig_bio->bi_io_vec;
6381         u64 start_sector = orig_bio->bi_sector;
6382         u64 file_offset = dip->logical_offset;
6383         u64 submit_len = 0;
6384         u64 map_length;
6385         int nr_pages = 0;
6386         int ret = 0;
6387         int async_submit = 0;
6388
6389         map_length = orig_bio->bi_size;
6390         ret = btrfs_map_block(root->fs_info, READ, start_sector << 9,
6391                               &map_length, NULL, 0);
6392         if (ret) {
6393                 bio_put(orig_bio);
6394                 return -EIO;
6395         }
6396
6397         if (map_length >= orig_bio->bi_size) {
6398                 bio = orig_bio;
6399                 goto submit;
6400         }
6401
6402         async_submit = 1;
6403         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev, start_sector, GFP_NOFS);
6404         if (!bio)
6405                 return -ENOMEM;
6406         bio->bi_private = dip;
6407         bio->bi_end_io = btrfs_end_dio_bio;
6408         atomic_inc(&dip->pending_bios);
6409
6410         while (bvec <= (orig_bio->bi_io_vec + orig_bio->bi_vcnt - 1)) {
6411                 if (unlikely(map_length < submit_len + bvec->bv_len ||
6412                     bio_add_page(bio, bvec->bv_page, bvec->bv_len,
6413                                  bvec->bv_offset) < bvec->bv_len)) {
6414                         /*
6415                          * inc the count before we submit the bio so
6416                          * we know the end IO handler won't happen before
6417                          * we inc the count. Otherwise, the dip might get freed
6418                          * before we're done setting it up
6419                          */
6420                         atomic_inc(&dip->pending_bios);
6421                         ret = __btrfs_submit_dio_bio(bio, inode, rw,
6422                                                      file_offset, skip_sum,
6423                                                      async_submit);
6424                         if (ret) {
6425                                 bio_put(bio);
6426                                 atomic_dec(&dip->pending_bios);
6427                                 goto out_err;
6428                         }
6429
6430                         start_sector += submit_len >> 9;
6431                         file_offset += submit_len;
6432
6433                         submit_len = 0;
6434                         nr_pages = 0;
6435
6436                         bio = btrfs_dio_bio_alloc(orig_bio->bi_bdev,
6437                                                   start_sector, GFP_NOFS);
6438                         if (!bio)
6439                                 goto out_err;
6440                         bio->bi_private = dip;
6441                         bio->bi_end_io = btrfs_end_dio_bio;
6442
6443                         map_length = orig_bio->bi_size;
6444                         ret = btrfs_map_block(root->fs_info, READ,
6445                                               start_sector << 9,
6446                                               &map_length, NULL, 0);
6447                         if (ret) {
6448                                 bio_put(bio);
6449                                 goto out_err;
6450                         }
6451                 } else {
6452                         submit_len += bvec->bv_len;
6453                         nr_pages ++;
6454                         bvec++;
6455                 }
6456         }
6457
6458 submit:
6459         ret = __btrfs_submit_dio_bio(bio, inode, rw, file_offset, skip_sum,
6460                                      async_submit);
6461         if (!ret)
6462                 return 0;
6463
6464         bio_put(bio);
6465 out_err:
6466         dip->errors = 1;
6467         /*
6468          * before atomic variable goto zero, we must
6469          * make sure dip->errors is perceived to be set.
6470          */
6471         smp_mb__before_atomic_dec();
6472         if (atomic_dec_and_test(&dip->pending_bios))
6473                 bio_io_error(dip->orig_bio);
6474
6475         /* bio_end_io() will handle error, so we needn't return it */
6476         return 0;
6477 }
6478
6479 static void btrfs_submit_direct(int rw, struct bio *bio, struct inode *inode,
6480                                 loff_t file_offset)
6481 {
6482         struct btrfs_root *root = BTRFS_I(inode)->root;
6483         struct btrfs_dio_private *dip;
6484         struct bio_vec *bvec = bio->bi_io_vec;
6485         int skip_sum;
6486         int write = rw & REQ_WRITE;
6487         int ret = 0;
6488
6489         skip_sum = BTRFS_I(inode)->flags & BTRFS_INODE_NODATASUM;
6490
6491         dip = kmalloc(sizeof(*dip), GFP_NOFS);
6492         if (!dip) {
6493                 ret = -ENOMEM;
6494                 goto free_ordered;
6495         }
6496
6497         dip->private = bio->bi_private;
6498         dip->inode = inode;
6499         dip->logical_offset = file_offset;
6500
6501         dip->bytes = 0;
6502         do {
6503                 dip->bytes += bvec->bv_len;
6504                 bvec++;
6505         } while (bvec <= (bio->bi_io_vec + bio->bi_vcnt - 1));
6506
6507         dip->disk_bytenr = (u64)bio->bi_sector << 9;
6508         bio->bi_private = dip;
6509         dip->errors = 0;
6510         dip->orig_bio = bio;
6511         atomic_set(&dip->pending_bios, 0);
6512
6513         if (write)
6514                 bio->bi_end_io = btrfs_endio_direct_write;
6515         else
6516                 bio->bi_end_io = btrfs_endio_direct_read;
6517
6518         ret = btrfs_submit_direct_hook(rw, dip, skip_sum);
6519         if (!ret)
6520                 return;
6521 free_ordered:
6522         /*
6523          * If this is a write, we need to clean up the reserved space and kill
6524          * the ordered extent.
6525          */
6526         if (write) {
6527                 struct btrfs_ordered_extent *ordered;
6528                 ordered = btrfs_lookup_ordered_extent(inode, file_offset);
6529                 if (!test_bit(BTRFS_ORDERED_PREALLOC, &ordered->flags) &&
6530                     !test_bit(BTRFS_ORDERED_NOCOW, &ordered->flags))
6531                         btrfs_free_reserved_extent(root, ordered->start,
6532                                                    ordered->disk_len);
6533                 btrfs_put_ordered_extent(ordered);
6534                 btrfs_put_ordered_extent(ordered);
6535         }
6536         bio_endio(bio, ret);
6537 }
6538
6539 static ssize_t check_direct_IO(struct btrfs_root *root, int rw, struct kiocb *iocb,
6540                         const struct iovec *iov, loff_t offset,
6541                         unsigned long nr_segs)
6542 {
6543         int seg;
6544         int i;
6545         size_t size;
6546         unsigned long addr;
6547         unsigned blocksize_mask = root->sectorsize - 1;
6548         ssize_t retval = -EINVAL;
6549         loff_t end = offset;
6550
6551         if (offset & blocksize_mask)
6552                 goto out;
6553
6554         /* Check the memory alignment.  Blocks cannot straddle pages */
6555         for (seg = 0; seg < nr_segs; seg++) {
6556                 addr = (unsigned long)iov[seg].iov_base;
6557                 size = iov[seg].iov_len;
6558                 end += size;
6559                 if ((addr & blocksize_mask) || (size & blocksize_mask))
6560                         goto out;
6561
6562                 /* If this is a write we don't need to check anymore */
6563                 if (rw & WRITE)
6564                         continue;
6565
6566                 /*
6567                  * Check to make sure we don't have duplicate iov_base's in this
6568                  * iovec, if so return EINVAL, otherwise we'll get csum errors
6569                  * when reading back.
6570                  */
6571                 for (i = seg + 1; i < nr_segs; i++) {
6572                         if (iov[seg].iov_base == iov[i].iov_base)
6573                                 goto out;
6574                 }
6575         }
6576         retval = 0;
6577 out:
6578         return retval;
6579 }
6580
6581 static ssize_t btrfs_direct_IO(int rw, struct kiocb *iocb,
6582                         const struct iovec *iov, loff_t offset,
6583                         unsigned long nr_segs)
6584 {
6585         struct file *file = iocb->ki_filp;
6586         struct inode *inode = file->f_mapping->host;
6587
6588         if (check_direct_IO(BTRFS_I(inode)->root, rw, iocb, iov,
6589                             offset, nr_segs))
6590                 return 0;
6591
6592         return __blockdev_direct_IO(rw, iocb, inode,
6593                    BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev,
6594                    iov, offset, nr_segs, btrfs_get_blocks_direct, NULL,
6595                    btrfs_submit_direct, 0);
6596 }
6597
6598 static int btrfs_fiemap(struct inode *inode, struct fiemap_extent_info *fieinfo,
6599                 __u64 start, __u64 len)
6600 {
6601         return extent_fiemap(inode, fieinfo, start, len, btrfs_get_extent_fiemap);
6602 }
6603
6604 int btrfs_readpage(struct file *file, struct page *page)
6605 {
6606         struct extent_io_tree *tree;
6607         tree = &BTRFS_I(page->mapping->host)->io_tree;
6608         return extent_read_full_page(tree, page, btrfs_get_extent, 0);
6609 }
6610
6611 static int btrfs_writepage(struct page *page, struct writeback_control *wbc)
6612 {
6613         struct extent_io_tree *tree;
6614
6615
6616         if (current->flags & PF_MEMALLOC) {
6617                 redirty_page_for_writepage(wbc, page);
6618                 unlock_page(page);
6619                 return 0;
6620         }
6621         tree = &BTRFS_I(page->mapping->host)->io_tree;
6622         return extent_write_full_page(tree, page, btrfs_get_extent, wbc);
6623 }
6624
6625 int btrfs_writepages(struct address_space *mapping,
6626                      struct writeback_control *wbc)
6627 {
6628         struct extent_io_tree *tree;
6629
6630         tree = &BTRFS_I(mapping->host)->io_tree;
6631         return extent_writepages(tree, mapping, btrfs_get_extent, wbc);
6632 }
6633
6634 static int
6635 btrfs_readpages(struct file *file, struct address_space *mapping,
6636                 struct list_head *pages, unsigned nr_pages)
6637 {
6638         struct extent_io_tree *tree;
6639         tree = &BTRFS_I(mapping->host)->io_tree;
6640         return extent_readpages(tree, mapping, pages, nr_pages,
6641                                 btrfs_get_extent);
6642 }
6643 static int __btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6644 {
6645         struct extent_io_tree *tree;
6646         struct extent_map_tree *map;
6647         int ret;
6648
6649         tree = &BTRFS_I(page->mapping->host)->io_tree;
6650         map = &BTRFS_I(page->mapping->host)->extent_tree;
6651         ret = try_release_extent_mapping(map, tree, page, gfp_flags);
6652         if (ret == 1) {
6653                 ClearPagePrivate(page);
6654                 set_page_private(page, 0);
6655                 page_cache_release(page);
6656         }
6657         return ret;
6658 }
6659
6660 static int btrfs_releasepage(struct page *page, gfp_t gfp_flags)
6661 {
6662         if (PageWriteback(page) || PageDirty(page))
6663                 return 0;
6664         return __btrfs_releasepage(page, gfp_flags & GFP_NOFS);
6665 }
6666
6667 static void btrfs_invalidatepage(struct page *page, unsigned long offset)
6668 {
6669         struct inode *inode = page->mapping->host;
6670         struct extent_io_tree *tree;
6671         struct btrfs_ordered_extent *ordered;
6672         struct extent_state *cached_state = NULL;
6673         u64 page_start = page_offset(page);
6674         u64 page_end = page_start + PAGE_CACHE_SIZE - 1;
6675
6676         /*
6677          * we have the page locked, so new writeback can't start,
6678          * and the dirty bit won't be cleared while we are here.
6679          *
6680          * Wait for IO on this page so that we can safely clear
6681          * the PagePrivate2 bit and do ordered accounting
6682          */
6683         wait_on_page_writeback(page);
6684
6685         tree = &BTRFS_I(inode)->io_tree;
6686         if (offset) {
6687                 btrfs_releasepage(page, GFP_NOFS);
6688                 return;
6689         }
6690         lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
6691         ordered = btrfs_lookup_ordered_extent(inode,
6692                                            page_offset(page));
6693         if (ordered) {
6694                 /*
6695                  * IO on this page will never be started, so we need
6696                  * to account for any ordered extents now
6697                  */
6698                 clear_extent_bit(tree, page_start, page_end,
6699                                  EXTENT_DIRTY | EXTENT_DELALLOC |
6700                                  EXTENT_LOCKED | EXTENT_DO_ACCOUNTING |
6701                                  EXTENT_DEFRAG, 1, 0, &cached_state, GFP_NOFS);
6702                 /*
6703                  * whoever cleared the private bit is responsible
6704                  * for the finish_ordered_io
6705                  */
6706                 if (TestClearPagePrivate2(page) &&
6707                     btrfs_dec_test_ordered_pending(inode, &ordered, page_start,
6708                                                    PAGE_CACHE_SIZE, 1)) {
6709                         btrfs_finish_ordered_io(ordered);
6710                 }
6711                 btrfs_put_ordered_extent(ordered);
6712                 cached_state = NULL;
6713                 lock_extent_bits(tree, page_start, page_end, 0, &cached_state);
6714         }
6715         clear_extent_bit(tree, page_start, page_end,
6716                  EXTENT_LOCKED | EXTENT_DIRTY | EXTENT_DELALLOC |
6717                  EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG, 1, 1,
6718                  &cached_state, GFP_NOFS);
6719         __btrfs_releasepage(page, GFP_NOFS);
6720
6721         ClearPageChecked(page);
6722         if (PagePrivate(page)) {
6723                 ClearPagePrivate(page);
6724                 set_page_private(page, 0);
6725                 page_cache_release(page);
6726         }
6727 }
6728
6729 /*
6730  * btrfs_page_mkwrite() is not allowed to change the file size as it gets
6731  * called from a page fault handler when a page is first dirtied. Hence we must
6732  * be careful to check for EOF conditions here. We set the page up correctly
6733  * for a written page which means we get ENOSPC checking when writing into
6734  * holes and correct delalloc and unwritten extent mapping on filesystems that
6735  * support these features.
6736  *
6737  * We are not allowed to take the i_mutex here so we have to play games to
6738  * protect against truncate races as the page could now be beyond EOF.  Because
6739  * vmtruncate() writes the inode size before removing pages, once we have the
6740  * page lock we can determine safely if the page is beyond EOF. If it is not
6741  * beyond EOF, then the page is guaranteed safe against truncation until we
6742  * unlock the page.
6743  */
6744 int btrfs_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
6745 {
6746         struct page *page = vmf->page;
6747         struct inode *inode = fdentry(vma->vm_file)->d_inode;
6748         struct btrfs_root *root = BTRFS_I(inode)->root;
6749         struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
6750         struct btrfs_ordered_extent *ordered;
6751         struct extent_state *cached_state = NULL;
6752         char *kaddr;
6753         unsigned long zero_start;
6754         loff_t size;
6755         int ret;
6756         int reserved = 0;
6757         u64 page_start;
6758         u64 page_end;
6759
6760         sb_start_pagefault(inode->i_sb);
6761         ret  = btrfs_delalloc_reserve_space(inode, PAGE_CACHE_SIZE);
6762         if (!ret) {
6763                 ret = file_update_time(vma->vm_file);
6764                 reserved = 1;
6765         }
6766         if (ret) {
6767                 if (ret == -ENOMEM)
6768                         ret = VM_FAULT_OOM;
6769                 else /* -ENOSPC, -EIO, etc */
6770                         ret = VM_FAULT_SIGBUS;
6771                 if (reserved)
6772                         goto out;
6773                 goto out_noreserve;
6774         }
6775
6776         ret = VM_FAULT_NOPAGE; /* make the VM retry the fault */
6777 again:
6778         lock_page(page);
6779         size = i_size_read(inode);
6780         page_start = page_offset(page);
6781         page_end = page_start + PAGE_CACHE_SIZE - 1;
6782
6783         if ((page->mapping != inode->i_mapping) ||
6784             (page_start >= size)) {
6785                 /* page got truncated out from underneath us */
6786                 goto out_unlock;
6787         }
6788         wait_on_page_writeback(page);
6789
6790         lock_extent_bits(io_tree, page_start, page_end, 0, &cached_state);
6791         set_page_extent_mapped(page);
6792
6793         /*
6794          * we can't set the delalloc bits if there are pending ordered
6795          * extents.  Drop our locks and wait for them to finish
6796          */
6797         ordered = btrfs_lookup_ordered_extent(inode, page_start);
6798         if (ordered) {
6799                 unlock_extent_cached(io_tree, page_start, page_end,
6800                                      &cached_state, GFP_NOFS);
6801                 unlock_page(page);
6802                 btrfs_start_ordered_extent(inode, ordered, 1);
6803                 btrfs_put_ordered_extent(ordered);
6804                 goto again;
6805         }
6806
6807         /*
6808          * XXX - page_mkwrite gets called every time the page is dirtied, even
6809          * if it was already dirty, so for space accounting reasons we need to
6810          * clear any delalloc bits for the range we are fixing to save.  There
6811          * is probably a better way to do this, but for now keep consistent with
6812          * prepare_pages in the normal write path.
6813          */
6814         clear_extent_bit(&BTRFS_I(inode)->io_tree, page_start, page_end,
6815                           EXTENT_DIRTY | EXTENT_DELALLOC |
6816                           EXTENT_DO_ACCOUNTING | EXTENT_DEFRAG,
6817                           0, 0, &cached_state, GFP_NOFS);
6818
6819         ret = btrfs_set_extent_delalloc(inode, page_start, page_end,
6820                                         &cached_state);
6821         if (ret) {
6822                 unlock_extent_cached(io_tree, page_start, page_end,
6823                                      &cached_state, GFP_NOFS);
6824                 ret = VM_FAULT_SIGBUS;
6825                 goto out_unlock;
6826         }
6827         ret = 0;
6828
6829         /* page is wholly or partially inside EOF */
6830         if (page_start + PAGE_CACHE_SIZE > size)
6831                 zero_start = size & ~PAGE_CACHE_MASK;
6832         else
6833                 zero_start = PAGE_CACHE_SIZE;
6834
6835         if (zero_start != PAGE_CACHE_SIZE) {
6836                 kaddr = kmap(page);
6837                 memset(kaddr + zero_start, 0, PAGE_CACHE_SIZE - zero_start);
6838                 flush_dcache_page(page);
6839                 kunmap(page);
6840         }
6841         ClearPageChecked(page);
6842         set_page_dirty(page);
6843         SetPageUptodate(page);
6844
6845         BTRFS_I(inode)->last_trans = root->fs_info->generation;
6846         BTRFS_I(inode)->last_sub_trans = BTRFS_I(inode)->root->log_transid;
6847         BTRFS_I(inode)->last_log_commit = BTRFS_I(inode)->root->last_log_commit;
6848
6849         unlock_extent_cached(io_tree, page_start, page_end, &cached_state, GFP_NOFS);
6850
6851 out_unlock:
6852         if (!ret) {
6853                 sb_end_pagefault(inode->i_sb);
6854                 return VM_FAULT_LOCKED;
6855         }
6856         unlock_page(page);
6857 out:
6858         btrfs_delalloc_release_space(inode, PAGE_CACHE_SIZE);
6859 out_noreserve:
6860         sb_end_pagefault(inode->i_sb);
6861         return ret;
6862 }
6863
6864 static int btrfs_truncate(struct inode *inode)
6865 {
6866         struct btrfs_root *root = BTRFS_I(inode)->root;
6867         struct btrfs_block_rsv *rsv;
6868         int ret;
6869         int err = 0;
6870         struct btrfs_trans_handle *trans;
6871         u64 mask = root->sectorsize - 1;
6872         u64 min_size = btrfs_calc_trunc_metadata_size(root, 1);
6873
6874         ret = btrfs_truncate_page(inode, inode->i_size, 0, 0);
6875         if (ret)
6876                 return ret;
6877
6878         btrfs_wait_ordered_range(inode, inode->i_size & (~mask), (u64)-1);
6879         btrfs_ordered_update_i_size(inode, inode->i_size, NULL);
6880
6881         /*
6882          * Yes ladies and gentelment, this is indeed ugly.  The fact is we have
6883          * 3 things going on here
6884          *
6885          * 1) We need to reserve space for our orphan item and the space to
6886          * delete our orphan item.  Lord knows we don't want to have a dangling
6887          * orphan item because we didn't reserve space to remove it.
6888          *
6889          * 2) We need to reserve space to update our inode.
6890          *
6891          * 3) We need to have something to cache all the space that is going to
6892          * be free'd up by the truncate operation, but also have some slack
6893          * space reserved in case it uses space during the truncate (thank you
6894          * very much snapshotting).
6895          *
6896          * And we need these to all be seperate.  The fact is we can use alot of
6897          * space doing the truncate, and we have no earthly idea how much space
6898          * we will use, so we need the truncate reservation to be seperate so it
6899          * doesn't end up using space reserved for updating the inode or
6900          * removing the orphan item.  We also need to be able to stop the
6901          * transaction and start a new one, which means we need to be able to
6902          * update the inode several times, and we have no idea of knowing how
6903          * many times that will be, so we can't just reserve 1 item for the
6904          * entirety of the opration, so that has to be done seperately as well.
6905          * Then there is the orphan item, which does indeed need to be held on
6906          * to for the whole operation, and we need nobody to touch this reserved
6907          * space except the orphan code.
6908          *
6909          * So that leaves us with
6910          *
6911          * 1) root->orphan_block_rsv - for the orphan deletion.
6912          * 2) rsv - for the truncate reservation, which we will steal from the
6913          * transaction reservation.
6914          * 3) fs_info->trans_block_rsv - this will have 1 items worth left for
6915          * updating the inode.
6916          */
6917         rsv = btrfs_alloc_block_rsv(root, BTRFS_BLOCK_RSV_TEMP);
6918         if (!rsv)
6919                 return -ENOMEM;
6920         rsv->size = min_size;
6921         rsv->failfast = 1;
6922
6923         /*
6924          * 1 for the truncate slack space
6925          * 1 for the orphan item we're going to add
6926          * 1 for the orphan item deletion
6927          * 1 for updating the inode.
6928          */
6929         trans = btrfs_start_transaction(root, 4);
6930         if (IS_ERR(trans)) {
6931                 err = PTR_ERR(trans);
6932                 goto out;
6933         }
6934
6935         /* Migrate the slack space for the truncate to our reserve */
6936         ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv, rsv,
6937                                       min_size);
6938         BUG_ON(ret);
6939
6940         ret = btrfs_orphan_add(trans, inode);
6941         if (ret) {
6942                 btrfs_end_transaction(trans, root);
6943                 goto out;
6944         }
6945
6946         /*
6947          * setattr is responsible for setting the ordered_data_close flag,
6948          * but that is only tested during the last file release.  That
6949          * could happen well after the next commit, leaving a great big
6950          * window where new writes may get lost if someone chooses to write
6951          * to this file after truncating to zero
6952          *
6953          * The inode doesn't have any dirty data here, and so if we commit
6954          * this is a noop.  If someone immediately starts writing to the inode
6955          * it is very likely we'll catch some of their writes in this
6956          * transaction, and the commit will find this file on the ordered
6957          * data list with good things to send down.
6958          *
6959          * This is a best effort solution, there is still a window where
6960          * using truncate to replace the contents of the file will
6961          * end up with a zero length file after a crash.
6962          */
6963         if (inode->i_size == 0 && test_bit(BTRFS_INODE_ORDERED_DATA_CLOSE,
6964                                            &BTRFS_I(inode)->runtime_flags))
6965                 btrfs_add_ordered_operation(trans, root, inode);
6966
6967         /*
6968          * So if we truncate and then write and fsync we normally would just
6969          * write the extents that changed, which is a problem if we need to
6970          * first truncate that entire inode.  So set this flag so we write out
6971          * all of the extents in the inode to the sync log so we're completely
6972          * safe.
6973          */
6974         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &BTRFS_I(inode)->runtime_flags);
6975         trans->block_rsv = rsv;
6976
6977         while (1) {
6978                 ret = btrfs_truncate_inode_items(trans, root, inode,
6979                                                  inode->i_size,
6980                                                  BTRFS_EXTENT_DATA_KEY);
6981                 if (ret != -ENOSPC) {
6982                         err = ret;
6983                         break;
6984                 }
6985
6986                 trans->block_rsv = &root->fs_info->trans_block_rsv;
6987                 ret = btrfs_update_inode(trans, root, inode);
6988                 if (ret) {
6989                         err = ret;
6990                         break;
6991                 }
6992
6993                 btrfs_end_transaction(trans, root);
6994                 btrfs_btree_balance_dirty(root);
6995
6996                 trans = btrfs_start_transaction(root, 2);
6997                 if (IS_ERR(trans)) {
6998                         ret = err = PTR_ERR(trans);
6999                         trans = NULL;
7000                         break;
7001                 }
7002
7003                 ret = btrfs_block_rsv_migrate(&root->fs_info->trans_block_rsv,
7004                                               rsv, min_size);
7005                 BUG_ON(ret);    /* shouldn't happen */
7006                 trans->block_rsv = rsv;
7007         }
7008
7009         if (ret == 0 && inode->i_nlink > 0) {
7010                 trans->block_rsv = root->orphan_block_rsv;
7011                 ret = btrfs_orphan_del(trans, inode);
7012                 if (ret)
7013                         err = ret;
7014         } else if (ret && inode->i_nlink > 0) {
7015                 /*
7016                  * Failed to do the truncate, remove us from the in memory
7017                  * orphan list.
7018                  */
7019                 ret = btrfs_orphan_del(NULL, inode);
7020         }
7021
7022         if (trans) {
7023                 trans->block_rsv = &root->fs_info->trans_block_rsv;
7024                 ret = btrfs_update_inode(trans, root, inode);
7025                 if (ret && !err)
7026                         err = ret;
7027
7028                 ret = btrfs_end_transaction(trans, root);
7029                 btrfs_btree_balance_dirty(root);
7030         }
7031
7032 out:
7033         btrfs_free_block_rsv(root, rsv);
7034
7035         if (ret && !err)
7036                 err = ret;
7037
7038         return err;
7039 }
7040
7041 /*
7042  * create a new subvolume directory/inode (helper for the ioctl).
7043  */
7044 int btrfs_create_subvol_root(struct btrfs_trans_handle *trans,
7045                              struct btrfs_root *new_root, u64 new_dirid)
7046 {
7047         struct inode *inode;
7048         int err;
7049         u64 index = 0;
7050
7051         inode = btrfs_new_inode(trans, new_root, NULL, "..", 2,
7052                                 new_dirid, new_dirid,
7053                                 S_IFDIR | (~current_umask() & S_IRWXUGO),
7054                                 &index);
7055         if (IS_ERR(inode))
7056                 return PTR_ERR(inode);
7057         inode->i_op = &btrfs_dir_inode_operations;
7058         inode->i_fop = &btrfs_dir_file_operations;
7059
7060         set_nlink(inode, 1);
7061         btrfs_i_size_write(inode, 0);
7062
7063         err = btrfs_update_inode(trans, new_root, inode);
7064
7065         iput(inode);
7066         return err;
7067 }
7068
7069 struct inode *btrfs_alloc_inode(struct super_block *sb)
7070 {
7071         struct btrfs_inode *ei;
7072         struct inode *inode;
7073
7074         ei = kmem_cache_alloc(btrfs_inode_cachep, GFP_NOFS);
7075         if (!ei)
7076                 return NULL;
7077
7078         ei->root = NULL;
7079         ei->generation = 0;
7080         ei->last_trans = 0;
7081         ei->last_sub_trans = 0;
7082         ei->logged_trans = 0;
7083         ei->delalloc_bytes = 0;
7084         ei->disk_i_size = 0;
7085         ei->flags = 0;
7086         ei->csum_bytes = 0;
7087         ei->index_cnt = (u64)-1;
7088         ei->last_unlink_trans = 0;
7089         ei->last_log_commit = 0;
7090
7091         spin_lock_init(&ei->lock);
7092         ei->outstanding_extents = 0;
7093         ei->reserved_extents = 0;
7094
7095         ei->runtime_flags = 0;
7096         ei->force_compress = BTRFS_COMPRESS_NONE;
7097
7098         ei->delayed_node = NULL;
7099
7100         inode = &ei->vfs_inode;
7101         extent_map_tree_init(&ei->extent_tree);
7102         extent_io_tree_init(&ei->io_tree, &inode->i_data);
7103         extent_io_tree_init(&ei->io_failure_tree, &inode->i_data);
7104         ei->io_tree.track_uptodate = 1;
7105         ei->io_failure_tree.track_uptodate = 1;
7106         mutex_init(&ei->log_mutex);
7107         mutex_init(&ei->delalloc_mutex);
7108         btrfs_ordered_inode_tree_init(&ei->ordered_tree);
7109         INIT_LIST_HEAD(&ei->delalloc_inodes);
7110         INIT_LIST_HEAD(&ei->ordered_operations);
7111         RB_CLEAR_NODE(&ei->rb_node);
7112
7113         return inode;
7114 }
7115
7116 static void btrfs_i_callback(struct rcu_head *head)
7117 {
7118         struct inode *inode = container_of(head, struct inode, i_rcu);
7119         kmem_cache_free(btrfs_inode_cachep, BTRFS_I(inode));
7120 }
7121
7122 void btrfs_destroy_inode(struct inode *inode)
7123 {
7124         struct btrfs_ordered_extent *ordered;
7125         struct btrfs_root *root = BTRFS_I(inode)->root;
7126
7127         WARN_ON(!hlist_empty(&inode->i_dentry));
7128         WARN_ON(inode->i_data.nrpages);
7129         WARN_ON(BTRFS_I(inode)->outstanding_extents);
7130         WARN_ON(BTRFS_I(inode)->reserved_extents);
7131         WARN_ON(BTRFS_I(inode)->delalloc_bytes);
7132         WARN_ON(BTRFS_I(inode)->csum_bytes);
7133
7134         /*
7135          * This can happen where we create an inode, but somebody else also
7136          * created the same inode and we need to destroy the one we already
7137          * created.
7138          */
7139         if (!root)
7140                 goto free;
7141
7142         /*
7143          * Make sure we're properly removed from the ordered operation
7144          * lists.
7145          */
7146         smp_mb();
7147         if (!list_empty(&BTRFS_I(inode)->ordered_operations)) {
7148                 spin_lock(&root->fs_info->ordered_extent_lock);
7149                 list_del_init(&BTRFS_I(inode)->ordered_operations);
7150                 spin_unlock(&root->fs_info->ordered_extent_lock);
7151         }
7152
7153         if (test_bit(BTRFS_INODE_HAS_ORPHAN_ITEM,
7154                      &BTRFS_I(inode)->runtime_flags)) {
7155                 printk(KERN_INFO "BTRFS: inode %llu still on the orphan list\n",
7156                        (unsigned long long)btrfs_ino(inode));
7157                 atomic_dec(&root->orphan_inodes);
7158         }
7159
7160         while (1) {
7161                 ordered = btrfs_lookup_first_ordered_extent(inode, (u64)-1);
7162                 if (!ordered)
7163                         break;
7164                 else {
7165                         printk(KERN_ERR "btrfs found ordered "
7166                                "extent %llu %llu on inode cleanup\n",
7167                                (unsigned long long)ordered->file_offset,
7168                                (unsigned long long)ordered->len);
7169                         btrfs_remove_ordered_extent(inode, ordered);
7170                         btrfs_put_ordered_extent(ordered);
7171                         btrfs_put_ordered_extent(ordered);
7172                 }
7173         }
7174         inode_tree_del(inode);
7175         btrfs_drop_extent_cache(inode, 0, (u64)-1, 0);
7176 free:
7177         btrfs_remove_delayed_node(inode);
7178         call_rcu(&inode->i_rcu, btrfs_i_callback);
7179 }
7180
7181 int btrfs_drop_inode(struct inode *inode)
7182 {
7183         struct btrfs_root *root = BTRFS_I(inode)->root;
7184
7185         if (btrfs_root_refs(&root->root_item) == 0 &&
7186             !btrfs_is_free_space_inode(inode))
7187                 return 1;
7188         else
7189                 return generic_drop_inode(inode);
7190 }
7191
7192 static void init_once(void *foo)
7193 {
7194         struct btrfs_inode *ei = (struct btrfs_inode *) foo;
7195
7196         inode_init_once(&ei->vfs_inode);
7197 }
7198
7199 void btrfs_destroy_cachep(void)
7200 {
7201         /*
7202          * Make sure all delayed rcu free inodes are flushed before we
7203          * destroy cache.
7204          */
7205         rcu_barrier();
7206         if (btrfs_inode_cachep)
7207                 kmem_cache_destroy(btrfs_inode_cachep);
7208         if (btrfs_trans_handle_cachep)
7209                 kmem_cache_destroy(btrfs_trans_handle_cachep);
7210         if (btrfs_transaction_cachep)
7211                 kmem_cache_destroy(btrfs_transaction_cachep);
7212         if (btrfs_path_cachep)
7213                 kmem_cache_destroy(btrfs_path_cachep);
7214         if (btrfs_free_space_cachep)
7215                 kmem_cache_destroy(btrfs_free_space_cachep);
7216         if (btrfs_delalloc_work_cachep)
7217                 kmem_cache_destroy(btrfs_delalloc_work_cachep);
7218 }
7219
7220 int btrfs_init_cachep(void)
7221 {
7222         btrfs_inode_cachep = kmem_cache_create("btrfs_inode",
7223                         sizeof(struct btrfs_inode), 0,
7224                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, init_once);
7225         if (!btrfs_inode_cachep)
7226                 goto fail;
7227
7228         btrfs_trans_handle_cachep = kmem_cache_create("btrfs_trans_handle",
7229                         sizeof(struct btrfs_trans_handle), 0,
7230                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7231         if (!btrfs_trans_handle_cachep)
7232                 goto fail;
7233
7234         btrfs_transaction_cachep = kmem_cache_create("btrfs_transaction",
7235                         sizeof(struct btrfs_transaction), 0,
7236                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7237         if (!btrfs_transaction_cachep)
7238                 goto fail;
7239
7240         btrfs_path_cachep = kmem_cache_create("btrfs_path",
7241                         sizeof(struct btrfs_path), 0,
7242                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7243         if (!btrfs_path_cachep)
7244                 goto fail;
7245
7246         btrfs_free_space_cachep = kmem_cache_create("btrfs_free_space",
7247                         sizeof(struct btrfs_free_space), 0,
7248                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD, NULL);
7249         if (!btrfs_free_space_cachep)
7250                 goto fail;
7251
7252         btrfs_delalloc_work_cachep = kmem_cache_create("btrfs_delalloc_work",
7253                         sizeof(struct btrfs_delalloc_work), 0,
7254                         SLAB_RECLAIM_ACCOUNT | SLAB_MEM_SPREAD,
7255                         NULL);
7256         if (!btrfs_delalloc_work_cachep)
7257                 goto fail;
7258
7259         return 0;
7260 fail:
7261         btrfs_destroy_cachep();
7262         return -ENOMEM;
7263 }
7264
7265 static int btrfs_getattr(struct vfsmount *mnt,
7266                          struct dentry *dentry, struct kstat *stat)
7267 {
7268         struct inode *inode = dentry->d_inode;
7269         u32 blocksize = inode->i_sb->s_blocksize;
7270
7271         generic_fillattr(inode, stat);
7272         stat->dev = BTRFS_I(inode)->root->anon_dev;
7273         stat->blksize = PAGE_CACHE_SIZE;
7274         stat->blocks = (ALIGN(inode_get_bytes(inode), blocksize) +
7275                 ALIGN(BTRFS_I(inode)->delalloc_bytes, blocksize)) >> 9;
7276         return 0;
7277 }
7278
7279 /*
7280  * If a file is moved, it will inherit the cow and compression flags of the new
7281  * directory.
7282  */
7283 static void fixup_inode_flags(struct inode *dir, struct inode *inode)
7284 {
7285         struct btrfs_inode *b_dir = BTRFS_I(dir);
7286         struct btrfs_inode *b_inode = BTRFS_I(inode);
7287
7288         if (b_dir->flags & BTRFS_INODE_NODATACOW)
7289                 b_inode->flags |= BTRFS_INODE_NODATACOW;
7290         else
7291                 b_inode->flags &= ~BTRFS_INODE_NODATACOW;
7292
7293         if (b_dir->flags & BTRFS_INODE_COMPRESS) {
7294                 b_inode->flags |= BTRFS_INODE_COMPRESS;
7295                 b_inode->flags &= ~BTRFS_INODE_NOCOMPRESS;
7296         } else {
7297                 b_inode->flags &= ~(BTRFS_INODE_COMPRESS |
7298                                     BTRFS_INODE_NOCOMPRESS);
7299         }
7300 }
7301
7302 static int btrfs_rename(struct inode *old_dir, struct dentry *old_dentry,
7303                            struct inode *new_dir, struct dentry *new_dentry)
7304 {
7305         struct btrfs_trans_handle *trans;
7306         struct btrfs_root *root = BTRFS_I(old_dir)->root;
7307         struct btrfs_root *dest = BTRFS_I(new_dir)->root;
7308         struct inode *new_inode = new_dentry->d_inode;
7309         struct inode *old_inode = old_dentry->d_inode;
7310         struct timespec ctime = CURRENT_TIME;
7311         u64 index = 0;
7312         u64 root_objectid;
7313         int ret;
7314         u64 old_ino = btrfs_ino(old_inode);
7315
7316         if (btrfs_ino(new_dir) == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)
7317                 return -EPERM;
7318
7319         /* we only allow rename subvolume link between subvolumes */
7320         if (old_ino != BTRFS_FIRST_FREE_OBJECTID && root != dest)
7321                 return -EXDEV;
7322
7323         if (old_ino == BTRFS_EMPTY_SUBVOL_DIR_OBJECTID ||
7324             (new_inode && btrfs_ino(new_inode) == BTRFS_FIRST_FREE_OBJECTID))
7325                 return -ENOTEMPTY;
7326
7327         if (S_ISDIR(old_inode->i_mode) && new_inode &&
7328             new_inode->i_size > BTRFS_EMPTY_DIR_SIZE)
7329                 return -ENOTEMPTY;
7330         /*
7331          * we're using rename to replace one file with another.
7332          * and the replacement file is large.  Start IO on it now so
7333          * we don't add too much work to the end of the transaction
7334          */
7335         if (new_inode && S_ISREG(old_inode->i_mode) && new_inode->i_size &&
7336             old_inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
7337                 filemap_flush(old_inode->i_mapping);
7338
7339         /* close the racy window with snapshot create/destroy ioctl */
7340         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7341                 down_read(&root->fs_info->subvol_sem);
7342         /*
7343          * We want to reserve the absolute worst case amount of items.  So if
7344          * both inodes are subvols and we need to unlink them then that would
7345          * require 4 item modifications, but if they are both normal inodes it
7346          * would require 5 item modifications, so we'll assume their normal
7347          * inodes.  So 5 * 2 is 10, plus 1 for the new link, so 11 total items
7348          * should cover the worst case number of items we'll modify.
7349          */
7350         trans = btrfs_start_transaction(root, 20);
7351         if (IS_ERR(trans)) {
7352                 ret = PTR_ERR(trans);
7353                 goto out_notrans;
7354         }
7355
7356         if (dest != root)
7357                 btrfs_record_root_in_trans(trans, dest);
7358
7359         ret = btrfs_set_inode_index(new_dir, &index);
7360         if (ret)
7361                 goto out_fail;
7362
7363         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7364                 /* force full log commit if subvolume involved. */
7365                 root->fs_info->last_trans_log_full_commit = trans->transid;
7366         } else {
7367                 ret = btrfs_insert_inode_ref(trans, dest,
7368                                              new_dentry->d_name.name,
7369                                              new_dentry->d_name.len,
7370                                              old_ino,
7371                                              btrfs_ino(new_dir), index);
7372                 if (ret)
7373                         goto out_fail;
7374                 /*
7375                  * this is an ugly little race, but the rename is required
7376                  * to make sure that if we crash, the inode is either at the
7377                  * old name or the new one.  pinning the log transaction lets
7378                  * us make sure we don't allow a log commit to come in after
7379                  * we unlink the name but before we add the new name back in.
7380                  */
7381                 btrfs_pin_log_trans(root);
7382         }
7383         /*
7384          * make sure the inode gets flushed if it is replacing
7385          * something.
7386          */
7387         if (new_inode && new_inode->i_size && S_ISREG(old_inode->i_mode))
7388                 btrfs_add_ordered_operation(trans, root, old_inode);
7389
7390         inode_inc_iversion(old_dir);
7391         inode_inc_iversion(new_dir);
7392         inode_inc_iversion(old_inode);
7393         old_dir->i_ctime = old_dir->i_mtime = ctime;
7394         new_dir->i_ctime = new_dir->i_mtime = ctime;
7395         old_inode->i_ctime = ctime;
7396
7397         if (old_dentry->d_parent != new_dentry->d_parent)
7398                 btrfs_record_unlink_dir(trans, old_dir, old_inode, 1);
7399
7400         if (unlikely(old_ino == BTRFS_FIRST_FREE_OBJECTID)) {
7401                 root_objectid = BTRFS_I(old_inode)->root->root_key.objectid;
7402                 ret = btrfs_unlink_subvol(trans, root, old_dir, root_objectid,
7403                                         old_dentry->d_name.name,
7404                                         old_dentry->d_name.len);
7405         } else {
7406                 ret = __btrfs_unlink_inode(trans, root, old_dir,
7407                                         old_dentry->d_inode,
7408                                         old_dentry->d_name.name,
7409                                         old_dentry->d_name.len);
7410                 if (!ret)
7411                         ret = btrfs_update_inode(trans, root, old_inode);
7412         }
7413         if (ret) {
7414                 btrfs_abort_transaction(trans, root, ret);
7415                 goto out_fail;
7416         }
7417
7418         if (new_inode) {
7419                 inode_inc_iversion(new_inode);
7420                 new_inode->i_ctime = CURRENT_TIME;
7421                 if (unlikely(btrfs_ino(new_inode) ==
7422                              BTRFS_EMPTY_SUBVOL_DIR_OBJECTID)) {
7423                         root_objectid = BTRFS_I(new_inode)->location.objectid;
7424                         ret = btrfs_unlink_subvol(trans, dest, new_dir,
7425                                                 root_objectid,
7426                                                 new_dentry->d_name.name,
7427                                                 new_dentry->d_name.len);
7428                         BUG_ON(new_inode->i_nlink == 0);
7429                 } else {
7430                         ret = btrfs_unlink_inode(trans, dest, new_dir,
7431                                                  new_dentry->d_inode,
7432                                                  new_dentry->d_name.name,
7433                                                  new_dentry->d_name.len);
7434                 }
7435                 if (!ret && new_inode->i_nlink == 0) {
7436                         ret = btrfs_orphan_add(trans, new_dentry->d_inode);
7437                         BUG_ON(ret);
7438                 }
7439                 if (ret) {
7440                         btrfs_abort_transaction(trans, root, ret);
7441                         goto out_fail;
7442                 }
7443         }
7444
7445         fixup_inode_flags(new_dir, old_inode);
7446
7447         ret = btrfs_add_link(trans, new_dir, old_inode,
7448                              new_dentry->d_name.name,
7449                              new_dentry->d_name.len, 0, index);
7450         if (ret) {
7451                 btrfs_abort_transaction(trans, root, ret);
7452                 goto out_fail;
7453         }
7454
7455         if (old_ino != BTRFS_FIRST_FREE_OBJECTID) {
7456                 struct dentry *parent = new_dentry->d_parent;
7457                 btrfs_log_new_name(trans, old_inode, old_dir, parent);
7458                 btrfs_end_log_trans(root);
7459         }
7460 out_fail:
7461         btrfs_end_transaction(trans, root);
7462 out_notrans:
7463         if (old_ino == BTRFS_FIRST_FREE_OBJECTID)
7464                 up_read(&root->fs_info->subvol_sem);
7465
7466         return ret;
7467 }
7468
7469 static void btrfs_run_delalloc_work(struct btrfs_work *work)
7470 {
7471         struct btrfs_delalloc_work *delalloc_work;
7472
7473         delalloc_work = container_of(work, struct btrfs_delalloc_work,
7474                                      work);
7475         if (delalloc_work->wait)
7476                 btrfs_wait_ordered_range(delalloc_work->inode, 0, (u64)-1);
7477         else
7478                 filemap_flush(delalloc_work->inode->i_mapping);
7479
7480         if (delalloc_work->delay_iput)
7481                 btrfs_add_delayed_iput(delalloc_work->inode);
7482         else
7483                 iput(delalloc_work->inode);
7484         complete(&delalloc_work->completion);
7485 }
7486
7487 struct btrfs_delalloc_work *btrfs_alloc_delalloc_work(struct inode *inode,
7488                                                     int wait, int delay_iput)
7489 {
7490         struct btrfs_delalloc_work *work;
7491
7492         work = kmem_cache_zalloc(btrfs_delalloc_work_cachep, GFP_NOFS);
7493         if (!work)
7494                 return NULL;
7495
7496         init_completion(&work->completion);
7497         INIT_LIST_HEAD(&work->list);
7498         work->inode = inode;
7499         work->wait = wait;
7500         work->delay_iput = delay_iput;
7501         work->work.func = btrfs_run_delalloc_work;
7502
7503         return work;
7504 }
7505
7506 void btrfs_wait_and_free_delalloc_work(struct btrfs_delalloc_work *work)
7507 {
7508         wait_for_completion(&work->completion);
7509         kmem_cache_free(btrfs_delalloc_work_cachep, work);
7510 }
7511
7512 /*
7513  * some fairly slow code that needs optimization. This walks the list
7514  * of all the inodes with pending delalloc and forces them to disk.
7515  */
7516 int btrfs_start_delalloc_inodes(struct btrfs_root *root, int delay_iput)
7517 {
7518         struct list_head *head = &root->fs_info->delalloc_inodes;
7519         struct btrfs_inode *binode;
7520         struct inode *inode;
7521         struct btrfs_delalloc_work *work, *next;
7522         struct list_head works;
7523         int ret = 0;
7524
7525         if (root->fs_info->sb->s_flags & MS_RDONLY)
7526                 return -EROFS;
7527
7528         INIT_LIST_HEAD(&works);
7529
7530         spin_lock(&root->fs_info->delalloc_lock);
7531         while (!list_empty(head)) {
7532                 binode = list_entry(head->next, struct btrfs_inode,
7533                                     delalloc_inodes);
7534                 inode = igrab(&binode->vfs_inode);
7535                 if (!inode)
7536                         list_del_init(&binode->delalloc_inodes);
7537                 spin_unlock(&root->fs_info->delalloc_lock);
7538                 if (inode) {
7539                         work = btrfs_alloc_delalloc_work(inode, 0, delay_iput);
7540                         if (!work) {
7541                                 ret = -ENOMEM;
7542                                 goto out;
7543                         }
7544                         list_add_tail(&work->list, &works);
7545                         btrfs_queue_worker(&root->fs_info->flush_workers,
7546                                            &work->work);
7547                 }
7548                 cond_resched();
7549                 spin_lock(&root->fs_info->delalloc_lock);
7550         }
7551         spin_unlock(&root->fs_info->delalloc_lock);
7552
7553         /* the filemap_flush will queue IO into the worker threads, but
7554          * we have to make sure the IO is actually started and that
7555          * ordered extents get created before we return
7556          */
7557         atomic_inc(&root->fs_info->async_submit_draining);
7558         while (atomic_read(&root->fs_info->nr_async_submits) ||
7559               atomic_read(&root->fs_info->async_delalloc_pages)) {
7560                 wait_event(root->fs_info->async_submit_wait,
7561                    (atomic_read(&root->fs_info->nr_async_submits) == 0 &&
7562                     atomic_read(&root->fs_info->async_delalloc_pages) == 0));
7563         }
7564         atomic_dec(&root->fs_info->async_submit_draining);
7565 out:
7566         list_for_each_entry_safe(work, next, &works, list) {
7567                 list_del_init(&work->list);
7568                 btrfs_wait_and_free_delalloc_work(work);
7569         }
7570         return ret;
7571 }
7572
7573 static int btrfs_symlink(struct inode *dir, struct dentry *dentry,
7574                          const char *symname)
7575 {
7576         struct btrfs_trans_handle *trans;
7577         struct btrfs_root *root = BTRFS_I(dir)->root;
7578         struct btrfs_path *path;
7579         struct btrfs_key key;
7580         struct inode *inode = NULL;
7581         int err;
7582         int drop_inode = 0;
7583         u64 objectid;
7584         u64 index = 0 ;
7585         int name_len;
7586         int datasize;
7587         unsigned long ptr;
7588         struct btrfs_file_extent_item *ei;
7589         struct extent_buffer *leaf;
7590
7591         name_len = strlen(symname) + 1;
7592         if (name_len > BTRFS_MAX_INLINE_DATA_SIZE(root))
7593                 return -ENAMETOOLONG;
7594
7595         /*
7596          * 2 items for inode item and ref
7597          * 2 items for dir items
7598          * 1 item for xattr if selinux is on
7599          */
7600         trans = btrfs_start_transaction(root, 5);
7601         if (IS_ERR(trans))
7602                 return PTR_ERR(trans);
7603
7604         err = btrfs_find_free_ino(root, &objectid);
7605         if (err)
7606                 goto out_unlock;
7607
7608         inode = btrfs_new_inode(trans, root, dir, dentry->d_name.name,
7609                                 dentry->d_name.len, btrfs_ino(dir), objectid,
7610                                 S_IFLNK|S_IRWXUGO, &index);
7611         if (IS_ERR(inode)) {
7612                 err = PTR_ERR(inode);
7613                 goto out_unlock;
7614         }
7615
7616         err = btrfs_init_inode_security(trans, inode, dir, &dentry->d_name);
7617         if (err) {
7618                 drop_inode = 1;
7619                 goto out_unlock;
7620         }
7621
7622         /*
7623         * If the active LSM wants to access the inode during
7624         * d_instantiate it needs these. Smack checks to see
7625         * if the filesystem supports xattrs by looking at the
7626         * ops vector.
7627         */
7628         inode->i_fop = &btrfs_file_operations;
7629         inode->i_op = &btrfs_file_inode_operations;
7630
7631         err = btrfs_add_nondir(trans, dir, dentry, inode, 0, index);
7632         if (err)
7633                 drop_inode = 1;
7634         else {
7635                 inode->i_mapping->a_ops = &btrfs_aops;
7636                 inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7637                 BTRFS_I(inode)->io_tree.ops = &btrfs_extent_io_ops;
7638         }
7639         if (drop_inode)
7640                 goto out_unlock;
7641
7642         path = btrfs_alloc_path();
7643         if (!path) {
7644                 err = -ENOMEM;
7645                 drop_inode = 1;
7646                 goto out_unlock;
7647         }
7648         key.objectid = btrfs_ino(inode);
7649         key.offset = 0;
7650         btrfs_set_key_type(&key, BTRFS_EXTENT_DATA_KEY);
7651         datasize = btrfs_file_extent_calc_inline_size(name_len);
7652         err = btrfs_insert_empty_item(trans, root, path, &key,
7653                                       datasize);
7654         if (err) {
7655                 drop_inode = 1;
7656                 btrfs_free_path(path);
7657                 goto out_unlock;
7658         }
7659         leaf = path->nodes[0];
7660         ei = btrfs_item_ptr(leaf, path->slots[0],
7661                             struct btrfs_file_extent_item);
7662         btrfs_set_file_extent_generation(leaf, ei, trans->transid);
7663         btrfs_set_file_extent_type(leaf, ei,
7664                                    BTRFS_FILE_EXTENT_INLINE);
7665         btrfs_set_file_extent_encryption(leaf, ei, 0);
7666         btrfs_set_file_extent_compression(leaf, ei, 0);
7667         btrfs_set_file_extent_other_encoding(leaf, ei, 0);
7668         btrfs_set_file_extent_ram_bytes(leaf, ei, name_len);
7669
7670         ptr = btrfs_file_extent_inline_start(ei);
7671         write_extent_buffer(leaf, symname, ptr, name_len);
7672         btrfs_mark_buffer_dirty(leaf);
7673         btrfs_free_path(path);
7674
7675         inode->i_op = &btrfs_symlink_inode_operations;
7676         inode->i_mapping->a_ops = &btrfs_symlink_aops;
7677         inode->i_mapping->backing_dev_info = &root->fs_info->bdi;
7678         inode_set_bytes(inode, name_len);
7679         btrfs_i_size_write(inode, name_len - 1);
7680         err = btrfs_update_inode(trans, root, inode);
7681         if (err)
7682                 drop_inode = 1;
7683
7684 out_unlock:
7685         if (!err)
7686                 d_instantiate(dentry, inode);
7687         btrfs_end_transaction(trans, root);
7688         if (drop_inode) {
7689                 inode_dec_link_count(inode);
7690                 iput(inode);
7691         }
7692         btrfs_btree_balance_dirty(root);
7693         return err;
7694 }
7695
7696 static int __btrfs_prealloc_file_range(struct inode *inode, int mode,
7697                                        u64 start, u64 num_bytes, u64 min_size,
7698                                        loff_t actual_len, u64 *alloc_hint,
7699                                        struct btrfs_trans_handle *trans)
7700 {
7701         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
7702         struct extent_map *em;
7703         struct btrfs_root *root = BTRFS_I(inode)->root;
7704         struct btrfs_key ins;
7705         u64 cur_offset = start;
7706         u64 i_size;
7707         int ret = 0;
7708         bool own_trans = true;
7709
7710         if (trans)
7711                 own_trans = false;
7712         while (num_bytes > 0) {
7713                 if (own_trans) {
7714                         trans = btrfs_start_transaction(root, 3);
7715                         if (IS_ERR(trans)) {
7716                                 ret = PTR_ERR(trans);
7717                                 break;
7718                         }
7719                 }
7720
7721                 ret = btrfs_reserve_extent(trans, root, num_bytes, min_size,
7722                                            0, *alloc_hint, &ins, 1);
7723                 if (ret) {
7724                         if (own_trans)
7725                                 btrfs_end_transaction(trans, root);
7726                         break;
7727                 }
7728
7729                 ret = insert_reserved_file_extent(trans, inode,
7730                                                   cur_offset, ins.objectid,
7731                                                   ins.offset, ins.offset,
7732                                                   ins.offset, 0, 0, 0,
7733                                                   BTRFS_FILE_EXTENT_PREALLOC);
7734                 if (ret) {
7735                         btrfs_abort_transaction(trans, root, ret);
7736                         if (own_trans)
7737                                 btrfs_end_transaction(trans, root);
7738                         break;
7739                 }
7740                 btrfs_drop_extent_cache(inode, cur_offset,
7741                                         cur_offset + ins.offset -1, 0);
7742
7743                 em = alloc_extent_map();
7744                 if (!em) {
7745                         set_bit(BTRFS_INODE_NEEDS_FULL_SYNC,
7746                                 &BTRFS_I(inode)->runtime_flags);
7747                         goto next;
7748                 }
7749
7750                 em->start = cur_offset;
7751                 em->orig_start = cur_offset;
7752                 em->len = ins.offset;
7753                 em->block_start = ins.objectid;
7754                 em->block_len = ins.offset;
7755                 em->bdev = root->fs_info->fs_devices->latest_bdev;
7756                 set_bit(EXTENT_FLAG_PREALLOC, &em->flags);
7757                 em->generation = trans->transid;
7758
7759                 while (1) {
7760                         write_lock(&em_tree->lock);
7761                         ret = add_extent_mapping(em_tree, em);
7762                         if (!ret)
7763                                 list_move(&em->list,
7764                                           &em_tree->modified_extents);
7765                         write_unlock(&em_tree->lock);
7766                         if (ret != -EEXIST)
7767                                 break;
7768                         btrfs_drop_extent_cache(inode, cur_offset,
7769                                                 cur_offset + ins.offset - 1,
7770                                                 0);
7771                 }
7772                 free_extent_map(em);
7773 next:
7774                 num_bytes -= ins.offset;
7775                 cur_offset += ins.offset;
7776                 *alloc_hint = ins.objectid + ins.offset;
7777
7778                 inode_inc_iversion(inode);
7779                 inode->i_ctime = CURRENT_TIME;
7780                 BTRFS_I(inode)->flags |= BTRFS_INODE_PREALLOC;
7781                 if (!(mode & FALLOC_FL_KEEP_SIZE) &&
7782                     (actual_len > inode->i_size) &&
7783                     (cur_offset > inode->i_size)) {
7784                         if (cur_offset > actual_len)
7785                                 i_size = actual_len;
7786                         else
7787                                 i_size = cur_offset;
7788                         i_size_write(inode, i_size);
7789                         btrfs_ordered_update_i_size(inode, i_size, NULL);
7790                 }
7791
7792                 ret = btrfs_update_inode(trans, root, inode);
7793
7794                 if (ret) {
7795                         btrfs_abort_transaction(trans, root, ret);
7796                         if (own_trans)
7797                                 btrfs_end_transaction(trans, root);
7798                         break;
7799                 }
7800
7801                 if (own_trans)
7802                         btrfs_end_transaction(trans, root);
7803         }
7804         return ret;
7805 }
7806
7807 int btrfs_prealloc_file_range(struct inode *inode, int mode,
7808                               u64 start, u64 num_bytes, u64 min_size,
7809                               loff_t actual_len, u64 *alloc_hint)
7810 {
7811         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7812                                            min_size, actual_len, alloc_hint,
7813                                            NULL);
7814 }
7815
7816 int btrfs_prealloc_file_range_trans(struct inode *inode,
7817                                     struct btrfs_trans_handle *trans, int mode,
7818                                     u64 start, u64 num_bytes, u64 min_size,
7819                                     loff_t actual_len, u64 *alloc_hint)
7820 {
7821         return __btrfs_prealloc_file_range(inode, mode, start, num_bytes,
7822                                            min_size, actual_len, alloc_hint, trans);
7823 }
7824
7825 static int btrfs_set_page_dirty(struct page *page)
7826 {
7827         return __set_page_dirty_nobuffers(page);
7828 }
7829
7830 static int btrfs_permission(struct inode *inode, int mask)
7831 {
7832         struct btrfs_root *root = BTRFS_I(inode)->root;
7833         umode_t mode = inode->i_mode;
7834
7835         if (mask & MAY_WRITE &&
7836             (S_ISREG(mode) || S_ISDIR(mode) || S_ISLNK(mode))) {
7837                 if (btrfs_root_readonly(root))
7838                         return -EROFS;
7839                 if (BTRFS_I(inode)->flags & BTRFS_INODE_READONLY)
7840                         return -EACCES;
7841         }
7842         return generic_permission(inode, mask);
7843 }
7844
7845 static const struct inode_operations btrfs_dir_inode_operations = {
7846         .getattr        = btrfs_getattr,
7847         .lookup         = btrfs_lookup,
7848         .create         = btrfs_create,
7849         .unlink         = btrfs_unlink,
7850         .link           = btrfs_link,
7851         .mkdir          = btrfs_mkdir,
7852         .rmdir          = btrfs_rmdir,
7853         .rename         = btrfs_rename,
7854         .symlink        = btrfs_symlink,
7855         .setattr        = btrfs_setattr,
7856         .mknod          = btrfs_mknod,
7857         .setxattr       = btrfs_setxattr,
7858         .getxattr       = btrfs_getxattr,
7859         .listxattr      = btrfs_listxattr,
7860         .removexattr    = btrfs_removexattr,
7861         .permission     = btrfs_permission,
7862         .get_acl        = btrfs_get_acl,
7863 };
7864 static const struct inode_operations btrfs_dir_ro_inode_operations = {
7865         .lookup         = btrfs_lookup,
7866         .permission     = btrfs_permission,
7867         .get_acl        = btrfs_get_acl,
7868 };
7869
7870 static const struct file_operations btrfs_dir_file_operations = {
7871         .llseek         = generic_file_llseek,
7872         .read           = generic_read_dir,
7873         .readdir        = btrfs_real_readdir,
7874         .unlocked_ioctl = btrfs_ioctl,
7875 #ifdef CONFIG_COMPAT
7876         .compat_ioctl   = btrfs_ioctl,
7877 #endif
7878         .release        = btrfs_release_file,
7879         .fsync          = btrfs_sync_file,
7880 };
7881
7882 static struct extent_io_ops btrfs_extent_io_ops = {
7883         .fill_delalloc = run_delalloc_range,
7884         .submit_bio_hook = btrfs_submit_bio_hook,
7885         .merge_bio_hook = btrfs_merge_bio_hook,
7886         .readpage_end_io_hook = btrfs_readpage_end_io_hook,
7887         .writepage_end_io_hook = btrfs_writepage_end_io_hook,
7888         .writepage_start_hook = btrfs_writepage_start_hook,
7889         .set_bit_hook = btrfs_set_bit_hook,
7890         .clear_bit_hook = btrfs_clear_bit_hook,
7891         .merge_extent_hook = btrfs_merge_extent_hook,
7892         .split_extent_hook = btrfs_split_extent_hook,
7893 };
7894
7895 /*
7896  * btrfs doesn't support the bmap operation because swapfiles
7897  * use bmap to make a mapping of extents in the file.  They assume
7898  * these extents won't change over the life of the file and they
7899  * use the bmap result to do IO directly to the drive.
7900  *
7901  * the btrfs bmap call would return logical addresses that aren't
7902  * suitable for IO and they also will change frequently as COW
7903  * operations happen.  So, swapfile + btrfs == corruption.
7904  *
7905  * For now we're avoiding this by dropping bmap.
7906  */
7907 static const struct address_space_operations btrfs_aops = {
7908         .readpage       = btrfs_readpage,
7909         .writepage      = btrfs_writepage,
7910         .writepages     = btrfs_writepages,
7911         .readpages      = btrfs_readpages,
7912         .direct_IO      = btrfs_direct_IO,
7913         .invalidatepage = btrfs_invalidatepage,
7914         .releasepage    = btrfs_releasepage,
7915         .set_page_dirty = btrfs_set_page_dirty,
7916         .error_remove_page = generic_error_remove_page,
7917 };
7918
7919 static const struct address_space_operations btrfs_symlink_aops = {
7920         .readpage       = btrfs_readpage,
7921         .writepage      = btrfs_writepage,
7922         .invalidatepage = btrfs_invalidatepage,
7923         .releasepage    = btrfs_releasepage,
7924 };
7925
7926 static const struct inode_operations btrfs_file_inode_operations = {
7927         .getattr        = btrfs_getattr,
7928         .setattr        = btrfs_setattr,
7929         .setxattr       = btrfs_setxattr,
7930         .getxattr       = btrfs_getxattr,
7931         .listxattr      = btrfs_listxattr,
7932         .removexattr    = btrfs_removexattr,
7933         .permission     = btrfs_permission,
7934         .fiemap         = btrfs_fiemap,
7935         .get_acl        = btrfs_get_acl,
7936         .update_time    = btrfs_update_time,
7937 };
7938 static const struct inode_operations btrfs_special_inode_operations = {
7939         .getattr        = btrfs_getattr,
7940         .setattr        = btrfs_setattr,
7941         .permission     = btrfs_permission,
7942         .setxattr       = btrfs_setxattr,
7943         .getxattr       = btrfs_getxattr,
7944         .listxattr      = btrfs_listxattr,
7945         .removexattr    = btrfs_removexattr,
7946         .get_acl        = btrfs_get_acl,
7947         .update_time    = btrfs_update_time,
7948 };
7949 static const struct inode_operations btrfs_symlink_inode_operations = {
7950         .readlink       = generic_readlink,
7951         .follow_link    = page_follow_link_light,
7952         .put_link       = page_put_link,
7953         .getattr        = btrfs_getattr,
7954         .setattr        = btrfs_setattr,
7955         .permission     = btrfs_permission,
7956         .setxattr       = btrfs_setxattr,
7957         .getxattr       = btrfs_getxattr,
7958         .listxattr      = btrfs_listxattr,
7959         .removexattr    = btrfs_removexattr,
7960         .get_acl        = btrfs_get_acl,
7961         .update_time    = btrfs_update_time,
7962 };
7963
7964 const struct dentry_operations btrfs_dentry_operations = {
7965         .d_delete       = btrfs_dentry_delete,
7966         .d_release      = btrfs_dentry_release,
7967 };