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/backing-dev.h>
  29 #include <linux/string.h>
  30 #include <linux/vmalloc.h>
  31 #include <linux/err.h>
  32 #include <linux/idr.h>
  33 #include <linux/sysfs.h>
  34 #include <linux/cpuhotplug.h>
  35
  36 #include "zram_drv.h"
  37
  38 static DEFINE_IDR(zram_index_idr);
  39 /* idr index must be protected */
  40 static DEFINE_MUTEX(zram_index_mutex);
  41
  42 static int zram_major;
  43 static const char *default_compressor = "lzo";
  44
  45 /* Module params (documentation at end) */
  46 static unsigned int num_devices = 1;
  47
  48 static inline void deprecated_attr_warn(const char *name)
  49 {
  50         pr_warn_once("%d (%s) Attribute %s (and others) will be removed. %s\n",
  51                         task_pid_nr(current),
  52                         current->comm,
  53                         name,
  54                         "See zram documentation.");
  55 }
  56
  57 #define ZRAM_ATTR_RO(name)                                              \
  58 static ssize_t name##_show(struct device *d,                            \
  59                                 struct device_attribute *attr, char *b) \
  60 {                                                                       \
  61         struct zram *zram = dev_to_zram(d);                             \
  62                                                                         \
  63         deprecated_attr_warn(__stringify(name));                        \
  64         return scnprintf(b, PAGE_SIZE, "%llu\n",                        \
  65                 (u64)atomic64_read(&zram->stats.name));                 \
  66 }                                                                       \
  67 static DEVICE_ATTR_RO(name);
  68
  69 static inline bool init_done(struct zram *zram)
  70 {
  71         return zram->disksize;
  72 }
  73
  74 static inline struct zram *dev_to_zram(struct device *dev)
  75 {
  76         return (struct zram *)dev_to_disk(dev)->private_data;
  77 }
  78
  79 /* flag operations require table entry bit_spin_lock() being held */
  80 static int zram_test_flag(struct zram_meta *meta, u32 index,
  81                         enum zram_pageflags flag)
  82 {
  83         return meta->table[index].value & BIT(flag);
  84 }
  85
  86 static void zram_set_flag(struct zram_meta *meta, u32 index,
  87                         enum zram_pageflags flag)
  88 {
  89         meta->table[index].value |= BIT(flag);
  90 }
  91
  92 static void zram_clear_flag(struct zram_meta *meta, u32 index,
  93                         enum zram_pageflags flag)
  94 {
  95         meta->table[index].value &= ~BIT(flag);
  96 }
  97
  98 static size_t zram_get_obj_size(struct zram_meta *meta, u32 index)
  99 {
 100         return meta->table[index].value & (BIT(ZRAM_FLAG_SHIFT) - 1);
 101 }
 102
 103 static void zram_set_obj_size(struct zram_meta *meta,
 104                                         u32 index, size_t size)
 105 {
 106         unsigned long flags = meta->table[index].value >> ZRAM_FLAG_SHIFT;
 107
 108         meta->table[index].value = (flags << ZRAM_FLAG_SHIFT) | size;
 109 }
 110
 111 static inline bool is_partial_io(struct bio_vec *bvec)
 112 {
 113         return bvec->bv_len != PAGE_SIZE;
 114 }
 115
 116 static void zram_revalidate_disk(struct zram *zram)
 117 {
 118         revalidate_disk(zram->disk);
 119         /* revalidate_disk reset the BDI_CAP_STABLE_WRITES so set again */
 120         zram->disk->queue->backing_dev_info->capabilities |=
 121                 BDI_CAP_STABLE_WRITES;
 122 }
 123
 124 /*
 125  * Check if request is within bounds and aligned on zram logical blocks.
 126  */
 127 static inline bool valid_io_request(struct zram *zram,
 128                 sector_t start, unsigned int size)
 129 {
 130         u64 end, bound;
 131
 132         /* unaligned request */
 133         if (unlikely(start & (ZRAM_SECTOR_PER_LOGICAL_BLOCK - 1)))
 134                 return false;
 135         if (unlikely(size & (ZRAM_LOGICAL_BLOCK_SIZE - 1)))
 136                 return false;
 137
 138         end = start + (size >> SECTOR_SHIFT);
 139         bound = zram->disksize >> SECTOR_SHIFT;
 140         /* out of range range */
 141         if (unlikely(start >= bound || end > bound || start > end))
 142                 return false;
 143
 144         /* I/O request is valid */
 145         return true;
 146 }
 147
 148 static void update_position(u32 *index, int *offset, struct bio_vec *bvec)
 149 {
 150         if (*offset + bvec->bv_len >= PAGE_SIZE)
 151                 (*index)++;
 152         *offset = (*offset + bvec->bv_len) % PAGE_SIZE;
 153 }
 154
 155 static inline void update_used_max(struct zram *zram,
 156                                         const unsigned long pages)
 157 {
 158         unsigned long old_max, cur_max;
 159
 160         old_max = atomic_long_read(&zram->stats.max_used_pages);
 161
 162         do {
 163                 cur_max = old_max;
 164                 if (pages > cur_max)
 165                         old_max = atomic_long_cmpxchg(
 166                                 &zram->stats.max_used_pages, cur_max, pages);
 167         } while (old_max != cur_max);
 168 }
 169
 170 static bool page_zero_filled(void *ptr)
 171 {
 172         unsigned int pos;
 173         unsigned long *page;
 174
 175         page = (unsigned long *)ptr;
 176
 177         for (pos = 0; pos != PAGE_SIZE / sizeof(*page); pos++) {
 178                 if (page[pos])
 179                         return false;
 180         }
 181
 182         return true;
 183 }
 184
 185 static void handle_zero_page(struct bio_vec *bvec)
 186 {
 187         struct page *page = bvec->bv_page;
 188         void *user_mem;
 189
 190         user_mem = kmap_atomic(page);
 191         if (is_partial_io(bvec))
 192                 memset(user_mem + bvec->bv_offset, 0, bvec->bv_len);
 193         else
 194                 clear_page(user_mem);
 195         kunmap_atomic(user_mem);
 196
 197         flush_dcache_page(page);
 198 }
 199
 200 static ssize_t initstate_show(struct device *dev,
 201                 struct device_attribute *attr, char *buf)
 202 {
 203         u32 val;
 204         struct zram *zram = dev_to_zram(dev);
 205
 206         down_read(&zram->init_lock);
 207         val = init_done(zram);
 208         up_read(&zram->init_lock);
 209
 210         return scnprintf(buf, PAGE_SIZE, "%u\n", val);
 211 }
 212
 213 static ssize_t disksize_show(struct device *dev,
 214                 struct device_attribute *attr, char *buf)
 215 {
 216         struct zram *zram = dev_to_zram(dev);
 217
 218         return scnprintf(buf, PAGE_SIZE, "%llu\n", zram->disksize);
 219 }
 220
 221 static ssize_t orig_data_size_show(struct device *dev,
 222                 struct device_attribute *attr, char *buf)
 223 {
 224         struct zram *zram = dev_to_zram(dev);
 225
 226         deprecated_attr_warn("orig_data_size");
 227         return scnprintf(buf, PAGE_SIZE, "%llu\n",
 228                 (u64)(atomic64_read(&zram->stats.pages_stored)) << PAGE_SHIFT);
 229 }
 230
 231 static ssize_t mem_used_total_show(struct device *dev,
 232                 struct device_attribute *attr, char *buf)
 233 {
 234         u64 val = 0;
 235         struct zram *zram = dev_to_zram(dev);
 236
 237         deprecated_attr_warn("mem_used_total");
 238         down_read(&zram->init_lock);
 239         if (init_done(zram)) {
 240                 struct zram_meta *meta = zram->meta;
 241                 val = zs_get_total_pages(meta->mem_pool);
 242         }
 243         up_read(&zram->init_lock);
 244
 245         return scnprintf(buf, PAGE_SIZE, "%llu\n", val << PAGE_SHIFT);
 246 }
 247
 248 static ssize_t mem_limit_show(struct device *dev,
 249                 struct device_attribute *attr, char *buf)
 250 {
 251         u64 val;
 252         struct zram *zram = dev_to_zram(dev);
 253
 254         deprecated_attr_warn("mem_limit");
 255         down_read(&zram->init_lock);
 256         val = zram->limit_pages;
 257         up_read(&zram->init_lock);
 258
 259         return scnprintf(buf, PAGE_SIZE, "%llu\n", val << PAGE_SHIFT);
 260 }
 261
 262 static ssize_t mem_limit_store(struct device *dev,
 263                 struct device_attribute *attr, const char *buf, size_t len)
 264 {
 265         u64 limit;
 266         char *tmp;
 267         struct zram *zram = dev_to_zram(dev);
 268
 269         limit = memparse(buf, &tmp);
 270         if (buf == tmp) /* no chars parsed, invalid input */
 271                 return -EINVAL;
 272
 273         down_write(&zram->init_lock);
 274         zram->limit_pages = PAGE_ALIGN(limit) >> PAGE_SHIFT;
 275         up_write(&zram->init_lock);
 276
 277         return len;
 278 }
 279
 280 static ssize_t mem_used_max_show(struct device *dev,
 281                 struct device_attribute *attr, char *buf)
 282 {
 283         u64 val = 0;
 284         struct zram *zram = dev_to_zram(dev);
 285
 286         deprecated_attr_warn("mem_used_max");
 287         down_read(&zram->init_lock);
 288         if (init_done(zram))
 289                 val = atomic_long_read(&zram->stats.max_used_pages);
 290         up_read(&zram->init_lock);
 291
 292         return scnprintf(buf, PAGE_SIZE, "%llu\n", val << PAGE_SHIFT);
 293 }
 294
 295 static ssize_t mem_used_max_store(struct device *dev,
 296                 struct device_attribute *attr, const char *buf, size_t len)
 297 {
 298         int err;
 299         unsigned long val;
 300         struct zram *zram = dev_to_zram(dev);
 301
 302         err = kstrtoul(buf, 10, &val);
 303         if (err || val != 0)
 304                 return -EINVAL;
 305
 306         down_read(&zram->init_lock);
 307         if (init_done(zram)) {
 308                 struct zram_meta *meta = zram->meta;
 309                 atomic_long_set(&zram->stats.max_used_pages,
 310                                 zs_get_total_pages(meta->mem_pool));
 311         }
 312         up_read(&zram->init_lock);
 313
 314         return len;
 315 }
 316
 317 /*
 318  * We switched to per-cpu streams and this attr is not needed anymore.
 319  * However, we will keep it around for some time, because:
 320  * a) we may revert per-cpu streams in the future
 321  * b) it's visible to user space and we need to follow our 2 years
 322  *    retirement rule; but we already have a number of 'soon to be
 323  *    altered' attrs, so max_comp_streams need to wait for the next
 324  *    layoff cycle.
 325  */
 326 static ssize_t max_comp_streams_show(struct device *dev,
 327                 struct device_attribute *attr, char *buf)
 328 {
 329         return scnprintf(buf, PAGE_SIZE, "%d\n", num_online_cpus());
 330 }
 331
 332 static ssize_t max_comp_streams_store(struct device *dev,
 333                 struct device_attribute *attr, const char *buf, size_t len)
 334 {
 335         return len;
 336 }
 337
 338 static ssize_t comp_algorithm_show(struct device *dev,
 339                 struct device_attribute *attr, char *buf)
 340 {
 341         size_t sz;
 342         struct zram *zram = dev_to_zram(dev);
 343
 344         down_read(&zram->init_lock);
 345         sz = zcomp_available_show(zram->compressor, buf);
 346         up_read(&zram->init_lock);
 347
 348         return sz;
 349 }
 350
 351 static ssize_t comp_algorithm_store(struct device *dev,
 352                 struct device_attribute *attr, const char *buf, size_t len)
 353 {
 354         struct zram *zram = dev_to_zram(dev);
 355         char compressor[CRYPTO_MAX_ALG_NAME];
 356         size_t sz;
 357
 358         strlcpy(compressor, buf, sizeof(compressor));
 359         /* ignore trailing newline */
 360         sz = strlen(compressor);
 361         if (sz > 0 && compressor[sz - 1] == '\n')
 362                 compressor[sz - 1] = 0x00;
 363
 364         if (!zcomp_available_algorithm(compressor))
 365                 return -EINVAL;
 366
 367         down_write(&zram->init_lock);
 368         if (init_done(zram)) {
 369                 up_write(&zram->init_lock);
 370                 pr_info("Can't change algorithm for initialized device\n");
 371                 return -EBUSY;
 372         }
 373
 374         strlcpy(zram->compressor, compressor, sizeof(compressor));
 375         up_write(&zram->init_lock);
 376         return len;
 377 }
 378
 379 static ssize_t compact_store(struct device *dev,
 380                 struct device_attribute *attr, const char *buf, size_t len)
 381 {
 382         struct zram *zram = dev_to_zram(dev);
 383         struct zram_meta *meta;
 384
 385         down_read(&zram->init_lock);
 386         if (!init_done(zram)) {
 387                 up_read(&zram->init_lock);
 388                 return -EINVAL;
 389         }
 390
 391         meta = zram->meta;
 392         zs_compact(meta->mem_pool);
 393         up_read(&zram->init_lock);
 394
 395         return len;
 396 }
 397
 398 static ssize_t io_stat_show(struct device *dev,
 399                 struct device_attribute *attr, char *buf)
 400 {
 401         struct zram *zram = dev_to_zram(dev);
 402         ssize_t ret;
 403
 404         down_read(&zram->init_lock);
 405         ret = scnprintf(buf, PAGE_SIZE,
 406                         "%8llu %8llu %8llu %8llu\n",
 407                         (u64)atomic64_read(&zram->stats.failed_reads),
 408                         (u64)atomic64_read(&zram->stats.failed_writes),
 409                         (u64)atomic64_read(&zram->stats.invalid_io),
 410                         (u64)atomic64_read(&zram->stats.notify_free));
 411         up_read(&zram->init_lock);
 412
 413         return ret;
 414 }
 415
 416 static ssize_t mm_stat_show(struct device *dev,
 417                 struct device_attribute *attr, char *buf)
 418 {
 419         struct zram *zram = dev_to_zram(dev);
 420         struct zs_pool_stats pool_stats;
 421         u64 orig_size, mem_used = 0;
 422         long max_used;
 423         ssize_t ret;
 424
 425         memset(&pool_stats, 0x00, sizeof(struct zs_pool_stats));
 426
 427         down_read(&zram->init_lock);
 428         if (init_done(zram)) {
 429                 mem_used = zs_get_total_pages(zram->meta->mem_pool);
 430                 zs_pool_stats(zram->meta->mem_pool, &pool_stats);
 431         }
 432
 433         orig_size = atomic64_read(&zram->stats.pages_stored);
 434         max_used = atomic_long_read(&zram->stats.max_used_pages);
 435
 436         ret = scnprintf(buf, PAGE_SIZE,
 437                         "%8llu %8llu %8llu %8lu %8ld %8llu %8lu\n",
 438                         orig_size << PAGE_SHIFT,
 439                         (u64)atomic64_read(&zram->stats.compr_data_size),
 440                         mem_used << PAGE_SHIFT,
 441                         zram->limit_pages << PAGE_SHIFT,
 442                         max_used << PAGE_SHIFT,
 443                         (u64)atomic64_read(&zram->stats.zero_pages),
 444                         pool_stats.pages_compacted);
 445         up_read(&zram->init_lock);
 446
 447         return ret;
 448 }
 449
 450 static ssize_t debug_stat_show(struct device *dev,
 451                 struct device_attribute *attr, char *buf)
 452 {
 453         int version = 1;
 454         struct zram *zram = dev_to_zram(dev);
 455         ssize_t ret;
 456
 457         down_read(&zram->init_lock);
 458         ret = scnprintf(buf, PAGE_SIZE,
 459                         "version: %d\n%8llu\n",
 460                         version,
 461                         (u64)atomic64_read(&zram->stats.writestall));
 462         up_read(&zram->init_lock);
 463
 464         return ret;
 465 }
 466
 467 static DEVICE_ATTR_RO(io_stat);
 468 static DEVICE_ATTR_RO(mm_stat);
 469 static DEVICE_ATTR_RO(debug_stat);
 470 ZRAM_ATTR_RO(num_reads);
 471 ZRAM_ATTR_RO(num_writes);
 472 ZRAM_ATTR_RO(failed_reads);
 473 ZRAM_ATTR_RO(failed_writes);
 474 ZRAM_ATTR_RO(invalid_io);
 475 ZRAM_ATTR_RO(notify_free);
 476 ZRAM_ATTR_RO(zero_pages);
 477 ZRAM_ATTR_RO(compr_data_size);
 478
 479 static inline bool zram_meta_get(struct zram *zram)
 480 {
 481         if (atomic_inc_not_zero(&zram->refcount))
 482                 return true;
 483         return false;
 484 }
 485
 486 static inline void zram_meta_put(struct zram *zram)
 487 {
 488         atomic_dec(&zram->refcount);
 489 }
 490
 491 static void zram_meta_free(struct zram_meta *meta, u64 disksize)
 492 {
 493         size_t num_pages = disksize >> PAGE_SHIFT;
 494         size_t index;
 495
 496         /* Free all pages that are still in this zram device */
 497         for (index = 0; index < num_pages; index++) {
 498                 unsigned long handle = meta->table[index].handle;
 499
 500                 if (!handle)
 501                         continue;
 502
 503                 zs_free(meta->mem_pool, handle);
 504         }
 505
 506         zs_destroy_pool(meta->mem_pool);
 507         vfree(meta->table);
 508         kfree(meta);
 509 }
 510
 511 static struct zram_meta *zram_meta_alloc(char *pool_name, u64 disksize)
 512 {
 513         size_t num_pages;
 514         struct zram_meta *meta = kmalloc(sizeof(*meta), GFP_KERNEL);
 515
 516         if (!meta)
 517                 return NULL;
 518
 519         num_pages = disksize >> PAGE_SHIFT;
 520         meta->table = vzalloc(num_pages * sizeof(*meta->table));
 521         if (!meta->table) {
 522                 pr_err("Error allocating zram address table\n");
 523                 goto out_error;
 524         }
 525
 526         meta->mem_pool = zs_create_pool(pool_name);
 527         if (!meta->mem_pool) {
 528                 pr_err("Error creating memory pool\n");
 529                 goto out_error;
 530         }
 531
 532         return meta;
 533
 534 out_error:
 535         vfree(meta->table);
 536         kfree(meta);
 537         return NULL;
 538 }
 539
 540 /*
 541  * To protect concurrent access to the same index entry,
 542  * caller should hold this table index entry's bit_spinlock to
 543  * indicate this index entry is accessing.
 544  */
 545 static void zram_free_page(struct zram *zram, size_t index)
 546 {
 547         struct zram_meta *meta = zram->meta;
 548         unsigned long handle = meta->table[index].handle;
 549
 550         if (unlikely(!handle)) {
 551                 /*
 552                  * No memory is allocated for zero filled pages.
 553                  * Simply clear zero page flag.
 554                  */
 555                 if (zram_test_flag(meta, index, ZRAM_ZERO)) {
 556                         zram_clear_flag(meta, index, ZRAM_ZERO);
 557                         atomic64_dec(&zram->stats.zero_pages);
 558                 }
 559                 return;
 560         }
 561
 562         zs_free(meta->mem_pool, handle);
 563
 564         atomic64_sub(zram_get_obj_size(meta, index),
 565                         &zram->stats.compr_data_size);
 566         atomic64_dec(&zram->stats.pages_stored);
 567
 568         meta->table[index].handle = 0;
 569         zram_set_obj_size(meta, index, 0);
 570 }
 571
 572 static int zram_decompress_page(struct zram *zram, char *mem, u32 index)
 573 {
 574         int ret = 0;
 575         unsigned char *cmem;
 576         struct zram_meta *meta = zram->meta;
 577         unsigned long handle;
 578         unsigned int size;
 579
 580         bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
 581         handle = meta->table[index].handle;
 582         size = zram_get_obj_size(meta, index);
 583
 584         if (!handle || zram_test_flag(meta, index, ZRAM_ZERO)) {
 585                 bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 586                 clear_page(mem);
 587                 return 0;
 588         }
 589
 590         cmem = zs_map_object(meta->mem_pool, handle, ZS_MM_RO);
 591         if (size == PAGE_SIZE) {
 592                 copy_page(mem, cmem);
 593         } else {
 594                 struct zcomp_strm *zstrm = zcomp_stream_get(zram->comp);
 595
 596                 ret = zcomp_decompress(zstrm, cmem, size, mem);
 597                 zcomp_stream_put(zram->comp);
 598         }
 599         zs_unmap_object(meta->mem_pool, handle);
 600         bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 601
 602         /* Should NEVER happen. Return bio error if it does. */
 603         if (unlikely(ret)) {
 604                 pr_err("Decompression failed! err=%d, page=%u\n", ret, index);
 605                 return ret;
 606         }
 607
 608         return 0;
 609 }
 610
 611 static int zram_bvec_read(struct zram *zram, struct bio_vec *bvec,
 612                           u32 index, int offset)
 613 {
 614         int ret;
 615         struct page *page;
 616         unsigned char *user_mem, *uncmem = NULL;
 617         struct zram_meta *meta = zram->meta;
 618         page = bvec->bv_page;
 619
 620         bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
 621         if (unlikely(!meta->table[index].handle) ||
 622                         zram_test_flag(meta, index, ZRAM_ZERO)) {
 623                 bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 624                 handle_zero_page(bvec);
 625                 return 0;
 626         }
 627         bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 628
 629         if (is_partial_io(bvec))
 630                 /* Use  a temporary buffer to decompress the page */
 631                 uncmem = kmalloc(PAGE_SIZE, GFP_NOIO);
 632
 633         user_mem = kmap_atomic(page);
 634         if (!is_partial_io(bvec))
 635                 uncmem = user_mem;
 636
 637         if (!uncmem) {
 638                 pr_err("Unable to allocate temp memory\n");
 639                 ret = -ENOMEM;
 640                 goto out_cleanup;
 641         }
 642
 643         ret = zram_decompress_page(zram, uncmem, index);
 644         /* Should NEVER happen. Return bio error if it does. */
 645         if (unlikely(ret))
 646                 goto out_cleanup;
 647
 648         if (is_partial_io(bvec))
 649                 memcpy(user_mem + bvec->bv_offset, uncmem + offset,
 650                                 bvec->bv_len);
 651
 652         flush_dcache_page(page);
 653         ret = 0;
 654 out_cleanup:
 655         kunmap_atomic(user_mem);
 656         if (is_partial_io(bvec))
 657                 kfree(uncmem);
 658         return ret;
 659 }
 660
 661 static int zram_bvec_write(struct zram *zram, struct bio_vec *bvec, u32 index,
 662                            int offset)
 663 {
 664         int ret = 0;
 665         unsigned int clen;
 666         unsigned long handle = 0;
 667         struct page *page;
 668         unsigned char *user_mem, *cmem, *src, *uncmem = NULL;
 669         struct zram_meta *meta = zram->meta;
 670         struct zcomp_strm *zstrm = NULL;
 671         unsigned long alloced_pages;
 672
 673         page = bvec->bv_page;
 674         if (is_partial_io(bvec)) {
 675                 /*
 676                  * This is a partial IO. We need to read the full page
 677                  * before to write the changes.
 678                  */
 679                 uncmem = kmalloc(PAGE_SIZE, GFP_NOIO);
 680                 if (!uncmem) {
 681                         ret = -ENOMEM;
 682                         goto out;
 683                 }
 684                 ret = zram_decompress_page(zram, uncmem, index);
 685                 if (ret)
 686                         goto out;
 687         }
 688
 689 compress_again:
 690         user_mem = kmap_atomic(page);
 691         if (is_partial_io(bvec)) {
 692                 memcpy(uncmem + offset, user_mem + bvec->bv_offset,
 693                        bvec->bv_len);
 694                 kunmap_atomic(user_mem);
 695                 user_mem = NULL;
 696         } else {
 697                 uncmem = user_mem;
 698         }
 699
 700         if (page_zero_filled(uncmem)) {
 701                 if (user_mem)
 702                         kunmap_atomic(user_mem);
 703                 /* Free memory associated with this sector now. */
 704                 bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
 705                 zram_free_page(zram, index);
 706                 zram_set_flag(meta, index, ZRAM_ZERO);
 707                 bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 708
 709                 atomic64_inc(&zram->stats.zero_pages);
 710                 ret = 0;
 711                 goto out;
 712         }
 713
 714         zstrm = zcomp_stream_get(zram->comp);
 715         ret = zcomp_compress(zstrm, uncmem, &clen);
 716         if (!is_partial_io(bvec)) {
 717                 kunmap_atomic(user_mem);
 718                 user_mem = NULL;
 719                 uncmem = NULL;
 720         }
 721
 722         if (unlikely(ret)) {
 723                 pr_err("Compression failed! err=%d\n", ret);
 724                 goto out;
 725         }
 726
 727         src = zstrm->buffer;
 728         if (unlikely(clen > max_zpage_size)) {
 729                 clen = PAGE_SIZE;
 730                 if (is_partial_io(bvec))
 731                         src = uncmem;
 732         }
 733
 734         /*
 735          * handle allocation has 2 paths:
 736          * a) fast path is executed with preemption disabled (for
 737          *  per-cpu streams) and has __GFP_DIRECT_RECLAIM bit clear,
 738          *  since we can't sleep;
 739          * b) slow path enables preemption and attempts to allocate
 740          *  the page with __GFP_DIRECT_RECLAIM bit set. we have to
 741          *  put per-cpu compression stream and, thus, to re-do
 742          *  the compression once handle is allocated.
 743          *
 744          * if we have a 'non-null' handle here then we are coming
 745          * from the slow path and handle has already been allocated.
 746          */
 747         if (!handle)
 748                 handle = zs_malloc(meta->mem_pool, clen,
 749                                 __GFP_KSWAPD_RECLAIM |
 750                                 __GFP_NOWARN |
 751                                 __GFP_HIGHMEM |
 752                                 __GFP_MOVABLE);
 753         if (!handle) {
 754                 zcomp_stream_put(zram->comp);
 755                 zstrm = NULL;
 756
 757                 atomic64_inc(&zram->stats.writestall);
 758
 759                 handle = zs_malloc(meta->mem_pool, clen,
 760                                 GFP_NOIO | __GFP_HIGHMEM |
 761                                 __GFP_MOVABLE);
 762                 if (handle)
 763                         goto compress_again;
 764
 765                 pr_err("Error allocating memory for compressed page: %u, size=%u\n",
 766                         index, clen);
 767                 ret = -ENOMEM;
 768                 goto out;
 769         }
 770
 771         alloced_pages = zs_get_total_pages(meta->mem_pool);
 772         update_used_max(zram, alloced_pages);
 773
 774         if (zram->limit_pages && alloced_pages > zram->limit_pages) {
 775                 zs_free(meta->mem_pool, handle);
 776                 ret = -ENOMEM;
 777                 goto out;
 778         }
 779
 780         cmem = zs_map_object(meta->mem_pool, handle, ZS_MM_WO);
 781
 782         if ((clen == PAGE_SIZE) && !is_partial_io(bvec)) {
 783                 src = kmap_atomic(page);
 784                 copy_page(cmem, src);
 785                 kunmap_atomic(src);
 786         } else {
 787                 memcpy(cmem, src, clen);
 788         }
 789
 790         zcomp_stream_put(zram->comp);
 791         zstrm = NULL;
 792         zs_unmap_object(meta->mem_pool, handle);
 793
 794         /*
 795          * Free memory associated with this sector
 796          * before overwriting unused sectors.
 797          */
 798         bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
 799         zram_free_page(zram, index);
 800
 801         meta->table[index].handle = handle;
 802         zram_set_obj_size(meta, index, clen);
 803         bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 804
 805         /* Update stats */
 806         atomic64_add(clen, &zram->stats.compr_data_size);
 807         atomic64_inc(&zram->stats.pages_stored);
 808 out:
 809         if (zstrm)
 810                 zcomp_stream_put(zram->comp);
 811         if (is_partial_io(bvec))
 812                 kfree(uncmem);
 813         return ret;
 814 }
 815
 816 /*
 817  * zram_bio_discard - handler on discard request
 818  * @index: physical block index in PAGE_SIZE units
 819  * @offset: byte offset within physical block
 820  */
 821 static void zram_bio_discard(struct zram *zram, u32 index,
 822                              int offset, struct bio *bio)
 823 {
 824         size_t n = bio->bi_iter.bi_size;
 825         struct zram_meta *meta = zram->meta;
 826
 827         /*
 828          * zram manages data in physical block size units. Because logical block
 829          * size isn't identical with physical block size on some arch, we
 830          * could get a discard request pointing to a specific offset within a
 831          * certain physical block.  Although we can handle this request by
 832          * reading that physiclal block and decompressing and partially zeroing
 833          * and re-compressing and then re-storing it, this isn't reasonable
 834          * because our intent with a discard request is to save memory.  So
 835          * skipping this logical block is appropriate here.
 836          */
 837         if (offset) {
 838                 if (n <= (PAGE_SIZE - offset))
 839                         return;
 840
 841                 n -= (PAGE_SIZE - offset);
 842                 index++;
 843         }
 844
 845         while (n >= PAGE_SIZE) {
 846                 bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
 847                 zram_free_page(zram, index);
 848                 bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 849                 atomic64_inc(&zram->stats.notify_free);
 850                 index++;
 851                 n -= PAGE_SIZE;
 852         }
 853 }
 854
 855 static int zram_bvec_rw(struct zram *zram, struct bio_vec *bvec, u32 index,
 856                         int offset, bool is_write)
 857 {
 858         unsigned long start_time = jiffies;
 859         int rw_acct = is_write ? REQ_OP_WRITE : REQ_OP_READ;
 860         int ret;
 861
 862         generic_start_io_acct(rw_acct, bvec->bv_len >> SECTOR_SHIFT,
 863                         &zram->disk->part0);
 864
 865         if (!is_write) {
 866                 atomic64_inc(&zram->stats.num_reads);
 867                 ret = zram_bvec_read(zram, bvec, index, offset);
 868         } else {
 869                 atomic64_inc(&zram->stats.num_writes);
 870                 ret = zram_bvec_write(zram, bvec, index, offset);
 871         }
 872
 873         generic_end_io_acct(rw_acct, &zram->disk->part0, start_time);
 874
 875         if (unlikely(ret)) {
 876                 if (!is_write)
 877                         atomic64_inc(&zram->stats.failed_reads);
 878                 else
 879                         atomic64_inc(&zram->stats.failed_writes);
 880         }
 881
 882         return ret;
 883 }
 884
 885 static void __zram_make_request(struct zram *zram, struct bio *bio)
 886 {
 887         int offset;
 888         u32 index;
 889         struct bio_vec bvec;
 890         struct bvec_iter iter;
 891
 892         index = bio->bi_iter.bi_sector >> SECTORS_PER_PAGE_SHIFT;
 893         offset = (bio->bi_iter.bi_sector &
 894                   (SECTORS_PER_PAGE - 1)) << SECTOR_SHIFT;
 895
 896         if (unlikely(bio_op(bio) == REQ_OP_DISCARD)) {
 897                 zram_bio_discard(zram, index, offset, bio);
 898                 bio_endio(bio);
 899                 return;
 900         }
 901
 902         bio_for_each_segment(bvec, bio, iter) {
 903                 int max_transfer_size = PAGE_SIZE - offset;
 904
 905                 if (bvec.bv_len > max_transfer_size) {
 906                         /*
 907                          * zram_bvec_rw() can only make operation on a single
 908                          * zram page. Split the bio vector.
 909                          */
 910                         struct bio_vec bv;
 911
 912                         bv.bv_page = bvec.bv_page;
 913                         bv.bv_len = max_transfer_size;
 914                         bv.bv_offset = bvec.bv_offset;
 915
 916                         if (zram_bvec_rw(zram, &bv, index, offset,
 917                                          op_is_write(bio_op(bio))) < 0)
 918                                 goto out;
 919
 920                         bv.bv_len = bvec.bv_len - max_transfer_size;
 921                         bv.bv_offset += max_transfer_size;
 922                         if (zram_bvec_rw(zram, &bv, index + 1, 0,
 923                                          op_is_write(bio_op(bio))) < 0)
 924                                 goto out;
 925                 } else
 926                         if (zram_bvec_rw(zram, &bvec, index, offset,
 927                                          op_is_write(bio_op(bio))) < 0)
 928                                 goto out;
 929
 930                 update_position(&index, &offset, &bvec);
 931         }
 932
 933         bio_endio(bio);
 934         return;
 935
 936 out:
 937         bio_io_error(bio);
 938 }
 939
 940 /*
 941  * Handler function for all zram I/O requests.
 942  */
 943 static blk_qc_t zram_make_request(struct request_queue *queue, struct bio *bio)
 944 {
 945         struct zram *zram = queue->queuedata;
 946
 947         if (unlikely(!zram_meta_get(zram)))
 948                 goto error;
 949
 950         blk_queue_split(queue, &bio, queue->bio_split);
 951
 952         if (!valid_io_request(zram, bio->bi_iter.bi_sector,
 953                                         bio->bi_iter.bi_size)) {
 954                 atomic64_inc(&zram->stats.invalid_io);
 955                 goto put_zram;
 956         }
 957
 958         __zram_make_request(zram, bio);
 959         zram_meta_put(zram);
 960         return BLK_QC_T_NONE;
 961 put_zram:
 962         zram_meta_put(zram);
 963 error:
 964         bio_io_error(bio);
 965         return BLK_QC_T_NONE;
 966 }
 967
 968 static void zram_slot_free_notify(struct block_device *bdev,
 969                                 unsigned long index)
 970 {
 971         struct zram *zram;
 972         struct zram_meta *meta;
 973
 974         zram = bdev->bd_disk->private_data;
 975         meta = zram->meta;
 976
 977         bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
 978         zram_free_page(zram, index);
 979         bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 980         atomic64_inc(&zram->stats.notify_free);
 981 }
 982
 983 static int zram_rw_page(struct block_device *bdev, sector_t sector,
 984                        struct page *page, bool is_write)
 985 {
 986         int offset, err = -EIO;
 987         u32 index;
 988         struct zram *zram;
 989         struct bio_vec bv;
 990
 991         zram = bdev->bd_disk->private_data;
 992         if (unlikely(!zram_meta_get(zram)))
 993                 goto out;
 994
 995         if (!valid_io_request(zram, sector, PAGE_SIZE)) {
 996                 atomic64_inc(&zram->stats.invalid_io);
 997                 err = -EINVAL;
 998                 goto put_zram;
 999         }
1000
1001         index = sector >> SECTORS_PER_PAGE_SHIFT;
1002         offset = sector & (SECTORS_PER_PAGE - 1) << SECTOR_SHIFT;
1003
1004         bv.bv_page = page;
1005         bv.bv_len = PAGE_SIZE;
1006         bv.bv_offset = 0;
1007
1008         err = zram_bvec_rw(zram, &bv, index, offset, is_write);
1009 put_zram:
1010         zram_meta_put(zram);
1011 out:
1012         /*
1013          * If I/O fails, just return error(ie, non-zero) without
1014          * calling page_endio.
1015          * It causes resubmit the I/O with bio request by upper functions
1016          * of rw_page(e.g., swap_readpage, __swap_writepage) and
1017          * bio->bi_end_io does things to handle the error
1018          * (e.g., SetPageError, set_page_dirty and extra works).
1019          */
1020         if (err == 0)
1021                 page_endio(page, is_write, 0);
1022         return err;
1023 }
1024
1025 static void zram_reset_device(struct zram *zram)
1026 {
1027         struct zram_meta *meta;
1028         struct zcomp *comp;
1029         u64 disksize;
1030
1031         down_write(&zram->init_lock);
1032
1033         zram->limit_pages = 0;
1034
1035         if (!init_done(zram)) {
1036                 up_write(&zram->init_lock);
1037                 return;
1038         }
1039
1040         meta = zram->meta;
1041         comp = zram->comp;
1042         disksize = zram->disksize;
1043         /*
1044          * Refcount will go down to 0 eventually and r/w handler
1045          * cannot handle further I/O so it will bail out by
1046          * check zram_meta_get.
1047          */
1048         zram_meta_put(zram);
1049         /*
1050          * We want to free zram_meta in process context to avoid
1051          * deadlock between reclaim path and any other locks.
1052          */
1053         wait_event(zram->io_done, atomic_read(&zram->refcount) == 0);
1054
1055         /* Reset stats */
1056         memset(&zram->stats, 0, sizeof(zram->stats));
1057         zram->disksize = 0;
1058
1059         set_capacity(zram->disk, 0);
1060         part_stat_set_all(&zram->disk->part0, 0);
1061
1062         up_write(&zram->init_lock);
1063         /* I/O operation under all of CPU are done so let's free */
1064         zram_meta_free(meta, disksize);
1065         zcomp_destroy(comp);
1066 }
1067
1068 static ssize_t disksize_store(struct device *dev,
1069                 struct device_attribute *attr, const char *buf, size_t len)
1070 {
1071         u64 disksize;
1072         struct zcomp *comp;
1073         struct zram_meta *meta;
1074         struct zram *zram = dev_to_zram(dev);
1075         int err;
1076
1077         disksize = memparse(buf, NULL);
1078         if (!disksize)
1079                 return -EINVAL;
1080
1081         disksize = PAGE_ALIGN(disksize);
1082         meta = zram_meta_alloc(zram->disk->disk_name, disksize);
1083         if (!meta)
1084                 return -ENOMEM;
1085
1086         comp = zcomp_create(zram->compressor);
1087         if (IS_ERR(comp)) {
1088                 pr_err("Cannot initialise %s compressing backend\n",
1089                                 zram->compressor);
1090                 err = PTR_ERR(comp);
1091                 goto out_free_meta;
1092         }
1093
1094         down_write(&zram->init_lock);
1095         if (init_done(zram)) {
1096                 pr_info("Cannot change disksize for initialized device\n");
1097                 err = -EBUSY;
1098                 goto out_destroy_comp;
1099         }
1100
1101         init_waitqueue_head(&zram->io_done);
1102         atomic_set(&zram->refcount, 1);
1103         zram->meta = meta;
1104         zram->comp = comp;
1105         zram->disksize = disksize;
1106         set_capacity(zram->disk, zram->disksize >> SECTOR_SHIFT);
1107         zram_revalidate_disk(zram);
1108         up_write(&zram->init_lock);
1109
1110         return len;
1111
1112 out_destroy_comp:
1113         up_write(&zram->init_lock);
1114         zcomp_destroy(comp);
1115 out_free_meta:
1116         zram_meta_free(meta, disksize);
1117         return err;
1118 }
1119
1120 static ssize_t reset_store(struct device *dev,
1121                 struct device_attribute *attr, const char *buf, size_t len)
1122 {
1123         int ret;
1124         unsigned short do_reset;
1125         struct zram *zram;
1126         struct block_device *bdev;
1127
1128         ret = kstrtou16(buf, 10, &do_reset);
1129         if (ret)
1130                 return ret;
1131
1132         if (!do_reset)
1133                 return -EINVAL;
1134
1135         zram = dev_to_zram(dev);
1136         bdev = bdget_disk(zram->disk, 0);
1137         if (!bdev)
1138                 return -ENOMEM;
1139
1140         mutex_lock(&bdev->bd_mutex);
1141         /* Do not reset an active device or claimed device */
1142         if (bdev->bd_openers || zram->claim) {
1143                 mutex_unlock(&bdev->bd_mutex);
1144                 bdput(bdev);
1145                 return -EBUSY;
1146         }
1147
1148         /* From now on, anyone can't open /dev/zram[0-9] */
1149         zram->claim = true;
1150         mutex_unlock(&bdev->bd_mutex);
1151
1152         /* Make sure all the pending I/O are finished */
1153         fsync_bdev(bdev);
1154         zram_reset_device(zram);
1155         zram_revalidate_disk(zram);
1156         bdput(bdev);
1157
1158         mutex_lock(&bdev->bd_mutex);
1159         zram->claim = false;
1160         mutex_unlock(&bdev->bd_mutex);
1161
1162         return len;
1163 }
1164
1165 static int zram_open(struct block_device *bdev, fmode_t mode)
1166 {
1167         int ret = 0;
1168         struct zram *zram;
1169
1170         WARN_ON(!mutex_is_locked(&bdev->bd_mutex));
1171
1172         zram = bdev->bd_disk->private_data;
1173         /* zram was claimed to reset so open request fails */
1174         if (zram->claim)
1175                 ret = -EBUSY;
1176
1177         return ret;
1178 }
1179
1180 static const struct block_device_operations zram_devops = {
1181         .open = zram_open,
1182         .swap_slot_free_notify = zram_slot_free_notify,
1183         .rw_page = zram_rw_page,
1184         .owner = THIS_MODULE
1185 };
1186
1187 static DEVICE_ATTR_WO(compact);
1188 static DEVICE_ATTR_RW(disksize);
1189 static DEVICE_ATTR_RO(initstate);
1190 static DEVICE_ATTR_WO(reset);
1191 static DEVICE_ATTR_RO(orig_data_size);
1192 static DEVICE_ATTR_RO(mem_used_total);
1193 static DEVICE_ATTR_RW(mem_limit);
1194 static DEVICE_ATTR_RW(mem_used_max);
1195 static DEVICE_ATTR_RW(max_comp_streams);
1196 static DEVICE_ATTR_RW(comp_algorithm);
1197
1198 static struct attribute *zram_disk_attrs[] = {
1199         &dev_attr_disksize.attr,
1200         &dev_attr_initstate.attr,
1201         &dev_attr_reset.attr,
1202         &dev_attr_num_reads.attr,
1203         &dev_attr_num_writes.attr,
1204         &dev_attr_failed_reads.attr,
1205         &dev_attr_failed_writes.attr,
1206         &dev_attr_compact.attr,
1207         &dev_attr_invalid_io.attr,
1208         &dev_attr_notify_free.attr,
1209         &dev_attr_zero_pages.attr,
1210         &dev_attr_orig_data_size.attr,
1211         &dev_attr_compr_data_size.attr,
1212         &dev_attr_mem_used_total.attr,
1213         &dev_attr_mem_limit.attr,
1214         &dev_attr_mem_used_max.attr,
1215         &dev_attr_max_comp_streams.attr,
1216         &dev_attr_comp_algorithm.attr,
1217         &dev_attr_io_stat.attr,
1218         &dev_attr_mm_stat.attr,
1219         &dev_attr_debug_stat.attr,
1220         NULL,
1221 };
1222
1223 static struct attribute_group zram_disk_attr_group = {
1224         .attrs = zram_disk_attrs,
1225 };
1226
1227 /*
1228  * Allocate and initialize new zram device. the function returns
1229  * '>= 0' device_id upon success, and negative value otherwise.
1230  */
1231 static int zram_add(void)
1232 {
1233         struct zram *zram;
1234         struct request_queue *queue;
1235         int ret, device_id;
1236
1237         zram = kzalloc(sizeof(struct zram), GFP_KERNEL);
1238         if (!zram)
1239                 return -ENOMEM;
1240
1241         ret = idr_alloc(&zram_index_idr, zram, 0, 0, GFP_KERNEL);
1242         if (ret < 0)
1243                 goto out_free_dev;
1244         device_id = ret;
1245
1246         init_rwsem(&zram->init_lock);
1247
1248         queue = blk_alloc_queue(GFP_KERNEL);
1249         if (!queue) {
1250                 pr_err("Error allocating disk queue for device %d\n",
1251                         device_id);
1252                 ret = -ENOMEM;
1253                 goto out_free_idr;
1254         }
1255
1256         blk_queue_make_request(queue, zram_make_request);
1257
1258         /* gendisk structure */
1259         zram->disk = alloc_disk(1);
1260         if (!zram->disk) {
1261                 pr_err("Error allocating disk structure for device %d\n",
1262                         device_id);
1263                 ret = -ENOMEM;
1264                 goto out_free_queue;
1265         }
1266
1267         zram->disk->major = zram_major;
1268         zram->disk->first_minor = device_id;
1269         zram->disk->fops = &zram_devops;
1270         zram->disk->queue = queue;
1271         zram->disk->queue->queuedata = zram;
1272         zram->disk->private_data = zram;
1273         snprintf(zram->disk->disk_name, 16, "zram%d", device_id);
1274
1275         /* Actual capacity set using syfs (/sys/block/zram<id>/disksize */
1276         set_capacity(zram->disk, 0);
1277         /* zram devices sort of resembles non-rotational disks */
1278         queue_flag_set_unlocked(QUEUE_FLAG_NONROT, zram->disk->queue);
1279         queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, zram->disk->queue);
1280         /*
1281          * To ensure that we always get PAGE_SIZE aligned
1282          * and n*PAGE_SIZED sized I/O requests.
1283          */
1284         blk_queue_physical_block_size(zram->disk->queue, PAGE_SIZE);
1285         blk_queue_logical_block_size(zram->disk->queue,
1286                                         ZRAM_LOGICAL_BLOCK_SIZE);
1287         blk_queue_io_min(zram->disk->queue, PAGE_SIZE);
1288         blk_queue_io_opt(zram->disk->queue, PAGE_SIZE);
1289         zram->disk->queue->limits.discard_granularity = PAGE_SIZE;
1290         blk_queue_max_discard_sectors(zram->disk->queue, UINT_MAX);
1291         /*
1292          * zram_bio_discard() will clear all logical blocks if logical block
1293          * size is identical with physical block size(PAGE_SIZE). But if it is
1294          * different, we will skip discarding some parts of logical blocks in
1295          * the part of the request range which isn't aligned to physical block
1296          * size.  So we can't ensure that all discarded logical blocks are
1297          * zeroed.
1298          */
1299         if (ZRAM_LOGICAL_BLOCK_SIZE == PAGE_SIZE)
1300                 zram->disk->queue->limits.discard_zeroes_data = 1;
1301         else
1302                 zram->disk->queue->limits.discard_zeroes_data = 0;
1303         queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, zram->disk->queue);
1304
1305         add_disk(zram->disk);
1306
1307         ret = sysfs_create_group(&disk_to_dev(zram->disk)->kobj,
1308                                 &zram_disk_attr_group);
1309         if (ret < 0) {
1310                 pr_err("Error creating sysfs group for device %d\n",
1311                                 device_id);
1312                 goto out_free_disk;
1313         }
1314         strlcpy(zram->compressor, default_compressor, sizeof(zram->compressor));
1315         zram->meta = NULL;
1316
1317         pr_info("Added device: %s\n", zram->disk->disk_name);
1318         return device_id;
1319
1320 out_free_disk:
1321         del_gendisk(zram->disk);
1322         put_disk(zram->disk);
1323 out_free_queue:
1324         blk_cleanup_queue(queue);
1325 out_free_idr:
1326         idr_remove(&zram_index_idr, device_id);
1327 out_free_dev:
1328         kfree(zram);
1329         return ret;
1330 }
1331
1332 static int zram_remove(struct zram *zram)
1333 {
1334         struct block_device *bdev;
1335
1336         bdev = bdget_disk(zram->disk, 0);
1337         if (!bdev)
1338                 return -ENOMEM;
1339
1340         mutex_lock(&bdev->bd_mutex);
1341         if (bdev->bd_openers || zram->claim) {
1342                 mutex_unlock(&bdev->bd_mutex);
1343                 bdput(bdev);
1344                 return -EBUSY;
1345         }
1346
1347         zram->claim = true;
1348         mutex_unlock(&bdev->bd_mutex);
1349
1350         /*
1351          * Remove sysfs first, so no one will perform a disksize
1352          * store while we destroy the devices. This also helps during
1353          * hot_remove -- zram_reset_device() is the last holder of
1354          * ->init_lock, no later/concurrent disksize_store() or any
1355          * other sysfs handlers are possible.
1356          */
1357         sysfs_remove_group(&disk_to_dev(zram->disk)->kobj,
1358                         &zram_disk_attr_group);
1359
1360         /* Make sure all the pending I/O are finished */
1361         fsync_bdev(bdev);
1362         zram_reset_device(zram);
1363         bdput(bdev);
1364
1365         pr_info("Removed device: %s\n", zram->disk->disk_name);
1366
1367         blk_cleanup_queue(zram->disk->queue);
1368         del_gendisk(zram->disk);
1369         put_disk(zram->disk);
1370         kfree(zram);
1371         return 0;
1372 }
1373
1374 /* zram-control sysfs attributes */
1375 static ssize_t hot_add_show(struct class *class,
1376                         struct class_attribute *attr,
1377                         char *buf)
1378 {
1379         int ret;
1380
1381         mutex_lock(&zram_index_mutex);
1382         ret = zram_add();
1383         mutex_unlock(&zram_index_mutex);
1384
1385         if (ret < 0)
1386                 return ret;
1387         return scnprintf(buf, PAGE_SIZE, "%d\n", ret);
1388 }
1389
1390 static ssize_t hot_remove_store(struct class *class,
1391                         struct class_attribute *attr,
1392                         const char *buf,
1393                         size_t count)
1394 {
1395         struct zram *zram;
1396         int ret, dev_id;
1397
1398         /* dev_id is gendisk->first_minor, which is `int' */
1399         ret = kstrtoint(buf, 10, &dev_id);
1400         if (ret)
1401                 return ret;
1402         if (dev_id < 0)
1403                 return -EINVAL;
1404
1405         mutex_lock(&zram_index_mutex);
1406
1407         zram = idr_find(&zram_index_idr, dev_id);
1408         if (zram) {
1409                 ret = zram_remove(zram);
1410                 if (!ret)
1411                         idr_remove(&zram_index_idr, dev_id);
1412         } else {
1413                 ret = -ENODEV;
1414         }
1415
1416         mutex_unlock(&zram_index_mutex);
1417         return ret ? ret : count;
1418 }
1419
1420 /*
1421  * NOTE: hot_add attribute is not the usual read-only sysfs attribute. In a
1422  * sense that reading from this file does alter the state of your system -- it
1423  * creates a new un-initialized zram device and returns back this device's
1424  * device_id (or an error code if it fails to create a new device).
1425  */
1426 static struct class_attribute zram_control_class_attrs[] = {
1427         __ATTR(hot_add, 0400, hot_add_show, NULL),
1428         __ATTR_WO(hot_remove),
1429         __ATTR_NULL,
1430 };
1431
1432 static struct class zram_control_class = {
1433         .name           = "zram-control",
1434         .owner          = THIS_MODULE,
1435         .class_attrs    = zram_control_class_attrs,
1436 };
1437
1438 static int zram_remove_cb(int id, void *ptr, void *data)
1439 {
1440         zram_remove(ptr);
1441         return 0;
1442 }
1443
1444 static void destroy_devices(void)
1445 {
1446         class_unregister(&zram_control_class);
1447         idr_for_each(&zram_index_idr, &zram_remove_cb, NULL);
1448         idr_destroy(&zram_index_idr);
1449         unregister_blkdev(zram_major, "zram");
1450         cpuhp_remove_multi_state(CPUHP_ZCOMP_PREPARE);
1451 }
1452
1453 static int __init zram_init(void)
1454 {
1455         int ret;
1456
1457         ret = cpuhp_setup_state_multi(CPUHP_ZCOMP_PREPARE, "block/zram:prepare",
1458                                       zcomp_cpu_up_prepare, zcomp_cpu_dead);
1459         if (ret < 0)
1460                 return ret;
1461
1462         ret = class_register(&zram_control_class);
1463         if (ret) {
1464                 pr_err("Unable to register zram-control class\n");
1465                 cpuhp_remove_multi_state(CPUHP_ZCOMP_PREPARE);
1466                 return ret;
1467         }
1468
1469         zram_major = register_blkdev(0, "zram");
1470         if (zram_major <= 0) {
1471                 pr_err("Unable to get major number\n");
1472                 class_unregister(&zram_control_class);
1473                 cpuhp_remove_multi_state(CPUHP_ZCOMP_PREPARE);
1474                 return -EBUSY;
1475         }
1476
1477         while (num_devices != 0) {
1478                 mutex_lock(&zram_index_mutex);
1479                 ret = zram_add();
1480                 mutex_unlock(&zram_index_mutex);
1481                 if (ret < 0)
1482                         goto out_error;
1483                 num_devices--;
1484         }
1485
1486         return 0;
1487
1488 out_error:
1489         destroy_devices();
1490         return ret;
1491 }
1492
1493 static void __exit zram_exit(void)
1494 {
1495         destroy_devices();
1496 }
1497
1498 module_init(zram_init);
1499 module_exit(zram_exit);
1500
1501 module_param(num_devices, uint, 0);
1502 MODULE_PARM_DESC(num_devices, "Number of pre-created zram devices");
1503
1504 MODULE_LICENSE("Dual BSD/GPL");
1505 MODULE_AUTHOR("Nitin Gupta <ngupta@vflare.org>");
1506 MODULE_DESCRIPTION("Compressed RAM Block Device");