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 int 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 int 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 0;
 125         if (unlikely(size & (ZRAM_LOGICAL_BLOCK_SIZE - 1)))
 126                 return 0;
 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 0;
 133
 134         /* I/O request is valid */
 135         return 1;
 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 int 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 0;
 170         }
 171
 172         return 1;
 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 static ssize_t max_comp_streams_show(struct device *dev,
 308                 struct device_attribute *attr, char *buf)
 309 {
 310         int val;
 311         struct zram *zram = dev_to_zram(dev);
 312
 313         down_read(&zram->init_lock);
 314         val = zram->max_comp_streams;
 315         up_read(&zram->init_lock);
 316
 317         return scnprintf(buf, PAGE_SIZE, "%d\n", val);
 318 }
 319
 320 static ssize_t max_comp_streams_store(struct device *dev,
 321                 struct device_attribute *attr, const char *buf, size_t len)
 322 {
 323         int num;
 324         struct zram *zram = dev_to_zram(dev);
 325         int ret;
 326
 327         ret = kstrtoint(buf, 0, &num);
 328         if (ret < 0)
 329                 return ret;
 330         if (num < 1)
 331                 return -EINVAL;
 332
 333         down_write(&zram->init_lock);
 334         if (init_done(zram)) {
 335                 if (!zcomp_set_max_streams(zram->comp, num)) {
 336                         pr_info("Cannot change max compression streams\n");
 337                         ret = -EINVAL;
 338                         goto out;
 339                 }
 340         }
 341
 342         zram->max_comp_streams = num;
 343         ret = len;
 344 out:
 345         up_write(&zram->init_lock);
 346         return ret;
 347 }
 348
 349 static ssize_t comp_algorithm_show(struct device *dev,
 350                 struct device_attribute *attr, char *buf)
 351 {
 352         size_t sz;
 353         struct zram *zram = dev_to_zram(dev);
 354
 355         down_read(&zram->init_lock);
 356         sz = zcomp_available_show(zram->compressor, buf);
 357         up_read(&zram->init_lock);
 358
 359         return sz;
 360 }
 361
 362 static ssize_t comp_algorithm_store(struct device *dev,
 363                 struct device_attribute *attr, const char *buf, size_t len)
 364 {
 365         struct zram *zram = dev_to_zram(dev);
 366         size_t sz;
 367
 368         down_write(&zram->init_lock);
 369         if (init_done(zram)) {
 370                 up_write(&zram->init_lock);
 371                 pr_info("Can't change algorithm for initialized device\n");
 372                 return -EBUSY;
 373         }
 374         strlcpy(zram->compressor, buf, sizeof(zram->compressor));
 375
 376         /* ignore trailing newline */
 377         sz = strlen(zram->compressor);
 378         if (sz > 0 && zram->compressor[sz - 1] == '\n')
 379                 zram->compressor[sz - 1] = 0x00;
 380
 381         if (!zcomp_available_algorithm(zram->compressor))
 382                 len = -EINVAL;
 383
 384         up_write(&zram->init_lock);
 385         return len;
 386 }
 387
 388 static ssize_t compact_store(struct device *dev,
 389                 struct device_attribute *attr, const char *buf, size_t len)
 390 {
 391         unsigned long nr_migrated;
 392         struct zram *zram = dev_to_zram(dev);
 393         struct zram_meta *meta;
 394
 395         down_read(&zram->init_lock);
 396         if (!init_done(zram)) {
 397                 up_read(&zram->init_lock);
 398                 return -EINVAL;
 399         }
 400
 401         meta = zram->meta;
 402         nr_migrated = zs_compact(meta->mem_pool);
 403         atomic64_add(nr_migrated, &zram->stats.num_migrated);
 404         up_read(&zram->init_lock);
 405
 406         return len;
 407 }
 408
 409 static ssize_t io_stat_show(struct device *dev,
 410                 struct device_attribute *attr, char *buf)
 411 {
 412         struct zram *zram = dev_to_zram(dev);
 413         ssize_t ret;
 414
 415         down_read(&zram->init_lock);
 416         ret = scnprintf(buf, PAGE_SIZE,
 417                         "%8llu %8llu %8llu %8llu\n",
 418                         (u64)atomic64_read(&zram->stats.failed_reads),
 419                         (u64)atomic64_read(&zram->stats.failed_writes),
 420                         (u64)atomic64_read(&zram->stats.invalid_io),
 421                         (u64)atomic64_read(&zram->stats.notify_free));
 422         up_read(&zram->init_lock);
 423
 424         return ret;
 425 }
 426
 427 static ssize_t mm_stat_show(struct device *dev,
 428                 struct device_attribute *attr, char *buf)
 429 {
 430         struct zram *zram = dev_to_zram(dev);
 431         u64 orig_size, mem_used = 0;
 432         long max_used;
 433         ssize_t ret;
 434
 435         down_read(&zram->init_lock);
 436         if (init_done(zram))
 437                 mem_used = zs_get_total_pages(zram->meta->mem_pool);
 438
 439         orig_size = atomic64_read(&zram->stats.pages_stored);
 440         max_used = atomic_long_read(&zram->stats.max_used_pages);
 441
 442         ret = scnprintf(buf, PAGE_SIZE,
 443                         "%8llu %8llu %8llu %8lu %8ld %8llu %8llu\n",
 444                         orig_size << PAGE_SHIFT,
 445                         (u64)atomic64_read(&zram->stats.compr_data_size),
 446                         mem_used << PAGE_SHIFT,
 447                         zram->limit_pages << PAGE_SHIFT,
 448                         max_used << PAGE_SHIFT,
 449                         (u64)atomic64_read(&zram->stats.zero_pages),
 450                         (u64)atomic64_read(&zram->stats.num_migrated));
 451         up_read(&zram->init_lock);
 452
 453         return ret;
 454 }
 455
 456 static DEVICE_ATTR_RO(io_stat);
 457 static DEVICE_ATTR_RO(mm_stat);
 458 ZRAM_ATTR_RO(num_reads);
 459 ZRAM_ATTR_RO(num_writes);
 460 ZRAM_ATTR_RO(failed_reads);
 461 ZRAM_ATTR_RO(failed_writes);
 462 ZRAM_ATTR_RO(invalid_io);
 463 ZRAM_ATTR_RO(notify_free);
 464 ZRAM_ATTR_RO(zero_pages);
 465 ZRAM_ATTR_RO(compr_data_size);
 466
 467 static inline bool zram_meta_get(struct zram *zram)
 468 {
 469         if (atomic_inc_not_zero(&zram->refcount))
 470                 return true;
 471         return false;
 472 }
 473
 474 static inline void zram_meta_put(struct zram *zram)
 475 {
 476         atomic_dec(&zram->refcount);
 477 }
 478
 479 static void zram_meta_free(struct zram_meta *meta, u64 disksize)
 480 {
 481         size_t num_pages = disksize >> PAGE_SHIFT;
 482         size_t index;
 483
 484         /* Free all pages that are still in this zram device */
 485         for (index = 0; index < num_pages; index++) {
 486                 unsigned long handle = meta->table[index].handle;
 487
 488                 if (!handle)
 489                         continue;
 490
 491                 zs_free(meta->mem_pool, handle);
 492         }
 493
 494         zs_destroy_pool(meta->mem_pool);
 495         vfree(meta->table);
 496         kfree(meta);
 497 }
 498
 499 static struct zram_meta *zram_meta_alloc(int device_id, u64 disksize)
 500 {
 501         size_t num_pages;
 502         char pool_name[8];
 503         struct zram_meta *meta = kmalloc(sizeof(*meta), GFP_KERNEL);
 504
 505         if (!meta)
 506                 return NULL;
 507
 508         num_pages = disksize >> PAGE_SHIFT;
 509         meta->table = vzalloc(num_pages * sizeof(*meta->table));
 510         if (!meta->table) {
 511                 pr_err("Error allocating zram address table\n");
 512                 goto out_error;
 513         }
 514
 515         snprintf(pool_name, sizeof(pool_name), "zram%d", device_id);
 516         meta->mem_pool = zs_create_pool(pool_name, GFP_NOIO | __GFP_HIGHMEM);
 517         if (!meta->mem_pool) {
 518                 pr_err("Error creating memory pool\n");
 519                 goto out_error;
 520         }
 521
 522         return meta;
 523
 524 out_error:
 525         vfree(meta->table);
 526         kfree(meta);
 527         return NULL;
 528 }
 529
 530 /*
 531  * To protect concurrent access to the same index entry,
 532  * caller should hold this table index entry's bit_spinlock to
 533  * indicate this index entry is accessing.
 534  */
 535 static void zram_free_page(struct zram *zram, size_t index)
 536 {
 537         struct zram_meta *meta = zram->meta;
 538         unsigned long handle = meta->table[index].handle;
 539
 540         if (unlikely(!handle)) {
 541                 /*
 542                  * No memory is allocated for zero filled pages.
 543                  * Simply clear zero page flag.
 544                  */
 545                 if (zram_test_flag(meta, index, ZRAM_ZERO)) {
 546                         zram_clear_flag(meta, index, ZRAM_ZERO);
 547                         atomic64_dec(&zram->stats.zero_pages);
 548                 }
 549                 return;
 550         }
 551
 552         zs_free(meta->mem_pool, handle);
 553
 554         atomic64_sub(zram_get_obj_size(meta, index),
 555                         &zram->stats.compr_data_size);
 556         atomic64_dec(&zram->stats.pages_stored);
 557
 558         meta->table[index].handle = 0;
 559         zram_set_obj_size(meta, index, 0);
 560 }
 561
 562 static int zram_decompress_page(struct zram *zram, char *mem, u32 index)
 563 {
 564         int ret = 0;
 565         unsigned char *cmem;
 566         struct zram_meta *meta = zram->meta;
 567         unsigned long handle;
 568         size_t size;
 569
 570         bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
 571         handle = meta->table[index].handle;
 572         size = zram_get_obj_size(meta, index);
 573
 574         if (!handle || zram_test_flag(meta, index, ZRAM_ZERO)) {
 575                 bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 576                 clear_page(mem);
 577                 return 0;
 578         }
 579
 580         cmem = zs_map_object(meta->mem_pool, handle, ZS_MM_RO);
 581         if (size == PAGE_SIZE)
 582                 copy_page(mem, cmem);
 583         else
 584                 ret = zcomp_decompress(zram->comp, cmem, size, mem);
 585         zs_unmap_object(meta->mem_pool, handle);
 586         bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 587
 588         /* Should NEVER happen. Return bio error if it does. */
 589         if (unlikely(ret)) {
 590                 pr_err("Decompression failed! err=%d, page=%u\n", ret, index);
 591                 return ret;
 592         }
 593
 594         return 0;
 595 }
 596
 597 static int zram_bvec_read(struct zram *zram, struct bio_vec *bvec,
 598                           u32 index, int offset)
 599 {
 600         int ret;
 601         struct page *page;
 602         unsigned char *user_mem, *uncmem = NULL;
 603         struct zram_meta *meta = zram->meta;
 604         page = bvec->bv_page;
 605
 606         bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
 607         if (unlikely(!meta->table[index].handle) ||
 608                         zram_test_flag(meta, index, ZRAM_ZERO)) {
 609                 bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 610                 handle_zero_page(bvec);
 611                 return 0;
 612         }
 613         bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 614
 615         if (is_partial_io(bvec))
 616                 /* Use  a temporary buffer to decompress the page */
 617                 uncmem = kmalloc(PAGE_SIZE, GFP_NOIO);
 618
 619         user_mem = kmap_atomic(page);
 620         if (!is_partial_io(bvec))
 621                 uncmem = user_mem;
 622
 623         if (!uncmem) {
 624                 pr_info("Unable to allocate temp memory\n");
 625                 ret = -ENOMEM;
 626                 goto out_cleanup;
 627         }
 628
 629         ret = zram_decompress_page(zram, uncmem, index);
 630         /* Should NEVER happen. Return bio error if it does. */
 631         if (unlikely(ret))
 632                 goto out_cleanup;
 633
 634         if (is_partial_io(bvec))
 635                 memcpy(user_mem + bvec->bv_offset, uncmem + offset,
 636                                 bvec->bv_len);
 637
 638         flush_dcache_page(page);
 639         ret = 0;
 640 out_cleanup:
 641         kunmap_atomic(user_mem);
 642         if (is_partial_io(bvec))
 643                 kfree(uncmem);
 644         return ret;
 645 }
 646
 647 static int zram_bvec_write(struct zram *zram, struct bio_vec *bvec, u32 index,
 648                            int offset)
 649 {
 650         int ret = 0;
 651         size_t clen;
 652         unsigned long handle;
 653         struct page *page;
 654         unsigned char *user_mem, *cmem, *src, *uncmem = NULL;
 655         struct zram_meta *meta = zram->meta;
 656         struct zcomp_strm *zstrm = NULL;
 657         unsigned long alloced_pages;
 658
 659         page = bvec->bv_page;
 660         if (is_partial_io(bvec)) {
 661                 /*
 662                  * This is a partial IO. We need to read the full page
 663                  * before to write the changes.
 664                  */
 665                 uncmem = kmalloc(PAGE_SIZE, GFP_NOIO);
 666                 if (!uncmem) {
 667                         ret = -ENOMEM;
 668                         goto out;
 669                 }
 670                 ret = zram_decompress_page(zram, uncmem, index);
 671                 if (ret)
 672                         goto out;
 673         }
 674
 675         zstrm = zcomp_strm_find(zram->comp);
 676         user_mem = kmap_atomic(page);
 677
 678         if (is_partial_io(bvec)) {
 679                 memcpy(uncmem + offset, user_mem + bvec->bv_offset,
 680                        bvec->bv_len);
 681                 kunmap_atomic(user_mem);
 682                 user_mem = NULL;
 683         } else {
 684                 uncmem = user_mem;
 685         }
 686
 687         if (page_zero_filled(uncmem)) {
 688                 if (user_mem)
 689                         kunmap_atomic(user_mem);
 690                 /* Free memory associated with this sector now. */
 691                 bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
 692                 zram_free_page(zram, index);
 693                 zram_set_flag(meta, index, ZRAM_ZERO);
 694                 bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 695
 696                 atomic64_inc(&zram->stats.zero_pages);
 697                 ret = 0;
 698                 goto out;
 699         }
 700
 701         ret = zcomp_compress(zram->comp, zstrm, uncmem, &clen);
 702         if (!is_partial_io(bvec)) {
 703                 kunmap_atomic(user_mem);
 704                 user_mem = NULL;
 705                 uncmem = NULL;
 706         }
 707
 708         if (unlikely(ret)) {
 709                 pr_err("Compression failed! err=%d\n", ret);
 710                 goto out;
 711         }
 712         src = zstrm->buffer;
 713         if (unlikely(clen > max_zpage_size)) {
 714                 clen = PAGE_SIZE;
 715                 if (is_partial_io(bvec))
 716                         src = uncmem;
 717         }
 718
 719         handle = zs_malloc(meta->mem_pool, clen);
 720         if (!handle) {
 721                 pr_info("Error allocating memory for compressed page: %u, size=%zu\n",
 722                         index, clen);
 723                 ret = -ENOMEM;
 724                 goto out;
 725         }
 726
 727         alloced_pages = zs_get_total_pages(meta->mem_pool);
 728         if (zram->limit_pages && alloced_pages > zram->limit_pages) {
 729                 zs_free(meta->mem_pool, handle);
 730                 ret = -ENOMEM;
 731                 goto out;
 732         }
 733
 734         update_used_max(zram, alloced_pages);
 735
 736         cmem = zs_map_object(meta->mem_pool, handle, ZS_MM_WO);
 737
 738         if ((clen == PAGE_SIZE) && !is_partial_io(bvec)) {
 739                 src = kmap_atomic(page);
 740                 copy_page(cmem, src);
 741                 kunmap_atomic(src);
 742         } else {
 743                 memcpy(cmem, src, clen);
 744         }
 745
 746         zcomp_strm_release(zram->comp, zstrm);
 747         zstrm = NULL;
 748         zs_unmap_object(meta->mem_pool, handle);
 749
 750         /*
 751          * Free memory associated with this sector
 752          * before overwriting unused sectors.
 753          */
 754         bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
 755         zram_free_page(zram, index);
 756
 757         meta->table[index].handle = handle;
 758         zram_set_obj_size(meta, index, clen);
 759         bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 760
 761         /* Update stats */
 762         atomic64_add(clen, &zram->stats.compr_data_size);
 763         atomic64_inc(&zram->stats.pages_stored);
 764 out:
 765         if (zstrm)
 766                 zcomp_strm_release(zram->comp, zstrm);
 767         if (is_partial_io(bvec))
 768                 kfree(uncmem);
 769         return ret;
 770 }
 771
 772 /*
 773  * zram_bio_discard - handler on discard request
 774  * @index: physical block index in PAGE_SIZE units
 775  * @offset: byte offset within physical block
 776  */
 777 static void zram_bio_discard(struct zram *zram, u32 index,
 778                              int offset, struct bio *bio)
 779 {
 780         size_t n = bio->bi_iter.bi_size;
 781         struct zram_meta *meta = zram->meta;
 782
 783         /*
 784          * zram manages data in physical block size units. Because logical block
 785          * size isn't identical with physical block size on some arch, we
 786          * could get a discard request pointing to a specific offset within a
 787          * certain physical block.  Although we can handle this request by
 788          * reading that physiclal block and decompressing and partially zeroing
 789          * and re-compressing and then re-storing it, this isn't reasonable
 790          * because our intent with a discard request is to save memory.  So
 791          * skipping this logical block is appropriate here.
 792          */
 793         if (offset) {
 794                 if (n <= (PAGE_SIZE - offset))
 795                         return;
 796
 797                 n -= (PAGE_SIZE - offset);
 798                 index++;
 799         }
 800
 801         while (n >= PAGE_SIZE) {
 802                 bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
 803                 zram_free_page(zram, index);
 804                 bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 805                 atomic64_inc(&zram->stats.notify_free);
 806                 index++;
 807                 n -= PAGE_SIZE;
 808         }
 809 }
 810
 811 static int zram_bvec_rw(struct zram *zram, struct bio_vec *bvec, u32 index,
 812                         int offset, int rw)
 813 {
 814         unsigned long start_time = jiffies;
 815         int ret;
 816
 817         generic_start_io_acct(rw, bvec->bv_len >> SECTOR_SHIFT,
 818                         &zram->disk->part0);
 819
 820         if (rw == READ) {
 821                 atomic64_inc(&zram->stats.num_reads);
 822                 ret = zram_bvec_read(zram, bvec, index, offset);
 823         } else {
 824                 atomic64_inc(&zram->stats.num_writes);
 825                 ret = zram_bvec_write(zram, bvec, index, offset);
 826         }
 827
 828         generic_end_io_acct(rw, &zram->disk->part0, start_time);
 829
 830         if (unlikely(ret)) {
 831                 if (rw == READ)
 832                         atomic64_inc(&zram->stats.failed_reads);
 833                 else
 834                         atomic64_inc(&zram->stats.failed_writes);
 835         }
 836
 837         return ret;
 838 }
 839
 840 static void __zram_make_request(struct zram *zram, struct bio *bio)
 841 {
 842         int offset, rw;
 843         u32 index;
 844         struct bio_vec bvec;
 845         struct bvec_iter iter;
 846
 847         index = bio->bi_iter.bi_sector >> SECTORS_PER_PAGE_SHIFT;
 848         offset = (bio->bi_iter.bi_sector &
 849                   (SECTORS_PER_PAGE - 1)) << SECTOR_SHIFT;
 850
 851         if (unlikely(bio->bi_rw & REQ_DISCARD)) {
 852                 zram_bio_discard(zram, index, offset, bio);
 853                 bio_endio(bio);
 854                 return;
 855         }
 856
 857         rw = bio_data_dir(bio);
 858         bio_for_each_segment(bvec, bio, iter) {
 859                 int max_transfer_size = PAGE_SIZE - offset;
 860
 861                 if (bvec.bv_len > max_transfer_size) {
 862                         /*
 863                          * zram_bvec_rw() can only make operation on a single
 864                          * zram page. Split the bio vector.
 865                          */
 866                         struct bio_vec bv;
 867
 868                         bv.bv_page = bvec.bv_page;
 869                         bv.bv_len = max_transfer_size;
 870                         bv.bv_offset = bvec.bv_offset;
 871
 872                         if (zram_bvec_rw(zram, &bv, index, offset, rw) < 0)
 873                                 goto out;
 874
 875                         bv.bv_len = bvec.bv_len - max_transfer_size;
 876                         bv.bv_offset += max_transfer_size;
 877                         if (zram_bvec_rw(zram, &bv, index + 1, 0, rw) < 0)
 878                                 goto out;
 879                 } else
 880                         if (zram_bvec_rw(zram, &bvec, index, offset, rw) < 0)
 881                                 goto out;
 882
 883                 update_position(&index, &offset, &bvec);
 884         }
 885
 886         bio_endio(bio);
 887         return;
 888
 889 out:
 890         bio_io_error(bio);
 891 }
 892
 893 /*
 894  * Handler function for all zram I/O requests.
 895  */
 896 static void zram_make_request(struct request_queue *queue, struct bio *bio)
 897 {
 898         struct zram *zram = queue->queuedata;
 899
 900         if (unlikely(!zram_meta_get(zram)))
 901                 goto error;
 902
 903         if (!valid_io_request(zram, bio->bi_iter.bi_sector,
 904                                         bio->bi_iter.bi_size)) {
 905                 atomic64_inc(&zram->stats.invalid_io);
 906                 goto put_zram;
 907         }
 908
 909         __zram_make_request(zram, bio);
 910         zram_meta_put(zram);
 911         return;
 912 put_zram:
 913         zram_meta_put(zram);
 914 error:
 915         bio_io_error(bio);
 916 }
 917
 918 static void zram_slot_free_notify(struct block_device *bdev,
 919                                 unsigned long index)
 920 {
 921         struct zram *zram;
 922         struct zram_meta *meta;
 923
 924         zram = bdev->bd_disk->private_data;
 925         meta = zram->meta;
 926
 927         bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
 928         zram_free_page(zram, index);
 929         bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 930         atomic64_inc(&zram->stats.notify_free);
 931 }
 932
 933 static int zram_rw_page(struct block_device *bdev, sector_t sector,
 934                        struct page *page, int rw)
 935 {
 936         int offset, err = -EIO;
 937         u32 index;
 938         struct zram *zram;
 939         struct bio_vec bv;
 940
 941         zram = bdev->bd_disk->private_data;
 942         if (unlikely(!zram_meta_get(zram)))
 943                 goto out;
 944
 945         if (!valid_io_request(zram, sector, PAGE_SIZE)) {
 946                 atomic64_inc(&zram->stats.invalid_io);
 947                 err = -EINVAL;
 948                 goto put_zram;
 949         }
 950
 951         index = sector >> SECTORS_PER_PAGE_SHIFT;
 952         offset = sector & (SECTORS_PER_PAGE - 1) << SECTOR_SHIFT;
 953
 954         bv.bv_page = page;
 955         bv.bv_len = PAGE_SIZE;
 956         bv.bv_offset = 0;
 957
 958         err = zram_bvec_rw(zram, &bv, index, offset, rw);
 959 put_zram:
 960         zram_meta_put(zram);
 961 out:
 962         /*
 963          * If I/O fails, just return error(ie, non-zero) without
 964          * calling page_endio.
 965          * It causes resubmit the I/O with bio request by upper functions
 966          * of rw_page(e.g., swap_readpage, __swap_writepage) and
 967          * bio->bi_end_io does things to handle the error
 968          * (e.g., SetPageError, set_page_dirty and extra works).
 969          */
 970         if (err == 0)
 971                 page_endio(page, rw, 0);
 972         return err;
 973 }
 974
 975 static void zram_reset_device(struct zram *zram)
 976 {
 977         struct zram_meta *meta;
 978         struct zcomp *comp;
 979         u64 disksize;
 980
 981         down_write(&zram->init_lock);
 982
 983         zram->limit_pages = 0;
 984
 985         if (!init_done(zram)) {
 986                 up_write(&zram->init_lock);
 987                 return;
 988         }
 989
 990         meta = zram->meta;
 991         comp = zram->comp;
 992         disksize = zram->disksize;
 993         /*
 994          * Refcount will go down to 0 eventually and r/w handler
 995          * cannot handle further I/O so it will bail out by
 996          * check zram_meta_get.
 997          */
 998         zram_meta_put(zram);
 999         /*
1000          * We want to free zram_meta in process context to avoid
1001          * deadlock between reclaim path and any other locks.
1002          */
1003         wait_event(zram->io_done, atomic_read(&zram->refcount) == 0);
1004
1005         /* Reset stats */
1006         memset(&zram->stats, 0, sizeof(zram->stats));
1007         zram->disksize = 0;
1008         zram->max_comp_streams = 1;
1009
1010         set_capacity(zram->disk, 0);
1011         part_stat_set_all(&zram->disk->part0, 0);
1012
1013         up_write(&zram->init_lock);
1014         /* I/O operation under all of CPU are done so let's free */
1015         zram_meta_free(meta, disksize);
1016         zcomp_destroy(comp);
1017 }
1018
1019 static ssize_t disksize_store(struct device *dev,
1020                 struct device_attribute *attr, const char *buf, size_t len)
1021 {
1022         u64 disksize;
1023         struct zcomp *comp;
1024         struct zram_meta *meta;
1025         struct zram *zram = dev_to_zram(dev);
1026         int err;
1027
1028         disksize = memparse(buf, NULL);
1029         if (!disksize)
1030                 return -EINVAL;
1031
1032         disksize = PAGE_ALIGN(disksize);
1033         meta = zram_meta_alloc(zram->disk->first_minor, disksize);
1034         if (!meta)
1035                 return -ENOMEM;
1036
1037         comp = zcomp_create(zram->compressor, zram->max_comp_streams);
1038         if (IS_ERR(comp)) {
1039                 pr_info("Cannot initialise %s compressing backend\n",
1040                                 zram->compressor);
1041                 err = PTR_ERR(comp);
1042                 goto out_free_meta;
1043         }
1044
1045         down_write(&zram->init_lock);
1046         if (init_done(zram)) {
1047                 pr_info("Cannot change disksize for initialized device\n");
1048                 err = -EBUSY;
1049                 goto out_destroy_comp;
1050         }
1051
1052         init_waitqueue_head(&zram->io_done);
1053         atomic_set(&zram->refcount, 1);
1054         zram->meta = meta;
1055         zram->comp = comp;
1056         zram->disksize = disksize;
1057         set_capacity(zram->disk, zram->disksize >> SECTOR_SHIFT);
1058         up_write(&zram->init_lock);
1059
1060         /*
1061          * Revalidate disk out of the init_lock to avoid lockdep splat.
1062          * It's okay because disk's capacity is protected by init_lock
1063          * so that revalidate_disk always sees up-to-date capacity.
1064          */
1065         revalidate_disk(zram->disk);
1066
1067         return len;
1068
1069 out_destroy_comp:
1070         up_write(&zram->init_lock);
1071         zcomp_destroy(comp);
1072 out_free_meta:
1073         zram_meta_free(meta, disksize);
1074         return err;
1075 }
1076
1077 static ssize_t reset_store(struct device *dev,
1078                 struct device_attribute *attr, const char *buf, size_t len)
1079 {
1080         int ret;
1081         unsigned short do_reset;
1082         struct zram *zram;
1083         struct block_device *bdev;
1084
1085         ret = kstrtou16(buf, 10, &do_reset);
1086         if (ret)
1087                 return ret;
1088
1089         if (!do_reset)
1090                 return -EINVAL;
1091
1092         zram = dev_to_zram(dev);
1093         bdev = bdget_disk(zram->disk, 0);
1094         if (!bdev)
1095                 return -ENOMEM;
1096
1097         mutex_lock(&bdev->bd_mutex);
1098         /* Do not reset an active device or claimed device */
1099         if (bdev->bd_openers || zram->claim) {
1100                 mutex_unlock(&bdev->bd_mutex);
1101                 bdput(bdev);
1102                 return -EBUSY;
1103         }
1104
1105         /* From now on, anyone can't open /dev/zram[0-9] */
1106         zram->claim = true;
1107         mutex_unlock(&bdev->bd_mutex);
1108
1109         /* Make sure all the pending I/O are finished */
1110         fsync_bdev(bdev);
1111         zram_reset_device(zram);
1112         revalidate_disk(zram->disk);
1113         bdput(bdev);
1114
1115         mutex_lock(&bdev->bd_mutex);
1116         zram->claim = false;
1117         mutex_unlock(&bdev->bd_mutex);
1118
1119         return len;
1120 }
1121
1122 static int zram_open(struct block_device *bdev, fmode_t mode)
1123 {
1124         int ret = 0;
1125         struct zram *zram;
1126
1127         WARN_ON(!mutex_is_locked(&bdev->bd_mutex));
1128
1129         zram = bdev->bd_disk->private_data;
1130         /* zram was claimed to reset so open request fails */
1131         if (zram->claim)
1132                 ret = -EBUSY;
1133
1134         return ret;
1135 }
1136
1137 static const struct block_device_operations zram_devops = {
1138         .open = zram_open,
1139         .swap_slot_free_notify = zram_slot_free_notify,
1140         .rw_page = zram_rw_page,
1141         .owner = THIS_MODULE
1142 };
1143
1144 static DEVICE_ATTR_WO(compact);
1145 static DEVICE_ATTR_RW(disksize);
1146 static DEVICE_ATTR_RO(initstate);
1147 static DEVICE_ATTR_WO(reset);
1148 static DEVICE_ATTR_RO(orig_data_size);
1149 static DEVICE_ATTR_RO(mem_used_total);
1150 static DEVICE_ATTR_RW(mem_limit);
1151 static DEVICE_ATTR_RW(mem_used_max);
1152 static DEVICE_ATTR_RW(max_comp_streams);
1153 static DEVICE_ATTR_RW(comp_algorithm);
1154
1155 static struct attribute *zram_disk_attrs[] = {
1156         &dev_attr_disksize.attr,
1157         &dev_attr_initstate.attr,
1158         &dev_attr_reset.attr,
1159         &dev_attr_num_reads.attr,
1160         &dev_attr_num_writes.attr,
1161         &dev_attr_failed_reads.attr,
1162         &dev_attr_failed_writes.attr,
1163         &dev_attr_compact.attr,
1164         &dev_attr_invalid_io.attr,
1165         &dev_attr_notify_free.attr,
1166         &dev_attr_zero_pages.attr,
1167         &dev_attr_orig_data_size.attr,
1168         &dev_attr_compr_data_size.attr,
1169         &dev_attr_mem_used_total.attr,
1170         &dev_attr_mem_limit.attr,
1171         &dev_attr_mem_used_max.attr,
1172         &dev_attr_max_comp_streams.attr,
1173         &dev_attr_comp_algorithm.attr,
1174         &dev_attr_io_stat.attr,
1175         &dev_attr_mm_stat.attr,
1176         NULL,
1177 };
1178
1179 static struct attribute_group zram_disk_attr_group = {
1180         .attrs = zram_disk_attrs,
1181 };
1182
1183 /*
1184  * Allocate and initialize new zram device. the function returns
1185  * '>= 0' device_id upon success, and negative value otherwise.
1186  */
1187 static int zram_add(void)
1188 {
1189         struct zram *zram;
1190         struct request_queue *queue;
1191         int ret, device_id;
1192
1193         zram = kzalloc(sizeof(struct zram), GFP_KERNEL);
1194         if (!zram)
1195                 return -ENOMEM;
1196
1197         ret = idr_alloc(&zram_index_idr, zram, 0, 0, GFP_KERNEL);
1198         if (ret < 0)
1199                 goto out_free_dev;
1200         device_id = ret;
1201
1202         init_rwsem(&zram->init_lock);
1203
1204         queue = blk_alloc_queue(GFP_KERNEL);
1205         if (!queue) {
1206                 pr_err("Error allocating disk queue for device %d\n",
1207                         device_id);
1208                 ret = -ENOMEM;
1209                 goto out_free_idr;
1210         }
1211
1212         blk_queue_make_request(queue, zram_make_request);
1213
1214         /* gendisk structure */
1215         zram->disk = alloc_disk(1);
1216         if (!zram->disk) {
1217                 pr_warn("Error allocating disk structure for device %d\n",
1218                         device_id);
1219                 ret = -ENOMEM;
1220                 goto out_free_queue;
1221         }
1222
1223         zram->disk->major = zram_major;
1224         zram->disk->first_minor = device_id;
1225         zram->disk->fops = &zram_devops;
1226         zram->disk->queue = queue;
1227         zram->disk->queue->queuedata = zram;
1228         zram->disk->private_data = zram;
1229         snprintf(zram->disk->disk_name, 16, "zram%d", device_id);
1230
1231         /* Actual capacity set using syfs (/sys/block/zram<id>/disksize */
1232         set_capacity(zram->disk, 0);
1233         /* zram devices sort of resembles non-rotational disks */
1234         queue_flag_set_unlocked(QUEUE_FLAG_NONROT, zram->disk->queue);
1235         queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, zram->disk->queue);
1236         /*
1237          * To ensure that we always get PAGE_SIZE aligned
1238          * and n*PAGE_SIZED sized I/O requests.
1239          */
1240         blk_queue_physical_block_size(zram->disk->queue, PAGE_SIZE);
1241         blk_queue_logical_block_size(zram->disk->queue,
1242                                         ZRAM_LOGICAL_BLOCK_SIZE);
1243         blk_queue_io_min(zram->disk->queue, PAGE_SIZE);
1244         blk_queue_io_opt(zram->disk->queue, PAGE_SIZE);
1245         zram->disk->queue->limits.discard_granularity = PAGE_SIZE;
1246         blk_queue_max_discard_sectors(zram->disk->queue, UINT_MAX);
1247         /*
1248          * zram_bio_discard() will clear all logical blocks if logical block
1249          * size is identical with physical block size(PAGE_SIZE). But if it is
1250          * different, we will skip discarding some parts of logical blocks in
1251          * the part of the request range which isn't aligned to physical block
1252          * size.  So we can't ensure that all discarded logical blocks are
1253          * zeroed.
1254          */
1255         if (ZRAM_LOGICAL_BLOCK_SIZE == PAGE_SIZE)
1256                 zram->disk->queue->limits.discard_zeroes_data = 1;
1257         else
1258                 zram->disk->queue->limits.discard_zeroes_data = 0;
1259         queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, zram->disk->queue);
1260
1261         add_disk(zram->disk);
1262
1263         ret = sysfs_create_group(&disk_to_dev(zram->disk)->kobj,
1264                                 &zram_disk_attr_group);
1265         if (ret < 0) {
1266                 pr_warn("Error creating sysfs group");
1267                 goto out_free_disk;
1268         }
1269         strlcpy(zram->compressor, default_compressor, sizeof(zram->compressor));
1270         zram->meta = NULL;
1271         zram->max_comp_streams = 1;
1272
1273         pr_info("Added device: %s\n", zram->disk->disk_name);
1274         return device_id;
1275
1276 out_free_disk:
1277         del_gendisk(zram->disk);
1278         put_disk(zram->disk);
1279 out_free_queue:
1280         blk_cleanup_queue(queue);
1281 out_free_idr:
1282         idr_remove(&zram_index_idr, device_id);
1283 out_free_dev:
1284         kfree(zram);
1285         return ret;
1286 }
1287
1288 static int zram_remove(struct zram *zram)
1289 {
1290         struct block_device *bdev;
1291
1292         bdev = bdget_disk(zram->disk, 0);
1293         if (!bdev)
1294                 return -ENOMEM;
1295
1296         mutex_lock(&bdev->bd_mutex);
1297         if (bdev->bd_openers || zram->claim) {
1298                 mutex_unlock(&bdev->bd_mutex);
1299                 bdput(bdev);
1300                 return -EBUSY;
1301         }
1302
1303         zram->claim = true;
1304         mutex_unlock(&bdev->bd_mutex);
1305
1306         /*
1307          * Remove sysfs first, so no one will perform a disksize
1308          * store while we destroy the devices. This also helps during
1309          * hot_remove -- zram_reset_device() is the last holder of
1310          * ->init_lock, no later/concurrent disksize_store() or any
1311          * other sysfs handlers are possible.
1312          */
1313         sysfs_remove_group(&disk_to_dev(zram->disk)->kobj,
1314                         &zram_disk_attr_group);
1315
1316         /* Make sure all the pending I/O are finished */
1317         fsync_bdev(bdev);
1318         zram_reset_device(zram);
1319         bdput(bdev);
1320
1321         pr_info("Removed device: %s\n", zram->disk->disk_name);
1322
1323         idr_remove(&zram_index_idr, zram->disk->first_minor);
1324         blk_cleanup_queue(zram->disk->queue);
1325         del_gendisk(zram->disk);
1326         put_disk(zram->disk);
1327         kfree(zram);
1328         return 0;
1329 }
1330
1331 /* zram-control sysfs attributes */
1332 static ssize_t hot_add_show(struct class *class,
1333                         struct class_attribute *attr,
1334                         char *buf)
1335 {
1336         int ret;
1337
1338         mutex_lock(&zram_index_mutex);
1339         ret = zram_add();
1340         mutex_unlock(&zram_index_mutex);
1341
1342         if (ret < 0)
1343                 return ret;
1344         return scnprintf(buf, PAGE_SIZE, "%d\n", ret);
1345 }
1346
1347 static ssize_t hot_remove_store(struct class *class,
1348                         struct class_attribute *attr,
1349                         const char *buf,
1350                         size_t count)
1351 {
1352         struct zram *zram;
1353         int ret, dev_id;
1354
1355         /* dev_id is gendisk->first_minor, which is `int' */
1356         ret = kstrtoint(buf, 10, &dev_id);
1357         if (ret)
1358                 return ret;
1359         if (dev_id < 0)
1360                 return -EINVAL;
1361
1362         mutex_lock(&zram_index_mutex);
1363
1364         zram = idr_find(&zram_index_idr, dev_id);
1365         if (zram)
1366                 ret = zram_remove(zram);
1367         else
1368                 ret = -ENODEV;
1369
1370         mutex_unlock(&zram_index_mutex);
1371         return ret ? ret : count;
1372 }
1373
1374 static struct class_attribute zram_control_class_attrs[] = {
1375         __ATTR_RO(hot_add),
1376         __ATTR_WO(hot_remove),
1377         __ATTR_NULL,
1378 };
1379
1380 static struct class zram_control_class = {
1381         .name           = "zram-control",
1382         .owner          = THIS_MODULE,
1383         .class_attrs    = zram_control_class_attrs,
1384 };
1385
1386 static int zram_remove_cb(int id, void *ptr, void *data)
1387 {
1388         zram_remove(ptr);
1389         return 0;
1390 }
1391
1392 static void destroy_devices(void)
1393 {
1394         class_unregister(&zram_control_class);
1395         idr_for_each(&zram_index_idr, &zram_remove_cb, NULL);
1396         idr_destroy(&zram_index_idr);
1397         unregister_blkdev(zram_major, "zram");
1398 }
1399
1400 static int __init zram_init(void)
1401 {
1402         int ret;
1403
1404         ret = class_register(&zram_control_class);
1405         if (ret) {
1406                 pr_warn("Unable to register zram-control class\n");
1407                 return ret;
1408         }
1409
1410         zram_major = register_blkdev(0, "zram");
1411         if (zram_major <= 0) {
1412                 pr_warn("Unable to get major number\n");
1413                 class_unregister(&zram_control_class);
1414                 return -EBUSY;
1415         }
1416
1417         while (num_devices != 0) {
1418                 mutex_lock(&zram_index_mutex);
1419                 ret = zram_add();
1420                 mutex_unlock(&zram_index_mutex);
1421                 if (ret < 0)
1422                         goto out_error;
1423                 num_devices--;
1424         }
1425
1426         return 0;
1427
1428 out_error:
1429         destroy_devices();
1430         return ret;
1431 }
1432
1433 static void __exit zram_exit(void)
1434 {
1435         destroy_devices();
1436 }
1437
1438 module_init(zram_init);
1439 module_exit(zram_exit);
1440
1441 module_param(num_devices, uint, 0);
1442 MODULE_PARM_DESC(num_devices, "Number of pre-created zram devices");
1443
1444 MODULE_LICENSE("Dual BSD/GPL");
1445 MODULE_AUTHOR("Nitin Gupta <ngupta@vflare.org>");
1446 MODULE_DESCRIPTION("Compressed RAM Block Device");