drivers/block/zram/zram_drv.c

   1 /*
   2  * Compressed RAM block device
   3  *
   4  * Copyright (C) 2008, 2009, 2010  Nitin Gupta
   5  *               2012, 2013 Minchan Kim
   6  *
   7  * This code is released using a dual license strategy: BSD/GPL
   8  * You can choose the licence that better fits your requirements.
   9  *
  10  * Released under the terms of 3-clause BSD License
  11  * Released under the terms of GNU General Public License Version 2.0
  12  *
  13  */
  14
  15 #define KMSG_COMPONENT "zram"
  16 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
  17
  18 #include <linux/module.h>
  19 #include <linux/kernel.h>
  20 #include <linux/bio.h>
  21 #include <linux/bitops.h>
  22 #include <linux/blkdev.h>
  23 #include <linux/buffer_head.h>
  24 #include <linux/device.h>
  25 #include <linux/genhd.h>
  26 #include <linux/highmem.h>
  27 #include <linux/slab.h>
  28 #include <linux/string.h>
  29 #include <linux/vmalloc.h>
  30 #include <linux/err.h>
  31 #include <linux/idr.h>
  32 #include <linux/sysfs.h>
  33
  34 #include "zram_drv.h"
  35
  36 static DEFINE_IDR(zram_index_idr);
  37 /* idr index must be protected */
  38 static DEFINE_MUTEX(zram_index_mutex);
  39
  40 static int zram_major;
  41 static const char *default_compressor = "lzo";
  42
  43 /* Module params (documentation at end) */
  44 static unsigned int num_devices = 1;
  45
  46 static inline void deprecated_attr_warn(const char *name)
  47 {
  48         pr_warn_once("%d (%s) Attribute %s (and others) will be removed. %s\n",
  49                         task_pid_nr(current),
  50                         current->comm,
  51                         name,
  52                         "See zram documentation.");
  53 }
  54
  55 #define ZRAM_ATTR_RO(name)                                              \
  56 static ssize_t name##_show(struct device *d,                            \
  57                                 struct device_attribute *attr, char *b) \
  58 {                                                                       \
  59         struct zram *zram = dev_to_zram(d);                             \
  60                                                                         \
  61         deprecated_attr_warn(__stringify(name));                        \
  62         return scnprintf(b, PAGE_SIZE, "%llu\n",                        \
  63                 (u64)atomic64_read(&zram->stats.name));                 \
  64 }                                                                       \
  65 static DEVICE_ATTR_RO(name);
  66
  67 static inline bool init_done(struct zram *zram)
  68 {
  69         return zram->disksize;
  70 }
  71
  72 static inline struct zram *dev_to_zram(struct device *dev)
  73 {
  74         return (struct zram *)dev_to_disk(dev)->private_data;
  75 }
  76
  77 /* flag operations require table entry bit_spin_lock() being held */
  78 static int zram_test_flag(struct zram_meta *meta, u32 index,
  79                         enum zram_pageflags flag)
  80 {
  81         return meta->table[index].value & BIT(flag);
  82 }
  83
  84 static void zram_set_flag(struct zram_meta *meta, u32 index,
  85                         enum zram_pageflags flag)
  86 {
  87         meta->table[index].value |= BIT(flag);
  88 }
  89
  90 static void zram_clear_flag(struct zram_meta *meta, u32 index,
  91                         enum zram_pageflags flag)
  92 {
  93         meta->table[index].value &= ~BIT(flag);
  94 }
  95
  96 static size_t zram_get_obj_size(struct zram_meta *meta, u32 index)
  97 {
  98         return meta->table[index].value & (BIT(ZRAM_FLAG_SHIFT) - 1);
  99 }
 100
 101 static void zram_set_obj_size(struct zram_meta *meta,
 102                                         u32 index, size_t size)
 103 {
 104         unsigned long flags = meta->table[index].value >> ZRAM_FLAG_SHIFT;
 105
 106         meta->table[index].value = (flags << ZRAM_FLAG_SHIFT) | size;
 107 }
 108
 109 static inline bool is_partial_io(struct bio_vec *bvec)
 110 {
 111         return bvec->bv_len != PAGE_SIZE;
 112 }
 113
 114 /*
 115  * Check if request is within bounds and aligned on zram logical blocks.
 116  */
 117 static inline bool valid_io_request(struct zram *zram,
 118                 sector_t start, unsigned int size)
 119 {
 120         u64 end, bound;
 121
 122         /* unaligned request */
 123         if (unlikely(start & (ZRAM_SECTOR_PER_LOGICAL_BLOCK - 1)))
 124                 return false;
 125         if (unlikely(size & (ZRAM_LOGICAL_BLOCK_SIZE - 1)))
 126                 return false;
 127
 128         end = start + (size >> SECTOR_SHIFT);
 129         bound = zram->disksize >> SECTOR_SHIFT;
 130         /* out of range range */
 131         if (unlikely(start >= bound || end > bound || start > end))
 132                 return false;
 133
 134         /* I/O request is valid */
 135         return true;
 136 }
 137
 138 static void update_position(u32 *index, int *offset, struct bio_vec *bvec)
 139 {
 140         if (*offset + bvec->bv_len >= PAGE_SIZE)
 141                 (*index)++;
 142         *offset = (*offset + bvec->bv_len) % PAGE_SIZE;
 143 }
 144
 145 static inline void update_used_max(struct zram *zram,
 146                                         const unsigned long pages)
 147 {
 148         unsigned long old_max, cur_max;
 149
 150         old_max = atomic_long_read(&zram->stats.max_used_pages);
 151
 152         do {
 153                 cur_max = old_max;
 154                 if (pages > cur_max)
 155                         old_max = atomic_long_cmpxchg(
 156                                 &zram->stats.max_used_pages, cur_max, pages);
 157         } while (old_max != cur_max);
 158 }
 159
 160 static bool page_zero_filled(void *ptr)
 161 {
 162         unsigned int pos;
 163         unsigned long *page;
 164
 165         page = (unsigned long *)ptr;
 166
 167         for (pos = 0; pos != PAGE_SIZE / sizeof(*page); pos++) {
 168                 if (page[pos])
 169                         return false;
 170         }
 171
 172         return true;
 173 }
 174
 175 static void handle_zero_page(struct bio_vec *bvec)
 176 {
 177         struct page *page = bvec->bv_page;
 178         void *user_mem;
 179
 180         user_mem = kmap_atomic(page);
 181         if (is_partial_io(bvec))
 182                 memset(user_mem + bvec->bv_offset, 0, bvec->bv_len);
 183         else
 184                 clear_page(user_mem);
 185         kunmap_atomic(user_mem);
 186
 187         flush_dcache_page(page);
 188 }
 189
 190 static ssize_t initstate_show(struct device *dev,
 191                 struct device_attribute *attr, char *buf)
 192 {
 193         u32 val;
 194         struct zram *zram = dev_to_zram(dev);
 195
 196         down_read(&zram->init_lock);
 197         val = init_done(zram);
 198         up_read(&zram->init_lock);
 199
 200         return scnprintf(buf, PAGE_SIZE, "%u\n", val);
 201 }
 202
 203 static ssize_t disksize_show(struct device *dev,
 204                 struct device_attribute *attr, char *buf)
 205 {
 206         struct zram *zram = dev_to_zram(dev);
 207
 208         return scnprintf(buf, PAGE_SIZE, "%llu\n", zram->disksize);
 209 }
 210
 211 static ssize_t orig_data_size_show(struct device *dev,
 212                 struct device_attribute *attr, char *buf)
 213 {
 214         struct zram *zram = dev_to_zram(dev);
 215
 216         deprecated_attr_warn("orig_data_size");
 217         return scnprintf(buf, PAGE_SIZE, "%llu\n",
 218                 (u64)(atomic64_read(&zram->stats.pages_stored)) << PAGE_SHIFT);
 219 }
 220
 221 static ssize_t mem_used_total_show(struct device *dev,
 222                 struct device_attribute *attr, char *buf)
 223 {
 224         u64 val = 0;
 225         struct zram *zram = dev_to_zram(dev);
 226
 227         deprecated_attr_warn("mem_used_total");
 228         down_read(&zram->init_lock);
 229         if (init_done(zram)) {
 230                 struct zram_meta *meta = zram->meta;
 231                 val = zs_get_total_pages(meta->mem_pool);
 232         }
 233         up_read(&zram->init_lock);
 234
 235         return scnprintf(buf, PAGE_SIZE, "%llu\n", val << PAGE_SHIFT);
 236 }
 237
 238 static ssize_t mem_limit_show(struct device *dev,
 239                 struct device_attribute *attr, char *buf)
 240 {
 241         u64 val;
 242         struct zram *zram = dev_to_zram(dev);
 243
 244         deprecated_attr_warn("mem_limit");
 245         down_read(&zram->init_lock);
 246         val = zram->limit_pages;
 247         up_read(&zram->init_lock);
 248
 249         return scnprintf(buf, PAGE_SIZE, "%llu\n", val << PAGE_SHIFT);
 250 }
 251
 252 static ssize_t mem_limit_store(struct device *dev,
 253                 struct device_attribute *attr, const char *buf, size_t len)
 254 {
 255         u64 limit;
 256         char *tmp;
 257         struct zram *zram = dev_to_zram(dev);
 258
 259         limit = memparse(buf, &tmp);
 260         if (buf == tmp) /* no chars parsed, invalid input */
 261                 return -EINVAL;
 262
 263         down_write(&zram->init_lock);
 264         zram->limit_pages = PAGE_ALIGN(limit) >> PAGE_SHIFT;
 265         up_write(&zram->init_lock);
 266
 267         return len;
 268 }
 269
 270 static ssize_t mem_used_max_show(struct device *dev,
 271                 struct device_attribute *attr, char *buf)
 272 {
 273         u64 val = 0;
 274         struct zram *zram = dev_to_zram(dev);
 275
 276         deprecated_attr_warn("mem_used_max");
 277         down_read(&zram->init_lock);
 278         if (init_done(zram))
 279                 val = atomic_long_read(&zram->stats.max_used_pages);
 280         up_read(&zram->init_lock);
 281
 282         return scnprintf(buf, PAGE_SIZE, "%llu\n", val << PAGE_SHIFT);
 283 }
 284
 285 static ssize_t mem_used_max_store(struct device *dev,
 286                 struct device_attribute *attr, const char *buf, size_t len)
 287 {
 288         int err;
 289         unsigned long val;
 290         struct zram *zram = dev_to_zram(dev);
 291
 292         err = kstrtoul(buf, 10, &val);
 293         if (err || val != 0)
 294                 return -EINVAL;
 295
 296         down_read(&zram->init_lock);
 297         if (init_done(zram)) {
 298                 struct zram_meta *meta = zram->meta;
 299                 atomic_long_set(&zram->stats.max_used_pages,
 300                                 zs_get_total_pages(meta->mem_pool));
 301         }
 302         up_read(&zram->init_lock);
 303
 304         return len;
 305 }
 306
 307 /*
 308  * We switched to per-cpu streams and this attr is not needed anymore.
 309  * However, we will keep it around for some time, because:
 310  * a) we may revert per-cpu streams in the future
 311  * b) it's visible to user space and we need to follow our 2 years
 312  *    retirement rule; but we already have a number of 'soon to be
 313  *    altered' attrs, so max_comp_streams need to wait for the next
 314  *    layoff cycle.
 315  */
 316 static ssize_t max_comp_streams_show(struct device *dev,
 317                 struct device_attribute *attr, char *buf)
 318 {
 319         return scnprintf(buf, PAGE_SIZE, "%d\n", num_online_cpus());
 320 }
 321
 322 static ssize_t max_comp_streams_store(struct device *dev,
 323                 struct device_attribute *attr, const char *buf, size_t len)
 324 {
 325         return len;
 326 }
 327
 328 static ssize_t comp_algorithm_show(struct device *dev,
 329                 struct device_attribute *attr, char *buf)
 330 {
 331         size_t sz;
 332         struct zram *zram = dev_to_zram(dev);
 333
 334         down_read(&zram->init_lock);
 335         sz = zcomp_available_show(zram->compressor, buf);
 336         up_read(&zram->init_lock);
 337
 338         return sz;
 339 }
 340
 341 static ssize_t comp_algorithm_store(struct device *dev,
 342                 struct device_attribute *attr, const char *buf, size_t len)
 343 {
 344         struct zram *zram = dev_to_zram(dev);
 345         char compressor[CRYPTO_MAX_ALG_NAME];
 346         size_t sz;
 347
 348         strlcpy(compressor, buf, sizeof(compressor));
 349         /* ignore trailing newline */
 350         sz = strlen(compressor);
 351         if (sz > 0 && compressor[sz - 1] == '\n')
 352                 compressor[sz - 1] = 0x00;
 353
 354         if (!zcomp_available_algorithm(compressor))
 355                 return -EINVAL;
 356
 357         down_write(&zram->init_lock);
 358         if (init_done(zram)) {
 359                 up_write(&zram->init_lock);
 360                 pr_info("Can't change algorithm for initialized device\n");
 361                 return -EBUSY;
 362         }
 363
 364         strlcpy(zram->compressor, compressor, sizeof(compressor));
 365         up_write(&zram->init_lock);
 366         return len;
 367 }
 368
 369 static ssize_t compact_store(struct device *dev,
 370                 struct device_attribute *attr, const char *buf, size_t len)
 371 {
 372         struct zram *zram = dev_to_zram(dev);
 373         struct zram_meta *meta;
 374
 375         down_read(&zram->init_lock);
 376         if (!init_done(zram)) {
 377                 up_read(&zram->init_lock);
 378                 return -EINVAL;
 379         }
 380
 381         meta = zram->meta;
 382         zs_compact(meta->mem_pool);
 383         up_read(&zram->init_lock);
 384
 385         return len;
 386 }
 387
 388 static ssize_t io_stat_show(struct device *dev,
 389                 struct device_attribute *attr, char *buf)
 390 {
 391         struct zram *zram = dev_to_zram(dev);
 392         ssize_t ret;
 393
 394         down_read(&zram->init_lock);
 395         ret = scnprintf(buf, PAGE_SIZE,
 396                         "%8llu %8llu %8llu %8llu\n",
 397                         (u64)atomic64_read(&zram->stats.failed_reads),
 398                         (u64)atomic64_read(&zram->stats.failed_writes),
 399                         (u64)atomic64_read(&zram->stats.invalid_io),
 400                         (u64)atomic64_read(&zram->stats.notify_free));
 401         up_read(&zram->init_lock);
 402
 403         return ret;
 404 }
 405
 406 static ssize_t mm_stat_show(struct device *dev,
 407                 struct device_attribute *attr, char *buf)
 408 {
 409         struct zram *zram = dev_to_zram(dev);
 410         struct zs_pool_stats pool_stats;
 411         u64 orig_size, mem_used = 0;
 412         long max_used;
 413         ssize_t ret;
 414
 415         memset(&pool_stats, 0x00, sizeof(struct zs_pool_stats));
 416
 417         down_read(&zram->init_lock);
 418         if (init_done(zram)) {
 419                 mem_used = zs_get_total_pages(zram->meta->mem_pool);
 420                 zs_pool_stats(zram->meta->mem_pool, &pool_stats);
 421         }
 422
 423         orig_size = atomic64_read(&zram->stats.pages_stored);
 424         max_used = atomic_long_read(&zram->stats.max_used_pages);
 425
 426         ret = scnprintf(buf, PAGE_SIZE,
 427                         "%8llu %8llu %8llu %8lu %8ld %8llu %8lu\n",
 428                         orig_size << PAGE_SHIFT,
 429                         (u64)atomic64_read(&zram->stats.compr_data_size),
 430                         mem_used << PAGE_SHIFT,
 431                         zram->limit_pages << PAGE_SHIFT,
 432                         max_used << PAGE_SHIFT,
 433                         (u64)atomic64_read(&zram->stats.zero_pages),
 434                         pool_stats.pages_compacted);
 435         up_read(&zram->init_lock);
 436
 437         return ret;
 438 }
 439
 440 static ssize_t debug_stat_show(struct device *dev,
 441                 struct device_attribute *attr, char *buf)
 442 {
 443         int version = 1;
 444         struct zram *zram = dev_to_zram(dev);
 445         ssize_t ret;
 446
 447         down_read(&zram->init_lock);
 448         ret = scnprintf(buf, PAGE_SIZE,
 449                         "version: %d\n%8llu\n",
 450                         version,
 451                         (u64)atomic64_read(&zram->stats.writestall));
 452         up_read(&zram->init_lock);
 453
 454         return ret;
 455 }
 456
 457 static DEVICE_ATTR_RO(io_stat);
 458 static DEVICE_ATTR_RO(mm_stat);
 459 static DEVICE_ATTR_RO(debug_stat);
 460 ZRAM_ATTR_RO(num_reads);
 461 ZRAM_ATTR_RO(num_writes);
 462 ZRAM_ATTR_RO(failed_reads);
 463 ZRAM_ATTR_RO(failed_writes);
 464 ZRAM_ATTR_RO(invalid_io);
 465 ZRAM_ATTR_RO(notify_free);
 466 ZRAM_ATTR_RO(zero_pages);
 467 ZRAM_ATTR_RO(compr_data_size);
 468
 469 static inline bool zram_meta_get(struct zram *zram)
 470 {
 471         if (atomic_inc_not_zero(&zram->refcount))
 472                 return true;
 473         return false;
 474 }
 475
 476 static inline void zram_meta_put(struct zram *zram)
 477 {
 478         atomic_dec(&zram->refcount);
 479 }
 480
 481 static void zram_meta_free(struct zram_meta *meta, u64 disksize)
 482 {
 483         size_t num_pages = disksize >> PAGE_SHIFT;
 484         size_t index;
 485
 486         /* Free all pages that are still in this zram device */
 487         for (index = 0; index < num_pages; index++) {
 488                 unsigned long handle = meta->table[index].handle;
 489
 490                 if (!handle)
 491                         continue;
 492
 493                 zs_free(meta->mem_pool, handle);
 494         }
 495
 496         zs_destroy_pool(meta->mem_pool);
 497         vfree(meta->table);
 498         kfree(meta);
 499 }
 500
 501 static struct zram_meta *zram_meta_alloc(char *pool_name, u64 disksize)
 502 {
 503         size_t num_pages;
 504         struct zram_meta *meta = kmalloc(sizeof(*meta), GFP_KERNEL);
 505
 506         if (!meta)
 507                 return NULL;
 508
 509         num_pages = disksize >> PAGE_SHIFT;
 510         meta->table = vzalloc(num_pages * sizeof(*meta->table));
 511         if (!meta->table) {
 512                 pr_err("Error allocating zram address table\n");
 513                 goto out_error;
 514         }
 515
 516         meta->mem_pool = zs_create_pool(pool_name);
 517         if (!meta->mem_pool) {
 518                 pr_err("Error creating memory pool\n");
 519                 goto out_error;
 520         }
 521
 522         zram_meta_init_table_locks(meta, disksize);
 523
 524         return meta;
 525
 526 out_error:
 527         vfree(meta->table);
 528         kfree(meta);
 529         return NULL;
 530 }
 531
 532 /*
 533  * To protect concurrent access to the same index entry,
 534  * caller should hold this table index entry's bit_spinlock to
 535  * indicate this index entry is accessing.
 536  */
 537 static void zram_free_page(struct zram *zram, size_t index)
 538 {
 539         struct zram_meta *meta = zram->meta;
 540         unsigned long handle = meta->table[index].handle;
 541
 542         if (unlikely(!handle)) {
 543                 /*
 544                  * No memory is allocated for zero filled pages.
 545                  * Simply clear zero page flag.
 546                  */
 547                 if (zram_test_flag(meta, index, ZRAM_ZERO)) {
 548                         zram_clear_flag(meta, index, ZRAM_ZERO);
 549                         atomic64_dec(&zram->stats.zero_pages);
 550                 }
 551                 return;
 552         }
 553
 554         zs_free(meta->mem_pool, handle);
 555
 556         atomic64_sub(zram_get_obj_size(meta, index),
 557                         &zram->stats.compr_data_size);
 558         atomic64_dec(&zram->stats.pages_stored);
 559
 560         meta->table[index].handle = 0;
 561         zram_set_obj_size(meta, index, 0);
 562 }
 563
 564 static int zram_decompress_page(struct zram *zram, char *mem, u32 index)
 565 {
 566         int ret = 0;
 567         unsigned char *cmem;
 568         struct zram_meta *meta = zram->meta;
 569         unsigned long handle;
 570         unsigned int size;
 571         struct zcomp_strm *zstrm;
 572
 573         zram_lock_table(&meta->table[index]);
 574         handle = meta->table[index].handle;
 575         size = zram_get_obj_size(meta, index);
 576
 577         if (!handle || zram_test_flag(meta, index, ZRAM_ZERO)) {
 578                 zram_unlock_table(&meta->table[index]);
 579                 clear_page(mem);
 580                 return 0;
 581         }
 582
 583         zstrm = zcomp_stream_get(zram->comp);
 584         cmem = zs_map_object(meta->mem_pool, handle, ZS_MM_RO);
 585         if (size == PAGE_SIZE) {
 586                 copy_page(mem, cmem);
 587         } else {
 588                 ret = zcomp_decompress(zstrm, cmem, size, mem);
 589         }
 590         zs_unmap_object(meta->mem_pool, handle);
 591         zcomp_stream_put(zram->comp);
 592         zram_unlock_table(&meta->table[index]);
 593
 594         /* Should NEVER happen. Return bio error if it does. */
 595         if (unlikely(ret)) {
 596                 pr_err("Decompression failed! err=%d, page=%u\n", ret, index);
 597                 return ret;
 598         }
 599
 600         return 0;
 601 }
 602
 603 static int zram_bvec_read(struct zram *zram, struct bio_vec *bvec,
 604                           u32 index, int offset)
 605 {
 606         int ret;
 607         struct page *page;
 608         unsigned char *user_mem, *uncmem = NULL;
 609         struct zram_meta *meta = zram->meta;
 610         page = bvec->bv_page;
 611
 612         zram_lock_table(&meta->table[index]);
 613         if (unlikely(!meta->table[index].handle) ||
 614                         zram_test_flag(meta, index, ZRAM_ZERO)) {
 615                 zram_unlock_table(&meta->table[index]);
 616                 handle_zero_page(bvec);
 617                 return 0;
 618         }
 619         zram_unlock_table(&meta->table[index]);
 620
 621         if (is_partial_io(bvec))
 622                 /* Use  a temporary buffer to decompress the page */
 623                 uncmem = kmalloc(PAGE_SIZE, GFP_NOIO);
 624
 625         user_mem = kmap_atomic(page);
 626         if (!is_partial_io(bvec))
 627                 uncmem = user_mem;
 628
 629         if (!uncmem) {
 630                 pr_err("Unable to allocate temp memory\n");
 631                 ret = -ENOMEM;
 632                 goto out_cleanup;
 633         }
 634
 635         ret = zram_decompress_page(zram, uncmem, index);
 636         /* Should NEVER happen. Return bio error if it does. */
 637         if (unlikely(ret))
 638                 goto out_cleanup;
 639
 640         if (is_partial_io(bvec))
 641                 memcpy(user_mem + bvec->bv_offset, uncmem + offset,
 642                                 bvec->bv_len);
 643
 644         flush_dcache_page(page);
 645         ret = 0;
 646 out_cleanup:
 647         kunmap_atomic(user_mem);
 648         if (is_partial_io(bvec))
 649                 kfree(uncmem);
 650         return ret;
 651 }
 652
 653 static int zram_bvec_write(struct zram *zram, struct bio_vec *bvec, u32 index,
 654                            int offset)
 655 {
 656         int ret = 0;
 657         unsigned int clen;
 658         unsigned long handle = 0;
 659         struct page *page;
 660         unsigned char *user_mem, *cmem, *src, *uncmem = NULL;
 661         struct zram_meta *meta = zram->meta;
 662         struct zcomp_strm *zstrm = NULL;
 663         unsigned long alloced_pages;
 664
 665         page = bvec->bv_page;
 666         if (is_partial_io(bvec)) {
 667                 /*
 668                  * This is a partial IO. We need to read the full page
 669                  * before to write the changes.
 670                  */
 671                 uncmem = kmalloc(PAGE_SIZE, GFP_NOIO);
 672                 if (!uncmem) {
 673                         ret = -ENOMEM;
 674                         goto out;
 675                 }
 676                 ret = zram_decompress_page(zram, uncmem, index);
 677                 if (ret)
 678                         goto out;
 679         }
 680
 681 compress_again:
 682         user_mem = kmap_atomic(page);
 683         if (is_partial_io(bvec)) {
 684                 memcpy(uncmem + offset, user_mem + bvec->bv_offset,
 685                        bvec->bv_len);
 686                 kunmap_atomic(user_mem);
 687                 user_mem = NULL;
 688         } else {
 689                 uncmem = user_mem;
 690         }
 691
 692         if (page_zero_filled(uncmem)) {
 693                 if (user_mem)
 694                         kunmap_atomic(user_mem);
 695                 /* Free memory associated with this sector now. */
 696                 zram_lock_table(&meta->table[index]);
 697                 zram_free_page(zram, index);
 698                 zram_set_flag(meta, index, ZRAM_ZERO);
 699                 zram_unlock_table(&meta->table[index]);
 700
 701                 atomic64_inc(&zram->stats.zero_pages);
 702                 ret = 0;
 703                 goto out;
 704         }
 705
 706         zstrm = zcomp_stream_get(zram->comp);
 707         ret = zcomp_compress(zstrm, uncmem, &clen);
 708         if (!is_partial_io(bvec)) {
 709                 kunmap_atomic(user_mem);
 710                 user_mem = NULL;
 711                 uncmem = NULL;
 712         }
 713
 714         if (unlikely(ret)) {
 715                 pr_err("Compression failed! err=%d\n", ret);
 716                 goto out;
 717         }
 718
 719         src = zstrm->buffer;
 720         if (unlikely(clen > max_zpage_size)) {
 721                 clen = PAGE_SIZE;
 722                 if (is_partial_io(bvec))
 723                         src = uncmem;
 724         }
 725
 726         /*
 727          * handle allocation has 2 paths:
 728          * a) fast path is executed with preemption disabled (for
 729          *  per-cpu streams) and has __GFP_DIRECT_RECLAIM bit clear,
 730          *  since we can't sleep;
 731          * b) slow path enables preemption and attempts to allocate
 732          *  the page with __GFP_DIRECT_RECLAIM bit set. we have to
 733          *  put per-cpu compression stream and, thus, to re-do
 734          *  the compression once handle is allocated.
 735          *
 736          * if we have a 'non-null' handle here then we are coming
 737          * from the slow path and handle has already been allocated.
 738          */
 739         if (!handle)
 740                 handle = zs_malloc(meta->mem_pool, clen,
 741                                 __GFP_KSWAPD_RECLAIM |
 742                                 __GFP_NOWARN |
 743                                 __GFP_HIGHMEM |
 744                                 __GFP_MOVABLE);
 745         if (!handle) {
 746                 zcomp_stream_put(zram->comp);
 747                 zstrm = NULL;
 748
 749                 atomic64_inc(&zram->stats.writestall);
 750
 751                 handle = zs_malloc(meta->mem_pool, clen,
 752                                 GFP_NOIO | __GFP_HIGHMEM |
 753                                 __GFP_MOVABLE);
 754                 if (handle)
 755                         goto compress_again;
 756
 757                 pr_err("Error allocating memory for compressed page: %u, size=%u\n",
 758                         index, clen);
 759                 ret = -ENOMEM;
 760                 goto out;
 761         }
 762
 763         alloced_pages = zs_get_total_pages(meta->mem_pool);
 764         update_used_max(zram, alloced_pages);
 765
 766         if (zram->limit_pages && alloced_pages > zram->limit_pages) {
 767                 zs_free(meta->mem_pool, handle);
 768                 ret = -ENOMEM;
 769                 goto out;
 770         }
 771
 772         cmem = zs_map_object(meta->mem_pool, handle, ZS_MM_WO);
 773
 774         if ((clen == PAGE_SIZE) && !is_partial_io(bvec)) {
 775                 src = kmap_atomic(page);
 776                 copy_page(cmem, src);
 777                 kunmap_atomic(src);
 778         } else {
 779                 memcpy(cmem, src, clen);
 780         }
 781
 782         zcomp_stream_put(zram->comp);
 783         zstrm = NULL;
 784         zs_unmap_object(meta->mem_pool, handle);
 785
 786         /*
 787          * Free memory associated with this sector
 788          * before overwriting unused sectors.
 789          */
 790         zram_lock_table(&meta->table[index]);
 791         zram_free_page(zram, index);
 792
 793         meta->table[index].handle = handle;
 794         zram_set_obj_size(meta, index, clen);
 795         zram_unlock_table(&meta->table[index]);
 796
 797         /* Update stats */
 798         atomic64_add(clen, &zram->stats.compr_data_size);
 799         atomic64_inc(&zram->stats.pages_stored);
 800 out:
 801         if (zstrm)
 802                 zcomp_stream_put(zram->comp);
 803         if (is_partial_io(bvec))
 804                 kfree(uncmem);
 805         return ret;
 806 }
 807
 808 /*
 809  * zram_bio_discard - handler on discard request
 810  * @index: physical block index in PAGE_SIZE units
 811  * @offset: byte offset within physical block
 812  */
 813 static void zram_bio_discard(struct zram *zram, u32 index,
 814                              int offset, struct bio *bio)
 815 {
 816         size_t n = bio->bi_iter.bi_size;
 817         struct zram_meta *meta = zram->meta;
 818
 819         /*
 820          * zram manages data in physical block size units. Because logical block
 821          * size isn't identical with physical block size on some arch, we
 822          * could get a discard request pointing to a specific offset within a
 823          * certain physical block.  Although we can handle this request by
 824          * reading that physiclal block and decompressing and partially zeroing
 825          * and re-compressing and then re-storing it, this isn't reasonable
 826          * because our intent with a discard request is to save memory.  So
 827          * skipping this logical block is appropriate here.
 828          */
 829         if (offset) {
 830                 if (n <= (PAGE_SIZE - offset))
 831                         return;
 832
 833                 n -= (PAGE_SIZE - offset);
 834                 index++;
 835         }
 836
 837         while (n >= PAGE_SIZE) {
 838                 zram_lock_table(&meta->table[index]);
 839                 zram_free_page(zram, index);
 840                 zram_unlock_table(&meta->table[index]);
 841                 atomic64_inc(&zram->stats.notify_free);
 842                 index++;
 843                 n -= PAGE_SIZE;
 844         }
 845 }
 846
 847 static int zram_bvec_rw(struct zram *zram, struct bio_vec *bvec, u32 index,
 848                         int offset, bool is_write)
 849 {
 850         unsigned long start_time = jiffies;
 851         int rw_acct = is_write ? REQ_OP_WRITE : REQ_OP_READ;
 852         int ret;
 853
 854         generic_start_io_acct(rw_acct, bvec->bv_len >> SECTOR_SHIFT,
 855                         &zram->disk->part0);
 856
 857         if (!is_write) {
 858                 atomic64_inc(&zram->stats.num_reads);
 859                 ret = zram_bvec_read(zram, bvec, index, offset);
 860         } else {
 861                 atomic64_inc(&zram->stats.num_writes);
 862                 ret = zram_bvec_write(zram, bvec, index, offset);
 863         }
 864
 865         generic_end_io_acct(rw_acct, &zram->disk->part0, start_time);
 866
 867         if (unlikely(ret)) {
 868                 if (!is_write)
 869                         atomic64_inc(&zram->stats.failed_reads);
 870                 else
 871                         atomic64_inc(&zram->stats.failed_writes);
 872         }
 873
 874         return ret;
 875 }
 876
 877 static void __zram_make_request(struct zram *zram, struct bio *bio)
 878 {
 879         int offset;
 880         u32 index;
 881         struct bio_vec bvec;
 882         struct bvec_iter iter;
 883
 884         index = bio->bi_iter.bi_sector >> SECTORS_PER_PAGE_SHIFT;
 885         offset = (bio->bi_iter.bi_sector &
 886                   (SECTORS_PER_PAGE - 1)) << SECTOR_SHIFT;
 887
 888         if (unlikely(bio_op(bio) == REQ_OP_DISCARD)) {
 889                 zram_bio_discard(zram, index, offset, bio);
 890                 bio_endio(bio);
 891                 return;
 892         }
 893
 894         bio_for_each_segment(bvec, bio, iter) {
 895                 int max_transfer_size = PAGE_SIZE - offset;
 896
 897                 if (bvec.bv_len > max_transfer_size) {
 898                         /*
 899                          * zram_bvec_rw() can only make operation on a single
 900                          * zram page. Split the bio vector.
 901                          */
 902                         struct bio_vec bv;
 903
 904                         bv.bv_page = bvec.bv_page;
 905                         bv.bv_len = max_transfer_size;
 906                         bv.bv_offset = bvec.bv_offset;
 907
 908                         if (zram_bvec_rw(zram, &bv, index, offset,
 909                                          op_is_write(bio_op(bio))) < 0)
 910                                 goto out;
 911
 912                         bv.bv_len = bvec.bv_len - max_transfer_size;
 913                         bv.bv_offset += max_transfer_size;
 914                         if (zram_bvec_rw(zram, &bv, index + 1, 0,
 915                                          op_is_write(bio_op(bio))) < 0)
 916                                 goto out;
 917                 } else
 918                         if (zram_bvec_rw(zram, &bvec, index, offset,
 919                                          op_is_write(bio_op(bio))) < 0)
 920                                 goto out;
 921
 922                 update_position(&index, &offset, &bvec);
 923         }
 924
 925         bio_endio(bio);
 926         return;
 927
 928 out:
 929         bio_io_error(bio);
 930 }
 931
 932 /*
 933  * Handler function for all zram I/O requests.
 934  */
 935 static blk_qc_t zram_make_request(struct request_queue *queue, struct bio *bio)
 936 {
 937         struct zram *zram = queue->queuedata;
 938
 939         if (unlikely(!zram_meta_get(zram)))
 940                 goto error;
 941
 942         blk_queue_split(queue, &bio, queue->bio_split);
 943
 944         if (!valid_io_request(zram, bio->bi_iter.bi_sector,
 945                                         bio->bi_iter.bi_size)) {
 946                 atomic64_inc(&zram->stats.invalid_io);
 947                 goto put_zram;
 948         }
 949
 950         __zram_make_request(zram, bio);
 951         zram_meta_put(zram);
 952         return BLK_QC_T_NONE;
 953 put_zram:
 954         zram_meta_put(zram);
 955 error:
 956         bio_io_error(bio);
 957         return BLK_QC_T_NONE;
 958 }
 959
 960 static void zram_slot_free_notify(struct block_device *bdev,
 961                                 unsigned long index)
 962 {
 963         struct zram *zram;
 964         struct zram_meta *meta;
 965
 966         zram = bdev->bd_disk->private_data;
 967         meta = zram->meta;
 968
 969         zram_lock_table(&meta->table[index]);
 970         zram_free_page(zram, index);
 971         zram_unlock_table(&meta->table[index]);
 972         atomic64_inc(&zram->stats.notify_free);
 973 }
 974
 975 static int zram_rw_page(struct block_device *bdev, sector_t sector,
 976                        struct page *page, bool is_write)
 977 {
 978         int offset, err = -EIO;
 979         u32 index;
 980         struct zram *zram;
 981         struct bio_vec bv;
 982
 983         zram = bdev->bd_disk->private_data;
 984         if (unlikely(!zram_meta_get(zram)))
 985                 goto out;
 986
 987         if (!valid_io_request(zram, sector, PAGE_SIZE)) {
 988                 atomic64_inc(&zram->stats.invalid_io);
 989                 err = -EINVAL;
 990                 goto put_zram;
 991         }
 992
 993         index = sector >> SECTORS_PER_PAGE_SHIFT;
 994         offset = sector & (SECTORS_PER_PAGE - 1) << SECTOR_SHIFT;
 995
 996         bv.bv_page = page;
 997         bv.bv_len = PAGE_SIZE;
 998         bv.bv_offset = 0;
 999
1000         err = zram_bvec_rw(zram, &bv, index, offset, is_write);
1001 put_zram:
1002         zram_meta_put(zram);
1003 out:
1004         /*
1005          * If I/O fails, just return error(ie, non-zero) without
1006          * calling page_endio.
1007          * It causes resubmit the I/O with bio request by upper functions
1008          * of rw_page(e.g., swap_readpage, __swap_writepage) and
1009          * bio->bi_end_io does things to handle the error
1010          * (e.g., SetPageError, set_page_dirty and extra works).
1011          */
1012         if (err == 0)
1013                 page_endio(page, is_write, 0);
1014         return err;
1015 }
1016
1017 static void zram_reset_device(struct zram *zram)
1018 {
1019         struct zram_meta *meta;
1020         struct zcomp *comp;
1021         u64 disksize;
1022
1023         down_write(&zram->init_lock);
1024
1025         zram->limit_pages = 0;
1026
1027         if (!init_done(zram)) {
1028                 up_write(&zram->init_lock);
1029                 return;
1030         }
1031
1032         meta = zram->meta;
1033         comp = zram->comp;
1034         disksize = zram->disksize;
1035         /*
1036          * Refcount will go down to 0 eventually and r/w handler
1037          * cannot handle further I/O so it will bail out by
1038          * check zram_meta_get.
1039          */
1040         zram_meta_put(zram);
1041         /*
1042          * We want to free zram_meta in process context to avoid
1043          * deadlock between reclaim path and any other locks.
1044          */
1045         wait_event(zram->io_done, atomic_read(&zram->refcount) == 0);
1046
1047         /* Reset stats */
1048         memset(&zram->stats, 0, sizeof(zram->stats));
1049         zram->disksize = 0;
1050
1051         set_capacity(zram->disk, 0);
1052         part_stat_set_all(&zram->disk->part0, 0);
1053
1054         up_write(&zram->init_lock);
1055         /* I/O operation under all of CPU are done so let's free */
1056         zram_meta_free(meta, disksize);
1057         zcomp_destroy(comp);
1058 }
1059
1060 static ssize_t disksize_store(struct device *dev,
1061                 struct device_attribute *attr, const char *buf, size_t len)
1062 {
1063         u64 disksize;
1064         struct zcomp *comp;
1065         struct zram_meta *meta;
1066         struct zram *zram = dev_to_zram(dev);
1067         int err;
1068
1069         disksize = memparse(buf, NULL);
1070         if (!disksize)
1071                 return -EINVAL;
1072
1073         disksize = PAGE_ALIGN(disksize);
1074         meta = zram_meta_alloc(zram->disk->disk_name, disksize);
1075         if (!meta)
1076                 return -ENOMEM;
1077
1078         comp = zcomp_create(zram->compressor);
1079         if (IS_ERR(comp)) {
1080                 pr_err("Cannot initialise %s compressing backend\n",
1081                                 zram->compressor);
1082                 err = PTR_ERR(comp);
1083                 goto out_free_meta;
1084         }
1085
1086         down_write(&zram->init_lock);
1087         if (init_done(zram)) {
1088                 pr_info("Cannot change disksize for initialized device\n");
1089                 err = -EBUSY;
1090                 goto out_destroy_comp;
1091         }
1092
1093         init_waitqueue_head(&zram->io_done);
1094         atomic_set(&zram->refcount, 1);
1095         zram->meta = meta;
1096         zram->comp = comp;
1097         zram->disksize = disksize;
1098         set_capacity(zram->disk, zram->disksize >> SECTOR_SHIFT);
1099         up_write(&zram->init_lock);
1100
1101         /*
1102          * Revalidate disk out of the init_lock to avoid lockdep splat.
1103          * It's okay because disk's capacity is protected by init_lock
1104          * so that revalidate_disk always sees up-to-date capacity.
1105          */
1106         revalidate_disk(zram->disk);
1107
1108         return len;
1109
1110 out_destroy_comp:
1111         up_write(&zram->init_lock);
1112         zcomp_destroy(comp);
1113 out_free_meta:
1114         zram_meta_free(meta, disksize);
1115         return err;
1116 }
1117
1118 static ssize_t reset_store(struct device *dev,
1119                 struct device_attribute *attr, const char *buf, size_t len)
1120 {
1121         int ret;
1122         unsigned short do_reset;
1123         struct zram *zram;
1124         struct block_device *bdev;
1125
1126         ret = kstrtou16(buf, 10, &do_reset);
1127         if (ret)
1128                 return ret;
1129
1130         if (!do_reset)
1131                 return -EINVAL;
1132
1133         zram = dev_to_zram(dev);
1134         bdev = bdget_disk(zram->disk, 0);
1135         if (!bdev)
1136                 return -ENOMEM;
1137
1138         mutex_lock(&bdev->bd_mutex);
1139         /* Do not reset an active device or claimed device */
1140         if (bdev->bd_openers || zram->claim) {
1141                 mutex_unlock(&bdev->bd_mutex);
1142                 bdput(bdev);
1143                 return -EBUSY;
1144         }
1145
1146         /* From now on, anyone can't open /dev/zram[0-9] */
1147         zram->claim = true;
1148         mutex_unlock(&bdev->bd_mutex);
1149
1150         /* Make sure all the pending I/O are finished */
1151         fsync_bdev(bdev);
1152         zram_reset_device(zram);
1153         revalidate_disk(zram->disk);
1154         bdput(bdev);
1155
1156         mutex_lock(&bdev->bd_mutex);
1157         zram->claim = false;
1158         mutex_unlock(&bdev->bd_mutex);
1159
1160         return len;
1161 }
1162
1163 static int zram_open(struct block_device *bdev, fmode_t mode)
1164 {
1165         int ret = 0;
1166         struct zram *zram;
1167
1168         WARN_ON(!mutex_is_locked(&bdev->bd_mutex));
1169
1170         zram = bdev->bd_disk->private_data;
1171         /* zram was claimed to reset so open request fails */
1172         if (zram->claim)
1173                 ret = -EBUSY;
1174
1175         return ret;
1176 }
1177
1178 static const struct block_device_operations zram_devops = {
1179         .open = zram_open,
1180         .swap_slot_free_notify = zram_slot_free_notify,
1181         .rw_page = zram_rw_page,
1182         .owner = THIS_MODULE
1183 };
1184
1185 static DEVICE_ATTR_WO(compact);
1186 static DEVICE_ATTR_RW(disksize);
1187 static DEVICE_ATTR_RO(initstate);
1188 static DEVICE_ATTR_WO(reset);
1189 static DEVICE_ATTR_RO(orig_data_size);
1190 static DEVICE_ATTR_RO(mem_used_total);
1191 static DEVICE_ATTR_RW(mem_limit);
1192 static DEVICE_ATTR_RW(mem_used_max);
1193 static DEVICE_ATTR_RW(max_comp_streams);
1194 static DEVICE_ATTR_RW(comp_algorithm);
1195
1196 static struct attribute *zram_disk_attrs[] = {
1197         &dev_attr_disksize.attr,
1198         &dev_attr_initstate.attr,
1199         &dev_attr_reset.attr,
1200         &dev_attr_num_reads.attr,
1201         &dev_attr_num_writes.attr,
1202         &dev_attr_failed_reads.attr,
1203         &dev_attr_failed_writes.attr,
1204         &dev_attr_compact.attr,
1205         &dev_attr_invalid_io.attr,
1206         &dev_attr_notify_free.attr,
1207         &dev_attr_zero_pages.attr,
1208         &dev_attr_orig_data_size.attr,
1209         &dev_attr_compr_data_size.attr,
1210         &dev_attr_mem_used_total.attr,
1211         &dev_attr_mem_limit.attr,
1212         &dev_attr_mem_used_max.attr,
1213         &dev_attr_max_comp_streams.attr,
1214         &dev_attr_comp_algorithm.attr,
1215         &dev_attr_io_stat.attr,
1216         &dev_attr_mm_stat.attr,
1217         &dev_attr_debug_stat.attr,
1218         NULL,
1219 };
1220
1221 static struct attribute_group zram_disk_attr_group = {
1222         .attrs = zram_disk_attrs,
1223 };
1224
1225 /*
1226  * Allocate and initialize new zram device. the function returns
1227  * '>= 0' device_id upon success, and negative value otherwise.
1228  */
1229 static int zram_add(void)
1230 {
1231         struct zram *zram;
1232         struct request_queue *queue;
1233         int ret, device_id;
1234
1235         zram = kzalloc(sizeof(struct zram), GFP_KERNEL);
1236         if (!zram)
1237                 return -ENOMEM;
1238
1239         ret = idr_alloc(&zram_index_idr, zram, 0, 0, GFP_KERNEL);
1240         if (ret < 0)
1241                 goto out_free_dev;
1242         device_id = ret;
1243
1244         init_rwsem(&zram->init_lock);
1245
1246         queue = blk_alloc_queue(GFP_KERNEL);
1247         if (!queue) {
1248                 pr_err("Error allocating disk queue for device %d\n",
1249                         device_id);
1250                 ret = -ENOMEM;
1251                 goto out_free_idr;
1252         }
1253
1254         blk_queue_make_request(queue, zram_make_request);
1255
1256         /* gendisk structure */
1257         zram->disk = alloc_disk(1);
1258         if (!zram->disk) {
1259                 pr_err("Error allocating disk structure for device %d\n",
1260                         device_id);
1261                 ret = -ENOMEM;
1262                 goto out_free_queue;
1263         }
1264
1265         zram->disk->major = zram_major;
1266         zram->disk->first_minor = device_id;
1267         zram->disk->fops = &zram_devops;
1268         zram->disk->queue = queue;
1269         zram->disk->queue->queuedata = zram;
1270         zram->disk->private_data = zram;
1271         snprintf(zram->disk->disk_name, 16, "zram%d", device_id);
1272
1273         /* Actual capacity set using syfs (/sys/block/zram<id>/disksize */
1274         set_capacity(zram->disk, 0);
1275         /* zram devices sort of resembles non-rotational disks */
1276         queue_flag_set_unlocked(QUEUE_FLAG_NONROT, zram->disk->queue);
1277         queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, zram->disk->queue);
1278         /*
1279          * To ensure that we always get PAGE_SIZE aligned
1280          * and n*PAGE_SIZED sized I/O requests.
1281          */
1282         blk_queue_physical_block_size(zram->disk->queue, PAGE_SIZE);
1283         blk_queue_logical_block_size(zram->disk->queue,
1284                                         ZRAM_LOGICAL_BLOCK_SIZE);
1285         blk_queue_io_min(zram->disk->queue, PAGE_SIZE);
1286         blk_queue_io_opt(zram->disk->queue, PAGE_SIZE);
1287         zram->disk->queue->limits.discard_granularity = PAGE_SIZE;
1288         blk_queue_max_discard_sectors(zram->disk->queue, UINT_MAX);
1289         /*
1290          * zram_bio_discard() will clear all logical blocks if logical block
1291          * size is identical with physical block size(PAGE_SIZE). But if it is
1292          * different, we will skip discarding some parts of logical blocks in
1293          * the part of the request range which isn't aligned to physical block
1294          * size.  So we can't ensure that all discarded logical blocks are
1295          * zeroed.
1296          */
1297         if (ZRAM_LOGICAL_BLOCK_SIZE == PAGE_SIZE)
1298                 zram->disk->queue->limits.discard_zeroes_data = 1;
1299         else
1300                 zram->disk->queue->limits.discard_zeroes_data = 0;
1301         queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, zram->disk->queue);
1302
1303         add_disk(zram->disk);
1304
1305         ret = sysfs_create_group(&disk_to_dev(zram->disk)->kobj,
1306                                 &zram_disk_attr_group);
1307         if (ret < 0) {
1308                 pr_err("Error creating sysfs group for device %d\n",
1309                                 device_id);
1310                 goto out_free_disk;
1311         }
1312         strlcpy(zram->compressor, default_compressor, sizeof(zram->compressor));
1313         zram->meta = NULL;
1314
1315         pr_info("Added device: %s\n", zram->disk->disk_name);
1316         return device_id;
1317
1318 out_free_disk:
1319         del_gendisk(zram->disk);
1320         put_disk(zram->disk);
1321 out_free_queue:
1322         blk_cleanup_queue(queue);
1323 out_free_idr:
1324         idr_remove(&zram_index_idr, device_id);
1325 out_free_dev:
1326         kfree(zram);
1327         return ret;
1328 }
1329
1330 static int zram_remove(struct zram *zram)
1331 {
1332         struct block_device *bdev;
1333
1334         bdev = bdget_disk(zram->disk, 0);
1335         if (!bdev)
1336                 return -ENOMEM;
1337
1338         mutex_lock(&bdev->bd_mutex);
1339         if (bdev->bd_openers || zram->claim) {
1340                 mutex_unlock(&bdev->bd_mutex);
1341                 bdput(bdev);
1342                 return -EBUSY;
1343         }
1344
1345         zram->claim = true;
1346         mutex_unlock(&bdev->bd_mutex);
1347
1348         /*
1349          * Remove sysfs first, so no one will perform a disksize
1350          * store while we destroy the devices. This also helps during
1351          * hot_remove -- zram_reset_device() is the last holder of
1352          * ->init_lock, no later/concurrent disksize_store() or any
1353          * other sysfs handlers are possible.
1354          */
1355         sysfs_remove_group(&disk_to_dev(zram->disk)->kobj,
1356                         &zram_disk_attr_group);
1357
1358         /* Make sure all the pending I/O are finished */
1359         fsync_bdev(bdev);
1360         zram_reset_device(zram);
1361         bdput(bdev);
1362
1363         pr_info("Removed device: %s\n", zram->disk->disk_name);
1364
1365         blk_cleanup_queue(zram->disk->queue);
1366         del_gendisk(zram->disk);
1367         put_disk(zram->disk);
1368         kfree(zram);
1369         return 0;
1370 }
1371
1372 /* zram-control sysfs attributes */
1373 static ssize_t hot_add_show(struct class *class,
1374                         struct class_attribute *attr,
1375                         char *buf)
1376 {
1377         int ret;
1378
1379         mutex_lock(&zram_index_mutex);
1380         ret = zram_add();
1381         mutex_unlock(&zram_index_mutex);
1382
1383         if (ret < 0)
1384                 return ret;
1385         return scnprintf(buf, PAGE_SIZE, "%d\n", ret);
1386 }
1387
1388 static ssize_t hot_remove_store(struct class *class,
1389                         struct class_attribute *attr,
1390                         const char *buf,
1391                         size_t count)
1392 {
1393         struct zram *zram;
1394         int ret, dev_id;
1395
1396         /* dev_id is gendisk->first_minor, which is `int' */
1397         ret = kstrtoint(buf, 10, &dev_id);
1398         if (ret)
1399                 return ret;
1400         if (dev_id < 0)
1401                 return -EINVAL;
1402
1403         mutex_lock(&zram_index_mutex);
1404
1405         zram = idr_find(&zram_index_idr, dev_id);
1406         if (zram) {
1407                 ret = zram_remove(zram);
1408                 if (!ret)
1409                         idr_remove(&zram_index_idr, dev_id);
1410         } else {
1411                 ret = -ENODEV;
1412         }
1413
1414         mutex_unlock(&zram_index_mutex);
1415         return ret ? ret : count;
1416 }
1417
1418 /*
1419  * NOTE: hot_add attribute is not the usual read-only sysfs attribute. In a
1420  * sense that reading from this file does alter the state of your system -- it
1421  * creates a new un-initialized zram device and returns back this device's
1422  * device_id (or an error code if it fails to create a new device).
1423  */
1424 static struct class_attribute zram_control_class_attrs[] = {
1425         __ATTR(hot_add, 0400, hot_add_show, NULL),
1426         __ATTR_WO(hot_remove),
1427         __ATTR_NULL,
1428 };
1429
1430 static struct class zram_control_class = {
1431         .name           = "zram-control",
1432         .owner          = THIS_MODULE,
1433         .class_attrs    = zram_control_class_attrs,
1434 };
1435
1436 static int zram_remove_cb(int id, void *ptr, void *data)
1437 {
1438         zram_remove(ptr);
1439         return 0;
1440 }
1441
1442 static void destroy_devices(void)
1443 {
1444         class_unregister(&zram_control_class);
1445         idr_for_each(&zram_index_idr, &zram_remove_cb, NULL);
1446         idr_destroy(&zram_index_idr);
1447         unregister_blkdev(zram_major, "zram");
1448 }
1449
1450 static int __init zram_init(void)
1451 {
1452         int ret;
1453
1454         ret = class_register(&zram_control_class);
1455         if (ret) {
1456                 pr_err("Unable to register zram-control class\n");
1457                 return ret;
1458         }
1459
1460         zram_major = register_blkdev(0, "zram");
1461         if (zram_major <= 0) {
1462                 pr_err("Unable to get major number\n");
1463                 class_unregister(&zram_control_class);
1464                 return -EBUSY;
1465         }
1466
1467         while (num_devices != 0) {
1468                 mutex_lock(&zram_index_mutex);
1469                 ret = zram_add();
1470                 mutex_unlock(&zram_index_mutex);
1471                 if (ret < 0)
1472                         goto out_error;
1473                 num_devices--;
1474         }
1475
1476         return 0;
1477
1478 out_error:
1479         destroy_devices();
1480         return ret;
1481 }
1482
1483 static void __exit zram_exit(void)
1484 {
1485         destroy_devices();
1486 }
1487
1488 module_init(zram_init);
1489 module_exit(zram_exit);
1490
1491 module_param(num_devices, uint, 0);
1492 MODULE_PARM_DESC(num_devices, "Number of pre-created zram devices");
1493
1494 MODULE_LICENSE("Dual BSD/GPL");
1495 MODULE_AUTHOR("Nitin Gupta <ngupta@vflare.org>");
1496 MODULE_DESCRIPTION("Compressed RAM Block Device");