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         size_t sz;
 346
 347         if (!zcomp_available_algorithm(buf))
 348                 return -EINVAL;
 349
 350         down_write(&zram->init_lock);
 351         if (init_done(zram)) {
 352                 up_write(&zram->init_lock);
 353                 pr_info("Can't change algorithm for initialized device\n");
 354                 return -EBUSY;
 355         }
 356         strlcpy(zram->compressor, buf, sizeof(zram->compressor));
 357
 358         /* ignore trailing newline */
 359         sz = strlen(zram->compressor);
 360         if (sz > 0 && zram->compressor[sz - 1] == '\n')
 361                 zram->compressor[sz - 1] = 0x00;
 362
 363         up_write(&zram->init_lock);
 364         return len;
 365 }
 366
 367 static ssize_t compact_store(struct device *dev,
 368                 struct device_attribute *attr, const char *buf, size_t len)
 369 {
 370         struct zram *zram = dev_to_zram(dev);
 371         struct zram_meta *meta;
 372
 373         down_read(&zram->init_lock);
 374         if (!init_done(zram)) {
 375                 up_read(&zram->init_lock);
 376                 return -EINVAL;
 377         }
 378
 379         meta = zram->meta;
 380         zs_compact(meta->mem_pool);
 381         up_read(&zram->init_lock);
 382
 383         return len;
 384 }
 385
 386 static ssize_t io_stat_show(struct device *dev,
 387                 struct device_attribute *attr, char *buf)
 388 {
 389         struct zram *zram = dev_to_zram(dev);
 390         ssize_t ret;
 391
 392         down_read(&zram->init_lock);
 393         ret = scnprintf(buf, PAGE_SIZE,
 394                         "%8llu %8llu %8llu %8llu\n",
 395                         (u64)atomic64_read(&zram->stats.failed_reads),
 396                         (u64)atomic64_read(&zram->stats.failed_writes),
 397                         (u64)atomic64_read(&zram->stats.invalid_io),
 398                         (u64)atomic64_read(&zram->stats.notify_free));
 399         up_read(&zram->init_lock);
 400
 401         return ret;
 402 }
 403
 404 static ssize_t mm_stat_show(struct device *dev,
 405                 struct device_attribute *attr, char *buf)
 406 {
 407         struct zram *zram = dev_to_zram(dev);
 408         struct zs_pool_stats pool_stats;
 409         u64 orig_size, mem_used = 0;
 410         long max_used;
 411         ssize_t ret;
 412
 413         memset(&pool_stats, 0x00, sizeof(struct zs_pool_stats));
 414
 415         down_read(&zram->init_lock);
 416         if (init_done(zram)) {
 417                 mem_used = zs_get_total_pages(zram->meta->mem_pool);
 418                 zs_pool_stats(zram->meta->mem_pool, &pool_stats);
 419         }
 420
 421         orig_size = atomic64_read(&zram->stats.pages_stored);
 422         max_used = atomic_long_read(&zram->stats.max_used_pages);
 423
 424         ret = scnprintf(buf, PAGE_SIZE,
 425                         "%8llu %8llu %8llu %8lu %8ld %8llu %8lu\n",
 426                         orig_size << PAGE_SHIFT,
 427                         (u64)atomic64_read(&zram->stats.compr_data_size),
 428                         mem_used << PAGE_SHIFT,
 429                         zram->limit_pages << PAGE_SHIFT,
 430                         max_used << PAGE_SHIFT,
 431                         (u64)atomic64_read(&zram->stats.zero_pages),
 432                         pool_stats.pages_compacted);
 433         up_read(&zram->init_lock);
 434
 435         return ret;
 436 }
 437
 438 static ssize_t debug_stat_show(struct device *dev,
 439                 struct device_attribute *attr, char *buf)
 440 {
 441         int version = 1;
 442         struct zram *zram = dev_to_zram(dev);
 443         ssize_t ret;
 444
 445         down_read(&zram->init_lock);
 446         ret = scnprintf(buf, PAGE_SIZE,
 447                         "version: %d\n%8llu\n",
 448                         version,
 449                         (u64)atomic64_read(&zram->stats.writestall));
 450         up_read(&zram->init_lock);
 451
 452         return ret;
 453 }
 454
 455 static DEVICE_ATTR_RO(io_stat);
 456 static DEVICE_ATTR_RO(mm_stat);
 457 static DEVICE_ATTR_RO(debug_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(char *pool_name, u64 disksize)
 500 {
 501         size_t num_pages;
 502         struct zram_meta *meta = kmalloc(sizeof(*meta), GFP_KERNEL);
 503
 504         if (!meta)
 505                 return NULL;
 506
 507         num_pages = disksize >> PAGE_SHIFT;
 508         meta->table = vzalloc(num_pages * sizeof(*meta->table));
 509         if (!meta->table) {
 510                 pr_err("Error allocating zram address table\n");
 511                 goto out_error;
 512         }
 513
 514         meta->mem_pool = zs_create_pool(pool_name);
 515         if (!meta->mem_pool) {
 516                 pr_err("Error creating memory pool\n");
 517                 goto out_error;
 518         }
 519
 520         return meta;
 521
 522 out_error:
 523         vfree(meta->table);
 524         kfree(meta);
 525         return NULL;
 526 }
 527
 528 /*
 529  * To protect concurrent access to the same index entry,
 530  * caller should hold this table index entry's bit_spinlock to
 531  * indicate this index entry is accessing.
 532  */
 533 static void zram_free_page(struct zram *zram, size_t index)
 534 {
 535         struct zram_meta *meta = zram->meta;
 536         unsigned long handle = meta->table[index].handle;
 537
 538         if (unlikely(!handle)) {
 539                 /*
 540                  * No memory is allocated for zero filled pages.
 541                  * Simply clear zero page flag.
 542                  */
 543                 if (zram_test_flag(meta, index, ZRAM_ZERO)) {
 544                         zram_clear_flag(meta, index, ZRAM_ZERO);
 545                         atomic64_dec(&zram->stats.zero_pages);
 546                 }
 547                 return;
 548         }
 549
 550         zs_free(meta->mem_pool, handle);
 551
 552         atomic64_sub(zram_get_obj_size(meta, index),
 553                         &zram->stats.compr_data_size);
 554         atomic64_dec(&zram->stats.pages_stored);
 555
 556         meta->table[index].handle = 0;
 557         zram_set_obj_size(meta, index, 0);
 558 }
 559
 560 static int zram_decompress_page(struct zram *zram, char *mem, u32 index)
 561 {
 562         int ret = 0;
 563         unsigned char *cmem;
 564         struct zram_meta *meta = zram->meta;
 565         unsigned long handle;
 566         size_t size;
 567
 568         bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
 569         handle = meta->table[index].handle;
 570         size = zram_get_obj_size(meta, index);
 571
 572         if (!handle || zram_test_flag(meta, index, ZRAM_ZERO)) {
 573                 bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 574                 clear_page(mem);
 575                 return 0;
 576         }
 577
 578         cmem = zs_map_object(meta->mem_pool, handle, ZS_MM_RO);
 579         if (size == PAGE_SIZE)
 580                 copy_page(mem, cmem);
 581         else
 582                 ret = zcomp_decompress(zram->comp, cmem, size, mem);
 583         zs_unmap_object(meta->mem_pool, handle);
 584         bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 585
 586         /* Should NEVER happen. Return bio error if it does. */
 587         if (unlikely(ret)) {
 588                 pr_err("Decompression failed! err=%d, page=%u\n", ret, index);
 589                 return ret;
 590         }
 591
 592         return 0;
 593 }
 594
 595 static int zram_bvec_read(struct zram *zram, struct bio_vec *bvec,
 596                           u32 index, int offset)
 597 {
 598         int ret;
 599         struct page *page;
 600         unsigned char *user_mem, *uncmem = NULL;
 601         struct zram_meta *meta = zram->meta;
 602         page = bvec->bv_page;
 603
 604         bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
 605         if (unlikely(!meta->table[index].handle) ||
 606                         zram_test_flag(meta, index, ZRAM_ZERO)) {
 607                 bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 608                 handle_zero_page(bvec);
 609                 return 0;
 610         }
 611         bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 612
 613         if (is_partial_io(bvec))
 614                 /* Use  a temporary buffer to decompress the page */
 615                 uncmem = kmalloc(PAGE_SIZE, GFP_NOIO);
 616
 617         user_mem = kmap_atomic(page);
 618         if (!is_partial_io(bvec))
 619                 uncmem = user_mem;
 620
 621         if (!uncmem) {
 622                 pr_err("Unable to allocate temp memory\n");
 623                 ret = -ENOMEM;
 624                 goto out_cleanup;
 625         }
 626
 627         ret = zram_decompress_page(zram, uncmem, index);
 628         /* Should NEVER happen. Return bio error if it does. */
 629         if (unlikely(ret))
 630                 goto out_cleanup;
 631
 632         if (is_partial_io(bvec))
 633                 memcpy(user_mem + bvec->bv_offset, uncmem + offset,
 634                                 bvec->bv_len);
 635
 636         flush_dcache_page(page);
 637         ret = 0;
 638 out_cleanup:
 639         kunmap_atomic(user_mem);
 640         if (is_partial_io(bvec))
 641                 kfree(uncmem);
 642         return ret;
 643 }
 644
 645 static int zram_bvec_write(struct zram *zram, struct bio_vec *bvec, u32 index,
 646                            int offset)
 647 {
 648         int ret = 0;
 649         size_t clen;
 650         unsigned long handle = 0;
 651         struct page *page;
 652         unsigned char *user_mem, *cmem, *src, *uncmem = NULL;
 653         struct zram_meta *meta = zram->meta;
 654         struct zcomp_strm *zstrm = NULL;
 655         unsigned long alloced_pages;
 656
 657         page = bvec->bv_page;
 658         if (is_partial_io(bvec)) {
 659                 /*
 660                  * This is a partial IO. We need to read the full page
 661                  * before to write the changes.
 662                  */
 663                 uncmem = kmalloc(PAGE_SIZE, GFP_NOIO);
 664                 if (!uncmem) {
 665                         ret = -ENOMEM;
 666                         goto out;
 667                 }
 668                 ret = zram_decompress_page(zram, uncmem, index);
 669                 if (ret)
 670                         goto out;
 671         }
 672
 673 compress_again:
 674         user_mem = kmap_atomic(page);
 675         if (is_partial_io(bvec)) {
 676                 memcpy(uncmem + offset, user_mem + bvec->bv_offset,
 677                        bvec->bv_len);
 678                 kunmap_atomic(user_mem);
 679                 user_mem = NULL;
 680         } else {
 681                 uncmem = user_mem;
 682         }
 683
 684         if (page_zero_filled(uncmem)) {
 685                 if (user_mem)
 686                         kunmap_atomic(user_mem);
 687                 /* Free memory associated with this sector now. */
 688                 bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
 689                 zram_free_page(zram, index);
 690                 zram_set_flag(meta, index, ZRAM_ZERO);
 691                 bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 692
 693                 atomic64_inc(&zram->stats.zero_pages);
 694                 ret = 0;
 695                 goto out;
 696         }
 697
 698         zstrm = zcomp_strm_find(zram->comp);
 699         ret = zcomp_compress(zram->comp, zstrm, uncmem, &clen);
 700         if (!is_partial_io(bvec)) {
 701                 kunmap_atomic(user_mem);
 702                 user_mem = NULL;
 703                 uncmem = NULL;
 704         }
 705
 706         if (unlikely(ret)) {
 707                 pr_err("Compression failed! err=%d\n", ret);
 708                 goto out;
 709         }
 710
 711         src = zstrm->buffer;
 712         if (unlikely(clen > max_zpage_size)) {
 713                 clen = PAGE_SIZE;
 714                 if (is_partial_io(bvec))
 715                         src = uncmem;
 716         }
 717
 718         /*
 719          * handle allocation has 2 paths:
 720          * a) fast path is executed with preemption disabled (for
 721          *  per-cpu streams) and has __GFP_DIRECT_RECLAIM bit clear,
 722          *  since we can't sleep;
 723          * b) slow path enables preemption and attempts to allocate
 724          *  the page with __GFP_DIRECT_RECLAIM bit set. we have to
 725          *  put per-cpu compression stream and, thus, to re-do
 726          *  the compression once handle is allocated.
 727          *
 728          * if we have a 'non-null' handle here then we are coming
 729          * from the slow path and handle has already been allocated.
 730          */
 731         if (!handle)
 732                 handle = zs_malloc(meta->mem_pool, clen,
 733                                 __GFP_KSWAPD_RECLAIM |
 734                                 __GFP_NOWARN |
 735                                 __GFP_HIGHMEM);
 736         if (!handle) {
 737                 zcomp_strm_release(zram->comp, zstrm);
 738                 zstrm = NULL;
 739
 740                 atomic64_inc(&zram->stats.writestall);
 741
 742                 handle = zs_malloc(meta->mem_pool, clen,
 743                                 GFP_NOIO | __GFP_HIGHMEM);
 744                 if (handle)
 745                         goto compress_again;
 746
 747                 pr_err("Error allocating memory for compressed page: %u, size=%zu\n",
 748                         index, clen);
 749                 ret = -ENOMEM;
 750                 goto out;
 751         }
 752
 753         alloced_pages = zs_get_total_pages(meta->mem_pool);
 754         update_used_max(zram, alloced_pages);
 755
 756         if (zram->limit_pages && alloced_pages > zram->limit_pages) {
 757                 zs_free(meta->mem_pool, handle);
 758                 ret = -ENOMEM;
 759                 goto out;
 760         }
 761
 762         cmem = zs_map_object(meta->mem_pool, handle, ZS_MM_WO);
 763
 764         if ((clen == PAGE_SIZE) && !is_partial_io(bvec)) {
 765                 src = kmap_atomic(page);
 766                 copy_page(cmem, src);
 767                 kunmap_atomic(src);
 768         } else {
 769                 memcpy(cmem, src, clen);
 770         }
 771
 772         zcomp_strm_release(zram->comp, zstrm);
 773         zstrm = NULL;
 774         zs_unmap_object(meta->mem_pool, handle);
 775
 776         /*
 777          * Free memory associated with this sector
 778          * before overwriting unused sectors.
 779          */
 780         bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
 781         zram_free_page(zram, index);
 782
 783         meta->table[index].handle = handle;
 784         zram_set_obj_size(meta, index, clen);
 785         bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 786
 787         /* Update stats */
 788         atomic64_add(clen, &zram->stats.compr_data_size);
 789         atomic64_inc(&zram->stats.pages_stored);
 790 out:
 791         if (zstrm)
 792                 zcomp_strm_release(zram->comp, zstrm);
 793         if (is_partial_io(bvec))
 794                 kfree(uncmem);
 795         return ret;
 796 }
 797
 798 /*
 799  * zram_bio_discard - handler on discard request
 800  * @index: physical block index in PAGE_SIZE units
 801  * @offset: byte offset within physical block
 802  */
 803 static void zram_bio_discard(struct zram *zram, u32 index,
 804                              int offset, struct bio *bio)
 805 {
 806         size_t n = bio->bi_iter.bi_size;
 807         struct zram_meta *meta = zram->meta;
 808
 809         /*
 810          * zram manages data in physical block size units. Because logical block
 811          * size isn't identical with physical block size on some arch, we
 812          * could get a discard request pointing to a specific offset within a
 813          * certain physical block.  Although we can handle this request by
 814          * reading that physiclal block and decompressing and partially zeroing
 815          * and re-compressing and then re-storing it, this isn't reasonable
 816          * because our intent with a discard request is to save memory.  So
 817          * skipping this logical block is appropriate here.
 818          */
 819         if (offset) {
 820                 if (n <= (PAGE_SIZE - offset))
 821                         return;
 822
 823                 n -= (PAGE_SIZE - offset);
 824                 index++;
 825         }
 826
 827         while (n >= PAGE_SIZE) {
 828                 bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
 829                 zram_free_page(zram, index);
 830                 bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 831                 atomic64_inc(&zram->stats.notify_free);
 832                 index++;
 833                 n -= PAGE_SIZE;
 834         }
 835 }
 836
 837 static int zram_bvec_rw(struct zram *zram, struct bio_vec *bvec, u32 index,
 838                         int offset, int rw)
 839 {
 840         unsigned long start_time = jiffies;
 841         int ret;
 842
 843         generic_start_io_acct(rw, bvec->bv_len >> SECTOR_SHIFT,
 844                         &zram->disk->part0);
 845
 846         if (rw == READ) {
 847                 atomic64_inc(&zram->stats.num_reads);
 848                 ret = zram_bvec_read(zram, bvec, index, offset);
 849         } else {
 850                 atomic64_inc(&zram->stats.num_writes);
 851                 ret = zram_bvec_write(zram, bvec, index, offset);
 852         }
 853
 854         generic_end_io_acct(rw, &zram->disk->part0, start_time);
 855
 856         if (unlikely(ret)) {
 857                 if (rw == READ)
 858                         atomic64_inc(&zram->stats.failed_reads);
 859                 else
 860                         atomic64_inc(&zram->stats.failed_writes);
 861         }
 862
 863         return ret;
 864 }
 865
 866 static void __zram_make_request(struct zram *zram, struct bio *bio)
 867 {
 868         int offset, rw;
 869         u32 index;
 870         struct bio_vec bvec;
 871         struct bvec_iter iter;
 872
 873         index = bio->bi_iter.bi_sector >> SECTORS_PER_PAGE_SHIFT;
 874         offset = (bio->bi_iter.bi_sector &
 875                   (SECTORS_PER_PAGE - 1)) << SECTOR_SHIFT;
 876
 877         if (unlikely(bio_op(bio) == REQ_OP_DISCARD)) {
 878                 zram_bio_discard(zram, index, offset, bio);
 879                 bio_endio(bio);
 880                 return;
 881         }
 882
 883         rw = bio_data_dir(bio);
 884         bio_for_each_segment(bvec, bio, iter) {
 885                 int max_transfer_size = PAGE_SIZE - offset;
 886
 887                 if (bvec.bv_len > max_transfer_size) {
 888                         /*
 889                          * zram_bvec_rw() can only make operation on a single
 890                          * zram page. Split the bio vector.
 891                          */
 892                         struct bio_vec bv;
 893
 894                         bv.bv_page = bvec.bv_page;
 895                         bv.bv_len = max_transfer_size;
 896                         bv.bv_offset = bvec.bv_offset;
 897
 898                         if (zram_bvec_rw(zram, &bv, index, offset, rw) < 0)
 899                                 goto out;
 900
 901                         bv.bv_len = bvec.bv_len - max_transfer_size;
 902                         bv.bv_offset += max_transfer_size;
 903                         if (zram_bvec_rw(zram, &bv, index + 1, 0, rw) < 0)
 904                                 goto out;
 905                 } else
 906                         if (zram_bvec_rw(zram, &bvec, index, offset, rw) < 0)
 907                                 goto out;
 908
 909                 update_position(&index, &offset, &bvec);
 910         }
 911
 912         bio_endio(bio);
 913         return;
 914
 915 out:
 916         bio_io_error(bio);
 917 }
 918
 919 /*
 920  * Handler function for all zram I/O requests.
 921  */
 922 static blk_qc_t zram_make_request(struct request_queue *queue, struct bio *bio)
 923 {
 924         struct zram *zram = queue->queuedata;
 925
 926         if (unlikely(!zram_meta_get(zram)))
 927                 goto error;
 928
 929         blk_queue_split(queue, &bio, queue->bio_split);
 930
 931         if (!valid_io_request(zram, bio->bi_iter.bi_sector,
 932                                         bio->bi_iter.bi_size)) {
 933                 atomic64_inc(&zram->stats.invalid_io);
 934                 goto put_zram;
 935         }
 936
 937         __zram_make_request(zram, bio);
 938         zram_meta_put(zram);
 939         return BLK_QC_T_NONE;
 940 put_zram:
 941         zram_meta_put(zram);
 942 error:
 943         bio_io_error(bio);
 944         return BLK_QC_T_NONE;
 945 }
 946
 947 static void zram_slot_free_notify(struct block_device *bdev,
 948                                 unsigned long index)
 949 {
 950         struct zram *zram;
 951         struct zram_meta *meta;
 952
 953         zram = bdev->bd_disk->private_data;
 954         meta = zram->meta;
 955
 956         bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
 957         zram_free_page(zram, index);
 958         bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 959         atomic64_inc(&zram->stats.notify_free);
 960 }
 961
 962 static int zram_rw_page(struct block_device *bdev, sector_t sector,
 963                        struct page *page, int rw)
 964 {
 965         int offset, err = -EIO;
 966         u32 index;
 967         struct zram *zram;
 968         struct bio_vec bv;
 969
 970         zram = bdev->bd_disk->private_data;
 971         if (unlikely(!zram_meta_get(zram)))
 972                 goto out;
 973
 974         if (!valid_io_request(zram, sector, PAGE_SIZE)) {
 975                 atomic64_inc(&zram->stats.invalid_io);
 976                 err = -EINVAL;
 977                 goto put_zram;
 978         }
 979
 980         index = sector >> SECTORS_PER_PAGE_SHIFT;
 981         offset = sector & (SECTORS_PER_PAGE - 1) << SECTOR_SHIFT;
 982
 983         bv.bv_page = page;
 984         bv.bv_len = PAGE_SIZE;
 985         bv.bv_offset = 0;
 986
 987         err = zram_bvec_rw(zram, &bv, index, offset, rw);
 988 put_zram:
 989         zram_meta_put(zram);
 990 out:
 991         /*
 992          * If I/O fails, just return error(ie, non-zero) without
 993          * calling page_endio.
 994          * It causes resubmit the I/O with bio request by upper functions
 995          * of rw_page(e.g., swap_readpage, __swap_writepage) and
 996          * bio->bi_end_io does things to handle the error
 997          * (e.g., SetPageError, set_page_dirty and extra works).
 998          */
 999         if (err == 0)
1000                 page_endio(page, rw, 0);
1001         return err;
1002 }
1003
1004 static void zram_reset_device(struct zram *zram)
1005 {
1006         struct zram_meta *meta;
1007         struct zcomp *comp;
1008         u64 disksize;
1009
1010         down_write(&zram->init_lock);
1011
1012         zram->limit_pages = 0;
1013
1014         if (!init_done(zram)) {
1015                 up_write(&zram->init_lock);
1016                 return;
1017         }
1018
1019         meta = zram->meta;
1020         comp = zram->comp;
1021         disksize = zram->disksize;
1022         /*
1023          * Refcount will go down to 0 eventually and r/w handler
1024          * cannot handle further I/O so it will bail out by
1025          * check zram_meta_get.
1026          */
1027         zram_meta_put(zram);
1028         /*
1029          * We want to free zram_meta in process context to avoid
1030          * deadlock between reclaim path and any other locks.
1031          */
1032         wait_event(zram->io_done, atomic_read(&zram->refcount) == 0);
1033
1034         /* Reset stats */
1035         memset(&zram->stats, 0, sizeof(zram->stats));
1036         zram->disksize = 0;
1037
1038         set_capacity(zram->disk, 0);
1039         part_stat_set_all(&zram->disk->part0, 0);
1040
1041         up_write(&zram->init_lock);
1042         /* I/O operation under all of CPU are done so let's free */
1043         zram_meta_free(meta, disksize);
1044         zcomp_destroy(comp);
1045 }
1046
1047 static ssize_t disksize_store(struct device *dev,
1048                 struct device_attribute *attr, const char *buf, size_t len)
1049 {
1050         u64 disksize;
1051         struct zcomp *comp;
1052         struct zram_meta *meta;
1053         struct zram *zram = dev_to_zram(dev);
1054         int err;
1055
1056         disksize = memparse(buf, NULL);
1057         if (!disksize)
1058                 return -EINVAL;
1059
1060         disksize = PAGE_ALIGN(disksize);
1061         meta = zram_meta_alloc(zram->disk->disk_name, disksize);
1062         if (!meta)
1063                 return -ENOMEM;
1064
1065         comp = zcomp_create(zram->compressor);
1066         if (IS_ERR(comp)) {
1067                 pr_err("Cannot initialise %s compressing backend\n",
1068                                 zram->compressor);
1069                 err = PTR_ERR(comp);
1070                 goto out_free_meta;
1071         }
1072
1073         down_write(&zram->init_lock);
1074         if (init_done(zram)) {
1075                 pr_info("Cannot change disksize for initialized device\n");
1076                 err = -EBUSY;
1077                 goto out_destroy_comp;
1078         }
1079
1080         init_waitqueue_head(&zram->io_done);
1081         atomic_set(&zram->refcount, 1);
1082         zram->meta = meta;
1083         zram->comp = comp;
1084         zram->disksize = disksize;
1085         set_capacity(zram->disk, zram->disksize >> SECTOR_SHIFT);
1086         up_write(&zram->init_lock);
1087
1088         /*
1089          * Revalidate disk out of the init_lock to avoid lockdep splat.
1090          * It's okay because disk's capacity is protected by init_lock
1091          * so that revalidate_disk always sees up-to-date capacity.
1092          */
1093         revalidate_disk(zram->disk);
1094
1095         return len;
1096
1097 out_destroy_comp:
1098         up_write(&zram->init_lock);
1099         zcomp_destroy(comp);
1100 out_free_meta:
1101         zram_meta_free(meta, disksize);
1102         return err;
1103 }
1104
1105 static ssize_t reset_store(struct device *dev,
1106                 struct device_attribute *attr, const char *buf, size_t len)
1107 {
1108         int ret;
1109         unsigned short do_reset;
1110         struct zram *zram;
1111         struct block_device *bdev;
1112
1113         ret = kstrtou16(buf, 10, &do_reset);
1114         if (ret)
1115                 return ret;
1116
1117         if (!do_reset)
1118                 return -EINVAL;
1119
1120         zram = dev_to_zram(dev);
1121         bdev = bdget_disk(zram->disk, 0);
1122         if (!bdev)
1123                 return -ENOMEM;
1124
1125         mutex_lock(&bdev->bd_mutex);
1126         /* Do not reset an active device or claimed device */
1127         if (bdev->bd_openers || zram->claim) {
1128                 mutex_unlock(&bdev->bd_mutex);
1129                 bdput(bdev);
1130                 return -EBUSY;
1131         }
1132
1133         /* From now on, anyone can't open /dev/zram[0-9] */
1134         zram->claim = true;
1135         mutex_unlock(&bdev->bd_mutex);
1136
1137         /* Make sure all the pending I/O are finished */
1138         fsync_bdev(bdev);
1139         zram_reset_device(zram);
1140         revalidate_disk(zram->disk);
1141         bdput(bdev);
1142
1143         mutex_lock(&bdev->bd_mutex);
1144         zram->claim = false;
1145         mutex_unlock(&bdev->bd_mutex);
1146
1147         return len;
1148 }
1149
1150 static int zram_open(struct block_device *bdev, fmode_t mode)
1151 {
1152         int ret = 0;
1153         struct zram *zram;
1154
1155         WARN_ON(!mutex_is_locked(&bdev->bd_mutex));
1156
1157         zram = bdev->bd_disk->private_data;
1158         /* zram was claimed to reset so open request fails */
1159         if (zram->claim)
1160                 ret = -EBUSY;
1161
1162         return ret;
1163 }
1164
1165 static const struct block_device_operations zram_devops = {
1166         .open = zram_open,
1167         .swap_slot_free_notify = zram_slot_free_notify,
1168         .rw_page = zram_rw_page,
1169         .owner = THIS_MODULE
1170 };
1171
1172 static DEVICE_ATTR_WO(compact);
1173 static DEVICE_ATTR_RW(disksize);
1174 static DEVICE_ATTR_RO(initstate);
1175 static DEVICE_ATTR_WO(reset);
1176 static DEVICE_ATTR_RO(orig_data_size);
1177 static DEVICE_ATTR_RO(mem_used_total);
1178 static DEVICE_ATTR_RW(mem_limit);
1179 static DEVICE_ATTR_RW(mem_used_max);
1180 static DEVICE_ATTR_RW(max_comp_streams);
1181 static DEVICE_ATTR_RW(comp_algorithm);
1182
1183 static struct attribute *zram_disk_attrs[] = {
1184         &dev_attr_disksize.attr,
1185         &dev_attr_initstate.attr,
1186         &dev_attr_reset.attr,
1187         &dev_attr_num_reads.attr,
1188         &dev_attr_num_writes.attr,
1189         &dev_attr_failed_reads.attr,
1190         &dev_attr_failed_writes.attr,
1191         &dev_attr_compact.attr,
1192         &dev_attr_invalid_io.attr,
1193         &dev_attr_notify_free.attr,
1194         &dev_attr_zero_pages.attr,
1195         &dev_attr_orig_data_size.attr,
1196         &dev_attr_compr_data_size.attr,
1197         &dev_attr_mem_used_total.attr,
1198         &dev_attr_mem_limit.attr,
1199         &dev_attr_mem_used_max.attr,
1200         &dev_attr_max_comp_streams.attr,
1201         &dev_attr_comp_algorithm.attr,
1202         &dev_attr_io_stat.attr,
1203         &dev_attr_mm_stat.attr,
1204         &dev_attr_debug_stat.attr,
1205         NULL,
1206 };
1207
1208 static struct attribute_group zram_disk_attr_group = {
1209         .attrs = zram_disk_attrs,
1210 };
1211
1212 /*
1213  * Allocate and initialize new zram device. the function returns
1214  * '>= 0' device_id upon success, and negative value otherwise.
1215  */
1216 static int zram_add(void)
1217 {
1218         struct zram *zram;
1219         struct request_queue *queue;
1220         int ret, device_id;
1221
1222         zram = kzalloc(sizeof(struct zram), GFP_KERNEL);
1223         if (!zram)
1224                 return -ENOMEM;
1225
1226         ret = idr_alloc(&zram_index_idr, zram, 0, 0, GFP_KERNEL);
1227         if (ret < 0)
1228                 goto out_free_dev;
1229         device_id = ret;
1230
1231         init_rwsem(&zram->init_lock);
1232
1233         queue = blk_alloc_queue(GFP_KERNEL);
1234         if (!queue) {
1235                 pr_err("Error allocating disk queue for device %d\n",
1236                         device_id);
1237                 ret = -ENOMEM;
1238                 goto out_free_idr;
1239         }
1240
1241         blk_queue_make_request(queue, zram_make_request);
1242
1243         /* gendisk structure */
1244         zram->disk = alloc_disk(1);
1245         if (!zram->disk) {
1246                 pr_err("Error allocating disk structure for device %d\n",
1247                         device_id);
1248                 ret = -ENOMEM;
1249                 goto out_free_queue;
1250         }
1251
1252         zram->disk->major = zram_major;
1253         zram->disk->first_minor = device_id;
1254         zram->disk->fops = &zram_devops;
1255         zram->disk->queue = queue;
1256         zram->disk->queue->queuedata = zram;
1257         zram->disk->private_data = zram;
1258         snprintf(zram->disk->disk_name, 16, "zram%d", device_id);
1259
1260         /* Actual capacity set using syfs (/sys/block/zram<id>/disksize */
1261         set_capacity(zram->disk, 0);
1262         /* zram devices sort of resembles non-rotational disks */
1263         queue_flag_set_unlocked(QUEUE_FLAG_NONROT, zram->disk->queue);
1264         queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, zram->disk->queue);
1265         /*
1266          * To ensure that we always get PAGE_SIZE aligned
1267          * and n*PAGE_SIZED sized I/O requests.
1268          */
1269         blk_queue_physical_block_size(zram->disk->queue, PAGE_SIZE);
1270         blk_queue_logical_block_size(zram->disk->queue,
1271                                         ZRAM_LOGICAL_BLOCK_SIZE);
1272         blk_queue_io_min(zram->disk->queue, PAGE_SIZE);
1273         blk_queue_io_opt(zram->disk->queue, PAGE_SIZE);
1274         zram->disk->queue->limits.discard_granularity = PAGE_SIZE;
1275         blk_queue_max_discard_sectors(zram->disk->queue, UINT_MAX);
1276         /*
1277          * zram_bio_discard() will clear all logical blocks if logical block
1278          * size is identical with physical block size(PAGE_SIZE). But if it is
1279          * different, we will skip discarding some parts of logical blocks in
1280          * the part of the request range which isn't aligned to physical block
1281          * size.  So we can't ensure that all discarded logical blocks are
1282          * zeroed.
1283          */
1284         if (ZRAM_LOGICAL_BLOCK_SIZE == PAGE_SIZE)
1285                 zram->disk->queue->limits.discard_zeroes_data = 1;
1286         else
1287                 zram->disk->queue->limits.discard_zeroes_data = 0;
1288         queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, zram->disk->queue);
1289
1290         add_disk(zram->disk);
1291
1292         ret = sysfs_create_group(&disk_to_dev(zram->disk)->kobj,
1293                                 &zram_disk_attr_group);
1294         if (ret < 0) {
1295                 pr_err("Error creating sysfs group for device %d\n",
1296                                 device_id);
1297                 goto out_free_disk;
1298         }
1299         strlcpy(zram->compressor, default_compressor, sizeof(zram->compressor));
1300         zram->meta = NULL;
1301
1302         pr_info("Added device: %s\n", zram->disk->disk_name);
1303         return device_id;
1304
1305 out_free_disk:
1306         del_gendisk(zram->disk);
1307         put_disk(zram->disk);
1308 out_free_queue:
1309         blk_cleanup_queue(queue);
1310 out_free_idr:
1311         idr_remove(&zram_index_idr, device_id);
1312 out_free_dev:
1313         kfree(zram);
1314         return ret;
1315 }
1316
1317 static int zram_remove(struct zram *zram)
1318 {
1319         struct block_device *bdev;
1320
1321         bdev = bdget_disk(zram->disk, 0);
1322         if (!bdev)
1323                 return -ENOMEM;
1324
1325         mutex_lock(&bdev->bd_mutex);
1326         if (bdev->bd_openers || zram->claim) {
1327                 mutex_unlock(&bdev->bd_mutex);
1328                 bdput(bdev);
1329                 return -EBUSY;
1330         }
1331
1332         zram->claim = true;
1333         mutex_unlock(&bdev->bd_mutex);
1334
1335         /*
1336          * Remove sysfs first, so no one will perform a disksize
1337          * store while we destroy the devices. This also helps during
1338          * hot_remove -- zram_reset_device() is the last holder of
1339          * ->init_lock, no later/concurrent disksize_store() or any
1340          * other sysfs handlers are possible.
1341          */
1342         sysfs_remove_group(&disk_to_dev(zram->disk)->kobj,
1343                         &zram_disk_attr_group);
1344
1345         /* Make sure all the pending I/O are finished */
1346         fsync_bdev(bdev);
1347         zram_reset_device(zram);
1348         bdput(bdev);
1349
1350         pr_info("Removed device: %s\n", zram->disk->disk_name);
1351
1352         blk_cleanup_queue(zram->disk->queue);
1353         del_gendisk(zram->disk);
1354         put_disk(zram->disk);
1355         kfree(zram);
1356         return 0;
1357 }
1358
1359 /* zram-control sysfs attributes */
1360 static ssize_t hot_add_show(struct class *class,
1361                         struct class_attribute *attr,
1362                         char *buf)
1363 {
1364         int ret;
1365
1366         mutex_lock(&zram_index_mutex);
1367         ret = zram_add();
1368         mutex_unlock(&zram_index_mutex);
1369
1370         if (ret < 0)
1371                 return ret;
1372         return scnprintf(buf, PAGE_SIZE, "%d\n", ret);
1373 }
1374
1375 static ssize_t hot_remove_store(struct class *class,
1376                         struct class_attribute *attr,
1377                         const char *buf,
1378                         size_t count)
1379 {
1380         struct zram *zram;
1381         int ret, dev_id;
1382
1383         /* dev_id is gendisk->first_minor, which is `int' */
1384         ret = kstrtoint(buf, 10, &dev_id);
1385         if (ret)
1386                 return ret;
1387         if (dev_id < 0)
1388                 return -EINVAL;
1389
1390         mutex_lock(&zram_index_mutex);
1391
1392         zram = idr_find(&zram_index_idr, dev_id);
1393         if (zram) {
1394                 ret = zram_remove(zram);
1395                 idr_remove(&zram_index_idr, dev_id);
1396         } else {
1397                 ret = -ENODEV;
1398         }
1399
1400         mutex_unlock(&zram_index_mutex);
1401         return ret ? ret : count;
1402 }
1403
1404 static struct class_attribute zram_control_class_attrs[] = {
1405         __ATTR_RO(hot_add),
1406         __ATTR_WO(hot_remove),
1407         __ATTR_NULL,
1408 };
1409
1410 static struct class zram_control_class = {
1411         .name           = "zram-control",
1412         .owner          = THIS_MODULE,
1413         .class_attrs    = zram_control_class_attrs,
1414 };
1415
1416 static int zram_remove_cb(int id, void *ptr, void *data)
1417 {
1418         zram_remove(ptr);
1419         return 0;
1420 }
1421
1422 static void destroy_devices(void)
1423 {
1424         class_unregister(&zram_control_class);
1425         idr_for_each(&zram_index_idr, &zram_remove_cb, NULL);
1426         idr_destroy(&zram_index_idr);
1427         unregister_blkdev(zram_major, "zram");
1428 }
1429
1430 static int __init zram_init(void)
1431 {
1432         int ret;
1433
1434         ret = class_register(&zram_control_class);
1435         if (ret) {
1436                 pr_err("Unable to register zram-control class\n");
1437                 return ret;
1438         }
1439
1440         zram_major = register_blkdev(0, "zram");
1441         if (zram_major <= 0) {
1442                 pr_err("Unable to get major number\n");
1443                 class_unregister(&zram_control_class);
1444                 return -EBUSY;
1445         }
1446
1447         while (num_devices != 0) {
1448                 mutex_lock(&zram_index_mutex);
1449                 ret = zram_add();
1450                 mutex_unlock(&zram_index_mutex);
1451                 if (ret < 0)
1452                         goto out_error;
1453                 num_devices--;
1454         }
1455
1456         return 0;
1457
1458 out_error:
1459         destroy_devices();
1460         return ret;
1461 }
1462
1463 static void __exit zram_exit(void)
1464 {
1465         destroy_devices();
1466 }
1467
1468 module_init(zram_init);
1469 module_exit(zram_exit);
1470
1471 module_param(num_devices, uint, 0);
1472 MODULE_PARM_DESC(num_devices, "Number of pre-created zram devices");
1473
1474 MODULE_LICENSE("Dual BSD/GPL");
1475 MODULE_AUTHOR("Nitin Gupta <ngupta@vflare.org>");
1476 MODULE_DESCRIPTION("Compressed RAM Block Device");