fs/btrfs/volumes.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 #include <linux/sched.h>
  19 #include <linux/bio.h>
  20 #include <linux/slab.h>
  21 #include <linux/buffer_head.h>
  22 #include <linux/blkdev.h>
  23 #include <linux/random.h>
  24 #include <linux/iocontext.h>
  25 #include <linux/capability.h>
  26 #include <linux/ratelimit.h>
  27 #include <linux/kthread.h>
  28 #include <linux/raid/pq.h>
  29 #include <asm/div64.h>
  30 #include "compat.h"
  31 #include "ctree.h"
  32 #include "extent_map.h"
  33 #include "disk-io.h"
  34 #include "transaction.h"
  35 #include "print-tree.h"
  36 #include "volumes.h"
  37 #include "raid56.h"
  38 #include "async-thread.h"
  39 #include "check-integrity.h"
  40 #include "rcu-string.h"
  41 #include "math.h"
  42 #include "dev-replace.h"
  43
  44 static int init_first_rw_device(struct btrfs_trans_handle *trans,
  45                                 struct btrfs_root *root,
  46                                 struct btrfs_device *device);
  47 static int btrfs_relocate_sys_chunks(struct btrfs_root *root);
  48 static void __btrfs_reset_dev_stats(struct btrfs_device *dev);
  49 static void btrfs_dev_stat_print_on_load(struct btrfs_device *device);
  50
  51 static DEFINE_MUTEX(uuid_mutex);
  52 static LIST_HEAD(fs_uuids);
  53
  54 static void lock_chunks(struct btrfs_root *root)
  55 {
  56         mutex_lock(&root->fs_info->chunk_mutex);
  57 }
  58
  59 static void unlock_chunks(struct btrfs_root *root)
  60 {
  61         mutex_unlock(&root->fs_info->chunk_mutex);
  62 }
  63
  64 static void free_fs_devices(struct btrfs_fs_devices *fs_devices)
  65 {
  66         struct btrfs_device *device;
  67         WARN_ON(fs_devices->opened);
  68         while (!list_empty(&fs_devices->devices)) {
  69                 device = list_entry(fs_devices->devices.next,
  70                                     struct btrfs_device, dev_list);
  71                 list_del(&device->dev_list);
  72                 rcu_string_free(device->name);
  73                 kfree(device);
  74         }
  75         kfree(fs_devices);
  76 }
  77
  78 static void btrfs_kobject_uevent(struct block_device *bdev,
  79                                  enum kobject_action action)
  80 {
  81         int ret;
  82
  83         ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action);
  84         if (ret)
  85                 pr_warn("Sending event '%d' to kobject: '%s' (%p): failed\n",
  86                         action,
  87                         kobject_name(&disk_to_dev(bdev->bd_disk)->kobj),
  88                         &disk_to_dev(bdev->bd_disk)->kobj);
  89 }
  90
  91 void btrfs_cleanup_fs_uuids(void)
  92 {
  93         struct btrfs_fs_devices *fs_devices;
  94
  95         while (!list_empty(&fs_uuids)) {
  96                 fs_devices = list_entry(fs_uuids.next,
  97                                         struct btrfs_fs_devices, list);
  98                 list_del(&fs_devices->list);
  99                 free_fs_devices(fs_devices);
 100         }
 101 }
 102
 103 static noinline struct btrfs_device *__find_device(struct list_head *head,
 104                                                    u64 devid, u8 *uuid)
 105 {
 106         struct btrfs_device *dev;
 107
 108         list_for_each_entry(dev, head, dev_list) {
 109                 if (dev->devid == devid &&
 110                     (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) {
 111                         return dev;
 112                 }
 113         }
 114         return NULL;
 115 }
 116
 117 static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid)
 118 {
 119         struct btrfs_fs_devices *fs_devices;
 120
 121         list_for_each_entry(fs_devices, &fs_uuids, list) {
 122                 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
 123                         return fs_devices;
 124         }
 125         return NULL;
 126 }
 127
 128 static int
 129 btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder,
 130                       int flush, struct block_device **bdev,
 131                       struct buffer_head **bh)
 132 {
 133         int ret;
 134
 135         *bdev = blkdev_get_by_path(device_path, flags, holder);
 136
 137         if (IS_ERR(*bdev)) {
 138                 ret = PTR_ERR(*bdev);
 139                 printk(KERN_INFO "btrfs: open %s failed\n", device_path);
 140                 goto error;
 141         }
 142
 143         if (flush)
 144                 filemap_write_and_wait((*bdev)->bd_inode->i_mapping);
 145         ret = set_blocksize(*bdev, 4096);
 146         if (ret) {
 147                 blkdev_put(*bdev, flags);
 148                 goto error;
 149         }
 150         invalidate_bdev(*bdev);
 151         *bh = btrfs_read_dev_super(*bdev);
 152         if (!*bh) {
 153                 ret = -EINVAL;
 154                 blkdev_put(*bdev, flags);
 155                 goto error;
 156         }
 157
 158         return 0;
 159
 160 error:
 161         *bdev = NULL;
 162         *bh = NULL;
 163         return ret;
 164 }
 165
 166 static void requeue_list(struct btrfs_pending_bios *pending_bios,
 167                         struct bio *head, struct bio *tail)
 168 {
 169
 170         struct bio *old_head;
 171
 172         old_head = pending_bios->head;
 173         pending_bios->head = head;
 174         if (pending_bios->tail)
 175                 tail->bi_next = old_head;
 176         else
 177                 pending_bios->tail = tail;
 178 }
 179
 180 /*
 181  * we try to collect pending bios for a device so we don't get a large
 182  * number of procs sending bios down to the same device.  This greatly
 183  * improves the schedulers ability to collect and merge the bios.
 184  *
 185  * But, it also turns into a long list of bios to process and that is sure
 186  * to eventually make the worker thread block.  The solution here is to
 187  * make some progress and then put this work struct back at the end of
 188  * the list if the block device is congested.  This way, multiple devices
 189  * can make progress from a single worker thread.
 190  */
 191 static noinline void run_scheduled_bios(struct btrfs_device *device)
 192 {
 193         struct bio *pending;
 194         struct backing_dev_info *bdi;
 195         struct btrfs_fs_info *fs_info;
 196         struct btrfs_pending_bios *pending_bios;
 197         struct bio *tail;
 198         struct bio *cur;
 199         int again = 0;
 200         unsigned long num_run;
 201         unsigned long batch_run = 0;
 202         unsigned long limit;
 203         unsigned long last_waited = 0;
 204         int force_reg = 0;
 205         int sync_pending = 0;
 206         struct blk_plug plug;
 207
 208         /*
 209          * this function runs all the bios we've collected for
 210          * a particular device.  We don't want to wander off to
 211          * another device without first sending all of these down.
 212          * So, setup a plug here and finish it off before we return
 213          */
 214         blk_start_plug(&plug);
 215
 216         bdi = blk_get_backing_dev_info(device->bdev);
 217         fs_info = device->dev_root->fs_info;
 218         limit = btrfs_async_submit_limit(fs_info);
 219         limit = limit * 2 / 3;
 220
 221 loop:
 222         spin_lock(&device->io_lock);
 223
 224 loop_lock:
 225         num_run = 0;
 226
 227         /* take all the bios off the list at once and process them
 228          * later on (without the lock held).  But, remember the
 229          * tail and other pointers so the bios can be properly reinserted
 230          * into the list if we hit congestion
 231          */
 232         if (!force_reg && device->pending_sync_bios.head) {
 233                 pending_bios = &device->pending_sync_bios;
 234                 force_reg = 1;
 235         } else {
 236                 pending_bios = &device->pending_bios;
 237                 force_reg = 0;
 238         }
 239
 240         pending = pending_bios->head;
 241         tail = pending_bios->tail;
 242         WARN_ON(pending && !tail);
 243
 244         /*
 245          * if pending was null this time around, no bios need processing
 246          * at all and we can stop.  Otherwise it'll loop back up again
 247          * and do an additional check so no bios are missed.
 248          *
 249          * device->running_pending is used to synchronize with the
 250          * schedule_bio code.
 251          */
 252         if (device->pending_sync_bios.head == NULL &&
 253             device->pending_bios.head == NULL) {
 254                 again = 0;
 255                 device->running_pending = 0;
 256         } else {
 257                 again = 1;
 258                 device->running_pending = 1;
 259         }
 260
 261         pending_bios->head = NULL;
 262         pending_bios->tail = NULL;
 263
 264         spin_unlock(&device->io_lock);
 265
 266         while (pending) {
 267
 268                 rmb();
 269                 /* we want to work on both lists, but do more bios on the
 270                  * sync list than the regular list
 271                  */
 272                 if ((num_run > 32 &&
 273                     pending_bios != &device->pending_sync_bios &&
 274                     device->pending_sync_bios.head) ||
 275                    (num_run > 64 && pending_bios == &device->pending_sync_bios &&
 276                     device->pending_bios.head)) {
 277                         spin_lock(&device->io_lock);
 278                         requeue_list(pending_bios, pending, tail);
 279                         goto loop_lock;
 280                 }
 281
 282                 cur = pending;
 283                 pending = pending->bi_next;
 284                 cur->bi_next = NULL;
 285
 286                 if (atomic_dec_return(&fs_info->nr_async_bios) < limit &&
 287                     waitqueue_active(&fs_info->async_submit_wait))
 288                         wake_up(&fs_info->async_submit_wait);
 289
 290                 BUG_ON(atomic_read(&cur->bi_cnt) == 0);
 291
 292                 /*
 293                  * if we're doing the sync list, record that our
 294                  * plug has some sync requests on it
 295                  *
 296                  * If we're doing the regular list and there are
 297                  * sync requests sitting around, unplug before
 298                  * we add more
 299                  */
 300                 if (pending_bios == &device->pending_sync_bios) {
 301                         sync_pending = 1;
 302                 } else if (sync_pending) {
 303                         blk_finish_plug(&plug);
 304                         blk_start_plug(&plug);
 305                         sync_pending = 0;
 306                 }
 307
 308                 btrfsic_submit_bio(cur->bi_rw, cur);
 309                 num_run++;
 310                 batch_run++;
 311                 if (need_resched())
 312                         cond_resched();
 313
 314                 /*
 315                  * we made progress, there is more work to do and the bdi
 316                  * is now congested.  Back off and let other work structs
 317                  * run instead
 318                  */
 319                 if (pending && bdi_write_congested(bdi) && batch_run > 8 &&
 320                     fs_info->fs_devices->open_devices > 1) {
 321                         struct io_context *ioc;
 322
 323                         ioc = current->io_context;
 324
 325                         /*
 326                          * the main goal here is that we don't want to
 327                          * block if we're going to be able to submit
 328                          * more requests without blocking.
 329                          *
 330                          * This code does two great things, it pokes into
 331                          * the elevator code from a filesystem _and_
 332                          * it makes assumptions about how batching works.
 333                          */
 334                         if (ioc && ioc->nr_batch_requests > 0 &&
 335                             time_before(jiffies, ioc->last_waited + HZ/50UL) &&
 336                             (last_waited == 0 ||
 337                              ioc->last_waited == last_waited)) {
 338                                 /*
 339                                  * we want to go through our batch of
 340                                  * requests and stop.  So, we copy out
 341                                  * the ioc->last_waited time and test
 342                                  * against it before looping
 343                                  */
 344                                 last_waited = ioc->last_waited;
 345                                 if (need_resched())
 346                                         cond_resched();
 347                                 continue;
 348                         }
 349                         spin_lock(&device->io_lock);
 350                         requeue_list(pending_bios, pending, tail);
 351                         device->running_pending = 1;
 352
 353                         spin_unlock(&device->io_lock);
 354                         btrfs_requeue_work(&device->work);
 355                         goto done;
 356                 }
 357                 /* unplug every 64 requests just for good measure */
 358                 if (batch_run % 64 == 0) {
 359                         blk_finish_plug(&plug);
 360                         blk_start_plug(&plug);
 361                         sync_pending = 0;
 362                 }
 363         }
 364
 365         cond_resched();
 366         if (again)
 367                 goto loop;
 368
 369         spin_lock(&device->io_lock);
 370         if (device->pending_bios.head || device->pending_sync_bios.head)
 371                 goto loop_lock;
 372         spin_unlock(&device->io_lock);
 373
 374 done:
 375         blk_finish_plug(&plug);
 376 }
 377
 378 static void pending_bios_fn(struct btrfs_work *work)
 379 {
 380         struct btrfs_device *device;
 381
 382         device = container_of(work, struct btrfs_device, work);
 383         run_scheduled_bios(device);
 384 }
 385
 386 static noinline int device_list_add(const char *path,
 387                            struct btrfs_super_block *disk_super,
 388                            u64 devid, struct btrfs_fs_devices **fs_devices_ret)
 389 {
 390         struct btrfs_device *device;
 391         struct btrfs_fs_devices *fs_devices;
 392         struct rcu_string *name;
 393         u64 found_transid = btrfs_super_generation(disk_super);
 394
 395         fs_devices = find_fsid(disk_super->fsid);
 396         if (!fs_devices) {
 397                 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
 398                 if (!fs_devices)
 399                         return -ENOMEM;
 400                 INIT_LIST_HEAD(&fs_devices->devices);
 401                 INIT_LIST_HEAD(&fs_devices->alloc_list);
 402                 list_add(&fs_devices->list, &fs_uuids);
 403                 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
 404                 fs_devices->latest_devid = devid;
 405                 fs_devices->latest_trans = found_transid;
 406                 mutex_init(&fs_devices->device_list_mutex);
 407                 device = NULL;
 408         } else {
 409                 device = __find_device(&fs_devices->devices, devid,
 410                                        disk_super->dev_item.uuid);
 411         }
 412         if (!device) {
 413                 if (fs_devices->opened)
 414                         return -EBUSY;
 415
 416                 device = kzalloc(sizeof(*device), GFP_NOFS);
 417                 if (!device) {
 418                         /* we can safely leave the fs_devices entry around */
 419                         return -ENOMEM;
 420                 }
 421                 device->devid = devid;
 422                 device->dev_stats_valid = 0;
 423                 device->work.func = pending_bios_fn;
 424                 memcpy(device->uuid, disk_super->dev_item.uuid,
 425                        BTRFS_UUID_SIZE);
 426                 spin_lock_init(&device->io_lock);
 427
 428                 name = rcu_string_strdup(path, GFP_NOFS);
 429                 if (!name) {
 430                         kfree(device);
 431                         return -ENOMEM;
 432                 }
 433                 rcu_assign_pointer(device->name, name);
 434                 INIT_LIST_HEAD(&device->dev_alloc_list);
 435
 436                 /* init readahead state */
 437                 spin_lock_init(&device->reada_lock);
 438                 device->reada_curr_zone = NULL;
 439                 atomic_set(&device->reada_in_flight, 0);
 440                 device->reada_next = 0;
 441                 INIT_RADIX_TREE(&device->reada_zones, GFP_NOFS & ~__GFP_WAIT);
 442                 INIT_RADIX_TREE(&device->reada_extents, GFP_NOFS & ~__GFP_WAIT);
 443
 444                 mutex_lock(&fs_devices->device_list_mutex);
 445                 list_add_rcu(&device->dev_list, &fs_devices->devices);
 446                 mutex_unlock(&fs_devices->device_list_mutex);
 447
 448                 device->fs_devices = fs_devices;
 449                 fs_devices->num_devices++;
 450         } else if (!device->name || strcmp(device->name->str, path)) {
 451                 name = rcu_string_strdup(path, GFP_NOFS);
 452                 if (!name)
 453                         return -ENOMEM;
 454                 rcu_string_free(device->name);
 455                 rcu_assign_pointer(device->name, name);
 456                 if (device->missing) {
 457                         fs_devices->missing_devices--;
 458                         device->missing = 0;
 459                 }
 460         }
 461
 462         if (found_transid > fs_devices->latest_trans) {
 463                 fs_devices->latest_devid = devid;
 464                 fs_devices->latest_trans = found_transid;
 465         }
 466         *fs_devices_ret = fs_devices;
 467         return 0;
 468 }
 469
 470 static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig)
 471 {
 472         struct btrfs_fs_devices *fs_devices;
 473         struct btrfs_device *device;
 474         struct btrfs_device *orig_dev;
 475
 476         fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
 477         if (!fs_devices)
 478                 return ERR_PTR(-ENOMEM);
 479
 480         INIT_LIST_HEAD(&fs_devices->devices);
 481         INIT_LIST_HEAD(&fs_devices->alloc_list);
 482         INIT_LIST_HEAD(&fs_devices->list);
 483         mutex_init(&fs_devices->device_list_mutex);
 484         fs_devices->latest_devid = orig->latest_devid;
 485         fs_devices->latest_trans = orig->latest_trans;
 486         fs_devices->total_devices = orig->total_devices;
 487         memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid));
 488
 489         /* We have held the volume lock, it is safe to get the devices. */
 490         list_for_each_entry(orig_dev, &orig->devices, dev_list) {
 491                 struct rcu_string *name;
 492
 493                 device = kzalloc(sizeof(*device), GFP_NOFS);
 494                 if (!device)
 495                         goto error;
 496
 497                 /*
 498                  * This is ok to do without rcu read locked because we hold the
 499                  * uuid mutex so nothing we touch in here is going to disappear.
 500                  */
 501                 name = rcu_string_strdup(orig_dev->name->str, GFP_NOFS);
 502                 if (!name) {
 503                         kfree(device);
 504                         goto error;
 505                 }
 506                 rcu_assign_pointer(device->name, name);
 507
 508                 device->devid = orig_dev->devid;
 509                 device->work.func = pending_bios_fn;
 510                 memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid));
 511                 spin_lock_init(&device->io_lock);
 512                 INIT_LIST_HEAD(&device->dev_list);
 513                 INIT_LIST_HEAD(&device->dev_alloc_list);
 514
 515                 list_add(&device->dev_list, &fs_devices->devices);
 516                 device->fs_devices = fs_devices;
 517                 fs_devices->num_devices++;
 518         }
 519         return fs_devices;
 520 error:
 521         free_fs_devices(fs_devices);
 522         return ERR_PTR(-ENOMEM);
 523 }
 524
 525 void btrfs_close_extra_devices(struct btrfs_fs_info *fs_info,
 526                                struct btrfs_fs_devices *fs_devices, int step)
 527 {
 528         struct btrfs_device *device, *next;
 529
 530         struct block_device *latest_bdev = NULL;
 531         u64 latest_devid = 0;
 532         u64 latest_transid = 0;
 533
 534         mutex_lock(&uuid_mutex);
 535 again:
 536         /* This is the initialized path, it is safe to release the devices. */
 537         list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) {
 538                 if (device->in_fs_metadata) {
 539                         if (!device->is_tgtdev_for_dev_replace &&
 540                             (!latest_transid ||
 541                              device->generation > latest_transid)) {
 542                                 latest_devid = device->devid;
 543                                 latest_transid = device->generation;
 544                                 latest_bdev = device->bdev;
 545                         }
 546                         continue;
 547                 }
 548
 549                 if (device->devid == BTRFS_DEV_REPLACE_DEVID) {
 550                         /*
 551                          * In the first step, keep the device which has
 552                          * the correct fsid and the devid that is used
 553                          * for the dev_replace procedure.
 554                          * In the second step, the dev_replace state is
 555                          * read from the device tree and it is known
 556                          * whether the procedure is really active or
 557                          * not, which means whether this device is
 558                          * used or whether it should be removed.
 559                          */
 560                         if (step == 0 || device->is_tgtdev_for_dev_replace) {
 561                                 continue;
 562                         }
 563                 }
 564                 if (device->bdev) {
 565                         blkdev_put(device->bdev, device->mode);
 566                         device->bdev = NULL;
 567                         fs_devices->open_devices--;
 568                 }
 569                 if (device->writeable) {
 570                         list_del_init(&device->dev_alloc_list);
 571                         device->writeable = 0;
 572                         if (!device->is_tgtdev_for_dev_replace)
 573                                 fs_devices->rw_devices--;
 574                 }
 575                 list_del_init(&device->dev_list);
 576                 fs_devices->num_devices--;
 577                 rcu_string_free(device->name);
 578                 kfree(device);
 579         }
 580
 581         if (fs_devices->seed) {
 582                 fs_devices = fs_devices->seed;
 583                 goto again;
 584         }
 585
 586         fs_devices->latest_bdev = latest_bdev;
 587         fs_devices->latest_devid = latest_devid;
 588         fs_devices->latest_trans = latest_transid;
 589
 590         mutex_unlock(&uuid_mutex);
 591 }
 592
 593 static void __free_device(struct work_struct *work)
 594 {
 595         struct btrfs_device *device;
 596
 597         device = container_of(work, struct btrfs_device, rcu_work);
 598
 599         if (device->bdev)
 600                 blkdev_put(device->bdev, device->mode);
 601
 602         rcu_string_free(device->name);
 603         kfree(device);
 604 }
 605
 606 static void free_device(struct rcu_head *head)
 607 {
 608         struct btrfs_device *device;
 609
 610         device = container_of(head, struct btrfs_device, rcu);
 611
 612         INIT_WORK(&device->rcu_work, __free_device);
 613         schedule_work(&device->rcu_work);
 614 }
 615
 616 static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 617 {
 618         struct btrfs_device *device;
 619
 620         if (--fs_devices->opened > 0)
 621                 return 0;
 622
 623         mutex_lock(&fs_devices->device_list_mutex);
 624         list_for_each_entry(device, &fs_devices->devices, dev_list) {
 625                 struct btrfs_device *new_device;
 626                 struct rcu_string *name;
 627
 628                 if (device->bdev)
 629                         fs_devices->open_devices--;
 630
 631                 if (device->writeable && !device->is_tgtdev_for_dev_replace) {
 632                         list_del_init(&device->dev_alloc_list);
 633                         fs_devices->rw_devices--;
 634                 }
 635
 636                 if (device->can_discard)
 637                         fs_devices->num_can_discard--;
 638
 639                 new_device = kmalloc(sizeof(*new_device), GFP_NOFS);
 640                 BUG_ON(!new_device); /* -ENOMEM */
 641                 memcpy(new_device, device, sizeof(*new_device));
 642
 643                 /* Safe because we are under uuid_mutex */
 644                 if (device->name) {
 645                         name = rcu_string_strdup(device->name->str, GFP_NOFS);
 646                         BUG_ON(device->name && !name); /* -ENOMEM */
 647                         rcu_assign_pointer(new_device->name, name);
 648                 }
 649                 new_device->bdev = NULL;
 650                 new_device->writeable = 0;
 651                 new_device->in_fs_metadata = 0;
 652                 new_device->can_discard = 0;
 653                 spin_lock_init(&new_device->io_lock);
 654                 list_replace_rcu(&device->dev_list, &new_device->dev_list);
 655
 656                 call_rcu(&device->rcu, free_device);
 657         }
 658         mutex_unlock(&fs_devices->device_list_mutex);
 659
 660         WARN_ON(fs_devices->open_devices);
 661         WARN_ON(fs_devices->rw_devices);
 662         fs_devices->opened = 0;
 663         fs_devices->seeding = 0;
 664
 665         return 0;
 666 }
 667
 668 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
 669 {
 670         struct btrfs_fs_devices *seed_devices = NULL;
 671         int ret;
 672
 673         mutex_lock(&uuid_mutex);
 674         ret = __btrfs_close_devices(fs_devices);
 675         if (!fs_devices->opened) {
 676                 seed_devices = fs_devices->seed;
 677                 fs_devices->seed = NULL;
 678         }
 679         mutex_unlock(&uuid_mutex);
 680
 681         while (seed_devices) {
 682                 fs_devices = seed_devices;
 683                 seed_devices = fs_devices->seed;
 684                 __btrfs_close_devices(fs_devices);
 685                 free_fs_devices(fs_devices);
 686         }
 687         return ret;
 688 }
 689
 690 static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 691                                 fmode_t flags, void *holder)
 692 {
 693         struct request_queue *q;
 694         struct block_device *bdev;
 695         struct list_head *head = &fs_devices->devices;
 696         struct btrfs_device *device;
 697         struct block_device *latest_bdev = NULL;
 698         struct buffer_head *bh;
 699         struct btrfs_super_block *disk_super;
 700         u64 latest_devid = 0;
 701         u64 latest_transid = 0;
 702         u64 devid;
 703         int seeding = 1;
 704         int ret = 0;
 705
 706         flags |= FMODE_EXCL;
 707
 708         list_for_each_entry(device, head, dev_list) {
 709                 if (device->bdev)
 710                         continue;
 711                 if (!device->name)
 712                         continue;
 713
 714                 ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1,
 715                                             &bdev, &bh);
 716                 if (ret)
 717                         continue;
 718
 719                 disk_super = (struct btrfs_super_block *)bh->b_data;
 720                 devid = btrfs_stack_device_id(&disk_super->dev_item);
 721                 if (devid != device->devid)
 722                         goto error_brelse;
 723
 724                 if (memcmp(device->uuid, disk_super->dev_item.uuid,
 725                            BTRFS_UUID_SIZE))
 726                         goto error_brelse;
 727
 728                 device->generation = btrfs_super_generation(disk_super);
 729                 if (!latest_transid || device->generation > latest_transid) {
 730                         latest_devid = devid;
 731                         latest_transid = device->generation;
 732                         latest_bdev = bdev;
 733                 }
 734
 735                 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) {
 736                         device->writeable = 0;
 737                 } else {
 738                         device->writeable = !bdev_read_only(bdev);
 739                         seeding = 0;
 740                 }
 741
 742                 q = bdev_get_queue(bdev);
 743                 if (blk_queue_discard(q)) {
 744                         device->can_discard = 1;
 745                         fs_devices->num_can_discard++;
 746                 }
 747
 748                 device->bdev = bdev;
 749                 device->in_fs_metadata = 0;
 750                 device->mode = flags;
 751
 752                 if (!blk_queue_nonrot(bdev_get_queue(bdev)))
 753                         fs_devices->rotating = 1;
 754
 755                 fs_devices->open_devices++;
 756                 if (device->writeable && !device->is_tgtdev_for_dev_replace) {
 757                         fs_devices->rw_devices++;
 758                         list_add(&device->dev_alloc_list,
 759                                  &fs_devices->alloc_list);
 760                 }
 761                 brelse(bh);
 762                 continue;
 763
 764 error_brelse:
 765                 brelse(bh);
 766                 blkdev_put(bdev, flags);
 767                 continue;
 768         }
 769         if (fs_devices->open_devices == 0) {
 770                 ret = -EINVAL;
 771                 goto out;
 772         }
 773         fs_devices->seeding = seeding;
 774         fs_devices->opened = 1;
 775         fs_devices->latest_bdev = latest_bdev;
 776         fs_devices->latest_devid = latest_devid;
 777         fs_devices->latest_trans = latest_transid;
 778         fs_devices->total_rw_bytes = 0;
 779 out:
 780         return ret;
 781 }
 782
 783 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices,
 784                        fmode_t flags, void *holder)
 785 {
 786         int ret;
 787
 788         mutex_lock(&uuid_mutex);
 789         if (fs_devices->opened) {
 790                 fs_devices->opened++;
 791                 ret = 0;
 792         } else {
 793                 ret = __btrfs_open_devices(fs_devices, flags, holder);
 794         }
 795         mutex_unlock(&uuid_mutex);
 796         return ret;
 797 }
 798
 799 /*
 800  * Look for a btrfs signature on a device. This may be called out of the mount path
 801  * and we are not allowed to call set_blocksize during the scan. The superblock
 802  * is read via pagecache
 803  */
 804 int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder,
 805                           struct btrfs_fs_devices **fs_devices_ret)
 806 {
 807         struct btrfs_super_block *disk_super;
 808         struct block_device *bdev;
 809         struct page *page;
 810         void *p;
 811         int ret = -EINVAL;
 812         u64 devid;
 813         u64 transid;
 814         u64 total_devices;
 815         u64 bytenr;
 816         pgoff_t index;
 817
 818         /*
 819          * we would like to check all the supers, but that would make
 820          * a btrfs mount succeed after a mkfs from a different FS.
 821          * So, we need to add a special mount option to scan for
 822          * later supers, using BTRFS_SUPER_MIRROR_MAX instead
 823          */
 824         bytenr = btrfs_sb_offset(0);
 825         flags |= FMODE_EXCL;
 826         mutex_lock(&uuid_mutex);
 827
 828         bdev = blkdev_get_by_path(path, flags, holder);
 829
 830         if (IS_ERR(bdev)) {
 831                 ret = PTR_ERR(bdev);
 832                 printk(KERN_INFO "btrfs: open %s failed\n", path);
 833                 goto error;
 834         }
 835
 836         /* make sure our super fits in the device */
 837         if (bytenr + PAGE_CACHE_SIZE >= i_size_read(bdev->bd_inode))
 838                 goto error_bdev_put;
 839
 840         /* make sure our super fits in the page */
 841         if (sizeof(*disk_super) > PAGE_CACHE_SIZE)
 842                 goto error_bdev_put;
 843
 844         /* make sure our super doesn't straddle pages on disk */
 845         index = bytenr >> PAGE_CACHE_SHIFT;
 846         if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_CACHE_SHIFT != index)
 847                 goto error_bdev_put;
 848
 849         /* pull in the page with our super */
 850         page = read_cache_page_gfp(bdev->bd_inode->i_mapping,
 851                                    index, GFP_NOFS);
 852
 853         if (IS_ERR_OR_NULL(page))
 854                 goto error_bdev_put;
 855
 856         p = kmap(page);
 857
 858         /* align our pointer to the offset of the super block */
 859         disk_super = p + (bytenr & ~PAGE_CACHE_MASK);
 860
 861         if (btrfs_super_bytenr(disk_super) != bytenr ||
 862             disk_super->magic != cpu_to_le64(BTRFS_MAGIC))
 863                 goto error_unmap;
 864
 865         devid = btrfs_stack_device_id(&disk_super->dev_item);
 866         transid = btrfs_super_generation(disk_super);
 867         total_devices = btrfs_super_num_devices(disk_super);
 868
 869         if (disk_super->label[0]) {
 870                 if (disk_super->label[BTRFS_LABEL_SIZE - 1])
 871                         disk_super->label[BTRFS_LABEL_SIZE - 1] = '\0';
 872                 printk(KERN_INFO "device label %s ", disk_super->label);
 873         } else {
 874                 printk(KERN_INFO "device fsid %pU ", disk_super->fsid);
 875         }
 876
 877         printk(KERN_CONT "devid %llu transid %llu %s\n",
 878                (unsigned long long)devid, (unsigned long long)transid, path);
 879
 880         ret = device_list_add(path, disk_super, devid, fs_devices_ret);
 881         if (!ret && fs_devices_ret)
 882                 (*fs_devices_ret)->total_devices = total_devices;
 883
 884 error_unmap:
 885         kunmap(page);
 886         page_cache_release(page);
 887
 888 error_bdev_put:
 889         blkdev_put(bdev, flags);
 890 error:
 891         mutex_unlock(&uuid_mutex);
 892         return ret;
 893 }
 894
 895 /* helper to account the used device space in the range */
 896 int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start,
 897                                    u64 end, u64 *length)
 898 {
 899         struct btrfs_key key;
 900         struct btrfs_root *root = device->dev_root;
 901         struct btrfs_dev_extent *dev_extent;
 902         struct btrfs_path *path;
 903         u64 extent_end;
 904         int ret;
 905         int slot;
 906         struct extent_buffer *l;
 907
 908         *length = 0;
 909
 910         if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace)
 911                 return 0;
 912
 913         path = btrfs_alloc_path();
 914         if (!path)
 915                 return -ENOMEM;
 916         path->reada = 2;
 917
 918         key.objectid = device->devid;
 919         key.offset = start;
 920         key.type = BTRFS_DEV_EXTENT_KEY;
 921
 922         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
 923         if (ret < 0)
 924                 goto out;
 925         if (ret > 0) {
 926                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
 927                 if (ret < 0)
 928                         goto out;
 929         }
 930
 931         while (1) {
 932                 l = path->nodes[0];
 933                 slot = path->slots[0];
 934                 if (slot >= btrfs_header_nritems(l)) {
 935                         ret = btrfs_next_leaf(root, path);
 936                         if (ret == 0)
 937                                 continue;
 938                         if (ret < 0)
 939                                 goto out;
 940
 941                         break;
 942                 }
 943                 btrfs_item_key_to_cpu(l, &key, slot);
 944
 945                 if (key.objectid < device->devid)
 946                         goto next;
 947
 948                 if (key.objectid > device->devid)
 949                         break;
 950
 951                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
 952                         goto next;
 953
 954                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
 955                 extent_end = key.offset + btrfs_dev_extent_length(l,
 956                                                                   dev_extent);
 957                 if (key.offset <= start && extent_end > end) {
 958                         *length = end - start + 1;
 959                         break;
 960                 } else if (key.offset <= start && extent_end > start)
 961                         *length += extent_end - start;
 962                 else if (key.offset > start && extent_end <= end)
 963                         *length += extent_end - key.offset;
 964                 else if (key.offset > start && key.offset <= end) {
 965                         *length += end - key.offset + 1;
 966                         break;
 967                 } else if (key.offset > end)
 968                         break;
 969
 970 next:
 971                 path->slots[0]++;
 972         }
 973         ret = 0;
 974 out:
 975         btrfs_free_path(path);
 976         return ret;
 977 }
 978
 979 /*
 980  * find_free_dev_extent - find free space in the specified device
 981  * @device:     the device which we search the free space in
 982  * @num_bytes:  the size of the free space that we need
 983  * @start:      store the start of the free space.
 984  * @len:        the size of the free space. that we find, or the size of the max
 985  *              free space if we don't find suitable free space
 986  *
 987  * this uses a pretty simple search, the expectation is that it is
 988  * called very infrequently and that a given device has a small number
 989  * of extents
 990  *
 991  * @start is used to store the start of the free space if we find. But if we
 992  * don't find suitable free space, it will be used to store the start position
 993  * of the max free space.
 994  *
 995  * @len is used to store the size of the free space that we find.
 996  * But if we don't find suitable free space, it is used to store the size of
 997  * the max free space.
 998  */
 999 int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes,
1000                          u64 *start, u64 *len)
1001 {
1002         struct btrfs_key key;
1003         struct btrfs_root *root = device->dev_root;
1004         struct btrfs_dev_extent *dev_extent;
1005         struct btrfs_path *path;
1006         u64 hole_size;
1007         u64 max_hole_start;
1008         u64 max_hole_size;
1009         u64 extent_end;
1010         u64 search_start;
1011         u64 search_end = device->total_bytes;
1012         int ret;
1013         int slot;
1014         struct extent_buffer *l;
1015
1016         /* FIXME use last free of some kind */
1017
1018         /* we don't want to overwrite the superblock on the drive,
1019          * so we make sure to start at an offset of at least 1MB
1020          */
1021         search_start = max(root->fs_info->alloc_start, 1024ull * 1024);
1022
1023         max_hole_start = search_start;
1024         max_hole_size = 0;
1025         hole_size = 0;
1026
1027         if (search_start >= search_end || device->is_tgtdev_for_dev_replace) {
1028                 ret = -ENOSPC;
1029                 goto error;
1030         }
1031
1032         path = btrfs_alloc_path();
1033         if (!path) {
1034                 ret = -ENOMEM;
1035                 goto error;
1036         }
1037         path->reada = 2;
1038
1039         key.objectid = device->devid;
1040         key.offset = search_start;
1041         key.type = BTRFS_DEV_EXTENT_KEY;
1042
1043         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1044         if (ret < 0)
1045                 goto out;
1046         if (ret > 0) {
1047                 ret = btrfs_previous_item(root, path, key.objectid, key.type);
1048                 if (ret < 0)
1049                         goto out;
1050         }
1051
1052         while (1) {
1053                 l = path->nodes[0];
1054                 slot = path->slots[0];
1055                 if (slot >= btrfs_header_nritems(l)) {
1056                         ret = btrfs_next_leaf(root, path);
1057                         if (ret == 0)
1058                                 continue;
1059                         if (ret < 0)
1060                                 goto out;
1061
1062                         break;
1063                 }
1064                 btrfs_item_key_to_cpu(l, &key, slot);
1065
1066                 if (key.objectid < device->devid)
1067                         goto next;
1068
1069                 if (key.objectid > device->devid)
1070                         break;
1071
1072                 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
1073                         goto next;
1074
1075                 if (key.offset > search_start) {
1076                         hole_size = key.offset - search_start;
1077
1078                         if (hole_size > max_hole_size) {
1079                                 max_hole_start = search_start;
1080                                 max_hole_size = hole_size;
1081                         }
1082
1083                         /*
1084                          * If this free space is greater than which we need,
1085                          * it must be the max free space that we have found
1086                          * until now, so max_hole_start must point to the start
1087                          * of this free space and the length of this free space
1088                          * is stored in max_hole_size. Thus, we return
1089                          * max_hole_start and max_hole_size and go back to the
1090                          * caller.
1091                          */
1092                         if (hole_size >= num_bytes) {
1093                                 ret = 0;
1094                                 goto out;
1095                         }
1096                 }
1097
1098                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
1099                 extent_end = key.offset + btrfs_dev_extent_length(l,
1100                                                                   dev_extent);
1101                 if (extent_end > search_start)
1102                         search_start = extent_end;
1103 next:
1104                 path->slots[0]++;
1105                 cond_resched();
1106         }
1107
1108         /*
1109          * At this point, search_start should be the end of
1110          * allocated dev extents, and when shrinking the device,
1111          * search_end may be smaller than search_start.
1112          */
1113         if (search_end > search_start)
1114                 hole_size = search_end - search_start;
1115
1116         if (hole_size > max_hole_size) {
1117                 max_hole_start = search_start;
1118                 max_hole_size = hole_size;
1119         }
1120
1121         /* See above. */
1122         if (hole_size < num_bytes)
1123                 ret = -ENOSPC;
1124         else
1125                 ret = 0;
1126
1127 out:
1128         btrfs_free_path(path);
1129 error:
1130         *start = max_hole_start;
1131         if (len)
1132                 *len = max_hole_size;
1133         return ret;
1134 }
1135
1136 static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans,
1137                           struct btrfs_device *device,
1138                           u64 start)
1139 {
1140         int ret;
1141         struct btrfs_path *path;
1142         struct btrfs_root *root = device->dev_root;
1143         struct btrfs_key key;
1144         struct btrfs_key found_key;
1145         struct extent_buffer *leaf = NULL;
1146         struct btrfs_dev_extent *extent = NULL;
1147
1148         path = btrfs_alloc_path();
1149         if (!path)
1150                 return -ENOMEM;
1151
1152         key.objectid = device->devid;
1153         key.offset = start;
1154         key.type = BTRFS_DEV_EXTENT_KEY;
1155 again:
1156         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1157         if (ret > 0) {
1158                 ret = btrfs_previous_item(root, path, key.objectid,
1159                                           BTRFS_DEV_EXTENT_KEY);
1160                 if (ret)
1161                         goto out;
1162                 leaf = path->nodes[0];
1163                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
1164                 extent = btrfs_item_ptr(leaf, path->slots[0],
1165                                         struct btrfs_dev_extent);
1166                 BUG_ON(found_key.offset > start || found_key.offset +
1167                        btrfs_dev_extent_length(leaf, extent) < start);
1168                 key = found_key;
1169                 btrfs_release_path(path);
1170                 goto again;
1171         } else if (ret == 0) {
1172                 leaf = path->nodes[0];
1173                 extent = btrfs_item_ptr(leaf, path->slots[0],
1174                                         struct btrfs_dev_extent);
1175         } else {
1176                 btrfs_error(root->fs_info, ret, "Slot search failed");
1177                 goto out;
1178         }
1179
1180         if (device->bytes_used > 0) {
1181                 u64 len = btrfs_dev_extent_length(leaf, extent);
1182                 device->bytes_used -= len;
1183                 spin_lock(&root->fs_info->free_chunk_lock);
1184                 root->fs_info->free_chunk_space += len;
1185                 spin_unlock(&root->fs_info->free_chunk_lock);
1186         }
1187         ret = btrfs_del_item(trans, root, path);
1188         if (ret) {
1189                 btrfs_error(root->fs_info, ret,
1190                             "Failed to remove dev extent item");
1191         }
1192 out:
1193         btrfs_free_path(path);
1194         return ret;
1195 }
1196
1197 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
1198                            struct btrfs_device *device,
1199                            u64 chunk_tree, u64 chunk_objectid,
1200                            u64 chunk_offset, u64 start, u64 num_bytes)
1201 {
1202         int ret;
1203         struct btrfs_path *path;
1204         struct btrfs_root *root = device->dev_root;
1205         struct btrfs_dev_extent *extent;
1206         struct extent_buffer *leaf;
1207         struct btrfs_key key;
1208
1209         WARN_ON(!device->in_fs_metadata);
1210         WARN_ON(device->is_tgtdev_for_dev_replace);
1211         path = btrfs_alloc_path();
1212         if (!path)
1213                 return -ENOMEM;
1214
1215         key.objectid = device->devid;
1216         key.offset = start;
1217         key.type = BTRFS_DEV_EXTENT_KEY;
1218         ret = btrfs_insert_empty_item(trans, root, path, &key,
1219                                       sizeof(*extent));
1220         if (ret)
1221                 goto out;
1222
1223         leaf = path->nodes[0];
1224         extent = btrfs_item_ptr(leaf, path->slots[0],
1225                                 struct btrfs_dev_extent);
1226         btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
1227         btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
1228         btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
1229
1230         write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
1231                     (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
1232                     BTRFS_UUID_SIZE);
1233
1234         btrfs_set_dev_extent_length(leaf, extent, num_bytes);
1235         btrfs_mark_buffer_dirty(leaf);
1236 out:
1237         btrfs_free_path(path);
1238         return ret;
1239 }
1240
1241 static noinline int find_next_chunk(struct btrfs_root *root,
1242                                     u64 objectid, u64 *offset)
1243 {
1244         struct btrfs_path *path;
1245         int ret;
1246         struct btrfs_key key;
1247         struct btrfs_chunk *chunk;
1248         struct btrfs_key found_key;
1249
1250         path = btrfs_alloc_path();
1251         if (!path)
1252                 return -ENOMEM;
1253
1254         key.objectid = objectid;
1255         key.offset = (u64)-1;
1256         key.type = BTRFS_CHUNK_ITEM_KEY;
1257
1258         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1259         if (ret < 0)
1260                 goto error;
1261
1262         BUG_ON(ret == 0); /* Corruption */
1263
1264         ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
1265         if (ret) {
1266                 *offset = 0;
1267         } else {
1268                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1269                                       path->slots[0]);
1270                 if (found_key.objectid != objectid)
1271                         *offset = 0;
1272                 else {
1273                         chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
1274                                                struct btrfs_chunk);
1275                         *offset = found_key.offset +
1276                                 btrfs_chunk_length(path->nodes[0], chunk);
1277                 }
1278         }
1279         ret = 0;
1280 error:
1281         btrfs_free_path(path);
1282         return ret;
1283 }
1284
1285 static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid)
1286 {
1287         int ret;
1288         struct btrfs_key key;
1289         struct btrfs_key found_key;
1290         struct btrfs_path *path;
1291
1292         root = root->fs_info->chunk_root;
1293
1294         path = btrfs_alloc_path();
1295         if (!path)
1296                 return -ENOMEM;
1297
1298         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1299         key.type = BTRFS_DEV_ITEM_KEY;
1300         key.offset = (u64)-1;
1301
1302         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1303         if (ret < 0)
1304                 goto error;
1305
1306         BUG_ON(ret == 0); /* Corruption */
1307
1308         ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
1309                                   BTRFS_DEV_ITEM_KEY);
1310         if (ret) {
1311                 *objectid = 1;
1312         } else {
1313                 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
1314                                       path->slots[0]);
1315                 *objectid = found_key.offset + 1;
1316         }
1317         ret = 0;
1318 error:
1319         btrfs_free_path(path);
1320         return ret;
1321 }
1322
1323 /*
1324  * the device information is stored in the chunk root
1325  * the btrfs_device struct should be fully filled in
1326  */
1327 int btrfs_add_device(struct btrfs_trans_handle *trans,
1328                      struct btrfs_root *root,
1329                      struct btrfs_device *device)
1330 {
1331         int ret;
1332         struct btrfs_path *path;
1333         struct btrfs_dev_item *dev_item;
1334         struct extent_buffer *leaf;
1335         struct btrfs_key key;
1336         unsigned long ptr;
1337
1338         root = root->fs_info->chunk_root;
1339
1340         path = btrfs_alloc_path();
1341         if (!path)
1342                 return -ENOMEM;
1343
1344         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1345         key.type = BTRFS_DEV_ITEM_KEY;
1346         key.offset = device->devid;
1347
1348         ret = btrfs_insert_empty_item(trans, root, path, &key,
1349                                       sizeof(*dev_item));
1350         if (ret)
1351                 goto out;
1352
1353         leaf = path->nodes[0];
1354         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
1355
1356         btrfs_set_device_id(leaf, dev_item, device->devid);
1357         btrfs_set_device_generation(leaf, dev_item, 0);
1358         btrfs_set_device_type(leaf, dev_item, device->type);
1359         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
1360         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
1361         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
1362         btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
1363         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
1364         btrfs_set_device_group(leaf, dev_item, 0);
1365         btrfs_set_device_seek_speed(leaf, dev_item, 0);
1366         btrfs_set_device_bandwidth(leaf, dev_item, 0);
1367         btrfs_set_device_start_offset(leaf, dev_item, 0);
1368
1369         ptr = (unsigned long)btrfs_device_uuid(dev_item);
1370         write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1371         ptr = (unsigned long)btrfs_device_fsid(dev_item);
1372         write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
1373         btrfs_mark_buffer_dirty(leaf);
1374
1375         ret = 0;
1376 out:
1377         btrfs_free_path(path);
1378         return ret;
1379 }
1380
1381 static int btrfs_rm_dev_item(struct btrfs_root *root,
1382                              struct btrfs_device *device)
1383 {
1384         int ret;
1385         struct btrfs_path *path;
1386         struct btrfs_key key;
1387         struct btrfs_trans_handle *trans;
1388
1389         root = root->fs_info->chunk_root;
1390
1391         path = btrfs_alloc_path();
1392         if (!path)
1393                 return -ENOMEM;
1394
1395         trans = btrfs_start_transaction(root, 0);
1396         if (IS_ERR(trans)) {
1397                 btrfs_free_path(path);
1398                 return PTR_ERR(trans);
1399         }
1400         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1401         key.type = BTRFS_DEV_ITEM_KEY;
1402         key.offset = device->devid;
1403         lock_chunks(root);
1404
1405         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
1406         if (ret < 0)
1407                 goto out;
1408
1409         if (ret > 0) {
1410                 ret = -ENOENT;
1411                 goto out;
1412         }
1413
1414         ret = btrfs_del_item(trans, root, path);
1415         if (ret)
1416                 goto out;
1417 out:
1418         btrfs_free_path(path);
1419         unlock_chunks(root);
1420         btrfs_commit_transaction(trans, root);
1421         return ret;
1422 }
1423
1424 int btrfs_rm_device(struct btrfs_root *root, char *device_path)
1425 {
1426         struct btrfs_device *device;
1427         struct btrfs_device *next_device;
1428         struct block_device *bdev;
1429         struct buffer_head *bh = NULL;
1430         struct btrfs_super_block *disk_super;
1431         struct btrfs_fs_devices *cur_devices;
1432         u64 all_avail;
1433         u64 devid;
1434         u64 num_devices;
1435         u8 *dev_uuid;
1436         unsigned seq;
1437         int ret = 0;
1438         bool clear_super = false;
1439
1440         mutex_lock(&uuid_mutex);
1441
1442         do {
1443                 seq = read_seqbegin(&root->fs_info->profiles_lock);
1444
1445                 all_avail = root->fs_info->avail_data_alloc_bits |
1446                             root->fs_info->avail_system_alloc_bits |
1447                             root->fs_info->avail_metadata_alloc_bits;
1448         } while (read_seqretry(&root->fs_info->profiles_lock, seq));
1449
1450         num_devices = root->fs_info->fs_devices->num_devices;
1451         btrfs_dev_replace_lock(&root->fs_info->dev_replace);
1452         if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) {
1453                 WARN_ON(num_devices < 1);
1454                 num_devices--;
1455         }
1456         btrfs_dev_replace_unlock(&root->fs_info->dev_replace);
1457
1458         if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) && num_devices <= 4) {
1459                 printk(KERN_ERR "btrfs: unable to go below four devices "
1460                        "on raid10\n");
1461                 ret = -EINVAL;
1462                 goto out;
1463         }
1464
1465         if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) && num_devices <= 2) {
1466                 printk(KERN_ERR "btrfs: unable to go below two "
1467                        "devices on raid1\n");
1468                 ret = -EINVAL;
1469                 goto out;
1470         }
1471
1472         if ((all_avail & BTRFS_BLOCK_GROUP_RAID5) &&
1473             root->fs_info->fs_devices->rw_devices <= 2) {
1474                 printk(KERN_ERR "btrfs: unable to go below two "
1475                        "devices on raid5\n");
1476                 ret = -EINVAL;
1477                 goto out;
1478         }
1479         if ((all_avail & BTRFS_BLOCK_GROUP_RAID6) &&
1480             root->fs_info->fs_devices->rw_devices <= 3) {
1481                 printk(KERN_ERR "btrfs: unable to go below three "
1482                        "devices on raid6\n");
1483                 ret = -EINVAL;
1484                 goto out;
1485         }
1486
1487         if (strcmp(device_path, "missing") == 0) {
1488                 struct list_head *devices;
1489                 struct btrfs_device *tmp;
1490
1491                 device = NULL;
1492                 devices = &root->fs_info->fs_devices->devices;
1493                 /*
1494                  * It is safe to read the devices since the volume_mutex
1495                  * is held.
1496                  */
1497                 list_for_each_entry(tmp, devices, dev_list) {
1498                         if (tmp->in_fs_metadata &&
1499                             !tmp->is_tgtdev_for_dev_replace &&
1500                             !tmp->bdev) {
1501                                 device = tmp;
1502                                 break;
1503                         }
1504                 }
1505                 bdev = NULL;
1506                 bh = NULL;
1507                 disk_super = NULL;
1508                 if (!device) {
1509                         printk(KERN_ERR "btrfs: no missing devices found to "
1510                                "remove\n");
1511                         goto out;
1512                 }
1513         } else {
1514                 ret = btrfs_get_bdev_and_sb(device_path,
1515                                             FMODE_WRITE | FMODE_EXCL,
1516                                             root->fs_info->bdev_holder, 0,
1517                                             &bdev, &bh);
1518                 if (ret)
1519                         goto out;
1520                 disk_super = (struct btrfs_super_block *)bh->b_data;
1521                 devid = btrfs_stack_device_id(&disk_super->dev_item);
1522                 dev_uuid = disk_super->dev_item.uuid;
1523                 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1524                                            disk_super->fsid);
1525                 if (!device) {
1526                         ret = -ENOENT;
1527                         goto error_brelse;
1528                 }
1529         }
1530
1531         if (device->is_tgtdev_for_dev_replace) {
1532                 pr_err("btrfs: unable to remove the dev_replace target dev\n");
1533                 ret = -EINVAL;
1534                 goto error_brelse;
1535         }
1536
1537         if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) {
1538                 printk(KERN_ERR "btrfs: unable to remove the only writeable "
1539                        "device\n");
1540                 ret = -EINVAL;
1541                 goto error_brelse;
1542         }
1543
1544         if (device->writeable) {
1545                 lock_chunks(root);
1546                 list_del_init(&device->dev_alloc_list);
1547                 unlock_chunks(root);
1548                 root->fs_info->fs_devices->rw_devices--;
1549                 clear_super = true;
1550         }
1551
1552         ret = btrfs_shrink_device(device, 0);
1553         if (ret)
1554                 goto error_undo;
1555
1556         /*
1557          * TODO: the superblock still includes this device in its num_devices
1558          * counter although write_all_supers() is not locked out. This
1559          * could give a filesystem state which requires a degraded mount.
1560          */
1561         ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device);
1562         if (ret)
1563                 goto error_undo;
1564
1565         spin_lock(&root->fs_info->free_chunk_lock);
1566         root->fs_info->free_chunk_space = device->total_bytes -
1567                 device->bytes_used;
1568         spin_unlock(&root->fs_info->free_chunk_lock);
1569
1570         device->in_fs_metadata = 0;
1571         btrfs_scrub_cancel_dev(root->fs_info, device);
1572
1573         /*
1574          * the device list mutex makes sure that we don't change
1575          * the device list while someone else is writing out all
1576          * the device supers.
1577          */
1578
1579         cur_devices = device->fs_devices;
1580         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1581         list_del_rcu(&device->dev_list);
1582
1583         device->fs_devices->num_devices--;
1584         device->fs_devices->total_devices--;
1585
1586         if (device->missing)
1587                 root->fs_info->fs_devices->missing_devices--;
1588
1589         next_device = list_entry(root->fs_info->fs_devices->devices.next,
1590                                  struct btrfs_device, dev_list);
1591         if (device->bdev == root->fs_info->sb->s_bdev)
1592                 root->fs_info->sb->s_bdev = next_device->bdev;
1593         if (device->bdev == root->fs_info->fs_devices->latest_bdev)
1594                 root->fs_info->fs_devices->latest_bdev = next_device->bdev;
1595
1596         if (device->bdev)
1597                 device->fs_devices->open_devices--;
1598
1599         call_rcu(&device->rcu, free_device);
1600         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1601
1602         num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1;
1603         btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices);
1604
1605         if (cur_devices->open_devices == 0) {
1606                 struct btrfs_fs_devices *fs_devices;
1607                 fs_devices = root->fs_info->fs_devices;
1608                 while (fs_devices) {
1609                         if (fs_devices->seed == cur_devices)
1610                                 break;
1611                         fs_devices = fs_devices->seed;
1612                 }
1613                 fs_devices->seed = cur_devices->seed;
1614                 cur_devices->seed = NULL;
1615                 lock_chunks(root);
1616                 __btrfs_close_devices(cur_devices);
1617                 unlock_chunks(root);
1618                 free_fs_devices(cur_devices);
1619         }
1620
1621         root->fs_info->num_tolerated_disk_barrier_failures =
1622                 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
1623
1624         /*
1625          * at this point, the device is zero sized.  We want to
1626          * remove it from the devices list and zero out the old super
1627          */
1628         if (clear_super && disk_super) {
1629                 /* make sure this device isn't detected as part of
1630                  * the FS anymore
1631                  */
1632                 memset(&disk_super->magic, 0, sizeof(disk_super->magic));
1633                 set_buffer_dirty(bh);
1634                 sync_dirty_buffer(bh);
1635         }
1636
1637         ret = 0;
1638
1639         /* Notify udev that device has changed */
1640         if (bdev)
1641                 btrfs_kobject_uevent(bdev, KOBJ_CHANGE);
1642
1643 error_brelse:
1644         brelse(bh);
1645         if (bdev)
1646                 blkdev_put(bdev, FMODE_READ | FMODE_EXCL);
1647 out:
1648         mutex_unlock(&uuid_mutex);
1649         return ret;
1650 error_undo:
1651         if (device->writeable) {
1652                 lock_chunks(root);
1653                 list_add(&device->dev_alloc_list,
1654                          &root->fs_info->fs_devices->alloc_list);
1655                 unlock_chunks(root);
1656                 root->fs_info->fs_devices->rw_devices++;
1657         }
1658         goto error_brelse;
1659 }
1660
1661 void btrfs_rm_dev_replace_srcdev(struct btrfs_fs_info *fs_info,
1662                                  struct btrfs_device *srcdev)
1663 {
1664         WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex));
1665         list_del_rcu(&srcdev->dev_list);
1666         list_del_rcu(&srcdev->dev_alloc_list);
1667         fs_info->fs_devices->num_devices--;
1668         if (srcdev->missing) {
1669                 fs_info->fs_devices->missing_devices--;
1670                 fs_info->fs_devices->rw_devices++;
1671         }
1672         if (srcdev->can_discard)
1673                 fs_info->fs_devices->num_can_discard--;
1674         if (srcdev->bdev)
1675                 fs_info->fs_devices->open_devices--;
1676
1677         call_rcu(&srcdev->rcu, free_device);
1678 }
1679
1680 void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info,
1681                                       struct btrfs_device *tgtdev)
1682 {
1683         struct btrfs_device *next_device;
1684
1685         WARN_ON(!tgtdev);
1686         mutex_lock(&fs_info->fs_devices->device_list_mutex);
1687         if (tgtdev->bdev) {
1688                 btrfs_scratch_superblock(tgtdev);
1689                 fs_info->fs_devices->open_devices--;
1690         }
1691         fs_info->fs_devices->num_devices--;
1692         if (tgtdev->can_discard)
1693                 fs_info->fs_devices->num_can_discard++;
1694
1695         next_device = list_entry(fs_info->fs_devices->devices.next,
1696                                  struct btrfs_device, dev_list);
1697         if (tgtdev->bdev == fs_info->sb->s_bdev)
1698                 fs_info->sb->s_bdev = next_device->bdev;
1699         if (tgtdev->bdev == fs_info->fs_devices->latest_bdev)
1700                 fs_info->fs_devices->latest_bdev = next_device->bdev;
1701         list_del_rcu(&tgtdev->dev_list);
1702
1703         call_rcu(&tgtdev->rcu, free_device);
1704
1705         mutex_unlock(&fs_info->fs_devices->device_list_mutex);
1706 }
1707
1708 int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path,
1709                               struct btrfs_device **device)
1710 {
1711         int ret = 0;
1712         struct btrfs_super_block *disk_super;
1713         u64 devid;
1714         u8 *dev_uuid;
1715         struct block_device *bdev;
1716         struct buffer_head *bh;
1717
1718         *device = NULL;
1719         ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ,
1720                                     root->fs_info->bdev_holder, 0, &bdev, &bh);
1721         if (ret)
1722                 return ret;
1723         disk_super = (struct btrfs_super_block *)bh->b_data;
1724         devid = btrfs_stack_device_id(&disk_super->dev_item);
1725         dev_uuid = disk_super->dev_item.uuid;
1726         *device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1727                                     disk_super->fsid);
1728         brelse(bh);
1729         if (!*device)
1730                 ret = -ENOENT;
1731         blkdev_put(bdev, FMODE_READ);
1732         return ret;
1733 }
1734
1735 int btrfs_find_device_missing_or_by_path(struct btrfs_root *root,
1736                                          char *device_path,
1737                                          struct btrfs_device **device)
1738 {
1739         *device = NULL;
1740         if (strcmp(device_path, "missing") == 0) {
1741                 struct list_head *devices;
1742                 struct btrfs_device *tmp;
1743
1744                 devices = &root->fs_info->fs_devices->devices;
1745                 /*
1746                  * It is safe to read the devices since the volume_mutex
1747                  * is held by the caller.
1748                  */
1749                 list_for_each_entry(tmp, devices, dev_list) {
1750                         if (tmp->in_fs_metadata && !tmp->bdev) {
1751                                 *device = tmp;
1752                                 break;
1753                         }
1754                 }
1755
1756                 if (!*device) {
1757                         pr_err("btrfs: no missing device found\n");
1758                         return -ENOENT;
1759                 }
1760
1761                 return 0;
1762         } else {
1763                 return btrfs_find_device_by_path(root, device_path, device);
1764         }
1765 }
1766
1767 /*
1768  * does all the dirty work required for changing file system's UUID.
1769  */
1770 static int btrfs_prepare_sprout(struct btrfs_root *root)
1771 {
1772         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
1773         struct btrfs_fs_devices *old_devices;
1774         struct btrfs_fs_devices *seed_devices;
1775         struct btrfs_super_block *disk_super = root->fs_info->super_copy;
1776         struct btrfs_device *device;
1777         u64 super_flags;
1778
1779         BUG_ON(!mutex_is_locked(&uuid_mutex));
1780         if (!fs_devices->seeding)
1781                 return -EINVAL;
1782
1783         seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1784         if (!seed_devices)
1785                 return -ENOMEM;
1786
1787         old_devices = clone_fs_devices(fs_devices);
1788         if (IS_ERR(old_devices)) {
1789                 kfree(seed_devices);
1790                 return PTR_ERR(old_devices);
1791         }
1792
1793         list_add(&old_devices->list, &fs_uuids);
1794
1795         memcpy(seed_devices, fs_devices, sizeof(*seed_devices));
1796         seed_devices->opened = 1;
1797         INIT_LIST_HEAD(&seed_devices->devices);
1798         INIT_LIST_HEAD(&seed_devices->alloc_list);
1799         mutex_init(&seed_devices->device_list_mutex);
1800
1801         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1802         list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices,
1803                               synchronize_rcu);
1804         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1805
1806         list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list);
1807         list_for_each_entry(device, &seed_devices->devices, dev_list) {
1808                 device->fs_devices = seed_devices;
1809         }
1810
1811         fs_devices->seeding = 0;
1812         fs_devices->num_devices = 0;
1813         fs_devices->open_devices = 0;
1814         fs_devices->total_devices = 0;
1815         fs_devices->seed = seed_devices;
1816
1817         generate_random_uuid(fs_devices->fsid);
1818         memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1819         memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE);
1820         super_flags = btrfs_super_flags(disk_super) &
1821                       ~BTRFS_SUPER_FLAG_SEEDING;
1822         btrfs_set_super_flags(disk_super, super_flags);
1823
1824         return 0;
1825 }
1826
1827 /*
1828  * strore the expected generation for seed devices in device items.
1829  */
1830 static int btrfs_finish_sprout(struct btrfs_trans_handle *trans,
1831                                struct btrfs_root *root)
1832 {
1833         struct btrfs_path *path;
1834         struct extent_buffer *leaf;
1835         struct btrfs_dev_item *dev_item;
1836         struct btrfs_device *device;
1837         struct btrfs_key key;
1838         u8 fs_uuid[BTRFS_UUID_SIZE];
1839         u8 dev_uuid[BTRFS_UUID_SIZE];
1840         u64 devid;
1841         int ret;
1842
1843         path = btrfs_alloc_path();
1844         if (!path)
1845                 return -ENOMEM;
1846
1847         root = root->fs_info->chunk_root;
1848         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1849         key.offset = 0;
1850         key.type = BTRFS_DEV_ITEM_KEY;
1851
1852         while (1) {
1853                 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
1854                 if (ret < 0)
1855                         goto error;
1856
1857                 leaf = path->nodes[0];
1858 next_slot:
1859                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
1860                         ret = btrfs_next_leaf(root, path);
1861                         if (ret > 0)
1862                                 break;
1863                         if (ret < 0)
1864                                 goto error;
1865                         leaf = path->nodes[0];
1866                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1867                         btrfs_release_path(path);
1868                         continue;
1869                 }
1870
1871                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
1872                 if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID ||
1873                     key.type != BTRFS_DEV_ITEM_KEY)
1874                         break;
1875
1876                 dev_item = btrfs_item_ptr(leaf, path->slots[0],
1877                                           struct btrfs_dev_item);
1878                 devid = btrfs_device_id(leaf, dev_item);
1879                 read_extent_buffer(leaf, dev_uuid,
1880                                    (unsigned long)btrfs_device_uuid(dev_item),
1881                                    BTRFS_UUID_SIZE);
1882                 read_extent_buffer(leaf, fs_uuid,
1883                                    (unsigned long)btrfs_device_fsid(dev_item),
1884                                    BTRFS_UUID_SIZE);
1885                 device = btrfs_find_device(root->fs_info, devid, dev_uuid,
1886                                            fs_uuid);
1887                 BUG_ON(!device); /* Logic error */
1888
1889                 if (device->fs_devices->seeding) {
1890                         btrfs_set_device_generation(leaf, dev_item,
1891                                                     device->generation);
1892                         btrfs_mark_buffer_dirty(leaf);
1893                 }
1894
1895                 path->slots[0]++;
1896                 goto next_slot;
1897         }
1898         ret = 0;
1899 error:
1900         btrfs_free_path(path);
1901         return ret;
1902 }
1903
1904 int btrfs_init_new_device(struct btrfs_root *root, char *device_path)
1905 {
1906         struct request_queue *q;
1907         struct btrfs_trans_handle *trans;
1908         struct btrfs_device *device;
1909         struct block_device *bdev;
1910         struct list_head *devices;
1911         struct super_block *sb = root->fs_info->sb;
1912         struct rcu_string *name;
1913         u64 total_bytes;
1914         int seeding_dev = 0;
1915         int ret = 0;
1916
1917         if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding)
1918                 return -EROFS;
1919
1920         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
1921                                   root->fs_info->bdev_holder);
1922         if (IS_ERR(bdev))
1923                 return PTR_ERR(bdev);
1924
1925         if (root->fs_info->fs_devices->seeding) {
1926                 seeding_dev = 1;
1927                 down_write(&sb->s_umount);
1928                 mutex_lock(&uuid_mutex);
1929         }
1930
1931         filemap_write_and_wait(bdev->bd_inode->i_mapping);
1932
1933         devices = &root->fs_info->fs_devices->devices;
1934
1935         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
1936         list_for_each_entry(device, devices, dev_list) {
1937                 if (device->bdev == bdev) {
1938                         ret = -EEXIST;
1939                         mutex_unlock(
1940                                 &root->fs_info->fs_devices->device_list_mutex);
1941                         goto error;
1942                 }
1943         }
1944         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
1945
1946         device = kzalloc(sizeof(*device), GFP_NOFS);
1947         if (!device) {
1948                 /* we can safely leave the fs_devices entry around */
1949                 ret = -ENOMEM;
1950                 goto error;
1951         }
1952
1953         name = rcu_string_strdup(device_path, GFP_NOFS);
1954         if (!name) {
1955                 kfree(device);
1956                 ret = -ENOMEM;
1957                 goto error;
1958         }
1959         rcu_assign_pointer(device->name, name);
1960
1961         ret = find_next_devid(root, &device->devid);
1962         if (ret) {
1963                 rcu_string_free(device->name);
1964                 kfree(device);
1965                 goto error;
1966         }
1967
1968         trans = btrfs_start_transaction(root, 0);
1969         if (IS_ERR(trans)) {
1970                 rcu_string_free(device->name);
1971                 kfree(device);
1972                 ret = PTR_ERR(trans);
1973                 goto error;
1974         }
1975
1976         lock_chunks(root);
1977
1978         q = bdev_get_queue(bdev);
1979         if (blk_queue_discard(q))
1980                 device->can_discard = 1;
1981         device->writeable = 1;
1982         device->work.func = pending_bios_fn;
1983         generate_random_uuid(device->uuid);
1984         spin_lock_init(&device->io_lock);
1985         device->generation = trans->transid;
1986         device->io_width = root->sectorsize;
1987         device->io_align = root->sectorsize;
1988         device->sector_size = root->sectorsize;
1989         device->total_bytes = i_size_read(bdev->bd_inode);
1990         device->disk_total_bytes = device->total_bytes;
1991         device->dev_root = root->fs_info->dev_root;
1992         device->bdev = bdev;
1993         device->in_fs_metadata = 1;
1994         device->is_tgtdev_for_dev_replace = 0;
1995         device->mode = FMODE_EXCL;
1996         set_blocksize(device->bdev, 4096);
1997
1998         if (seeding_dev) {
1999                 sb->s_flags &= ~MS_RDONLY;
2000                 ret = btrfs_prepare_sprout(root);
2001                 BUG_ON(ret); /* -ENOMEM */
2002         }
2003
2004         device->fs_devices = root->fs_info->fs_devices;
2005
2006         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2007         list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices);
2008         list_add(&device->dev_alloc_list,
2009                  &root->fs_info->fs_devices->alloc_list);
2010         root->fs_info->fs_devices->num_devices++;
2011         root->fs_info->fs_devices->open_devices++;
2012         root->fs_info->fs_devices->rw_devices++;
2013         root->fs_info->fs_devices->total_devices++;
2014         if (device->can_discard)
2015                 root->fs_info->fs_devices->num_can_discard++;
2016         root->fs_info->fs_devices->total_rw_bytes += device->total_bytes;
2017
2018         spin_lock(&root->fs_info->free_chunk_lock);
2019         root->fs_info->free_chunk_space += device->total_bytes;
2020         spin_unlock(&root->fs_info->free_chunk_lock);
2021
2022         if (!blk_queue_nonrot(bdev_get_queue(bdev)))
2023                 root->fs_info->fs_devices->rotating = 1;
2024
2025         total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy);
2026         btrfs_set_super_total_bytes(root->fs_info->super_copy,
2027                                     total_bytes + device->total_bytes);
2028
2029         total_bytes = btrfs_super_num_devices(root->fs_info->super_copy);
2030         btrfs_set_super_num_devices(root->fs_info->super_copy,
2031                                     total_bytes + 1);
2032         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2033
2034         if (seeding_dev) {
2035                 ret = init_first_rw_device(trans, root, device);
2036                 if (ret) {
2037                         btrfs_abort_transaction(trans, root, ret);
2038                         goto error_trans;
2039                 }
2040                 ret = btrfs_finish_sprout(trans, root);
2041                 if (ret) {
2042                         btrfs_abort_transaction(trans, root, ret);
2043                         goto error_trans;
2044                 }
2045         } else {
2046                 ret = btrfs_add_device(trans, root, device);
2047                 if (ret) {
2048                         btrfs_abort_transaction(trans, root, ret);
2049                         goto error_trans;
2050                 }
2051         }
2052
2053         /*
2054          * we've got more storage, clear any full flags on the space
2055          * infos
2056          */
2057         btrfs_clear_space_info_full(root->fs_info);
2058
2059         unlock_chunks(root);
2060         root->fs_info->num_tolerated_disk_barrier_failures =
2061                 btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info);
2062         ret = btrfs_commit_transaction(trans, root);
2063
2064         if (seeding_dev) {
2065                 mutex_unlock(&uuid_mutex);
2066                 up_write(&sb->s_umount);
2067
2068                 if (ret) /* transaction commit */
2069                         return ret;
2070
2071                 ret = btrfs_relocate_sys_chunks(root);
2072                 if (ret < 0)
2073                         btrfs_error(root->fs_info, ret,
2074                                     "Failed to relocate sys chunks after "
2075                                     "device initialization. This can be fixed "
2076                                     "using the \"btrfs balance\" command.");
2077                 trans = btrfs_attach_transaction(root);
2078                 if (IS_ERR(trans)) {
2079                         if (PTR_ERR(trans) == -ENOENT)
2080                                 return 0;
2081                         return PTR_ERR(trans);
2082                 }
2083                 ret = btrfs_commit_transaction(trans, root);
2084         }
2085
2086         return ret;
2087
2088 error_trans:
2089         unlock_chunks(root);
2090         btrfs_end_transaction(trans, root);
2091         rcu_string_free(device->name);
2092         kfree(device);
2093 error:
2094         blkdev_put(bdev, FMODE_EXCL);
2095         if (seeding_dev) {
2096                 mutex_unlock(&uuid_mutex);
2097                 up_write(&sb->s_umount);
2098         }
2099         return ret;
2100 }
2101
2102 int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path,
2103                                   struct btrfs_device **device_out)
2104 {
2105         struct request_queue *q;
2106         struct btrfs_device *device;
2107         struct block_device *bdev;
2108         struct btrfs_fs_info *fs_info = root->fs_info;
2109         struct list_head *devices;
2110         struct rcu_string *name;
2111         int ret = 0;
2112
2113         *device_out = NULL;
2114         if (fs_info->fs_devices->seeding)
2115                 return -EINVAL;
2116
2117         bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL,
2118                                   fs_info->bdev_holder);
2119         if (IS_ERR(bdev))
2120                 return PTR_ERR(bdev);
2121
2122         filemap_write_and_wait(bdev->bd_inode->i_mapping);
2123
2124         devices = &fs_info->fs_devices->devices;
2125         list_for_each_entry(device, devices, dev_list) {
2126                 if (device->bdev == bdev) {
2127                         ret = -EEXIST;
2128                         goto error;
2129                 }
2130         }
2131
2132         device = kzalloc(sizeof(*device), GFP_NOFS);
2133         if (!device) {
2134                 ret = -ENOMEM;
2135                 goto error;
2136         }
2137
2138         name = rcu_string_strdup(device_path, GFP_NOFS);
2139         if (!name) {
2140                 kfree(device);
2141                 ret = -ENOMEM;
2142                 goto error;
2143         }
2144         rcu_assign_pointer(device->name, name);
2145
2146         q = bdev_get_queue(bdev);
2147         if (blk_queue_discard(q))
2148                 device->can_discard = 1;
2149         mutex_lock(&root->fs_info->fs_devices->device_list_mutex);
2150         device->writeable = 1;
2151         device->work.func = pending_bios_fn;
2152         generate_random_uuid(device->uuid);
2153         device->devid = BTRFS_DEV_REPLACE_DEVID;
2154         spin_lock_init(&device->io_lock);
2155         device->generation = 0;
2156         device->io_width = root->sectorsize;
2157         device->io_align = root->sectorsize;
2158         device->sector_size = root->sectorsize;
2159         device->total_bytes = i_size_read(bdev->bd_inode);
2160         device->disk_total_bytes = device->total_bytes;
2161         device->dev_root = fs_info->dev_root;
2162         device->bdev = bdev;
2163         device->in_fs_metadata = 1;
2164         device->is_tgtdev_for_dev_replace = 1;
2165         device->mode = FMODE_EXCL;
2166         set_blocksize(device->bdev, 4096);
2167         device->fs_devices = fs_info->fs_devices;
2168         list_add(&device->dev_list, &fs_info->fs_devices->devices);
2169         fs_info->fs_devices->num_devices++;
2170         fs_info->fs_devices->open_devices++;
2171         if (device->can_discard)
2172                 fs_info->fs_devices->num_can_discard++;
2173         mutex_unlock(&root->fs_info->fs_devices->device_list_mutex);
2174
2175         *device_out = device;
2176         return ret;
2177
2178 error:
2179         blkdev_put(bdev, FMODE_EXCL);
2180         return ret;
2181 }
2182
2183 void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info,
2184                                               struct btrfs_device *tgtdev)
2185 {
2186         WARN_ON(fs_info->fs_devices->rw_devices == 0);
2187         tgtdev->io_width = fs_info->dev_root->sectorsize;
2188         tgtdev->io_align = fs_info->dev_root->sectorsize;
2189         tgtdev->sector_size = fs_info->dev_root->sectorsize;
2190         tgtdev->dev_root = fs_info->dev_root;
2191         tgtdev->in_fs_metadata = 1;
2192 }
2193
2194 static noinline int btrfs_update_device(struct btrfs_trans_handle *trans,
2195                                         struct btrfs_device *device)
2196 {
2197         int ret;
2198         struct btrfs_path *path;
2199         struct btrfs_root *root;
2200         struct btrfs_dev_item *dev_item;
2201         struct extent_buffer *leaf;
2202         struct btrfs_key key;
2203
2204         root = device->dev_root->fs_info->chunk_root;
2205
2206         path = btrfs_alloc_path();
2207         if (!path)
2208                 return -ENOMEM;
2209
2210         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2211         key.type = BTRFS_DEV_ITEM_KEY;
2212         key.offset = device->devid;
2213
2214         ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
2215         if (ret < 0)
2216                 goto out;
2217
2218         if (ret > 0) {
2219                 ret = -ENOENT;
2220                 goto out;
2221         }
2222
2223         leaf = path->nodes[0];
2224         dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
2225
2226         btrfs_set_device_id(leaf, dev_item, device->devid);
2227         btrfs_set_device_type(leaf, dev_item, device->type);
2228         btrfs_set_device_io_align(leaf, dev_item, device->io_align);
2229         btrfs_set_device_io_width(leaf, dev_item, device->io_width);
2230         btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
2231         btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes);
2232         btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
2233         btrfs_mark_buffer_dirty(leaf);
2234
2235 out:
2236         btrfs_free_path(path);
2237         return ret;
2238 }
2239
2240 static int __btrfs_grow_device(struct btrfs_trans_handle *trans,
2241                       struct btrfs_device *device, u64 new_size)
2242 {
2243         struct btrfs_super_block *super_copy =
2244                 device->dev_root->fs_info->super_copy;
2245         u64 old_total = btrfs_super_total_bytes(super_copy);
2246         u64 diff = new_size - device->total_bytes;
2247
2248         if (!device->writeable)
2249                 return -EACCES;
2250         if (new_size <= device->total_bytes ||
2251             device->is_tgtdev_for_dev_replace)
2252                 return -EINVAL;
2253
2254         btrfs_set_super_total_bytes(super_copy, old_total + diff);
2255         device->fs_devices->total_rw_bytes += diff;
2256
2257         device->total_bytes = new_size;
2258         device->disk_total_bytes = new_size;
2259         btrfs_clear_space_info_full(device->dev_root->fs_info);
2260
2261         return btrfs_update_device(trans, device);
2262 }
2263
2264 int btrfs_grow_device(struct btrfs_trans_handle *trans,
2265                       struct btrfs_device *device, u64 new_size)
2266 {
2267         int ret;
2268         lock_chunks(device->dev_root);
2269         ret = __btrfs_grow_device(trans, device, new_size);
2270         unlock_chunks(device->dev_root);
2271         return ret;
2272 }
2273
2274 static int btrfs_free_chunk(struct btrfs_trans_handle *trans,
2275                             struct btrfs_root *root,
2276                             u64 chunk_tree, u64 chunk_objectid,
2277                             u64 chunk_offset)
2278 {
2279         int ret;
2280         struct btrfs_path *path;
2281         struct btrfs_key key;
2282
2283         root = root->fs_info->chunk_root;
2284         path = btrfs_alloc_path();
2285         if (!path)
2286                 return -ENOMEM;
2287
2288         key.objectid = chunk_objectid;
2289         key.offset = chunk_offset;
2290         key.type = BTRFS_CHUNK_ITEM_KEY;
2291
2292         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2293         if (ret < 0)
2294                 goto out;
2295         else if (ret > 0) { /* Logic error or corruption */
2296                 btrfs_error(root->fs_info, -ENOENT,
2297                             "Failed lookup while freeing chunk.");
2298                 ret = -ENOENT;
2299                 goto out;
2300         }
2301
2302         ret = btrfs_del_item(trans, root, path);
2303         if (ret < 0)
2304                 btrfs_error(root->fs_info, ret,
2305                             "Failed to delete chunk item.");
2306 out:
2307         btrfs_free_path(path);
2308         return ret;
2309 }
2310
2311 static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64
2312                         chunk_offset)
2313 {
2314         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
2315         struct btrfs_disk_key *disk_key;
2316         struct btrfs_chunk *chunk;
2317         u8 *ptr;
2318         int ret = 0;
2319         u32 num_stripes;
2320         u32 array_size;
2321         u32 len = 0;
2322         u32 cur;
2323         struct btrfs_key key;
2324
2325         array_size = btrfs_super_sys_array_size(super_copy);
2326
2327         ptr = super_copy->sys_chunk_array;
2328         cur = 0;
2329
2330         while (cur < array_size) {
2331                 disk_key = (struct btrfs_disk_key *)ptr;
2332                 btrfs_disk_key_to_cpu(&key, disk_key);
2333
2334                 len = sizeof(*disk_key);
2335
2336                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
2337                         chunk = (struct btrfs_chunk *)(ptr + len);
2338                         num_stripes = btrfs_stack_chunk_num_stripes(chunk);
2339                         len += btrfs_chunk_item_size(num_stripes);
2340                 } else {
2341                         ret = -EIO;
2342                         break;
2343                 }
2344                 if (key.objectid == chunk_objectid &&
2345                     key.offset == chunk_offset) {
2346                         memmove(ptr, ptr + len, array_size - (cur + len));
2347                         array_size -= len;
2348                         btrfs_set_super_sys_array_size(super_copy, array_size);
2349                 } else {
2350                         ptr += len;
2351                         cur += len;
2352                 }
2353         }
2354         return ret;
2355 }
2356
2357 static int btrfs_relocate_chunk(struct btrfs_root *root,
2358                          u64 chunk_tree, u64 chunk_objectid,
2359                          u64 chunk_offset)
2360 {
2361         struct extent_map_tree *em_tree;
2362         struct btrfs_root *extent_root;
2363         struct btrfs_trans_handle *trans;
2364         struct extent_map *em;
2365         struct map_lookup *map;
2366         int ret;
2367         int i;
2368
2369         root = root->fs_info->chunk_root;
2370         extent_root = root->fs_info->extent_root;
2371         em_tree = &root->fs_info->mapping_tree.map_tree;
2372
2373         ret = btrfs_can_relocate(extent_root, chunk_offset);
2374         if (ret)
2375                 return -ENOSPC;
2376
2377         /* step one, relocate all the extents inside this chunk */
2378         ret = btrfs_relocate_block_group(extent_root, chunk_offset);
2379         if (ret)
2380                 return ret;
2381
2382         trans = btrfs_start_transaction(root, 0);
2383         BUG_ON(IS_ERR(trans));
2384
2385         lock_chunks(root);
2386
2387         /*
2388          * step two, delete the device extents and the
2389          * chunk tree entries
2390          */
2391         read_lock(&em_tree->lock);
2392         em = lookup_extent_mapping(em_tree, chunk_offset, 1);
2393         read_unlock(&em_tree->lock);
2394
2395         BUG_ON(!em || em->start > chunk_offset ||
2396                em->start + em->len < chunk_offset);
2397         map = (struct map_lookup *)em->bdev;
2398
2399         for (i = 0; i < map->num_stripes; i++) {
2400                 ret = btrfs_free_dev_extent(trans, map->stripes[i].dev,
2401                                             map->stripes[i].physical);
2402                 BUG_ON(ret);
2403
2404                 if (map->stripes[i].dev) {
2405                         ret = btrfs_update_device(trans, map->stripes[i].dev);
2406                         BUG_ON(ret);
2407                 }
2408         }
2409         ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid,
2410                                chunk_offset);
2411
2412         BUG_ON(ret);
2413
2414         trace_btrfs_chunk_free(root, map, chunk_offset, em->len);
2415
2416         if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
2417                 ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset);
2418                 BUG_ON(ret);
2419         }
2420
2421         ret = btrfs_remove_block_group(trans, extent_root, chunk_offset);
2422         BUG_ON(ret);
2423
2424         write_lock(&em_tree->lock);
2425         remove_extent_mapping(em_tree, em);
2426         write_unlock(&em_tree->lock);
2427
2428         kfree(map);
2429         em->bdev = NULL;
2430
2431         /* once for the tree */
2432         free_extent_map(em);
2433         /* once for us */
2434         free_extent_map(em);
2435
2436         unlock_chunks(root);
2437         btrfs_end_transaction(trans, root);
2438         return 0;
2439 }
2440
2441 static int btrfs_relocate_sys_chunks(struct btrfs_root *root)
2442 {
2443         struct btrfs_root *chunk_root = root->fs_info->chunk_root;
2444         struct btrfs_path *path;
2445         struct extent_buffer *leaf;
2446         struct btrfs_chunk *chunk;
2447         struct btrfs_key key;
2448         struct btrfs_key found_key;
2449         u64 chunk_tree = chunk_root->root_key.objectid;
2450         u64 chunk_type;
2451         bool retried = false;
2452         int failed = 0;
2453         int ret;
2454
2455         path = btrfs_alloc_path();
2456         if (!path)
2457                 return -ENOMEM;
2458
2459 again:
2460         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2461         key.offset = (u64)-1;
2462         key.type = BTRFS_CHUNK_ITEM_KEY;
2463
2464         while (1) {
2465                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2466                 if (ret < 0)
2467                         goto error;
2468                 BUG_ON(ret == 0); /* Corruption */
2469
2470                 ret = btrfs_previous_item(chunk_root, path, key.objectid,
2471                                           key.type);
2472                 if (ret < 0)
2473                         goto error;
2474                 if (ret > 0)
2475                         break;
2476
2477                 leaf = path->nodes[0];
2478                 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
2479
2480                 chunk = btrfs_item_ptr(leaf, path->slots[0],
2481                                        struct btrfs_chunk);
2482                 chunk_type = btrfs_chunk_type(leaf, chunk);
2483                 btrfs_release_path(path);
2484
2485                 if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) {
2486                         ret = btrfs_relocate_chunk(chunk_root, chunk_tree,
2487                                                    found_key.objectid,
2488                                                    found_key.offset);
2489                         if (ret == -ENOSPC)
2490                                 failed++;
2491                         else if (ret)
2492                                 BUG();
2493                 }
2494
2495                 if (found_key.offset == 0)
2496                         break;
2497                 key.offset = found_key.offset - 1;
2498         }
2499         ret = 0;
2500         if (failed && !retried) {
2501                 failed = 0;
2502                 retried = true;
2503                 goto again;
2504         } else if (failed && retried) {
2505                 WARN_ON(1);
2506                 ret = -ENOSPC;
2507         }
2508 error:
2509         btrfs_free_path(path);
2510         return ret;
2511 }
2512
2513 static int insert_balance_item(struct btrfs_root *root,
2514                                struct btrfs_balance_control *bctl)
2515 {
2516         struct btrfs_trans_handle *trans;
2517         struct btrfs_balance_item *item;
2518         struct btrfs_disk_balance_args disk_bargs;
2519         struct btrfs_path *path;
2520         struct extent_buffer *leaf;
2521         struct btrfs_key key;
2522         int ret, err;
2523
2524         path = btrfs_alloc_path();
2525         if (!path)
2526                 return -ENOMEM;
2527
2528         trans = btrfs_start_transaction(root, 0);
2529         if (IS_ERR(trans)) {
2530                 btrfs_free_path(path);
2531                 return PTR_ERR(trans);
2532         }
2533
2534         key.objectid = BTRFS_BALANCE_OBJECTID;
2535         key.type = BTRFS_BALANCE_ITEM_KEY;
2536         key.offset = 0;
2537
2538         ret = btrfs_insert_empty_item(trans, root, path, &key,
2539                                       sizeof(*item));
2540         if (ret)
2541                 goto out;
2542
2543         leaf = path->nodes[0];
2544         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
2545
2546         memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item));
2547
2548         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data);
2549         btrfs_set_balance_data(leaf, item, &disk_bargs);
2550         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta);
2551         btrfs_set_balance_meta(leaf, item, &disk_bargs);
2552         btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys);
2553         btrfs_set_balance_sys(leaf, item, &disk_bargs);
2554
2555         btrfs_set_balance_flags(leaf, item, bctl->flags);
2556
2557         btrfs_mark_buffer_dirty(leaf);
2558 out:
2559         btrfs_free_path(path);
2560         err = btrfs_commit_transaction(trans, root);
2561         if (err && !ret)
2562                 ret = err;
2563         return ret;
2564 }
2565
2566 static int del_balance_item(struct btrfs_root *root)
2567 {
2568         struct btrfs_trans_handle *trans;
2569         struct btrfs_path *path;
2570         struct btrfs_key key;
2571         int ret, err;
2572
2573         path = btrfs_alloc_path();
2574         if (!path)
2575                 return -ENOMEM;
2576
2577         trans = btrfs_start_transaction(root, 0);
2578         if (IS_ERR(trans)) {
2579                 btrfs_free_path(path);
2580                 return PTR_ERR(trans);
2581         }
2582
2583         key.objectid = BTRFS_BALANCE_OBJECTID;
2584         key.type = BTRFS_BALANCE_ITEM_KEY;
2585         key.offset = 0;
2586
2587         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
2588         if (ret < 0)
2589                 goto out;
2590         if (ret > 0) {
2591                 ret = -ENOENT;
2592                 goto out;
2593         }
2594
2595         ret = btrfs_del_item(trans, root, path);
2596 out:
2597         btrfs_free_path(path);
2598         err = btrfs_commit_transaction(trans, root);
2599         if (err && !ret)
2600                 ret = err;
2601         return ret;
2602 }
2603
2604 /*
2605  * This is a heuristic used to reduce the number of chunks balanced on
2606  * resume after balance was interrupted.
2607  */
2608 static void update_balance_args(struct btrfs_balance_control *bctl)
2609 {
2610         /*
2611          * Turn on soft mode for chunk types that were being converted.
2612          */
2613         if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)
2614                 bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT;
2615         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)
2616                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT;
2617         if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)
2618                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT;
2619
2620         /*
2621          * Turn on usage filter if is not already used.  The idea is
2622          * that chunks that we have already balanced should be
2623          * reasonably full.  Don't do it for chunks that are being
2624          * converted - that will keep us from relocating unconverted
2625          * (albeit full) chunks.
2626          */
2627         if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2628             !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2629                 bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE;
2630                 bctl->data.usage = 90;
2631         }
2632         if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2633             !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2634                 bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE;
2635                 bctl->sys.usage = 90;
2636         }
2637         if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) &&
2638             !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) {
2639                 bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE;
2640                 bctl->meta.usage = 90;
2641         }
2642 }
2643
2644 /*
2645  * Should be called with both balance and volume mutexes held to
2646  * serialize other volume operations (add_dev/rm_dev/resize) with
2647  * restriper.  Same goes for unset_balance_control.
2648  */
2649 static void set_balance_control(struct btrfs_balance_control *bctl)
2650 {
2651         struct btrfs_fs_info *fs_info = bctl->fs_info;
2652
2653         BUG_ON(fs_info->balance_ctl);
2654
2655         spin_lock(&fs_info->balance_lock);
2656         fs_info->balance_ctl = bctl;
2657         spin_unlock(&fs_info->balance_lock);
2658 }
2659
2660 static void unset_balance_control(struct btrfs_fs_info *fs_info)
2661 {
2662         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2663
2664         BUG_ON(!fs_info->balance_ctl);
2665
2666         spin_lock(&fs_info->balance_lock);
2667         fs_info->balance_ctl = NULL;
2668         spin_unlock(&fs_info->balance_lock);
2669
2670         kfree(bctl);
2671 }
2672
2673 /*
2674  * Balance filters.  Return 1 if chunk should be filtered out
2675  * (should not be balanced).
2676  */
2677 static int chunk_profiles_filter(u64 chunk_type,
2678                                  struct btrfs_balance_args *bargs)
2679 {
2680         chunk_type = chunk_to_extended(chunk_type) &
2681                                 BTRFS_EXTENDED_PROFILE_MASK;
2682
2683         if (bargs->profiles & chunk_type)
2684                 return 0;
2685
2686         return 1;
2687 }
2688
2689 static int chunk_usage_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset,
2690                               struct btrfs_balance_args *bargs)
2691 {
2692         struct btrfs_block_group_cache *cache;
2693         u64 chunk_used, user_thresh;
2694         int ret = 1;
2695
2696         cache = btrfs_lookup_block_group(fs_info, chunk_offset);
2697         chunk_used = btrfs_block_group_used(&cache->item);
2698
2699         if (bargs->usage == 0)
2700                 user_thresh = 1;
2701         else if (bargs->usage > 100)
2702                 user_thresh = cache->key.offset;
2703         else
2704                 user_thresh = div_factor_fine(cache->key.offset,
2705                                               bargs->usage);
2706
2707         if (chunk_used < user_thresh)
2708                 ret = 0;
2709
2710         btrfs_put_block_group(cache);
2711         return ret;
2712 }
2713
2714 static int chunk_devid_filter(struct extent_buffer *leaf,
2715                               struct btrfs_chunk *chunk,
2716                               struct btrfs_balance_args *bargs)
2717 {
2718         struct btrfs_stripe *stripe;
2719         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2720         int i;
2721
2722         for (i = 0; i < num_stripes; i++) {
2723                 stripe = btrfs_stripe_nr(chunk, i);
2724                 if (btrfs_stripe_devid(leaf, stripe) == bargs->devid)
2725                         return 0;
2726         }
2727
2728         return 1;
2729 }
2730
2731 /* [pstart, pend) */
2732 static int chunk_drange_filter(struct extent_buffer *leaf,
2733                                struct btrfs_chunk *chunk,
2734                                u64 chunk_offset,
2735                                struct btrfs_balance_args *bargs)
2736 {
2737         struct btrfs_stripe *stripe;
2738         int num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
2739         u64 stripe_offset;
2740         u64 stripe_length;
2741         int factor;
2742         int i;
2743
2744         if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID))
2745                 return 0;
2746
2747         if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP |
2748              BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) {
2749                 factor = num_stripes / 2;
2750         } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) {
2751                 factor = num_stripes - 1;
2752         } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) {
2753                 factor = num_stripes - 2;
2754         } else {
2755                 factor = num_stripes;
2756         }
2757
2758         for (i = 0; i < num_stripes; i++) {
2759                 stripe = btrfs_stripe_nr(chunk, i);
2760                 if (btrfs_stripe_devid(leaf, stripe) != bargs->devid)
2761                         continue;
2762
2763                 stripe_offset = btrfs_stripe_offset(leaf, stripe);
2764                 stripe_length = btrfs_chunk_length(leaf, chunk);
2765                 do_div(stripe_length, factor);
2766
2767                 if (stripe_offset < bargs->pend &&
2768                     stripe_offset + stripe_length > bargs->pstart)
2769                         return 0;
2770         }
2771
2772         return 1;
2773 }
2774
2775 /* [vstart, vend) */
2776 static int chunk_vrange_filter(struct extent_buffer *leaf,
2777                                struct btrfs_chunk *chunk,
2778                                u64 chunk_offset,
2779                                struct btrfs_balance_args *bargs)
2780 {
2781         if (chunk_offset < bargs->vend &&
2782             chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart)
2783                 /* at least part of the chunk is inside this vrange */
2784                 return 0;
2785
2786         return 1;
2787 }
2788
2789 static int chunk_soft_convert_filter(u64 chunk_type,
2790                                      struct btrfs_balance_args *bargs)
2791 {
2792         if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT))
2793                 return 0;
2794
2795         chunk_type = chunk_to_extended(chunk_type) &
2796                                 BTRFS_EXTENDED_PROFILE_MASK;
2797
2798         if (bargs->target == chunk_type)
2799                 return 1;
2800
2801         return 0;
2802 }
2803
2804 static int should_balance_chunk(struct btrfs_root *root,
2805                                 struct extent_buffer *leaf,
2806                                 struct btrfs_chunk *chunk, u64 chunk_offset)
2807 {
2808         struct btrfs_balance_control *bctl = root->fs_info->balance_ctl;
2809         struct btrfs_balance_args *bargs = NULL;
2810         u64 chunk_type = btrfs_chunk_type(leaf, chunk);
2811
2812         /* type filter */
2813         if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) &
2814               (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) {
2815                 return 0;
2816         }
2817
2818         if (chunk_type & BTRFS_BLOCK_GROUP_DATA)
2819                 bargs = &bctl->data;
2820         else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM)
2821                 bargs = &bctl->sys;
2822         else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA)
2823                 bargs = &bctl->meta;
2824
2825         /* profiles filter */
2826         if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) &&
2827             chunk_profiles_filter(chunk_type, bargs)) {
2828                 return 0;
2829         }
2830
2831         /* usage filter */
2832         if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) &&
2833             chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) {
2834                 return 0;
2835         }
2836
2837         /* devid filter */
2838         if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) &&
2839             chunk_devid_filter(leaf, chunk, bargs)) {
2840                 return 0;
2841         }
2842
2843         /* drange filter, makes sense only with devid filter */
2844         if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) &&
2845             chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) {
2846                 return 0;
2847         }
2848
2849         /* vrange filter */
2850         if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) &&
2851             chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) {
2852                 return 0;
2853         }
2854
2855         /* soft profile changing mode */
2856         if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) &&
2857             chunk_soft_convert_filter(chunk_type, bargs)) {
2858                 return 0;
2859         }
2860
2861         return 1;
2862 }
2863
2864 static int __btrfs_balance(struct btrfs_fs_info *fs_info)
2865 {
2866         struct btrfs_balance_control *bctl = fs_info->balance_ctl;
2867         struct btrfs_root *chunk_root = fs_info->chunk_root;
2868         struct btrfs_root *dev_root = fs_info->dev_root;
2869         struct list_head *devices;
2870         struct btrfs_device *device;
2871         u64 old_size;
2872         u64 size_to_free;
2873         struct btrfs_chunk *chunk;
2874         struct btrfs_path *path;
2875         struct btrfs_key key;
2876         struct btrfs_key found_key;
2877         struct btrfs_trans_handle *trans;
2878         struct extent_buffer *leaf;
2879         int slot;
2880         int ret;
2881         int enospc_errors = 0;
2882         bool counting = true;
2883
2884         /* step one make some room on all the devices */
2885         devices = &fs_info->fs_devices->devices;
2886         list_for_each_entry(device, devices, dev_list) {
2887                 old_size = device->total_bytes;
2888                 size_to_free = div_factor(old_size, 1);
2889                 size_to_free = min(size_to_free, (u64)1 * 1024 * 1024);
2890                 if (!device->writeable ||
2891                     device->total_bytes - device->bytes_used > size_to_free ||
2892                     device->is_tgtdev_for_dev_replace)
2893                         continue;
2894
2895                 ret = btrfs_shrink_device(device, old_size - size_to_free);
2896                 if (ret == -ENOSPC)
2897                         break;
2898                 BUG_ON(ret);
2899
2900                 trans = btrfs_start_transaction(dev_root, 0);
2901                 BUG_ON(IS_ERR(trans));
2902
2903                 ret = btrfs_grow_device(trans, device, old_size);
2904                 BUG_ON(ret);
2905
2906                 btrfs_end_transaction(trans, dev_root);
2907         }
2908
2909         /* step two, relocate all the chunks */
2910         path = btrfs_alloc_path();
2911         if (!path) {
2912                 ret = -ENOMEM;
2913                 goto error;
2914         }
2915
2916         /* zero out stat counters */
2917         spin_lock(&fs_info->balance_lock);
2918         memset(&bctl->stat, 0, sizeof(bctl->stat));
2919         spin_unlock(&fs_info->balance_lock);
2920 again:
2921         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
2922         key.offset = (u64)-1;
2923         key.type = BTRFS_CHUNK_ITEM_KEY;
2924
2925         while (1) {
2926                 if ((!counting && atomic_read(&fs_info->balance_pause_req)) ||
2927                     atomic_read(&fs_info->balance_cancel_req)) {
2928                         ret = -ECANCELED;
2929                         goto error;
2930                 }
2931
2932                 ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0);
2933                 if (ret < 0)
2934                         goto error;
2935
2936                 /*
2937                  * this shouldn't happen, it means the last relocate
2938                  * failed
2939                  */
2940                 if (ret == 0)
2941                         BUG(); /* FIXME break ? */
2942
2943                 ret = btrfs_previous_item(chunk_root, path, 0,
2944                                           BTRFS_CHUNK_ITEM_KEY);
2945                 if (ret) {
2946                         ret = 0;
2947                         break;
2948                 }
2949
2950                 leaf = path->nodes[0];
2951                 slot = path->slots[0];
2952                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2953
2954                 if (found_key.objectid != key.objectid)
2955                         break;
2956
2957                 /* chunk zero is special */
2958                 if (found_key.offset == 0)
2959                         break;
2960
2961                 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
2962
2963                 if (!counting) {
2964                         spin_lock(&fs_info->balance_lock);
2965                         bctl->stat.considered++;
2966                         spin_unlock(&fs_info->balance_lock);
2967                 }
2968
2969                 ret = should_balance_chunk(chunk_root, leaf, chunk,
2970                                            found_key.offset);
2971                 btrfs_release_path(path);
2972                 if (!ret)
2973                         goto loop;
2974
2975                 if (counting) {
2976                         spin_lock(&fs_info->balance_lock);
2977                         bctl->stat.expected++;
2978                         spin_unlock(&fs_info->balance_lock);
2979                         goto loop;
2980                 }
2981
2982                 ret = btrfs_relocate_chunk(chunk_root,
2983                                            chunk_root->root_key.objectid,
2984                                            found_key.objectid,
2985                                            found_key.offset);
2986                 if (ret && ret != -ENOSPC)
2987                         goto error;
2988                 if (ret == -ENOSPC) {
2989                         enospc_errors++;
2990                 } else {
2991                         spin_lock(&fs_info->balance_lock);
2992                         bctl->stat.completed++;
2993                         spin_unlock(&fs_info->balance_lock);
2994                 }
2995 loop:
2996                 key.offset = found_key.offset - 1;
2997         }
2998
2999         if (counting) {
3000                 btrfs_release_path(path);
3001                 counting = false;
3002                 goto again;
3003         }
3004 error:
3005         btrfs_free_path(path);
3006         if (enospc_errors) {
3007                 printk(KERN_INFO "btrfs: %d enospc errors during balance\n",
3008                        enospc_errors);
3009                 if (!ret)
3010                         ret = -ENOSPC;
3011         }
3012
3013         return ret;
3014 }
3015
3016 /**
3017  * alloc_profile_is_valid - see if a given profile is valid and reduced
3018  * @flags: profile to validate
3019  * @extended: if true @flags is treated as an extended profile
3020  */
3021 static int alloc_profile_is_valid(u64 flags, int extended)
3022 {
3023         u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK :
3024                                BTRFS_BLOCK_GROUP_PROFILE_MASK);
3025
3026         flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK;
3027
3028         /* 1) check that all other bits are zeroed */
3029         if (flags & ~mask)
3030                 return 0;
3031
3032         /* 2) see if profile is reduced */
3033         if (flags == 0)
3034                 return !extended; /* "0" is valid for usual profiles */
3035
3036         /* true if exactly one bit set */
3037         return (flags & (flags - 1)) == 0;
3038 }
3039
3040 static inline int balance_need_close(struct btrfs_fs_info *fs_info)
3041 {
3042         /* cancel requested || normal exit path */
3043         return atomic_read(&fs_info->balance_cancel_req) ||
3044                 (atomic_read(&fs_info->balance_pause_req) == 0 &&
3045                  atomic_read(&fs_info->balance_cancel_req) == 0);
3046 }
3047
3048 static void __cancel_balance(struct btrfs_fs_info *fs_info)
3049 {
3050         int ret;
3051
3052         unset_balance_control(fs_info);
3053         ret = del_balance_item(fs_info->tree_root);
3054         BUG_ON(ret);
3055
3056         atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3057 }
3058
3059 void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock,
3060                                struct btrfs_ioctl_balance_args *bargs);
3061
3062 /*
3063  * Should be called with both balance and volume mutexes held
3064  */
3065 int btrfs_balance(struct btrfs_balance_control *bctl,
3066                   struct btrfs_ioctl_balance_args *bargs)
3067 {
3068         struct btrfs_fs_info *fs_info = bctl->fs_info;
3069         u64 allowed;
3070         int mixed = 0;
3071         int ret;
3072         u64 num_devices;
3073         unsigned seq;
3074
3075         if (btrfs_fs_closing(fs_info) ||
3076             atomic_read(&fs_info->balance_pause_req) ||
3077             atomic_read(&fs_info->balance_cancel_req)) {
3078                 ret = -EINVAL;
3079                 goto out;
3080         }
3081
3082         allowed = btrfs_super_incompat_flags(fs_info->super_copy);
3083         if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS)
3084                 mixed = 1;
3085
3086         /*
3087          * In case of mixed groups both data and meta should be picked,
3088          * and identical options should be given for both of them.
3089          */
3090         allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA;
3091         if (mixed && (bctl->flags & allowed)) {
3092                 if (!(bctl->flags & BTRFS_BALANCE_DATA) ||
3093                     !(bctl->flags & BTRFS_BALANCE_METADATA) ||
3094                     memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) {
3095                         printk(KERN_ERR "btrfs: with mixed groups data and "
3096                                "metadata balance options must be the same\n");
3097                         ret = -EINVAL;
3098                         goto out;
3099                 }
3100         }
3101
3102         num_devices = fs_info->fs_devices->num_devices;
3103         btrfs_dev_replace_lock(&fs_info->dev_replace);
3104         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) {
3105                 BUG_ON(num_devices < 1);
3106                 num_devices--;
3107         }
3108         btrfs_dev_replace_unlock(&fs_info->dev_replace);
3109         allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE;
3110         if (num_devices == 1)
3111                 allowed |= BTRFS_BLOCK_GROUP_DUP;
3112         else if (num_devices < 4)
3113                 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1);
3114         else
3115                 allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
3116                                 BTRFS_BLOCK_GROUP_RAID10 |
3117                                 BTRFS_BLOCK_GROUP_RAID5 |
3118                                 BTRFS_BLOCK_GROUP_RAID6);
3119
3120         if ((bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3121             (!alloc_profile_is_valid(bctl->data.target, 1) ||
3122              (bctl->data.target & ~allowed))) {
3123                 printk(KERN_ERR "btrfs: unable to start balance with target "
3124                        "data profile %llu\n",
3125                        (unsigned long long)bctl->data.target);
3126                 ret = -EINVAL;
3127                 goto out;
3128         }
3129         if ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3130             (!alloc_profile_is_valid(bctl->meta.target, 1) ||
3131              (bctl->meta.target & ~allowed))) {
3132                 printk(KERN_ERR "btrfs: unable to start balance with target "
3133                        "metadata profile %llu\n",
3134                        (unsigned long long)bctl->meta.target);
3135                 ret = -EINVAL;
3136                 goto out;
3137         }
3138         if ((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3139             (!alloc_profile_is_valid(bctl->sys.target, 1) ||
3140              (bctl->sys.target & ~allowed))) {
3141                 printk(KERN_ERR "btrfs: unable to start balance with target "
3142                        "system profile %llu\n",
3143                        (unsigned long long)bctl->sys.target);
3144                 ret = -EINVAL;
3145                 goto out;
3146         }
3147
3148         /* allow dup'ed data chunks only in mixed mode */
3149         if (!mixed && (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3150             (bctl->data.target & BTRFS_BLOCK_GROUP_DUP)) {
3151                 printk(KERN_ERR "btrfs: dup for data is not allowed\n");
3152                 ret = -EINVAL;
3153                 goto out;
3154         }
3155
3156         /* allow to reduce meta or sys integrity only if force set */
3157         allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 |
3158                         BTRFS_BLOCK_GROUP_RAID10 |
3159                         BTRFS_BLOCK_GROUP_RAID5 |
3160                         BTRFS_BLOCK_GROUP_RAID6;
3161         do {
3162                 seq = read_seqbegin(&fs_info->profiles_lock);
3163
3164                 if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3165                      (fs_info->avail_system_alloc_bits & allowed) &&
3166                      !(bctl->sys.target & allowed)) ||
3167                     ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) &&
3168                      (fs_info->avail_metadata_alloc_bits & allowed) &&
3169                      !(bctl->meta.target & allowed))) {
3170                         if (bctl->flags & BTRFS_BALANCE_FORCE) {
3171                                 printk(KERN_INFO "btrfs: force reducing metadata "
3172                                        "integrity\n");
3173                         } else {
3174                                 printk(KERN_ERR "btrfs: balance will reduce metadata "
3175                                        "integrity, use force if you want this\n");
3176                                 ret = -EINVAL;
3177                                 goto out;
3178                         }
3179                 }
3180         } while (read_seqretry(&fs_info->profiles_lock, seq));
3181
3182         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3183                 int num_tolerated_disk_barrier_failures;
3184                 u64 target = bctl->sys.target;
3185
3186                 num_tolerated_disk_barrier_failures =
3187                         btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3188                 if (num_tolerated_disk_barrier_failures > 0 &&
3189                     (target &
3190                      (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 |
3191                       BTRFS_AVAIL_ALLOC_BIT_SINGLE)))
3192                         num_tolerated_disk_barrier_failures = 0;
3193                 else if (num_tolerated_disk_barrier_failures > 1 &&
3194                          (target &
3195                           (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)))
3196                         num_tolerated_disk_barrier_failures = 1;
3197
3198                 fs_info->num_tolerated_disk_barrier_failures =
3199                         num_tolerated_disk_barrier_failures;
3200         }
3201
3202         ret = insert_balance_item(fs_info->tree_root, bctl);
3203         if (ret && ret != -EEXIST)
3204                 goto out;
3205
3206         if (!(bctl->flags & BTRFS_BALANCE_RESUME)) {
3207                 BUG_ON(ret == -EEXIST);
3208                 set_balance_control(bctl);
3209         } else {
3210                 BUG_ON(ret != -EEXIST);
3211                 spin_lock(&fs_info->balance_lock);
3212                 update_balance_args(bctl);
3213                 spin_unlock(&fs_info->balance_lock);
3214         }
3215
3216         atomic_inc(&fs_info->balance_running);
3217         mutex_unlock(&fs_info->balance_mutex);
3218
3219         ret = __btrfs_balance(fs_info);
3220
3221         mutex_lock(&fs_info->balance_mutex);
3222         atomic_dec(&fs_info->balance_running);
3223
3224         if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) {
3225                 fs_info->num_tolerated_disk_barrier_failures =
3226                         btrfs_calc_num_tolerated_disk_barrier_failures(fs_info);
3227         }
3228
3229         if (bargs) {
3230                 memset(bargs, 0, sizeof(*bargs));
3231                 update_ioctl_balance_args(fs_info, 0, bargs);
3232         }
3233
3234         wake_up(&fs_info->balance_wait_q);
3235
3236         return ret;
3237 out:
3238         if (bctl->flags & BTRFS_BALANCE_RESUME)
3239                 __cancel_balance(fs_info);
3240         else {
3241                 kfree(bctl);
3242                 atomic_set(&fs_info->mutually_exclusive_operation_running, 0);
3243         }
3244         return ret;
3245 }
3246
3247 static int balance_kthread(void *data)
3248 {
3249         struct btrfs_fs_info *fs_info = data;
3250         int ret = 0;
3251
3252         mutex_lock(&fs_info->volume_mutex);
3253         mutex_lock(&fs_info->balance_mutex);
3254
3255         if (fs_info->balance_ctl) {
3256                 printk(KERN_INFO "btrfs: continuing balance\n");
3257                 ret = btrfs_balance(fs_info->balance_ctl, NULL);
3258         }
3259
3260         mutex_unlock(&fs_info->balance_mutex);
3261         mutex_unlock(&fs_info->volume_mutex);
3262
3263         return ret;
3264 }
3265
3266 int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info)
3267 {
3268         struct task_struct *tsk;
3269
3270         spin_lock(&fs_info->balance_lock);
3271         if (!fs_info->balance_ctl) {
3272                 spin_unlock(&fs_info->balance_lock);
3273                 return 0;
3274         }
3275         spin_unlock(&fs_info->balance_lock);
3276
3277         if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) {
3278                 printk(KERN_INFO "btrfs: force skipping balance\n");
3279                 return 0;
3280         }
3281
3282         tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance");
3283         if (IS_ERR(tsk))
3284                 return PTR_ERR(tsk);
3285
3286         return 0;
3287 }
3288
3289 int btrfs_recover_balance(struct btrfs_fs_info *fs_info)
3290 {
3291         struct btrfs_balance_control *bctl;
3292         struct btrfs_balance_item *item;
3293         struct btrfs_disk_balance_args disk_bargs;
3294         struct btrfs_path *path;
3295         struct extent_buffer *leaf;
3296         struct btrfs_key key;
3297         int ret;
3298
3299         path = btrfs_alloc_path();
3300         if (!path)
3301                 return -ENOMEM;
3302
3303         key.objectid = BTRFS_BALANCE_OBJECTID;
3304         key.type = BTRFS_BALANCE_ITEM_KEY;
3305         key.offset = 0;
3306
3307         ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0);
3308         if (ret < 0)
3309                 goto out;
3310         if (ret > 0) { /* ret = -ENOENT; */
3311                 ret = 0;
3312                 goto out;
3313         }
3314
3315         bctl = kzalloc(sizeof(*bctl), GFP_NOFS);
3316         if (!bctl) {
3317                 ret = -ENOMEM;
3318                 goto out;
3319         }
3320
3321         leaf = path->nodes[0];
3322         item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item);
3323
3324         bctl->fs_info = fs_info;
3325         bctl->flags = btrfs_balance_flags(leaf, item);
3326         bctl->flags |= BTRFS_BALANCE_RESUME;
3327
3328         btrfs_balance_data(leaf, item, &disk_bargs);
3329         btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs);
3330         btrfs_balance_meta(leaf, item, &disk_bargs);
3331         btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs);
3332         btrfs_balance_sys(leaf, item, &disk_bargs);
3333         btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs);
3334
3335         WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1));
3336
3337         mutex_lock(&fs_info->volume_mutex);
3338         mutex_lock(&fs_info->balance_mutex);
3339
3340         set_balance_control(bctl);
3341
3342         mutex_unlock(&fs_info->balance_mutex);
3343         mutex_unlock(&fs_info->volume_mutex);
3344 out:
3345         btrfs_free_path(path);
3346         return ret;
3347 }
3348
3349 int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
3350 {
3351         int ret = 0;
3352
3353         mutex_lock(&fs_info->balance_mutex);
3354         if (!fs_info->balance_ctl) {
3355                 mutex_unlock(&fs_info->balance_mutex);
3356                 return -ENOTCONN;
3357         }
3358
3359         if (atomic_read(&fs_info->balance_running)) {
3360                 atomic_inc(&fs_info->balance_pause_req);
3361                 mutex_unlock(&fs_info->balance_mutex);
3362
3363                 wait_event(fs_info->balance_wait_q,
3364                            atomic_read(&fs_info->balance_running) == 0);
3365
3366                 mutex_lock(&fs_info->balance_mutex);
3367                 /* we are good with balance_ctl ripped off from under us */
3368                 BUG_ON(atomic_read(&fs_info->balance_running));
3369                 atomic_dec(&fs_info->balance_pause_req);
3370         } else {
3371                 ret = -ENOTCONN;
3372         }
3373
3374         mutex_unlock(&fs_info->balance_mutex);
3375         return ret;
3376 }
3377
3378 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
3379 {
3380         mutex_lock(&fs_info->balance_mutex);
3381         if (!fs_info->balance_ctl) {
3382                 mutex_unlock(&fs_info->balance_mutex);
3383                 return -ENOTCONN;
3384         }
3385
3386         atomic_inc(&fs_info->balance_cancel_req);
3387         /*
3388          * if we are running just wait and return, balance item is
3389          * deleted in btrfs_balance in this case
3390          */
3391         if (atomic_read(&fs_info->balance_running)) {
3392                 mutex_unlock(&fs_info->balance_mutex);
3393                 wait_event(fs_info->balance_wait_q,
3394                            atomic_read(&fs_info->balance_running) == 0);
3395                 mutex_lock(&fs_info->balance_mutex);
3396         } else {
3397                 /* __cancel_balance needs volume_mutex */
3398                 mutex_unlock(&fs_info->balance_mutex);
3399                 mutex_lock(&fs_info->volume_mutex);
3400                 mutex_lock(&fs_info->balance_mutex);
3401
3402                 if (fs_info->balance_ctl)
3403                         __cancel_balance(fs_info);
3404
3405                 mutex_unlock(&fs_info->volume_mutex);
3406         }
3407
3408         BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running));
3409         atomic_dec(&fs_info->balance_cancel_req);
3410         mutex_unlock(&fs_info->balance_mutex);
3411         return 0;
3412 }
3413
3414 /*
3415  * shrinking a device means finding all of the device extents past
3416  * the new size, and then following the back refs to the chunks.
3417  * The chunk relocation code actually frees the device extent
3418  */
3419 int btrfs_shrink_device(struct btrfs_device *device, u64 new_size)
3420 {
3421         struct btrfs_trans_handle *trans;
3422         struct btrfs_root *root = device->dev_root;
3423         struct btrfs_dev_extent *dev_extent = NULL;
3424         struct btrfs_path *path;
3425         u64 length;
3426         u64 chunk_tree;
3427         u64 chunk_objectid;
3428         u64 chunk_offset;
3429         int ret;
3430         int slot;
3431         int failed = 0;
3432         bool retried = false;
3433         struct extent_buffer *l;
3434         struct btrfs_key key;
3435         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3436         u64 old_total = btrfs_super_total_bytes(super_copy);
3437         u64 old_size = device->total_bytes;
3438         u64 diff = device->total_bytes - new_size;
3439
3440         if (device->is_tgtdev_for_dev_replace)
3441                 return -EINVAL;
3442
3443         path = btrfs_alloc_path();
3444         if (!path)
3445                 return -ENOMEM;
3446
3447         path->reada = 2;
3448
3449         lock_chunks(root);
3450
3451         device->total_bytes = new_size;
3452         if (device->writeable) {
3453                 device->fs_devices->total_rw_bytes -= diff;
3454                 spin_lock(&root->fs_info->free_chunk_lock);
3455                 root->fs_info->free_chunk_space -= diff;
3456                 spin_unlock(&root->fs_info->free_chunk_lock);
3457         }
3458         unlock_chunks(root);
3459
3460 again:
3461         key.objectid = device->devid;
3462         key.offset = (u64)-1;
3463         key.type = BTRFS_DEV_EXTENT_KEY;
3464
3465         do {
3466                 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3467                 if (ret < 0)
3468                         goto done;
3469
3470                 ret = btrfs_previous_item(root, path, 0, key.type);
3471                 if (ret < 0)
3472                         goto done;
3473                 if (ret) {
3474                         ret = 0;
3475                         btrfs_release_path(path);
3476                         break;
3477                 }
3478
3479                 l = path->nodes[0];
3480                 slot = path->slots[0];
3481                 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
3482
3483                 if (key.objectid != device->devid) {
3484                         btrfs_release_path(path);
3485                         break;
3486                 }
3487
3488                 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
3489                 length = btrfs_dev_extent_length(l, dev_extent);
3490
3491                 if (key.offset + length <= new_size) {
3492                         btrfs_release_path(path);
3493                         break;
3494                 }
3495
3496                 chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent);
3497                 chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent);
3498                 chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent);
3499                 btrfs_release_path(path);
3500
3501                 ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid,
3502                                            chunk_offset);
3503                 if (ret && ret != -ENOSPC)
3504                         goto done;
3505                 if (ret == -ENOSPC)
3506                         failed++;
3507         } while (key.offset-- > 0);
3508
3509         if (failed && !retried) {
3510                 failed = 0;
3511                 retried = true;
3512                 goto again;
3513         } else if (failed && retried) {
3514                 ret = -ENOSPC;
3515                 lock_chunks(root);
3516
3517                 device->total_bytes = old_size;
3518                 if (device->writeable)
3519                         device->fs_devices->total_rw_bytes += diff;
3520                 spin_lock(&root->fs_info->free_chunk_lock);
3521                 root->fs_info->free_chunk_space += diff;
3522                 spin_unlock(&root->fs_info->free_chunk_lock);
3523                 unlock_chunks(root);
3524                 goto done;
3525         }
3526
3527         /* Shrinking succeeded, else we would be at "done". */
3528         trans = btrfs_start_transaction(root, 0);
3529         if (IS_ERR(trans)) {
3530                 ret = PTR_ERR(trans);
3531                 goto done;
3532         }
3533
3534         lock_chunks(root);
3535
3536         device->disk_total_bytes = new_size;
3537         /* Now btrfs_update_device() will change the on-disk size. */
3538         ret = btrfs_update_device(trans, device);
3539         if (ret) {
3540                 unlock_chunks(root);
3541                 btrfs_end_transaction(trans, root);
3542                 goto done;
3543         }
3544         WARN_ON(diff > old_total);
3545         btrfs_set_super_total_bytes(super_copy, old_total - diff);
3546         unlock_chunks(root);
3547         btrfs_end_transaction(trans, root);
3548 done:
3549         btrfs_free_path(path);
3550         return ret;
3551 }
3552
3553 static int btrfs_add_system_chunk(struct btrfs_root *root,
3554                            struct btrfs_key *key,
3555                            struct btrfs_chunk *chunk, int item_size)
3556 {
3557         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
3558         struct btrfs_disk_key disk_key;
3559         u32 array_size;
3560         u8 *ptr;
3561
3562         array_size = btrfs_super_sys_array_size(super_copy);
3563         if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
3564                 return -EFBIG;
3565
3566         ptr = super_copy->sys_chunk_array + array_size;
3567         btrfs_cpu_key_to_disk(&disk_key, key);
3568         memcpy(ptr, &disk_key, sizeof(disk_key));
3569         ptr += sizeof(disk_key);
3570         memcpy(ptr, chunk, item_size);
3571         item_size += sizeof(disk_key);
3572         btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
3573         return 0;
3574 }
3575
3576 /*
3577  * sort the devices in descending order by max_avail, total_avail
3578  */
3579 static int btrfs_cmp_device_info(const void *a, const void *b)
3580 {
3581         const struct btrfs_device_info *di_a = a;
3582         const struct btrfs_device_info *di_b = b;
3583
3584         if (di_a->max_avail > di_b->max_avail)
3585                 return -1;
3586         if (di_a->max_avail < di_b->max_avail)
3587                 return 1;
3588         if (di_a->total_avail > di_b->total_avail)
3589                 return -1;
3590         if (di_a->total_avail < di_b->total_avail)
3591                 return 1;
3592         return 0;
3593 }
3594
3595 struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = {
3596         [BTRFS_RAID_RAID10] = {
3597                 .sub_stripes    = 2,
3598                 .dev_stripes    = 1,
3599                 .devs_max       = 0,    /* 0 == as many as possible */
3600                 .devs_min       = 4,
3601                 .devs_increment = 2,
3602                 .ncopies        = 2,
3603         },
3604         [BTRFS_RAID_RAID1] = {
3605                 .sub_stripes    = 1,
3606                 .dev_stripes    = 1,
3607                 .devs_max       = 2,
3608                 .devs_min       = 2,
3609                 .devs_increment = 2,
3610                 .ncopies        = 2,
3611         },
3612         [BTRFS_RAID_DUP] = {
3613                 .sub_stripes    = 1,
3614                 .dev_stripes    = 2,
3615                 .devs_max       = 1,
3616                 .devs_min       = 1,
3617                 .devs_increment = 1,
3618                 .ncopies        = 2,
3619         },
3620         [BTRFS_RAID_RAID0] = {
3621                 .sub_stripes    = 1,
3622                 .dev_stripes    = 1,
3623                 .devs_max       = 0,
3624                 .devs_min       = 2,
3625                 .devs_increment = 1,
3626                 .ncopies        = 1,
3627         },
3628         [BTRFS_RAID_SINGLE] = {
3629                 .sub_stripes    = 1,
3630                 .dev_stripes    = 1,
3631                 .devs_max       = 1,
3632                 .devs_min       = 1,
3633                 .devs_increment = 1,
3634                 .ncopies        = 1,
3635         },
3636         [BTRFS_RAID_RAID5] = {
3637                 .sub_stripes    = 1,
3638                 .dev_stripes    = 1,
3639                 .devs_max       = 0,
3640                 .devs_min       = 2,
3641                 .devs_increment = 1,
3642                 .ncopies        = 2,
3643         },
3644         [BTRFS_RAID_RAID6] = {
3645                 .sub_stripes    = 1,
3646                 .dev_stripes    = 1,
3647                 .devs_max       = 0,
3648                 .devs_min       = 3,
3649                 .devs_increment = 1,
3650                 .ncopies        = 3,
3651         },
3652 };
3653
3654 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
3655 {
3656         /* TODO allow them to set a preferred stripe size */
3657         return 64 * 1024;
3658 }
3659
3660 static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type)
3661 {
3662         u64 features;
3663
3664         if (!(type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)))
3665                 return;
3666
3667         features = btrfs_super_incompat_flags(info->super_copy);
3668         if (features & BTRFS_FEATURE_INCOMPAT_RAID56)
3669                 return;
3670
3671         features |= BTRFS_FEATURE_INCOMPAT_RAID56;
3672         btrfs_set_super_incompat_flags(info->super_copy, features);
3673         printk(KERN_INFO "btrfs: setting RAID5/6 feature flag\n");
3674 }
3675
3676 static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
3677                                struct btrfs_root *extent_root,
3678                                struct map_lookup **map_ret,
3679                                u64 *num_bytes_out, u64 *stripe_size_out,
3680                                u64 start, u64 type)
3681 {
3682         struct btrfs_fs_info *info = extent_root->fs_info;
3683         struct btrfs_fs_devices *fs_devices = info->fs_devices;
3684         struct list_head *cur;
3685         struct map_lookup *map = NULL;
3686         struct extent_map_tree *em_tree;
3687         struct extent_map *em;
3688         struct btrfs_device_info *devices_info = NULL;
3689         u64 total_avail;
3690         int num_stripes;        /* total number of stripes to allocate */
3691         int data_stripes;       /* number of stripes that count for
3692                                    block group size */
3693         int sub_stripes;        /* sub_stripes info for map */
3694         int dev_stripes;        /* stripes per dev */
3695         int devs_max;           /* max devs to use */
3696         int devs_min;           /* min devs needed */
3697         int devs_increment;     /* ndevs has to be a multiple of this */
3698         int ncopies;            /* how many copies to data has */
3699         int ret;
3700         u64 max_stripe_size;
3701         u64 max_chunk_size;
3702         u64 stripe_size;
3703         u64 num_bytes;
3704         u64 raid_stripe_len = BTRFS_STRIPE_LEN;
3705         int ndevs;
3706         int i;
3707         int j;
3708         int index;
3709
3710         BUG_ON(!alloc_profile_is_valid(type, 0));
3711
3712         if (list_empty(&fs_devices->alloc_list))
3713                 return -ENOSPC;
3714
3715         index = __get_raid_index(type);
3716
3717         sub_stripes = btrfs_raid_array[index].sub_stripes;
3718         dev_stripes = btrfs_raid_array[index].dev_stripes;
3719         devs_max = btrfs_raid_array[index].devs_max;
3720         devs_min = btrfs_raid_array[index].devs_min;
3721         devs_increment = btrfs_raid_array[index].devs_increment;
3722         ncopies = btrfs_raid_array[index].ncopies;
3723
3724         if (type & BTRFS_BLOCK_GROUP_DATA) {
3725                 max_stripe_size = 1024 * 1024 * 1024;
3726                 max_chunk_size = 10 * max_stripe_size;
3727         } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
3728                 /* for larger filesystems, use larger metadata chunks */
3729                 if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024)
3730                         max_stripe_size = 1024 * 1024 * 1024;
3731                 else
3732                         max_stripe_size = 256 * 1024 * 1024;
3733                 max_chunk_size = max_stripe_size;
3734         } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
3735                 max_stripe_size = 32 * 1024 * 1024;
3736                 max_chunk_size = 2 * max_stripe_size;
3737         } else {
3738                 printk(KERN_ERR "btrfs: invalid chunk type 0x%llx requested\n",
3739                        type);
3740                 BUG_ON(1);
3741         }
3742
3743         /* we don't want a chunk larger than 10% of writeable space */
3744         max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1),
3745                              max_chunk_size);
3746
3747         devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices,
3748                                GFP_NOFS);
3749         if (!devices_info)
3750                 return -ENOMEM;
3751
3752         cur = fs_devices->alloc_list.next;
3753
3754         /*
3755          * in the first pass through the devices list, we gather information
3756          * about the available holes on each device.
3757          */
3758         ndevs = 0;
3759         while (cur != &fs_devices->alloc_list) {
3760                 struct btrfs_device *device;
3761                 u64 max_avail;
3762                 u64 dev_offset;
3763
3764                 device = list_entry(cur, struct btrfs_device, dev_alloc_list);
3765
3766                 cur = cur->next;
3767
3768                 if (!device->writeable) {
3769                         WARN(1, KERN_ERR
3770                                "btrfs: read-only device in alloc_list\n");
3771                         continue;
3772                 }
3773
3774                 if (!device->in_fs_metadata ||
3775                     device->is_tgtdev_for_dev_replace)
3776                         continue;
3777
3778                 if (device->total_bytes > device->bytes_used)
3779                         total_avail = device->total_bytes - device->bytes_used;
3780                 else
3781                         total_avail = 0;
3782
3783                 /* If there is no space on this device, skip it. */
3784                 if (total_avail == 0)
3785                         continue;
3786
3787                 ret = find_free_dev_extent(device,
3788                                            max_stripe_size * dev_stripes,
3789                                            &dev_offset, &max_avail);
3790                 if (ret && ret != -ENOSPC)
3791                         goto error;
3792
3793                 if (ret == 0)
3794                         max_avail = max_stripe_size * dev_stripes;
3795
3796                 if (max_avail < BTRFS_STRIPE_LEN * dev_stripes)
3797                         continue;
3798
3799                 if (ndevs == fs_devices->rw_devices) {
3800                         WARN(1, "%s: found more than %llu devices\n",
3801                              __func__, fs_devices->rw_devices);
3802                         break;
3803                 }
3804                 devices_info[ndevs].dev_offset = dev_offset;
3805                 devices_info[ndevs].max_avail = max_avail;
3806                 devices_info[ndevs].total_avail = total_avail;
3807                 devices_info[ndevs].dev = device;
3808                 ++ndevs;
3809         }
3810
3811         /*
3812          * now sort the devices by hole size / available space
3813          */
3814         sort(devices_info, ndevs, sizeof(struct btrfs_device_info),
3815              btrfs_cmp_device_info, NULL);
3816
3817         /* round down to number of usable stripes */
3818         ndevs -= ndevs % devs_increment;
3819
3820         if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) {
3821                 ret = -ENOSPC;
3822                 goto error;
3823         }
3824
3825         if (devs_max && ndevs > devs_max)
3826                 ndevs = devs_max;
3827         /*
3828          * the primary goal is to maximize the number of stripes, so use as many
3829          * devices as possible, even if the stripes are not maximum sized.
3830          */
3831         stripe_size = devices_info[ndevs-1].max_avail;
3832         num_stripes = ndevs * dev_stripes;
3833
3834         /*
3835          * this will have to be fixed for RAID1 and RAID10 over
3836          * more drives
3837          */
3838         data_stripes = num_stripes / ncopies;
3839
3840         if (stripe_size * ndevs > max_chunk_size * ncopies) {
3841                 stripe_size = max_chunk_size * ncopies;
3842                 do_div(stripe_size, ndevs);
3843         }
3844         if (type & BTRFS_BLOCK_GROUP_RAID5) {
3845                 raid_stripe_len = find_raid56_stripe_len(ndevs - 1,
3846                                  btrfs_super_stripesize(info->super_copy));
3847                 data_stripes = num_stripes - 1;
3848         }
3849         if (type & BTRFS_BLOCK_GROUP_RAID6) {
3850                 raid_stripe_len = find_raid56_stripe_len(ndevs - 2,
3851                                  btrfs_super_stripesize(info->super_copy));
3852                 data_stripes = num_stripes - 2;
3853         }
3854         do_div(stripe_size, dev_stripes);
3855
3856         /* align to BTRFS_STRIPE_LEN */
3857         do_div(stripe_size, raid_stripe_len);
3858         stripe_size *= raid_stripe_len;
3859
3860         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
3861         if (!map) {
3862                 ret = -ENOMEM;
3863                 goto error;
3864         }
3865         map->num_stripes = num_stripes;
3866
3867         for (i = 0; i < ndevs; ++i) {
3868                 for (j = 0; j < dev_stripes; ++j) {
3869                         int s = i * dev_stripes + j;
3870                         map->stripes[s].dev = devices_info[i].dev;
3871                         map->stripes[s].physical = devices_info[i].dev_offset +
3872                                                    j * stripe_size;
3873                 }
3874         }
3875         map->sector_size = extent_root->sectorsize;
3876         map->stripe_len = raid_stripe_len;
3877         map->io_align = raid_stripe_len;
3878         map->io_width = raid_stripe_len;
3879         map->type = type;
3880         map->sub_stripes = sub_stripes;
3881
3882         *map_ret = map;
3883         num_bytes = stripe_size * data_stripes;
3884
3885         *stripe_size_out = stripe_size;
3886         *num_bytes_out = num_bytes;
3887
3888         trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes);
3889
3890         em = alloc_extent_map();
3891         if (!em) {
3892                 ret = -ENOMEM;
3893                 goto error;
3894         }
3895         em->bdev = (struct block_device *)map;
3896         em->start = start;
3897         em->len = num_bytes;
3898         em->block_start = 0;
3899         em->block_len = em->len;
3900
3901         em_tree = &extent_root->fs_info->mapping_tree.map_tree;
3902         write_lock(&em_tree->lock);
3903         ret = add_extent_mapping(em_tree, em);
3904         write_unlock(&em_tree->lock);
3905         if (ret) {
3906                 free_extent_map(em);
3907                 goto error;
3908         }
3909
3910         for (i = 0; i < map->num_stripes; ++i) {
3911                 struct btrfs_device *device;
3912                 u64 dev_offset;
3913
3914                 device = map->stripes[i].dev;
3915                 dev_offset = map->stripes[i].physical;
3916
3917                 ret = btrfs_alloc_dev_extent(trans, device,
3918                                 info->chunk_root->root_key.objectid,
3919                                 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3920                                 start, dev_offset, stripe_size);
3921                 if (ret)
3922                         goto error_dev_extent;
3923         }
3924
3925         ret = btrfs_make_block_group(trans, extent_root, 0, type,
3926                                      BTRFS_FIRST_CHUNK_TREE_OBJECTID,
3927                                      start, num_bytes);
3928         if (ret) {
3929                 i = map->num_stripes - 1;
3930                 goto error_dev_extent;
3931         }
3932
3933         free_extent_map(em);
3934         check_raid56_incompat_flag(extent_root->fs_info, type);
3935
3936         kfree(devices_info);
3937         return 0;
3938
3939 error_dev_extent:
3940         for (; i >= 0; i--) {
3941                 struct btrfs_device *device;
3942                 int err;
3943
3944                 device = map->stripes[i].dev;
3945                 err = btrfs_free_dev_extent(trans, device, start);
3946                 if (err) {
3947                         btrfs_abort_transaction(trans, extent_root, err);
3948                         break;
3949                 }
3950         }
3951         write_lock(&em_tree->lock);
3952         remove_extent_mapping(em_tree, em);
3953         write_unlock(&em_tree->lock);
3954
3955         /* One for our allocation */
3956         free_extent_map(em);
3957         /* One for the tree reference */
3958         free_extent_map(em);
3959 error:
3960         kfree(map);
3961         kfree(devices_info);
3962         return ret;
3963 }
3964
3965 static int __finish_chunk_alloc(struct btrfs_trans_handle *trans,
3966                                 struct btrfs_root *extent_root,
3967                                 struct map_lookup *map, u64 chunk_offset,
3968                                 u64 chunk_size, u64 stripe_size)
3969 {
3970         u64 dev_offset;
3971         struct btrfs_key key;
3972         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
3973         struct btrfs_device *device;
3974         struct btrfs_chunk *chunk;
3975         struct btrfs_stripe *stripe;
3976         size_t item_size = btrfs_chunk_item_size(map->num_stripes);
3977         int index = 0;
3978         int ret;
3979
3980         chunk = kzalloc(item_size, GFP_NOFS);
3981         if (!chunk)
3982                 return -ENOMEM;
3983
3984         index = 0;
3985         while (index < map->num_stripes) {
3986                 device = map->stripes[index].dev;
3987                 device->bytes_used += stripe_size;
3988                 ret = btrfs_update_device(trans, device);
3989                 if (ret)
3990                         goto out_free;
3991                 index++;
3992         }
3993
3994         spin_lock(&extent_root->fs_info->free_chunk_lock);
3995         extent_root->fs_info->free_chunk_space -= (stripe_size *
3996                                                    map->num_stripes);
3997         spin_unlock(&extent_root->fs_info->free_chunk_lock);
3998
3999         index = 0;
4000         stripe = &chunk->stripe;
4001         while (index < map->num_stripes) {
4002                 device = map->stripes[index].dev;
4003                 dev_offset = map->stripes[index].physical;
4004
4005                 btrfs_set_stack_stripe_devid(stripe, device->devid);
4006                 btrfs_set_stack_stripe_offset(stripe, dev_offset);
4007                 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
4008                 stripe++;
4009                 index++;
4010         }
4011
4012         btrfs_set_stack_chunk_length(chunk, chunk_size);
4013         btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
4014         btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len);
4015         btrfs_set_stack_chunk_type(chunk, map->type);
4016         btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes);
4017         btrfs_set_stack_chunk_io_align(chunk, map->stripe_len);
4018         btrfs_set_stack_chunk_io_width(chunk, map->stripe_len);
4019         btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
4020         btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes);
4021
4022         key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
4023         key.type = BTRFS_CHUNK_ITEM_KEY;
4024         key.offset = chunk_offset;
4025
4026         ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size);
4027
4028         if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) {
4029                 /*
4030                  * TODO: Cleanup of inserted chunk root in case of
4031                  * failure.
4032                  */
4033                 ret = btrfs_add_system_chunk(chunk_root, &key, chunk,
4034                                              item_size);
4035         }
4036
4037 out_free:
4038         kfree(chunk);
4039         return ret;
4040 }
4041
4042 /*
4043  * Chunk allocation falls into two parts. The first part does works
4044  * that make the new allocated chunk useable, but not do any operation
4045  * that modifies the chunk tree. The second part does the works that
4046  * require modifying the chunk tree. This division is important for the
4047  * bootstrap process of adding storage to a seed btrfs.
4048  */
4049 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
4050                       struct btrfs_root *extent_root, u64 type)
4051 {
4052         u64 chunk_offset;
4053         u64 chunk_size;
4054         u64 stripe_size;
4055         struct map_lookup *map;
4056         struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
4057         int ret;
4058
4059         ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
4060                               &chunk_offset);
4061         if (ret)
4062                 return ret;
4063
4064         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
4065                                   &stripe_size, chunk_offset, type);
4066         if (ret)
4067                 return ret;
4068
4069         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
4070                                    chunk_size, stripe_size);
4071         if (ret)
4072                 return ret;
4073         return 0;
4074 }
4075
4076 static noinline int init_first_rw_device(struct btrfs_trans_handle *trans,
4077                                          struct btrfs_root *root,
4078                                          struct btrfs_device *device)
4079 {
4080         u64 chunk_offset;
4081         u64 sys_chunk_offset;
4082         u64 chunk_size;
4083         u64 sys_chunk_size;
4084         u64 stripe_size;
4085         u64 sys_stripe_size;
4086         u64 alloc_profile;
4087         struct map_lookup *map;
4088         struct map_lookup *sys_map;
4089         struct btrfs_fs_info *fs_info = root->fs_info;
4090         struct btrfs_root *extent_root = fs_info->extent_root;
4091         int ret;
4092
4093         ret = find_next_chunk(fs_info->chunk_root,
4094                               BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset);
4095         if (ret)
4096                 return ret;
4097
4098         alloc_profile = btrfs_get_alloc_profile(extent_root, 0);
4099         ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size,
4100                                   &stripe_size, chunk_offset, alloc_profile);
4101         if (ret)
4102                 return ret;
4103
4104         sys_chunk_offset = chunk_offset + chunk_size;
4105
4106         alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0);
4107         ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map,
4108                                   &sys_chunk_size, &sys_stripe_size,
4109                                   sys_chunk_offset, alloc_profile);
4110         if (ret) {
4111                 btrfs_abort_transaction(trans, root, ret);
4112                 goto out;
4113         }
4114
4115         ret = btrfs_add_device(trans, fs_info->chunk_root, device);
4116         if (ret) {
4117                 btrfs_abort_transaction(trans, root, ret);
4118                 goto out;
4119         }
4120
4121         /*
4122          * Modifying chunk tree needs allocating new blocks from both
4123          * system block group and metadata block group. So we only can
4124          * do operations require modifying the chunk tree after both
4125          * block groups were created.
4126          */
4127         ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset,
4128                                    chunk_size, stripe_size);
4129         if (ret) {
4130                 btrfs_abort_transaction(trans, root, ret);
4131                 goto out;
4132         }
4133
4134         ret = __finish_chunk_alloc(trans, extent_root, sys_map,
4135                                    sys_chunk_offset, sys_chunk_size,
4136                                    sys_stripe_size);
4137         if (ret)
4138                 btrfs_abort_transaction(trans, root, ret);
4139
4140 out:
4141
4142         return ret;
4143 }
4144
4145 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
4146 {
4147         struct extent_map *em;
4148         struct map_lookup *map;
4149         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
4150         int readonly = 0;
4151         int i;
4152
4153         read_lock(&map_tree->map_tree.lock);
4154         em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1);
4155         read_unlock(&map_tree->map_tree.lock);
4156         if (!em)
4157                 return 1;
4158
4159         if (btrfs_test_opt(root, DEGRADED)) {
4160                 free_extent_map(em);
4161                 return 0;
4162         }
4163
4164         map = (struct map_lookup *)em->bdev;
4165         for (i = 0; i < map->num_stripes; i++) {
4166                 if (!map->stripes[i].dev->writeable) {
4167                         readonly = 1;
4168                         break;
4169                 }
4170         }
4171         free_extent_map(em);
4172         return readonly;
4173 }
4174
4175 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
4176 {
4177         extent_map_tree_init(&tree->map_tree);
4178 }
4179
4180 void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree)
4181 {
4182         struct extent_map *em;
4183
4184         while (1) {
4185                 write_lock(&tree->map_tree.lock);
4186                 em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1);
4187                 if (em)
4188                         remove_extent_mapping(&tree->map_tree, em);
4189                 write_unlock(&tree->map_tree.lock);
4190                 if (!em)
4191                         break;
4192                 kfree(em->bdev);
4193                 /* once for us */
4194                 free_extent_map(em);
4195                 /* once for the tree */
4196                 free_extent_map(em);
4197         }
4198 }
4199
4200 int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len)
4201 {
4202         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4203         struct extent_map *em;
4204         struct map_lookup *map;
4205         struct extent_map_tree *em_tree = &map_tree->map_tree;
4206         int ret;
4207
4208         read_lock(&em_tree->lock);
4209         em = lookup_extent_mapping(em_tree, logical, len);
4210         read_unlock(&em_tree->lock);
4211         BUG_ON(!em);
4212
4213         BUG_ON(em->start > logical || em->start + em->len < logical);
4214         map = (struct map_lookup *)em->bdev;
4215         if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
4216                 ret = map->num_stripes;
4217         else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
4218                 ret = map->sub_stripes;
4219         else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
4220                 ret = 2;
4221         else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
4222                 ret = 3;
4223         else
4224                 ret = 1;
4225         free_extent_map(em);
4226
4227         btrfs_dev_replace_lock(&fs_info->dev_replace);
4228         if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace))
4229                 ret++;
4230         btrfs_dev_replace_unlock(&fs_info->dev_replace);
4231
4232         return ret;
4233 }
4234
4235 unsigned long btrfs_full_stripe_len(struct btrfs_root *root,
4236                                     struct btrfs_mapping_tree *map_tree,
4237                                     u64 logical)
4238 {
4239         struct extent_map *em;
4240         struct map_lookup *map;
4241         struct extent_map_tree *em_tree = &map_tree->map_tree;
4242         unsigned long len = root->sectorsize;
4243
4244         read_lock(&em_tree->lock);
4245         em = lookup_extent_mapping(em_tree, logical, len);
4246         read_unlock(&em_tree->lock);
4247         BUG_ON(!em);
4248
4249         BUG_ON(em->start > logical || em->start + em->len < logical);
4250         map = (struct map_lookup *)em->bdev;
4251         if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4252                          BTRFS_BLOCK_GROUP_RAID6)) {
4253                 len = map->stripe_len * nr_data_stripes(map);
4254         }
4255         free_extent_map(em);
4256         return len;
4257 }
4258
4259 int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree,
4260                            u64 logical, u64 len, int mirror_num)
4261 {
4262         struct extent_map *em;
4263         struct map_lookup *map;
4264         struct extent_map_tree *em_tree = &map_tree->map_tree;
4265         int ret = 0;
4266
4267         read_lock(&em_tree->lock);
4268         em = lookup_extent_mapping(em_tree, logical, len);
4269         read_unlock(&em_tree->lock);
4270         BUG_ON(!em);
4271
4272         BUG_ON(em->start > logical || em->start + em->len < logical);
4273         map = (struct map_lookup *)em->bdev;
4274         if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4275                          BTRFS_BLOCK_GROUP_RAID6))
4276                 ret = 1;
4277         free_extent_map(em);
4278         return ret;
4279 }
4280
4281 static int find_live_mirror(struct btrfs_fs_info *fs_info,
4282                             struct map_lookup *map, int first, int num,
4283                             int optimal, int dev_replace_is_ongoing)
4284 {
4285         int i;
4286         int tolerance;
4287         struct btrfs_device *srcdev;
4288
4289         if (dev_replace_is_ongoing &&
4290             fs_info->dev_replace.cont_reading_from_srcdev_mode ==
4291              BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID)
4292                 srcdev = fs_info->dev_replace.srcdev;
4293         else
4294                 srcdev = NULL;
4295
4296         /*
4297          * try to avoid the drive that is the source drive for a
4298          * dev-replace procedure, only choose it if no other non-missing
4299          * mirror is available
4300          */
4301         for (tolerance = 0; tolerance < 2; tolerance++) {
4302                 if (map->stripes[optimal].dev->bdev &&
4303                     (tolerance || map->stripes[optimal].dev != srcdev))
4304                         return optimal;
4305                 for (i = first; i < first + num; i++) {
4306                         if (map->stripes[i].dev->bdev &&
4307                             (tolerance || map->stripes[i].dev != srcdev))
4308                                 return i;
4309                 }
4310         }
4311
4312         /* we couldn't find one that doesn't fail.  Just return something
4313          * and the io error handling code will clean up eventually
4314          */
4315         return optimal;
4316 }
4317
4318 static inline int parity_smaller(u64 a, u64 b)
4319 {
4320         return a > b;
4321 }
4322
4323 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
4324 static void sort_parity_stripes(struct btrfs_bio *bbio, u64 *raid_map)
4325 {
4326         struct btrfs_bio_stripe s;
4327         int i;
4328         u64 l;
4329         int again = 1;
4330
4331         while (again) {
4332                 again = 0;
4333                 for (i = 0; i < bbio->num_stripes - 1; i++) {
4334                         if (parity_smaller(raid_map[i], raid_map[i+1])) {
4335                                 s = bbio->stripes[i];
4336                                 l = raid_map[i];
4337                                 bbio->stripes[i] = bbio->stripes[i+1];
4338                                 raid_map[i] = raid_map[i+1];
4339                                 bbio->stripes[i+1] = s;
4340                                 raid_map[i+1] = l;
4341                                 again = 1;
4342                         }
4343                 }
4344         }
4345 }
4346
4347 static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
4348                              u64 logical, u64 *length,
4349                              struct btrfs_bio **bbio_ret,
4350                              int mirror_num, u64 **raid_map_ret)
4351 {
4352         struct extent_map *em;
4353         struct map_lookup *map;
4354         struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree;
4355         struct extent_map_tree *em_tree = &map_tree->map_tree;
4356         u64 offset;
4357         u64 stripe_offset;
4358         u64 stripe_end_offset;
4359         u64 stripe_nr;
4360         u64 stripe_nr_orig;
4361         u64 stripe_nr_end;
4362         u64 stripe_len;
4363         u64 *raid_map = NULL;
4364         int stripe_index;
4365         int i;
4366         int ret = 0;
4367         int num_stripes;
4368         int max_errors = 0;
4369         struct btrfs_bio *bbio = NULL;
4370         struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace;
4371         int dev_replace_is_ongoing = 0;
4372         int num_alloc_stripes;
4373         int patch_the_first_stripe_for_dev_replace = 0;
4374         u64 physical_to_patch_in_first_stripe = 0;
4375         u64 raid56_full_stripe_start = (u64)-1;
4376
4377         read_lock(&em_tree->lock);
4378         em = lookup_extent_mapping(em_tree, logical, *length);
4379         read_unlock(&em_tree->lock);
4380
4381         if (!em) {
4382                 printk(KERN_CRIT "btrfs: unable to find logical %llu len %llu\n",
4383                        (unsigned long long)logical,
4384                        (unsigned long long)*length);
4385                 BUG();
4386         }
4387
4388         BUG_ON(em->start > logical || em->start + em->len < logical);
4389         map = (struct map_lookup *)em->bdev;
4390         offset = logical - em->start;
4391
4392         if (mirror_num > map->num_stripes)
4393                 mirror_num = 0;
4394
4395         stripe_len = map->stripe_len;
4396         stripe_nr = offset;
4397         /*
4398          * stripe_nr counts the total number of stripes we have to stride
4399          * to get to this block
4400          */
4401         do_div(stripe_nr, stripe_len);
4402
4403         stripe_offset = stripe_nr * stripe_len;
4404         BUG_ON(offset < stripe_offset);
4405
4406         /* stripe_offset is the offset of this block in its stripe*/
4407         stripe_offset = offset - stripe_offset;
4408
4409         /* if we're here for raid56, we need to know the stripe aligned start */
4410         if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
4411                 unsigned long full_stripe_len = stripe_len * nr_data_stripes(map);
4412                 raid56_full_stripe_start = offset;
4413
4414                 /* allow a write of a full stripe, but make sure we don't
4415                  * allow straddling of stripes
4416                  */
4417                 do_div(raid56_full_stripe_start, full_stripe_len);
4418                 raid56_full_stripe_start *= full_stripe_len;
4419         }
4420
4421         if (rw & REQ_DISCARD) {
4422                 /* we don't discard raid56 yet */
4423                 if (map->type &
4424                     (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)) {
4425                         ret = -EOPNOTSUPP;
4426                         goto out;
4427                 }
4428                 *length = min_t(u64, em->len - offset, *length);
4429         } else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) {
4430                 u64 max_len;
4431                 /* For writes to RAID[56], allow a full stripeset across all disks.
4432                    For other RAID types and for RAID[56] reads, just allow a single
4433                    stripe (on a single disk). */
4434                 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6) &&
4435                     (rw & REQ_WRITE)) {
4436                         max_len = stripe_len * nr_data_stripes(map) -
4437                                 (offset - raid56_full_stripe_start);
4438                 } else {
4439                         /* we limit the length of each bio to what fits in a stripe */
4440                         max_len = stripe_len - stripe_offset;
4441                 }
4442                 *length = min_t(u64, em->len - offset, max_len);
4443         } else {
4444                 *length = em->len - offset;
4445         }
4446
4447         /* This is for when we're called from btrfs_merge_bio_hook() and all
4448            it cares about is the length */
4449         if (!bbio_ret)
4450                 goto out;
4451
4452         btrfs_dev_replace_lock(dev_replace);
4453         dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace);
4454         if (!dev_replace_is_ongoing)
4455                 btrfs_dev_replace_unlock(dev_replace);
4456
4457         if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 &&
4458             !(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) &&
4459             dev_replace->tgtdev != NULL) {
4460                 /*
4461                  * in dev-replace case, for repair case (that's the only
4462                  * case where the mirror is selected explicitly when
4463                  * calling btrfs_map_block), blocks left of the left cursor
4464                  * can also be read from the target drive.
4465                  * For REQ_GET_READ_MIRRORS, the target drive is added as
4466                  * the last one to the array of stripes. For READ, it also
4467                  * needs to be supported using the same mirror number.
4468                  * If the requested block is not left of the left cursor,
4469                  * EIO is returned. This can happen because btrfs_num_copies()
4470                  * returns one more in the dev-replace case.
4471                  */
4472                 u64 tmp_length = *length;
4473                 struct btrfs_bio *tmp_bbio = NULL;
4474                 int tmp_num_stripes;
4475                 u64 srcdev_devid = dev_replace->srcdev->devid;
4476                 int index_srcdev = 0;
4477                 int found = 0;
4478                 u64 physical_of_found = 0;
4479
4480                 ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS,
4481                              logical, &tmp_length, &tmp_bbio, 0, NULL);
4482                 if (ret) {
4483                         WARN_ON(tmp_bbio != NULL);
4484                         goto out;
4485                 }
4486
4487                 tmp_num_stripes = tmp_bbio->num_stripes;
4488                 if (mirror_num > tmp_num_stripes) {
4489                         /*
4490                          * REQ_GET_READ_MIRRORS does not contain this
4491                          * mirror, that means that the requested area
4492                          * is not left of the left cursor
4493                          */
4494                         ret = -EIO;
4495                         kfree(tmp_bbio);
4496                         goto out;
4497                 }
4498
4499                 /*
4500                  * process the rest of the function using the mirror_num
4501                  * of the source drive. Therefore look it up first.
4502                  * At the end, patch the device pointer to the one of the
4503                  * target drive.
4504                  */
4505                 for (i = 0; i < tmp_num_stripes; i++) {
4506                         if (tmp_bbio->stripes[i].dev->devid == srcdev_devid) {
4507                                 /*
4508                                  * In case of DUP, in order to keep it
4509                                  * simple, only add the mirror with the
4510                                  * lowest physical address
4511                                  */
4512                                 if (found &&
4513                                     physical_of_found <=
4514                                      tmp_bbio->stripes[i].physical)
4515                                         continue;
4516                                 index_srcdev = i;
4517                                 found = 1;
4518                                 physical_of_found =
4519                                         tmp_bbio->stripes[i].physical;
4520                         }
4521                 }
4522
4523                 if (found) {
4524                         mirror_num = index_srcdev + 1;
4525                         patch_the_first_stripe_for_dev_replace = 1;
4526                         physical_to_patch_in_first_stripe = physical_of_found;
4527                 } else {
4528                         WARN_ON(1);
4529                         ret = -EIO;
4530                         kfree(tmp_bbio);
4531                         goto out;
4532                 }
4533
4534                 kfree(tmp_bbio);
4535         } else if (mirror_num > map->num_stripes) {
4536                 mirror_num = 0;
4537         }
4538
4539         num_stripes = 1;
4540         stripe_index = 0;
4541         stripe_nr_orig = stripe_nr;
4542         stripe_nr_end = (offset + *length + map->stripe_len - 1) &
4543                         (~(map->stripe_len - 1));
4544         do_div(stripe_nr_end, map->stripe_len);
4545         stripe_end_offset = stripe_nr_end * map->stripe_len -
4546                             (offset + *length);
4547
4548         if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
4549                 if (rw & REQ_DISCARD)
4550                         num_stripes = min_t(u64, map->num_stripes,
4551                                             stripe_nr_end - stripe_nr_orig);
4552                 stripe_index = do_div(stripe_nr, map->num_stripes);
4553         } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
4554                 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS))
4555                         num_stripes = map->num_stripes;
4556                 else if (mirror_num)
4557                         stripe_index = mirror_num - 1;
4558                 else {
4559                         stripe_index = find_live_mirror(fs_info, map, 0,
4560                                             map->num_stripes,
4561                                             current->pid % map->num_stripes,
4562                                             dev_replace_is_ongoing);
4563                         mirror_num = stripe_index + 1;
4564                 }
4565
4566         } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
4567                 if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) {
4568                         num_stripes = map->num_stripes;
4569                 } else if (mirror_num) {
4570                         stripe_index = mirror_num - 1;
4571                 } else {
4572                         mirror_num = 1;
4573                 }
4574
4575         } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
4576                 int factor = map->num_stripes / map->sub_stripes;
4577
4578                 stripe_index = do_div(stripe_nr, factor);
4579                 stripe_index *= map->sub_stripes;
4580
4581                 if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS))
4582                         num_stripes = map->sub_stripes;
4583                 else if (rw & REQ_DISCARD)
4584                         num_stripes = min_t(u64, map->sub_stripes *
4585                                             (stripe_nr_end - stripe_nr_orig),
4586                                             map->num_stripes);
4587                 else if (mirror_num)
4588                         stripe_index += mirror_num - 1;
4589                 else {
4590                         int old_stripe_index = stripe_index;
4591                         stripe_index = find_live_mirror(fs_info, map,
4592                                               stripe_index,
4593                                               map->sub_stripes, stripe_index +
4594                                               current->pid % map->sub_stripes,
4595                                               dev_replace_is_ongoing);
4596                         mirror_num = stripe_index - old_stripe_index + 1;
4597                 }
4598
4599         } else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4600                                 BTRFS_BLOCK_GROUP_RAID6)) {
4601                 u64 tmp;
4602
4603                 if (bbio_ret && ((rw & REQ_WRITE) || mirror_num > 1)
4604                     && raid_map_ret) {
4605                         int i, rot;
4606
4607                         /* push stripe_nr back to the start of the full stripe */
4608                         stripe_nr = raid56_full_stripe_start;
4609                         do_div(stripe_nr, stripe_len);
4610
4611                         stripe_index = do_div(stripe_nr, nr_data_stripes(map));
4612
4613                         /* RAID[56] write or recovery. Return all stripes */
4614                         num_stripes = map->num_stripes;
4615                         max_errors = nr_parity_stripes(map);
4616
4617                         raid_map = kmalloc(sizeof(u64) * num_stripes,
4618                                            GFP_NOFS);
4619                         if (!raid_map) {
4620                                 ret = -ENOMEM;
4621                                 goto out;
4622                         }
4623
4624                         /* Work out the disk rotation on this stripe-set */
4625                         tmp = stripe_nr;
4626                         rot = do_div(tmp, num_stripes);
4627
4628                         /* Fill in the logical address of each stripe */
4629                         tmp = stripe_nr * nr_data_stripes(map);
4630                         for (i = 0; i < nr_data_stripes(map); i++)
4631                                 raid_map[(i+rot) % num_stripes] =
4632                                         em->start + (tmp + i) * map->stripe_len;
4633
4634                         raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE;
4635                         if (map->type & BTRFS_BLOCK_GROUP_RAID6)
4636                                 raid_map[(i+rot+1) % num_stripes] =
4637                                         RAID6_Q_STRIPE;
4638
4639                         *length = map->stripe_len;
4640                         stripe_index = 0;
4641                         stripe_offset = 0;
4642                 } else {
4643                         /*
4644                          * Mirror #0 or #1 means the original data block.
4645                          * Mirror #2 is RAID5 parity block.
4646                          * Mirror #3 is RAID6 Q block.
4647                          */
4648                         stripe_index = do_div(stripe_nr, nr_data_stripes(map));
4649                         if (mirror_num > 1)
4650                                 stripe_index = nr_data_stripes(map) +
4651                                                 mirror_num - 2;
4652
4653                         /* We distribute the parity blocks across stripes */
4654                         tmp = stripe_nr + stripe_index;
4655                         stripe_index = do_div(tmp, map->num_stripes);
4656                 }
4657         } else {
4658                 /*
4659                  * after this do_div call, stripe_nr is the number of stripes
4660                  * on this device we have to walk to find the data, and
4661                  * stripe_index is the number of our device in the stripe array
4662                  */
4663                 stripe_index = do_div(stripe_nr, map->num_stripes);
4664                 mirror_num = stripe_index + 1;
4665         }
4666         BUG_ON(stripe_index >= map->num_stripes);
4667
4668         num_alloc_stripes = num_stripes;
4669         if (dev_replace_is_ongoing) {
4670                 if (rw & (REQ_WRITE | REQ_DISCARD))
4671                         num_alloc_stripes <<= 1;
4672                 if (rw & REQ_GET_READ_MIRRORS)
4673                         num_alloc_stripes++;
4674         }
4675         bbio = kzalloc(btrfs_bio_size(num_alloc_stripes), GFP_NOFS);
4676         if (!bbio) {
4677                 ret = -ENOMEM;
4678                 goto out;
4679         }
4680         atomic_set(&bbio->error, 0);
4681
4682         if (rw & REQ_DISCARD) {
4683                 int factor = 0;
4684                 int sub_stripes = 0;
4685                 u64 stripes_per_dev = 0;
4686                 u32 remaining_stripes = 0;
4687                 u32 last_stripe = 0;
4688
4689                 if (map->type &
4690                     (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) {
4691                         if (map->type & BTRFS_BLOCK_GROUP_RAID0)
4692                                 sub_stripes = 1;
4693                         else
4694                                 sub_stripes = map->sub_stripes;
4695
4696                         factor = map->num_stripes / sub_stripes;
4697                         stripes_per_dev = div_u64_rem(stripe_nr_end -
4698                                                       stripe_nr_orig,
4699                                                       factor,
4700                                                       &remaining_stripes);
4701                         div_u64_rem(stripe_nr_end - 1, factor, &last_stripe);
4702                         last_stripe *= sub_stripes;
4703                 }
4704
4705                 for (i = 0; i < num_stripes; i++) {
4706                         bbio->stripes[i].physical =
4707                                 map->stripes[stripe_index].physical +
4708                                 stripe_offset + stripe_nr * map->stripe_len;
4709                         bbio->stripes[i].dev = map->stripes[stripe_index].dev;
4710
4711                         if (map->type & (BTRFS_BLOCK_GROUP_RAID0 |
4712                                          BTRFS_BLOCK_GROUP_RAID10)) {
4713                                 bbio->stripes[i].length = stripes_per_dev *
4714                                                           map->stripe_len;
4715
4716                                 if (i / sub_stripes < remaining_stripes)
4717                                         bbio->stripes[i].length +=
4718                                                 map->stripe_len;
4719
4720                                 /*
4721                                  * Special for the first stripe and
4722                                  * the last stripe:
4723                                  *
4724                                  * |-------|...|-------|
4725                                  *     |----------|
4726                                  *    off     end_off
4727                                  */
4728                                 if (i < sub_stripes)
4729                                         bbio->stripes[i].length -=
4730                                                 stripe_offset;
4731
4732                                 if (stripe_index >= last_stripe &&
4733                                     stripe_index <= (last_stripe +
4734                                                      sub_stripes - 1))
4735                                         bbio->stripes[i].length -=
4736                                                 stripe_end_offset;
4737
4738                                 if (i == sub_stripes - 1)
4739                                         stripe_offset = 0;
4740                         } else
4741                                 bbio->stripes[i].length = *length;
4742
4743                         stripe_index++;
4744                         if (stripe_index == map->num_stripes) {
4745                                 /* This could only happen for RAID0/10 */
4746                                 stripe_index = 0;
4747                                 stripe_nr++;
4748                         }
4749                 }
4750         } else {
4751                 for (i = 0; i < num_stripes; i++) {
4752                         bbio->stripes[i].physical =
4753                                 map->stripes[stripe_index].physical +
4754                                 stripe_offset +
4755                                 stripe_nr * map->stripe_len;
4756                         bbio->stripes[i].dev =
4757                                 map->stripes[stripe_index].dev;
4758                         stripe_index++;
4759                 }
4760         }
4761
4762         if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) {
4763                 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
4764                                  BTRFS_BLOCK_GROUP_RAID10 |
4765                                  BTRFS_BLOCK_GROUP_RAID5 |
4766                                  BTRFS_BLOCK_GROUP_DUP)) {
4767                         max_errors = 1;
4768                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) {
4769                         max_errors = 2;
4770                 }
4771         }
4772
4773         if (dev_replace_is_ongoing && (rw & (REQ_WRITE | REQ_DISCARD)) &&
4774             dev_replace->tgtdev != NULL) {
4775                 int index_where_to_add;
4776                 u64 srcdev_devid = dev_replace->srcdev->devid;
4777
4778                 /*
4779                  * duplicate the write operations while the dev replace
4780                  * procedure is running. Since the copying of the old disk
4781                  * to the new disk takes place at run time while the
4782                  * filesystem is mounted writable, the regular write
4783                  * operations to the old disk have to be duplicated to go
4784                  * to the new disk as well.
4785                  * Note that device->missing is handled by the caller, and
4786                  * that the write to the old disk is already set up in the
4787                  * stripes array.
4788                  */
4789                 index_where_to_add = num_stripes;
4790                 for (i = 0; i < num_stripes; i++) {
4791                         if (bbio->stripes[i].dev->devid == srcdev_devid) {
4792                                 /* write to new disk, too */
4793                                 struct btrfs_bio_stripe *new =
4794                                         bbio->stripes + index_where_to_add;
4795                                 struct btrfs_bio_stripe *old =
4796                                         bbio->stripes + i;
4797
4798                                 new->physical = old->physical;
4799                                 new->length = old->length;
4800                                 new->dev = dev_replace->tgtdev;
4801                                 index_where_to_add++;
4802                                 max_errors++;
4803                         }
4804                 }
4805                 num_stripes = index_where_to_add;
4806         } else if (dev_replace_is_ongoing && (rw & REQ_GET_READ_MIRRORS) &&
4807                    dev_replace->tgtdev != NULL) {
4808                 u64 srcdev_devid = dev_replace->srcdev->devid;
4809                 int index_srcdev = 0;
4810                 int found = 0;
4811                 u64 physical_of_found = 0;
4812
4813                 /*
4814                  * During the dev-replace procedure, the target drive can
4815                  * also be used to read data in case it is needed to repair
4816                  * a corrupt block elsewhere. This is possible if the
4817                  * requested area is left of the left cursor. In this area,
4818                  * the target drive is a full copy of the source drive.
4819                  */
4820                 for (i = 0; i < num_stripes; i++) {
4821                         if (bbio->stripes[i].dev->devid == srcdev_devid) {
4822                                 /*
4823                                  * In case of DUP, in order to keep it
4824                                  * simple, only add the mirror with the
4825                                  * lowest physical address
4826                                  */
4827                                 if (found &&
4828                                     physical_of_found <=
4829                                      bbio->stripes[i].physical)
4830                                         continue;
4831                                 index_srcdev = i;
4832                                 found = 1;
4833                                 physical_of_found = bbio->stripes[i].physical;
4834                         }
4835                 }
4836                 if (found) {
4837                         u64 length = map->stripe_len;
4838
4839                         if (physical_of_found + length <=
4840                             dev_replace->cursor_left) {
4841                                 struct btrfs_bio_stripe *tgtdev_stripe =
4842                                         bbio->stripes + num_stripes;
4843
4844                                 tgtdev_stripe->physical = physical_of_found;
4845                                 tgtdev_stripe->length =
4846                                         bbio->stripes[index_srcdev].length;
4847                                 tgtdev_stripe->dev = dev_replace->tgtdev;
4848
4849                                 num_stripes++;
4850                         }
4851                 }
4852         }
4853
4854         *bbio_ret = bbio;
4855         bbio->num_stripes = num_stripes;
4856         bbio->max_errors = max_errors;
4857         bbio->mirror_num = mirror_num;
4858
4859         /*
4860          * this is the case that REQ_READ && dev_replace_is_ongoing &&
4861          * mirror_num == num_stripes + 1 && dev_replace target drive is
4862          * available as a mirror
4863          */
4864         if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) {
4865                 WARN_ON(num_stripes > 1);
4866                 bbio->stripes[0].dev = dev_replace->tgtdev;
4867                 bbio->stripes[0].physical = physical_to_patch_in_first_stripe;
4868                 bbio->mirror_num = map->num_stripes + 1;
4869         }
4870         if (raid_map) {
4871                 sort_parity_stripes(bbio, raid_map);
4872                 *raid_map_ret = raid_map;
4873         }
4874 out:
4875         if (dev_replace_is_ongoing)
4876                 btrfs_dev_replace_unlock(dev_replace);
4877         free_extent_map(em);
4878         return ret;
4879 }
4880
4881 int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw,
4882                       u64 logical, u64 *length,
4883                       struct btrfs_bio **bbio_ret, int mirror_num)
4884 {
4885         return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret,
4886                                  mirror_num, NULL);
4887 }
4888
4889 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
4890                      u64 chunk_start, u64 physical, u64 devid,
4891                      u64 **logical, int *naddrs, int *stripe_len)
4892 {
4893         struct extent_map_tree *em_tree = &map_tree->map_tree;
4894         struct extent_map *em;
4895         struct map_lookup *map;
4896         u64 *buf;
4897         u64 bytenr;
4898         u64 length;
4899         u64 stripe_nr;
4900         u64 rmap_len;
4901         int i, j, nr = 0;
4902
4903         read_lock(&em_tree->lock);
4904         em = lookup_extent_mapping(em_tree, chunk_start, 1);
4905         read_unlock(&em_tree->lock);
4906
4907         BUG_ON(!em || em->start != chunk_start);
4908         map = (struct map_lookup *)em->bdev;
4909
4910         length = em->len;
4911         rmap_len = map->stripe_len;
4912
4913         if (map->type & BTRFS_BLOCK_GROUP_RAID10)
4914                 do_div(length, map->num_stripes / map->sub_stripes);
4915         else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
4916                 do_div(length, map->num_stripes);
4917         else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
4918                               BTRFS_BLOCK_GROUP_RAID6)) {
4919                 do_div(length, nr_data_stripes(map));
4920                 rmap_len = map->stripe_len * nr_data_stripes(map);
4921         }
4922
4923         buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
4924         BUG_ON(!buf); /* -ENOMEM */
4925
4926         for (i = 0; i < map->num_stripes; i++) {
4927                 if (devid && map->stripes[i].dev->devid != devid)
4928                         continue;
4929                 if (map->stripes[i].physical > physical ||
4930                     map->stripes[i].physical + length <= physical)
4931                         continue;
4932
4933                 stripe_nr = physical - map->stripes[i].physical;
4934                 do_div(stripe_nr, map->stripe_len);
4935
4936                 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
4937                         stripe_nr = stripe_nr * map->num_stripes + i;
4938                         do_div(stripe_nr, map->sub_stripes);
4939                 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
4940                         stripe_nr = stripe_nr * map->num_stripes + i;
4941                 } /* else if RAID[56], multiply by nr_data_stripes().
4942                    * Alternatively, just use rmap_len below instead of
4943                    * map->stripe_len */
4944
4945                 bytenr = chunk_start + stripe_nr * rmap_len;
4946                 WARN_ON(nr >= map->num_stripes);
4947                 for (j = 0; j < nr; j++) {
4948                         if (buf[j] == bytenr)
4949                                 break;
4950                 }
4951                 if (j == nr) {
4952                         WARN_ON(nr >= map->num_stripes);
4953                         buf[nr++] = bytenr;
4954                 }
4955         }
4956
4957         *logical = buf;
4958         *naddrs = nr;
4959         *stripe_len = rmap_len;
4960
4961         free_extent_map(em);
4962         return 0;
4963 }
4964
4965 static void *merge_stripe_index_into_bio_private(void *bi_private,
4966                                                  unsigned int stripe_index)
4967 {
4968         /*
4969          * with single, dup, RAID0, RAID1 and RAID10, stripe_index is
4970          * at most 1.
4971          * The alternative solution (instead of stealing bits from the
4972          * pointer) would be to allocate an intermediate structure
4973          * that contains the old private pointer plus the stripe_index.
4974          */
4975         BUG_ON((((uintptr_t)bi_private) & 3) != 0);
4976         BUG_ON(stripe_index > 3);
4977         return (void *)(((uintptr_t)bi_private) | stripe_index);
4978 }
4979
4980 static struct btrfs_bio *extract_bbio_from_bio_private(void *bi_private)
4981 {
4982         return (struct btrfs_bio *)(((uintptr_t)bi_private) & ~((uintptr_t)3));
4983 }
4984
4985 static unsigned int extract_stripe_index_from_bio_private(void *bi_private)
4986 {
4987         return (unsigned int)((uintptr_t)bi_private) & 3;
4988 }
4989
4990 static void btrfs_end_bio(struct bio *bio, int err)
4991 {
4992         struct btrfs_bio *bbio = extract_bbio_from_bio_private(bio->bi_private);
4993         int is_orig_bio = 0;
4994
4995         if (err) {
4996                 atomic_inc(&bbio->error);
4997                 if (err == -EIO || err == -EREMOTEIO) {
4998                         unsigned int stripe_index =
4999                                 extract_stripe_index_from_bio_private(
5000                                         bio->bi_private);
5001                         struct btrfs_device *dev;
5002
5003                         BUG_ON(stripe_index >= bbio->num_stripes);
5004                         dev = bbio->stripes[stripe_index].dev;
5005                         if (dev->bdev) {
5006                                 if (bio->bi_rw & WRITE)
5007                                         btrfs_dev_stat_inc(dev,
5008                                                 BTRFS_DEV_STAT_WRITE_ERRS);
5009                                 else
5010                                         btrfs_dev_stat_inc(dev,
5011                                                 BTRFS_DEV_STAT_READ_ERRS);
5012                                 if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH)
5013                                         btrfs_dev_stat_inc(dev,
5014                                                 BTRFS_DEV_STAT_FLUSH_ERRS);
5015                                 btrfs_dev_stat_print_on_error(dev);
5016                         }
5017                 }
5018         }
5019
5020         if (bio == bbio->orig_bio)
5021                 is_orig_bio = 1;
5022
5023         if (atomic_dec_and_test(&bbio->stripes_pending)) {
5024                 if (!is_orig_bio) {
5025                         bio_put(bio);
5026                         bio = bbio->orig_bio;
5027                 }
5028                 bio->bi_private = bbio->private;
5029                 bio->bi_end_io = bbio->end_io;
5030                 bio->bi_bdev = (struct block_device *)
5031                                         (unsigned long)bbio->mirror_num;
5032                 /* only send an error to the higher layers if it is
5033                  * beyond the tolerance of the btrfs bio
5034                  */
5035                 if (atomic_read(&bbio->error) > bbio->max_errors) {
5036                         err = -EIO;
5037                 } else {
5038                         /*
5039                          * this bio is actually up to date, we didn't
5040                          * go over the max number of errors
5041                          */
5042                         set_bit(BIO_UPTODATE, &bio->bi_flags);
5043                         err = 0;
5044                 }
5045                 kfree(bbio);
5046
5047                 bio_endio(bio, err);
5048         } else if (!is_orig_bio) {
5049                 bio_put(bio);
5050         }
5051 }
5052
5053 struct async_sched {
5054         struct bio *bio;
5055         int rw;
5056         struct btrfs_fs_info *info;
5057         struct btrfs_work work;
5058 };
5059
5060 /*
5061  * see run_scheduled_bios for a description of why bios are collected for
5062  * async submit.
5063  *
5064  * This will add one bio to the pending list for a device and make sure
5065  * the work struct is scheduled.
5066  */
5067 noinline void btrfs_schedule_bio(struct btrfs_root *root,
5068                                  struct btrfs_device *device,
5069                                  int rw, struct bio *bio)
5070 {
5071         int should_queue = 1;
5072         struct btrfs_pending_bios *pending_bios;
5073
5074         if (device->missing || !device->bdev) {
5075                 bio_endio(bio, -EIO);
5076                 return;
5077         }
5078
5079         /* don't bother with additional async steps for reads, right now */
5080         if (!(rw & REQ_WRITE)) {
5081                 bio_get(bio);
5082                 btrfsic_submit_bio(rw, bio);
5083                 bio_put(bio);
5084                 return;
5085         }
5086
5087         /*
5088          * nr_async_bios allows us to reliably return congestion to the
5089          * higher layers.  Otherwise, the async bio makes it appear we have
5090          * made progress against dirty pages when we've really just put it
5091          * on a queue for later
5092          */
5093         atomic_inc(&root->fs_info->nr_async_bios);
5094         WARN_ON(bio->bi_next);
5095         bio->bi_next = NULL;
5096         bio->bi_rw |= rw;
5097
5098         spin_lock(&device->io_lock);
5099         if (bio->bi_rw & REQ_SYNC)
5100                 pending_bios = &device->pending_sync_bios;
5101         else
5102                 pending_bios = &device->pending_bios;
5103
5104         if (pending_bios->tail)
5105                 pending_bios->tail->bi_next = bio;
5106
5107         pending_bios->tail = bio;
5108         if (!pending_bios->head)
5109                 pending_bios->head = bio;
5110         if (device->running_pending)
5111                 should_queue = 0;
5112
5113         spin_unlock(&device->io_lock);
5114
5115         if (should_queue)
5116                 btrfs_queue_worker(&root->fs_info->submit_workers,
5117                                    &device->work);
5118 }
5119
5120 static int bio_size_ok(struct block_device *bdev, struct bio *bio,
5121                        sector_t sector)
5122 {
5123         struct bio_vec *prev;
5124         struct request_queue *q = bdev_get_queue(bdev);
5125         unsigned short max_sectors = queue_max_sectors(q);
5126         struct bvec_merge_data bvm = {
5127                 .bi_bdev = bdev,
5128                 .bi_sector = sector,
5129                 .bi_rw = bio->bi_rw,
5130         };
5131
5132         if (bio->bi_vcnt == 0) {
5133                 WARN_ON(1);
5134                 return 1;
5135         }
5136
5137         prev = &bio->bi_io_vec[bio->bi_vcnt - 1];
5138         if ((bio->bi_size >> 9) > max_sectors)
5139                 return 0;
5140
5141         if (!q->merge_bvec_fn)
5142                 return 1;
5143
5144         bvm.bi_size = bio->bi_size - prev->bv_len;
5145         if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len)
5146                 return 0;
5147         return 1;
5148 }
5149
5150 static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5151                               struct bio *bio, u64 physical, int dev_nr,
5152                               int rw, int async)
5153 {
5154         struct btrfs_device *dev = bbio->stripes[dev_nr].dev;
5155
5156         bio->bi_private = bbio;
5157         bio->bi_private = merge_stripe_index_into_bio_private(
5158                         bio->bi_private, (unsigned int)dev_nr);
5159         bio->bi_end_io = btrfs_end_bio;
5160         bio->bi_sector = physical >> 9;
5161 #ifdef DEBUG
5162         {
5163                 struct rcu_string *name;
5164
5165                 rcu_read_lock();
5166                 name = rcu_dereference(dev->name);
5167                 pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu "
5168                          "(%s id %llu), size=%u\n", rw,
5169                          (u64)bio->bi_sector, (u_long)dev->bdev->bd_dev,
5170                          name->str, dev->devid, bio->bi_size);
5171                 rcu_read_unlock();
5172         }
5173 #endif
5174         bio->bi_bdev = dev->bdev;
5175         if (async)
5176                 btrfs_schedule_bio(root, dev, rw, bio);
5177         else
5178                 btrfsic_submit_bio(rw, bio);
5179 }
5180
5181 static int breakup_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio,
5182                               struct bio *first_bio, struct btrfs_device *dev,
5183                               int dev_nr, int rw, int async)
5184 {
5185         struct bio_vec *bvec = first_bio->bi_io_vec;
5186         struct bio *bio;
5187         int nr_vecs = bio_get_nr_vecs(dev->bdev);
5188         u64 physical = bbio->stripes[dev_nr].physical;
5189
5190 again:
5191         bio = btrfs_bio_alloc(dev->bdev, physical >> 9, nr_vecs, GFP_NOFS);
5192         if (!bio)
5193                 return -ENOMEM;
5194
5195         while (bvec <= (first_bio->bi_io_vec + first_bio->bi_vcnt - 1)) {
5196                 if (bio_add_page(bio, bvec->bv_page, bvec->bv_len,
5197                                  bvec->bv_offset) < bvec->bv_len) {
5198                         u64 len = bio->bi_size;
5199
5200                         atomic_inc(&bbio->stripes_pending);
5201                         submit_stripe_bio(root, bbio, bio, physical, dev_nr,
5202                                           rw, async);
5203                         physical += len;
5204                         goto again;
5205                 }
5206                 bvec++;
5207         }
5208
5209         submit_stripe_bio(root, bbio, bio, physical, dev_nr, rw, async);
5210         return 0;
5211 }
5212
5213 static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical)
5214 {
5215         atomic_inc(&bbio->error);
5216         if (atomic_dec_and_test(&bbio->stripes_pending)) {
5217                 bio->bi_private = bbio->private;
5218                 bio->bi_end_io = bbio->end_io;
5219                 bio->bi_bdev = (struct block_device *)
5220                         (unsigned long)bbio->mirror_num;
5221                 bio->bi_sector = logical >> 9;
5222                 kfree(bbio);
5223                 bio_endio(bio, -EIO);
5224         }
5225 }
5226
5227 int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio,
5228                   int mirror_num, int async_submit)
5229 {
5230         struct btrfs_device *dev;
5231         struct bio *first_bio = bio;
5232         u64 logical = (u64)bio->bi_sector << 9;
5233         u64 length = 0;
5234         u64 map_length;
5235         u64 *raid_map = NULL;
5236         int ret;
5237         int dev_nr = 0;
5238         int total_devs = 1;
5239         struct btrfs_bio *bbio = NULL;
5240
5241         length = bio->bi_size;
5242         map_length = length;
5243
5244         ret = __btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio,
5245                               mirror_num, &raid_map);
5246         if (ret) /* -ENOMEM */
5247                 return ret;
5248
5249         total_devs = bbio->num_stripes;
5250         bbio->orig_bio = first_bio;
5251         bbio->private = first_bio->bi_private;
5252         bbio->end_io = first_bio->bi_end_io;
5253         atomic_set(&bbio->stripes_pending, bbio->num_stripes);
5254
5255         if (raid_map) {
5256                 /* In this case, map_length has been set to the length of
5257                    a single stripe; not the whole write */
5258                 if (rw & WRITE) {
5259                         return raid56_parity_write(root, bio, bbio,
5260                                                    raid_map, map_length);
5261                 } else {
5262                         return raid56_parity_recover(root, bio, bbio,
5263                                                      raid_map, map_length,
5264                                                      mirror_num);
5265                 }
5266         }
5267
5268         if (map_length < length) {
5269                 printk(KERN_CRIT "btrfs: mapping failed logical %llu bio len %llu "
5270                        "len %llu\n", (unsigned long long)logical,
5271                        (unsigned long long)length,
5272                        (unsigned long long)map_length);
5273                 BUG();
5274         }
5275
5276         while (dev_nr < total_devs) {
5277                 dev = bbio->stripes[dev_nr].dev;
5278                 if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) {
5279                         bbio_error(bbio, first_bio, logical);
5280                         dev_nr++;
5281                         continue;
5282                 }
5283
5284                 /*
5285                  * Check and see if we're ok with this bio based on it's size
5286                  * and offset with the given device.
5287                  */
5288                 if (!bio_size_ok(dev->bdev, first_bio,
5289                                  bbio->stripes[dev_nr].physical >> 9)) {
5290                         ret = breakup_stripe_bio(root, bbio, first_bio, dev,
5291                                                  dev_nr, rw, async_submit);
5292                         BUG_ON(ret);
5293                         dev_nr++;
5294                         continue;
5295                 }
5296
5297                 if (dev_nr < total_devs - 1) {
5298                         bio = bio_clone(first_bio, GFP_NOFS);
5299                         BUG_ON(!bio); /* -ENOMEM */
5300                 } else {
5301                         bio = first_bio;
5302                 }
5303
5304                 submit_stripe_bio(root, bbio, bio,
5305                                   bbio->stripes[dev_nr].physical, dev_nr, rw,
5306                                   async_submit);
5307                 dev_nr++;
5308         }
5309         return 0;
5310 }
5311
5312 struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid,
5313                                        u8 *uuid, u8 *fsid)
5314 {
5315         struct btrfs_device *device;
5316         struct btrfs_fs_devices *cur_devices;
5317
5318         cur_devices = fs_info->fs_devices;
5319         while (cur_devices) {
5320                 if (!fsid ||
5321                     !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
5322                         device = __find_device(&cur_devices->devices,
5323                                                devid, uuid);
5324                         if (device)
5325                                 return device;
5326                 }
5327                 cur_devices = cur_devices->seed;
5328         }
5329         return NULL;
5330 }
5331
5332 static struct btrfs_device *add_missing_dev(struct btrfs_root *root,
5333                                             u64 devid, u8 *dev_uuid)
5334 {
5335         struct btrfs_device *device;
5336         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
5337
5338         device = kzalloc(sizeof(*device), GFP_NOFS);
5339         if (!device)
5340                 return NULL;
5341         list_add(&device->dev_list,
5342                  &fs_devices->devices);
5343         device->dev_root = root->fs_info->dev_root;
5344         device->devid = devid;
5345         device->work.func = pending_bios_fn;
5346         device->fs_devices = fs_devices;
5347         device->missing = 1;
5348         fs_devices->num_devices++;
5349         fs_devices->missing_devices++;
5350         spin_lock_init(&device->io_lock);
5351         INIT_LIST_HEAD(&device->dev_alloc_list);
5352         memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE);
5353         return device;
5354 }
5355
5356 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
5357                           struct extent_buffer *leaf,
5358                           struct btrfs_chunk *chunk)
5359 {
5360         struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
5361         struct map_lookup *map;
5362         struct extent_map *em;
5363         u64 logical;
5364         u64 length;
5365         u64 devid;
5366         u8 uuid[BTRFS_UUID_SIZE];
5367         int num_stripes;
5368         int ret;
5369         int i;
5370
5371         logical = key->offset;
5372         length = btrfs_chunk_length(leaf, chunk);
5373
5374         read_lock(&map_tree->map_tree.lock);
5375         em = lookup_extent_mapping(&map_tree->map_tree, logical, 1);
5376         read_unlock(&map_tree->map_tree.lock);
5377
5378         /* already mapped? */
5379         if (em && em->start <= logical && em->start + em->len > logical) {
5380                 free_extent_map(em);
5381                 return 0;
5382         } else if (em) {
5383                 free_extent_map(em);
5384         }
5385
5386         em = alloc_extent_map();
5387         if (!em)
5388                 return -ENOMEM;
5389         num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
5390         map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
5391         if (!map) {
5392                 free_extent_map(em);
5393                 return -ENOMEM;
5394         }
5395
5396         em->bdev = (struct block_device *)map;
5397         em->start = logical;
5398         em->len = length;
5399         em->orig_start = 0;
5400         em->block_start = 0;
5401         em->block_len = em->len;
5402
5403         map->num_stripes = num_stripes;
5404         map->io_width = btrfs_chunk_io_width(leaf, chunk);
5405         map->io_align = btrfs_chunk_io_align(leaf, chunk);
5406         map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
5407         map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
5408         map->type = btrfs_chunk_type(leaf, chunk);
5409         map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
5410         for (i = 0; i < num_stripes; i++) {
5411                 map->stripes[i].physical =
5412                         btrfs_stripe_offset_nr(leaf, chunk, i);
5413                 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
5414                 read_extent_buffer(leaf, uuid, (unsigned long)
5415                                    btrfs_stripe_dev_uuid_nr(chunk, i),
5416                                    BTRFS_UUID_SIZE);
5417                 map->stripes[i].dev = btrfs_find_device(root->fs_info, devid,
5418                                                         uuid, NULL);
5419                 if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) {
5420                         kfree(map);
5421                         free_extent_map(em);
5422                         return -EIO;
5423                 }
5424                 if (!map->stripes[i].dev) {
5425                         map->stripes[i].dev =
5426                                 add_missing_dev(root, devid, uuid);
5427                         if (!map->stripes[i].dev) {
5428                                 kfree(map);
5429                                 free_extent_map(em);
5430                                 return -EIO;
5431                         }
5432                 }
5433                 map->stripes[i].dev->in_fs_metadata = 1;
5434         }
5435
5436         write_lock(&map_tree->map_tree.lock);
5437         ret = add_extent_mapping(&map_tree->map_tree, em);
5438         write_unlock(&map_tree->map_tree.lock);
5439         BUG_ON(ret); /* Tree corruption */
5440         free_extent_map(em);
5441
5442         return 0;
5443 }
5444
5445 static void fill_device_from_item(struct extent_buffer *leaf,
5446                                  struct btrfs_dev_item *dev_item,
5447                                  struct btrfs_device *device)
5448 {
5449         unsigned long ptr;
5450
5451         device->devid = btrfs_device_id(leaf, dev_item);
5452         device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item);
5453         device->total_bytes = device->disk_total_bytes;
5454         device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
5455         device->type = btrfs_device_type(leaf, dev_item);
5456         device->io_align = btrfs_device_io_align(leaf, dev_item);
5457         device->io_width = btrfs_device_io_width(leaf, dev_item);
5458         device->sector_size = btrfs_device_sector_size(leaf, dev_item);
5459         WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID);
5460         device->is_tgtdev_for_dev_replace = 0;
5461
5462         ptr = (unsigned long)btrfs_device_uuid(dev_item);
5463         read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
5464 }
5465
5466 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
5467 {
5468         struct btrfs_fs_devices *fs_devices;
5469         int ret;
5470
5471         BUG_ON(!mutex_is_locked(&uuid_mutex));
5472
5473         fs_devices = root->fs_info->fs_devices->seed;
5474         while (fs_devices) {
5475                 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
5476                         ret = 0;
5477                         goto out;
5478                 }
5479                 fs_devices = fs_devices->seed;
5480         }
5481
5482         fs_devices = find_fsid(fsid);
5483         if (!fs_devices) {
5484                 ret = -ENOENT;
5485                 goto out;
5486         }
5487
5488         fs_devices = clone_fs_devices(fs_devices);
5489         if (IS_ERR(fs_devices)) {
5490                 ret = PTR_ERR(fs_devices);
5491                 goto out;
5492         }
5493
5494         ret = __btrfs_open_devices(fs_devices, FMODE_READ,
5495                                    root->fs_info->bdev_holder);
5496         if (ret) {
5497                 free_fs_devices(fs_devices);
5498                 goto out;
5499         }
5500
5501         if (!fs_devices->seeding) {
5502                 __btrfs_close_devices(fs_devices);
5503                 free_fs_devices(fs_devices);
5504                 ret = -EINVAL;
5505                 goto out;
5506         }
5507
5508         fs_devices->seed = root->fs_info->fs_devices->seed;
5509         root->fs_info->fs_devices->seed = fs_devices;
5510 out:
5511         return ret;
5512 }
5513
5514 static int read_one_dev(struct btrfs_root *root,
5515                         struct extent_buffer *leaf,
5516                         struct btrfs_dev_item *dev_item)
5517 {
5518         struct btrfs_device *device;
5519         u64 devid;
5520         int ret;
5521         u8 fs_uuid[BTRFS_UUID_SIZE];
5522         u8 dev_uuid[BTRFS_UUID_SIZE];
5523
5524         devid = btrfs_device_id(leaf, dev_item);
5525         read_extent_buffer(leaf, dev_uuid,
5526                            (unsigned long)btrfs_device_uuid(dev_item),
5527                            BTRFS_UUID_SIZE);
5528         read_extent_buffer(leaf, fs_uuid,
5529                            (unsigned long)btrfs_device_fsid(dev_item),
5530                            BTRFS_UUID_SIZE);
5531
5532         if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
5533                 ret = open_seed_devices(root, fs_uuid);
5534                 if (ret && !btrfs_test_opt(root, DEGRADED))
5535                         return ret;
5536         }
5537
5538         device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid);
5539         if (!device || !device->bdev) {
5540                 if (!btrfs_test_opt(root, DEGRADED))
5541                         return -EIO;
5542
5543                 if (!device) {
5544                         printk(KERN_WARNING "warning devid %llu missing\n",
5545                                (unsigned long long)devid);
5546                         device = add_missing_dev(root, devid, dev_uuid);
5547                         if (!device)
5548                                 return -ENOMEM;
5549                 } else if (!device->missing) {
5550                         /*
5551                          * this happens when a device that was properly setup
5552                          * in the device info lists suddenly goes bad.
5553                          * device->bdev is NULL, and so we have to set
5554                          * device->missing to one here
5555                          */
5556                         root->fs_info->fs_devices->missing_devices++;
5557                         device->missing = 1;
5558                 }
5559         }
5560
5561         if (device->fs_devices != root->fs_info->fs_devices) {
5562                 BUG_ON(device->writeable);
5563                 if (device->generation !=
5564                     btrfs_device_generation(leaf, dev_item))
5565                         return -EINVAL;
5566         }
5567
5568         fill_device_from_item(leaf, dev_item, device);
5569         device->dev_root = root->fs_info->dev_root;
5570         device->in_fs_metadata = 1;
5571         if (device->writeable && !device->is_tgtdev_for_dev_replace) {
5572                 device->fs_devices->total_rw_bytes += device->total_bytes;
5573                 spin_lock(&root->fs_info->free_chunk_lock);
5574                 root->fs_info->free_chunk_space += device->total_bytes -
5575                         device->bytes_used;
5576                 spin_unlock(&root->fs_info->free_chunk_lock);
5577         }
5578         ret = 0;
5579         return ret;
5580 }
5581
5582 int btrfs_read_sys_array(struct btrfs_root *root)
5583 {
5584         struct btrfs_super_block *super_copy = root->fs_info->super_copy;
5585         struct extent_buffer *sb;
5586         struct btrfs_disk_key *disk_key;
5587         struct btrfs_chunk *chunk;
5588         u8 *ptr;
5589         unsigned long sb_ptr;
5590         int ret = 0;
5591         u32 num_stripes;
5592         u32 array_size;
5593         u32 len = 0;
5594         u32 cur;
5595         struct btrfs_key key;
5596
5597         sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
5598                                           BTRFS_SUPER_INFO_SIZE);
5599         if (!sb)
5600                 return -ENOMEM;
5601         btrfs_set_buffer_uptodate(sb);
5602         btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0);
5603         /*
5604          * The sb extent buffer is artifical and just used to read the system array.
5605          * btrfs_set_buffer_uptodate() call does not properly mark all it's
5606          * pages up-to-date when the page is larger: extent does not cover the
5607          * whole page and consequently check_page_uptodate does not find all
5608          * the page's extents up-to-date (the hole beyond sb),
5609          * write_extent_buffer then triggers a WARN_ON.
5610          *
5611          * Regular short extents go through mark_extent_buffer_dirty/writeback cycle,
5612          * but sb spans only this function. Add an explicit SetPageUptodate call
5613          * to silence the warning eg. on PowerPC 64.
5614          */
5615         if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE)
5616                 SetPageUptodate(sb->pages[0]);
5617
5618         write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE);
5619         array_size = btrfs_super_sys_array_size(super_copy);
5620
5621         ptr = super_copy->sys_chunk_array;
5622         sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
5623         cur = 0;
5624
5625         while (cur < array_size) {
5626                 disk_key = (struct btrfs_disk_key *)ptr;
5627                 btrfs_disk_key_to_cpu(&key, disk_key);
5628
5629                 len = sizeof(*disk_key); ptr += len;
5630                 sb_ptr += len;
5631                 cur += len;
5632
5633                 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
5634                         chunk = (struct btrfs_chunk *)sb_ptr;
5635                         ret = read_one_chunk(root, &key, sb, chunk);
5636                         if (ret)
5637                                 break;
5638                         num_stripes = btrfs_chunk_num_stripes(sb, chunk);
5639                         len = btrfs_chunk_item_size(num_stripes);
5640                 } else {
5641                         ret = -EIO;
5642                         break;
5643                 }
5644                 ptr += len;
5645                 sb_ptr += len;
5646                 cur += len;
5647         }
5648         free_extent_buffer(sb);
5649         return ret;
5650 }
5651
5652 int btrfs_read_chunk_tree(struct btrfs_root *root)
5653 {
5654         struct btrfs_path *path;
5655         struct extent_buffer *leaf;
5656         struct btrfs_key key;
5657         struct btrfs_key found_key;
5658         int ret;
5659         int slot;
5660
5661         root = root->fs_info->chunk_root;
5662
5663         path = btrfs_alloc_path();
5664         if (!path)
5665                 return -ENOMEM;
5666
5667         mutex_lock(&uuid_mutex);
5668         lock_chunks(root);
5669
5670         /* first we search for all of the device items, and then we
5671          * read in all of the chunk items.  This way we can create chunk
5672          * mappings that reference all of the devices that are afound
5673          */
5674         key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
5675         key.offset = 0;
5676         key.type = 0;
5677 again:
5678         ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5679         if (ret < 0)
5680                 goto error;
5681         while (1) {
5682                 leaf = path->nodes[0];
5683                 slot = path->slots[0];
5684                 if (slot >= btrfs_header_nritems(leaf)) {
5685                         ret = btrfs_next_leaf(root, path);
5686                         if (ret == 0)
5687                                 continue;
5688                         if (ret < 0)
5689                                 goto error;
5690                         break;
5691                 }
5692                 btrfs_item_key_to_cpu(leaf, &found_key, slot);
5693                 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
5694                         if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
5695                                 break;
5696                         if (found_key.type == BTRFS_DEV_ITEM_KEY) {
5697                                 struct btrfs_dev_item *dev_item;
5698                                 dev_item = btrfs_item_ptr(leaf, slot,
5699                                                   struct btrfs_dev_item);
5700                                 ret = read_one_dev(root, leaf, dev_item);
5701                                 if (ret)
5702                                         goto error;
5703                         }
5704                 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
5705                         struct btrfs_chunk *chunk;
5706                         chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
5707                         ret = read_one_chunk(root, &found_key, leaf, chunk);
5708                         if (ret)
5709                                 goto error;
5710                 }
5711                 path->slots[0]++;
5712         }
5713         if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
5714                 key.objectid = 0;
5715                 btrfs_release_path(path);
5716                 goto again;
5717         }
5718         ret = 0;
5719 error:
5720         unlock_chunks(root);
5721         mutex_unlock(&uuid_mutex);
5722
5723         btrfs_free_path(path);
5724         return ret;
5725 }
5726
5727 static void __btrfs_reset_dev_stats(struct btrfs_device *dev)
5728 {
5729         int i;
5730
5731         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
5732                 btrfs_dev_stat_reset(dev, i);
5733 }
5734
5735 int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info)
5736 {
5737         struct btrfs_key key;
5738         struct btrfs_key found_key;
5739         struct btrfs_root *dev_root = fs_info->dev_root;
5740         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
5741         struct extent_buffer *eb;
5742         int slot;
5743         int ret = 0;
5744         struct btrfs_device *device;
5745         struct btrfs_path *path = NULL;
5746         int i;
5747
5748         path = btrfs_alloc_path();
5749         if (!path) {
5750                 ret = -ENOMEM;
5751                 goto out;
5752         }
5753
5754         mutex_lock(&fs_devices->device_list_mutex);
5755         list_for_each_entry(device, &fs_devices->devices, dev_list) {
5756                 int item_size;
5757                 struct btrfs_dev_stats_item *ptr;
5758
5759                 key.objectid = 0;
5760                 key.type = BTRFS_DEV_STATS_KEY;
5761                 key.offset = device->devid;
5762                 ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0);
5763                 if (ret) {
5764                         __btrfs_reset_dev_stats(device);
5765                         device->dev_stats_valid = 1;
5766                         btrfs_release_path(path);
5767                         continue;
5768                 }
5769                 slot = path->slots[0];
5770                 eb = path->nodes[0];
5771                 btrfs_item_key_to_cpu(eb, &found_key, slot);
5772                 item_size = btrfs_item_size_nr(eb, slot);
5773
5774                 ptr = btrfs_item_ptr(eb, slot,
5775                                      struct btrfs_dev_stats_item);
5776
5777                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
5778                         if (item_size >= (1 + i) * sizeof(__le64))
5779                                 btrfs_dev_stat_set(device, i,
5780                                         btrfs_dev_stats_value(eb, ptr, i));
5781                         else
5782                                 btrfs_dev_stat_reset(device, i);
5783                 }
5784
5785                 device->dev_stats_valid = 1;
5786                 btrfs_dev_stat_print_on_load(device);
5787                 btrfs_release_path(path);
5788         }
5789         mutex_unlock(&fs_devices->device_list_mutex);
5790
5791 out:
5792         btrfs_free_path(path);
5793         return ret < 0 ? ret : 0;
5794 }
5795
5796 static int update_dev_stat_item(struct btrfs_trans_handle *trans,
5797                                 struct btrfs_root *dev_root,
5798                                 struct btrfs_device *device)
5799 {
5800         struct btrfs_path *path;
5801         struct btrfs_key key;
5802         struct extent_buffer *eb;
5803         struct btrfs_dev_stats_item *ptr;
5804         int ret;
5805         int i;
5806
5807         key.objectid = 0;
5808         key.type = BTRFS_DEV_STATS_KEY;
5809         key.offset = device->devid;
5810
5811         path = btrfs_alloc_path();
5812         BUG_ON(!path);
5813         ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1);
5814         if (ret < 0) {
5815                 printk_in_rcu(KERN_WARNING "btrfs: error %d while searching for dev_stats item for device %s!\n",
5816                               ret, rcu_str_deref(device->name));
5817                 goto out;
5818         }
5819
5820         if (ret == 0 &&
5821             btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) {
5822                 /* need to delete old one and insert a new one */
5823                 ret = btrfs_del_item(trans, dev_root, path);
5824                 if (ret != 0) {
5825                         printk_in_rcu(KERN_WARNING "btrfs: delete too small dev_stats item for device %s failed %d!\n",
5826                                       rcu_str_deref(device->name), ret);
5827                         goto out;
5828                 }
5829                 ret = 1;
5830         }
5831
5832         if (ret == 1) {
5833                 /* need to insert a new item */
5834                 btrfs_release_path(path);
5835                 ret = btrfs_insert_empty_item(trans, dev_root, path,
5836                                               &key, sizeof(*ptr));
5837                 if (ret < 0) {
5838                         printk_in_rcu(KERN_WARNING "btrfs: insert dev_stats item for device %s failed %d!\n",
5839                                       rcu_str_deref(device->name), ret);
5840                         goto out;
5841                 }
5842         }
5843
5844         eb = path->nodes[0];
5845         ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item);
5846         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
5847                 btrfs_set_dev_stats_value(eb, ptr, i,
5848                                           btrfs_dev_stat_read(device, i));
5849         btrfs_mark_buffer_dirty(eb);
5850
5851 out:
5852         btrfs_free_path(path);
5853         return ret;
5854 }
5855
5856 /*
5857  * called from commit_transaction. Writes all changed device stats to disk.
5858  */
5859 int btrfs_run_dev_stats(struct btrfs_trans_handle *trans,
5860                         struct btrfs_fs_info *fs_info)
5861 {
5862         struct btrfs_root *dev_root = fs_info->dev_root;
5863         struct btrfs_fs_devices *fs_devices = fs_info->fs_devices;
5864         struct btrfs_device *device;
5865         int ret = 0;
5866
5867         mutex_lock(&fs_devices->device_list_mutex);
5868         list_for_each_entry(device, &fs_devices->devices, dev_list) {
5869                 if (!device->dev_stats_valid || !device->dev_stats_dirty)
5870                         continue;
5871
5872                 ret = update_dev_stat_item(trans, dev_root, device);
5873                 if (!ret)
5874                         device->dev_stats_dirty = 0;
5875         }
5876         mutex_unlock(&fs_devices->device_list_mutex);
5877
5878         return ret;
5879 }
5880
5881 void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index)
5882 {
5883         btrfs_dev_stat_inc(dev, index);
5884         btrfs_dev_stat_print_on_error(dev);
5885 }
5886
5887 void btrfs_dev_stat_print_on_error(struct btrfs_device *dev)
5888 {
5889         if (!dev->dev_stats_valid)
5890                 return;
5891         printk_ratelimited_in_rcu(KERN_ERR
5892                            "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
5893                            rcu_str_deref(dev->name),
5894                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
5895                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
5896                            btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
5897                            btrfs_dev_stat_read(dev,
5898                                                BTRFS_DEV_STAT_CORRUPTION_ERRS),
5899                            btrfs_dev_stat_read(dev,
5900                                                BTRFS_DEV_STAT_GENERATION_ERRS));
5901 }
5902
5903 static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev)
5904 {
5905         int i;
5906
5907         for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
5908                 if (btrfs_dev_stat_read(dev, i) != 0)
5909                         break;
5910         if (i == BTRFS_DEV_STAT_VALUES_MAX)
5911                 return; /* all values == 0, suppress message */
5912
5913         printk_in_rcu(KERN_INFO "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n",
5914                rcu_str_deref(dev->name),
5915                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS),
5916                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS),
5917                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS),
5918                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS),
5919                btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS));
5920 }
5921
5922 int btrfs_get_dev_stats(struct btrfs_root *root,
5923                         struct btrfs_ioctl_get_dev_stats *stats)
5924 {
5925         struct btrfs_device *dev;
5926         struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
5927         int i;
5928
5929         mutex_lock(&fs_devices->device_list_mutex);
5930         dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL);
5931         mutex_unlock(&fs_devices->device_list_mutex);
5932
5933         if (!dev) {
5934                 printk(KERN_WARNING
5935                        "btrfs: get dev_stats failed, device not found\n");
5936                 return -ENODEV;
5937         } else if (!dev->dev_stats_valid) {
5938                 printk(KERN_WARNING
5939                        "btrfs: get dev_stats failed, not yet valid\n");
5940                 return -ENODEV;
5941         } else if (stats->flags & BTRFS_DEV_STATS_RESET) {
5942                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) {
5943                         if (stats->nr_items > i)
5944                                 stats->values[i] =
5945                                         btrfs_dev_stat_read_and_reset(dev, i);
5946                         else
5947                                 btrfs_dev_stat_reset(dev, i);
5948                 }
5949         } else {
5950                 for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++)
5951                         if (stats->nr_items > i)
5952                                 stats->values[i] = btrfs_dev_stat_read(dev, i);
5953         }
5954         if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX)
5955                 stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX;
5956         return 0;
5957 }
5958
5959 int btrfs_scratch_superblock(struct btrfs_device *device)
5960 {
5961         struct buffer_head *bh;
5962         struct btrfs_super_block *disk_super;
5963
5964         bh = btrfs_read_dev_super(device->bdev);
5965         if (!bh)
5966                 return -EINVAL;
5967         disk_super = (struct btrfs_super_block *)bh->b_data;
5968
5969         memset(&disk_super->magic, 0, sizeof(disk_super->magic));
5970         set_buffer_dirty(bh);
5971         sync_dirty_buffer(bh);
5972         brelse(bh);
5973
5974         return 0;
5975 }