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 static ssize_t max_comp_streams_show(struct device *dev,
 308                 struct device_attribute *attr, char *buf)
 309 {
 310         int val;
 311         struct zram *zram = dev_to_zram(dev);
 312
 313         down_read(&zram->init_lock);
 314         val = zram->max_comp_streams;
 315         up_read(&zram->init_lock);
 316
 317         return scnprintf(buf, PAGE_SIZE, "%d\n", val);
 318 }
 319
 320 static ssize_t max_comp_streams_store(struct device *dev,
 321                 struct device_attribute *attr, const char *buf, size_t len)
 322 {
 323         int num;
 324         struct zram *zram = dev_to_zram(dev);
 325         int ret;
 326
 327         ret = kstrtoint(buf, 0, &num);
 328         if (ret < 0)
 329                 return ret;
 330         if (num < 1)
 331                 return -EINVAL;
 332
 333         down_write(&zram->init_lock);
 334         if (init_done(zram)) {
 335                 if (!zcomp_set_max_streams(zram->comp, num)) {
 336                         pr_info("Cannot change max compression streams\n");
 337                         ret = -EINVAL;
 338                         goto out;
 339                 }
 340         }
 341
 342         zram->max_comp_streams = num;
 343         ret = len;
 344 out:
 345         up_write(&zram->init_lock);
 346         return ret;
 347 }
 348
 349 static ssize_t comp_algorithm_show(struct device *dev,
 350                 struct device_attribute *attr, char *buf)
 351 {
 352         size_t sz;
 353         struct zram *zram = dev_to_zram(dev);
 354
 355         down_read(&zram->init_lock);
 356         sz = zcomp_available_show(zram->compressor, buf);
 357         up_read(&zram->init_lock);
 358
 359         return sz;
 360 }
 361
 362 static ssize_t comp_algorithm_store(struct device *dev,
 363                 struct device_attribute *attr, const char *buf, size_t len)
 364 {
 365         struct zram *zram = dev_to_zram(dev);
 366         size_t sz;
 367
 368         if (!zcomp_available_algorithm(buf))
 369                 return -EINVAL;
 370
 371         down_write(&zram->init_lock);
 372         if (init_done(zram)) {
 373                 up_write(&zram->init_lock);
 374                 pr_info("Can't change algorithm for initialized device\n");
 375                 return -EBUSY;
 376         }
 377         strlcpy(zram->compressor, buf, sizeof(zram->compressor));
 378
 379         /* ignore trailing newline */
 380         sz = strlen(zram->compressor);
 381         if (sz > 0 && zram->compressor[sz - 1] == '\n')
 382                 zram->compressor[sz - 1] = 0x00;
 383
 384         up_write(&zram->init_lock);
 385         return len;
 386 }
 387
 388 static ssize_t compact_store(struct device *dev,
 389                 struct device_attribute *attr, const char *buf, size_t len)
 390 {
 391         struct zram *zram = dev_to_zram(dev);
 392         struct zram_meta *meta;
 393
 394         down_read(&zram->init_lock);
 395         if (!init_done(zram)) {
 396                 up_read(&zram->init_lock);
 397                 return -EINVAL;
 398         }
 399
 400         meta = zram->meta;
 401         zs_compact(meta->mem_pool);
 402         up_read(&zram->init_lock);
 403
 404         return len;
 405 }
 406
 407 static ssize_t io_stat_show(struct device *dev,
 408                 struct device_attribute *attr, char *buf)
 409 {
 410         struct zram *zram = dev_to_zram(dev);
 411         ssize_t ret;
 412
 413         down_read(&zram->init_lock);
 414         ret = scnprintf(buf, PAGE_SIZE,
 415                         "%8llu %8llu %8llu %8llu\n",
 416                         (u64)atomic64_read(&zram->stats.failed_reads),
 417                         (u64)atomic64_read(&zram->stats.failed_writes),
 418                         (u64)atomic64_read(&zram->stats.invalid_io),
 419                         (u64)atomic64_read(&zram->stats.notify_free));
 420         up_read(&zram->init_lock);
 421
 422         return ret;
 423 }
 424
 425 static ssize_t mm_stat_show(struct device *dev,
 426                 struct device_attribute *attr, char *buf)
 427 {
 428         struct zram *zram = dev_to_zram(dev);
 429         struct zs_pool_stats pool_stats;
 430         u64 orig_size, mem_used = 0;
 431         long max_used;
 432         ssize_t ret;
 433
 434         memset(&pool_stats, 0x00, sizeof(struct zs_pool_stats));
 435
 436         down_read(&zram->init_lock);
 437         if (init_done(zram)) {
 438                 mem_used = zs_get_total_pages(zram->meta->mem_pool);
 439                 zs_pool_stats(zram->meta->mem_pool, &pool_stats);
 440         }
 441
 442         orig_size = atomic64_read(&zram->stats.pages_stored);
 443         max_used = atomic_long_read(&zram->stats.max_used_pages);
 444
 445         ret = scnprintf(buf, PAGE_SIZE,
 446                         "%8llu %8llu %8llu %8lu %8ld %8llu %8lu\n",
 447                         orig_size << PAGE_SHIFT,
 448                         (u64)atomic64_read(&zram->stats.compr_data_size),
 449                         mem_used << PAGE_SHIFT,
 450                         zram->limit_pages << PAGE_SHIFT,
 451                         max_used << PAGE_SHIFT,
 452                         (u64)atomic64_read(&zram->stats.zero_pages),
 453                         atomic_long_read(&pool_stats.pages_compacted));
 454         up_read(&zram->init_lock);
 455
 456         return ret;
 457 }
 458
 459 static DEVICE_ATTR_RO(io_stat);
 460 static DEVICE_ATTR_RO(mm_stat);
 461 ZRAM_ATTR_RO(num_reads);
 462 ZRAM_ATTR_RO(num_writes);
 463 ZRAM_ATTR_RO(failed_reads);
 464 ZRAM_ATTR_RO(failed_writes);
 465 ZRAM_ATTR_RO(invalid_io);
 466 ZRAM_ATTR_RO(notify_free);
 467 ZRAM_ATTR_RO(zero_pages);
 468 ZRAM_ATTR_RO(compr_data_size);
 469
 470 static inline bool zram_meta_get(struct zram *zram)
 471 {
 472         if (atomic_inc_not_zero(&zram->refcount))
 473                 return true;
 474         return false;
 475 }
 476
 477 static inline void zram_meta_put(struct zram *zram)
 478 {
 479         atomic_dec(&zram->refcount);
 480 }
 481
 482 static void zram_meta_free(struct zram_meta *meta, u64 disksize)
 483 {
 484         size_t num_pages = disksize >> PAGE_SHIFT;
 485         size_t index;
 486
 487         /* Free all pages that are still in this zram device */
 488         for (index = 0; index < num_pages; index++) {
 489                 unsigned long handle = meta->table[index].handle;
 490
 491                 if (!handle)
 492                         continue;
 493
 494                 zs_free(meta->mem_pool, handle);
 495         }
 496
 497         zs_destroy_pool(meta->mem_pool);
 498         vfree(meta->table);
 499         kfree(meta);
 500 }
 501
 502 static struct zram_meta *zram_meta_alloc(char *pool_name, u64 disksize)
 503 {
 504         size_t num_pages;
 505         struct zram_meta *meta = kmalloc(sizeof(*meta), GFP_KERNEL);
 506
 507         if (!meta)
 508                 return NULL;
 509
 510         num_pages = disksize >> PAGE_SHIFT;
 511         meta->table = vzalloc(num_pages * sizeof(*meta->table));
 512         if (!meta->table) {
 513                 pr_err("Error allocating zram address table\n");
 514                 goto out_error;
 515         }
 516
 517         meta->mem_pool = zs_create_pool(pool_name, GFP_NOIO | __GFP_HIGHMEM);
 518         if (!meta->mem_pool) {
 519                 pr_err("Error creating memory pool\n");
 520                 goto out_error;
 521         }
 522
 523         return meta;
 524
 525 out_error:
 526         vfree(meta->table);
 527         kfree(meta);
 528         return NULL;
 529 }
 530
 531 /*
 532  * To protect concurrent access to the same index entry,
 533  * caller should hold this table index entry's bit_spinlock to
 534  * indicate this index entry is accessing.
 535  */
 536 static void zram_free_page(struct zram *zram, size_t index)
 537 {
 538         struct zram_meta *meta = zram->meta;
 539         unsigned long handle = meta->table[index].handle;
 540
 541         if (unlikely(!handle)) {
 542                 /*
 543                  * No memory is allocated for zero filled pages.
 544                  * Simply clear zero page flag.
 545                  */
 546                 if (zram_test_flag(meta, index, ZRAM_ZERO)) {
 547                         zram_clear_flag(meta, index, ZRAM_ZERO);
 548                         atomic64_dec(&zram->stats.zero_pages);
 549                 }
 550                 return;
 551         }
 552
 553         zs_free(meta->mem_pool, handle);
 554
 555         atomic64_sub(zram_get_obj_size(meta, index),
 556                         &zram->stats.compr_data_size);
 557         atomic64_dec(&zram->stats.pages_stored);
 558
 559         meta->table[index].handle = 0;
 560         zram_set_obj_size(meta, index, 0);
 561 }
 562
 563 static int zram_decompress_page(struct zram *zram, char *mem, u32 index)
 564 {
 565         int ret = 0;
 566         unsigned char *cmem;
 567         struct zram_meta *meta = zram->meta;
 568         unsigned long handle;
 569         size_t size;
 570
 571         bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
 572         handle = meta->table[index].handle;
 573         size = zram_get_obj_size(meta, index);
 574
 575         if (!handle || zram_test_flag(meta, index, ZRAM_ZERO)) {
 576                 bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 577                 memset(mem, 0, PAGE_SIZE);
 578                 return 0;
 579         }
 580
 581         cmem = zs_map_object(meta->mem_pool, handle, ZS_MM_RO);
 582         if (size == PAGE_SIZE)
 583                 memcpy(mem, cmem, PAGE_SIZE);
 584         else
 585                 ret = zcomp_decompress(zram->comp, cmem, size, mem);
 586         zs_unmap_object(meta->mem_pool, handle);
 587         bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 588
 589         /* Should NEVER happen. Return bio error if it does. */
 590         if (unlikely(ret)) {
 591                 pr_err("Decompression failed! err=%d, page=%u\n", ret, index);
 592                 return ret;
 593         }
 594
 595         return 0;
 596 }
 597
 598 static int zram_bvec_read(struct zram *zram, struct bio_vec *bvec,
 599                           u32 index, int offset)
 600 {
 601         int ret;
 602         struct page *page;
 603         unsigned char *user_mem, *uncmem = NULL;
 604         struct zram_meta *meta = zram->meta;
 605         page = bvec->bv_page;
 606
 607         bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
 608         if (unlikely(!meta->table[index].handle) ||
 609                         zram_test_flag(meta, index, ZRAM_ZERO)) {
 610                 bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 611                 handle_zero_page(bvec);
 612                 return 0;
 613         }
 614         bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 615
 616         if (is_partial_io(bvec))
 617                 /* Use  a temporary buffer to decompress the page */
 618                 uncmem = kmalloc(PAGE_SIZE, GFP_NOIO);
 619
 620         user_mem = kmap_atomic(page);
 621         if (!is_partial_io(bvec))
 622                 uncmem = user_mem;
 623
 624         if (!uncmem) {
 625                 pr_err("Unable to allocate temp memory\n");
 626                 ret = -ENOMEM;
 627                 goto out_cleanup;
 628         }
 629
 630         ret = zram_decompress_page(zram, uncmem, index);
 631         /* Should NEVER happen. Return bio error if it does. */
 632         if (unlikely(ret))
 633                 goto out_cleanup;
 634
 635         if (is_partial_io(bvec))
 636                 memcpy(user_mem + bvec->bv_offset, uncmem + offset,
 637                                 bvec->bv_len);
 638
 639         flush_dcache_page(page);
 640         ret = 0;
 641 out_cleanup:
 642         kunmap_atomic(user_mem);
 643         if (is_partial_io(bvec))
 644                 kfree(uncmem);
 645         return ret;
 646 }
 647
 648 static int zram_bvec_write(struct zram *zram, struct bio_vec *bvec, u32 index,
 649                            int offset)
 650 {
 651         int ret = 0;
 652         size_t clen;
 653         unsigned long handle;
 654         struct page *page;
 655         unsigned char *user_mem, *cmem, *src, *uncmem = NULL;
 656         struct zram_meta *meta = zram->meta;
 657         struct zcomp_strm *zstrm = NULL;
 658         unsigned long alloced_pages;
 659
 660         page = bvec->bv_page;
 661         if (is_partial_io(bvec)) {
 662                 /*
 663                  * This is a partial IO. We need to read the full page
 664                  * before to write the changes.
 665                  */
 666                 uncmem = kmalloc(PAGE_SIZE, GFP_NOIO);
 667                 if (!uncmem) {
 668                         ret = -ENOMEM;
 669                         goto out;
 670                 }
 671                 ret = zram_decompress_page(zram, uncmem, index);
 672                 if (ret)
 673                         goto out;
 674         }
 675
 676         zstrm = zcomp_strm_find(zram->comp);
 677         user_mem = kmap_atomic(page);
 678
 679         if (is_partial_io(bvec)) {
 680                 memcpy(uncmem + offset, user_mem + bvec->bv_offset,
 681                        bvec->bv_len);
 682                 kunmap_atomic(user_mem);
 683                 user_mem = NULL;
 684         } else {
 685                 uncmem = user_mem;
 686         }
 687
 688         if (page_zero_filled(uncmem)) {
 689                 if (user_mem)
 690                         kunmap_atomic(user_mem);
 691                 /* Free memory associated with this sector now. */
 692                 bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
 693                 zram_free_page(zram, index);
 694                 zram_set_flag(meta, index, ZRAM_ZERO);
 695                 bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 696
 697                 atomic64_inc(&zram->stats.zero_pages);
 698                 ret = 0;
 699                 goto out;
 700         }
 701
 702         ret = zcomp_compress(zram->comp, zstrm, uncmem, &clen);
 703         if (!is_partial_io(bvec)) {
 704                 kunmap_atomic(user_mem);
 705                 user_mem = NULL;
 706                 uncmem = NULL;
 707         }
 708
 709         if (unlikely(ret)) {
 710                 pr_err("Compression failed! err=%d\n", ret);
 711                 goto out;
 712         }
 713         src = zstrm->buffer;
 714         if (unlikely(clen > max_zpage_size)) {
 715                 clen = PAGE_SIZE;
 716                 if (is_partial_io(bvec))
 717                         src = uncmem;
 718         }
 719
 720         handle = zs_malloc(meta->mem_pool, clen);
 721         if (!handle) {
 722                 pr_err("Error allocating memory for compressed page: %u, size=%zu\n",
 723                         index, clen);
 724                 ret = -ENOMEM;
 725                 goto out;
 726         }
 727
 728         alloced_pages = zs_get_total_pages(meta->mem_pool);
 729         update_used_max(zram, alloced_pages);
 730
 731         if (zram->limit_pages && alloced_pages > zram->limit_pages) {
 732                 zs_free(meta->mem_pool, handle);
 733                 ret = -ENOMEM;
 734                 goto out;
 735         }
 736
 737         cmem = zs_map_object(meta->mem_pool, handle, ZS_MM_WO);
 738
 739         if ((clen == PAGE_SIZE) && !is_partial_io(bvec)) {
 740                 src = kmap_atomic(page);
 741                 memcpy(cmem, src, PAGE_SIZE);
 742                 kunmap_atomic(src);
 743         } else {
 744                 memcpy(cmem, src, clen);
 745         }
 746
 747         zcomp_strm_release(zram->comp, zstrm);
 748         zstrm = NULL;
 749         zs_unmap_object(meta->mem_pool, handle);
 750
 751         /*
 752          * Free memory associated with this sector
 753          * before overwriting unused sectors.
 754          */
 755         bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
 756         zram_free_page(zram, index);
 757
 758         meta->table[index].handle = handle;
 759         zram_set_obj_size(meta, index, clen);
 760         bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 761
 762         /* Update stats */
 763         atomic64_add(clen, &zram->stats.compr_data_size);
 764         atomic64_inc(&zram->stats.pages_stored);
 765 out:
 766         if (zstrm)
 767                 zcomp_strm_release(zram->comp, zstrm);
 768         if (is_partial_io(bvec))
 769                 kfree(uncmem);
 770         return ret;
 771 }
 772
 773 /*
 774  * zram_bio_discard - handler on discard request
 775  * @index: physical block index in PAGE_SIZE units
 776  * @offset: byte offset within physical block
 777  */
 778 static void zram_bio_discard(struct zram *zram, u32 index,
 779                              int offset, struct bio *bio)
 780 {
 781         size_t n = bio->bi_iter.bi_size;
 782         struct zram_meta *meta = zram->meta;
 783
 784         /*
 785          * zram manages data in physical block size units. Because logical block
 786          * size isn't identical with physical block size on some arch, we
 787          * could get a discard request pointing to a specific offset within a
 788          * certain physical block.  Although we can handle this request by
 789          * reading that physiclal block and decompressing and partially zeroing
 790          * and re-compressing and then re-storing it, this isn't reasonable
 791          * because our intent with a discard request is to save memory.  So
 792          * skipping this logical block is appropriate here.
 793          */
 794         if (offset) {
 795                 if (n <= (PAGE_SIZE - offset))
 796                         return;
 797
 798                 n -= (PAGE_SIZE - offset);
 799                 index++;
 800         }
 801
 802         while (n >= PAGE_SIZE) {
 803                 bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
 804                 zram_free_page(zram, index);
 805                 bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 806                 atomic64_inc(&zram->stats.notify_free);
 807                 index++;
 808                 n -= PAGE_SIZE;
 809         }
 810 }
 811
 812 static int zram_bvec_rw(struct zram *zram, struct bio_vec *bvec, u32 index,
 813                         int offset, int rw)
 814 {
 815         unsigned long start_time = jiffies;
 816         int ret;
 817
 818         generic_start_io_acct(rw, bvec->bv_len >> SECTOR_SHIFT,
 819                         &zram->disk->part0);
 820
 821         if (rw == READ) {
 822                 atomic64_inc(&zram->stats.num_reads);
 823                 ret = zram_bvec_read(zram, bvec, index, offset);
 824         } else {
 825                 atomic64_inc(&zram->stats.num_writes);
 826                 ret = zram_bvec_write(zram, bvec, index, offset);
 827         }
 828
 829         generic_end_io_acct(rw, &zram->disk->part0, start_time);
 830
 831         if (unlikely(ret)) {
 832                 if (rw == READ)
 833                         atomic64_inc(&zram->stats.failed_reads);
 834                 else
 835                         atomic64_inc(&zram->stats.failed_writes);
 836         }
 837
 838         return ret;
 839 }
 840
 841 static void __zram_make_request(struct zram *zram, struct bio *bio)
 842 {
 843         int offset, rw;
 844         u32 index;
 845         struct bio_vec bvec;
 846         struct bvec_iter iter;
 847
 848         index = bio->bi_iter.bi_sector >> SECTORS_PER_PAGE_SHIFT;
 849         offset = (bio->bi_iter.bi_sector &
 850                   (SECTORS_PER_PAGE - 1)) << SECTOR_SHIFT;
 851
 852         if (unlikely(bio->bi_rw & REQ_DISCARD)) {
 853                 zram_bio_discard(zram, index, offset, bio);
 854                 bio_endio(bio);
 855                 return;
 856         }
 857
 858         rw = bio_data_dir(bio);
 859         bio_for_each_segment(bvec, bio, iter) {
 860                 int max_transfer_size = PAGE_SIZE - offset;
 861
 862                 if (bvec.bv_len > max_transfer_size) {
 863                         /*
 864                          * zram_bvec_rw() can only make operation on a single
 865                          * zram page. Split the bio vector.
 866                          */
 867                         struct bio_vec bv;
 868
 869                         bv.bv_page = bvec.bv_page;
 870                         bv.bv_len = max_transfer_size;
 871                         bv.bv_offset = bvec.bv_offset;
 872
 873                         if (zram_bvec_rw(zram, &bv, index, offset, rw) < 0)
 874                                 goto out;
 875
 876                         bv.bv_len = bvec.bv_len - max_transfer_size;
 877                         bv.bv_offset += max_transfer_size;
 878                         if (zram_bvec_rw(zram, &bv, index + 1, 0, rw) < 0)
 879                                 goto out;
 880                 } else
 881                         if (zram_bvec_rw(zram, &bvec, index, offset, rw) < 0)
 882                                 goto out;
 883
 884                 update_position(&index, &offset, &bvec);
 885         }
 886
 887         bio_endio(bio);
 888         return;
 889
 890 out:
 891         bio_io_error(bio);
 892 }
 893
 894 /*
 895  * Handler function for all zram I/O requests.
 896  */
 897 static blk_qc_t zram_make_request(struct request_queue *queue, struct bio *bio)
 898 {
 899         struct zram *zram = queue->queuedata;
 900
 901         if (unlikely(!zram_meta_get(zram)))
 902                 goto error;
 903
 904         blk_queue_split(queue, &bio, queue->bio_split);
 905
 906         if (!valid_io_request(zram, bio->bi_iter.bi_sector,
 907                                         bio->bi_iter.bi_size)) {
 908                 atomic64_inc(&zram->stats.invalid_io);
 909                 goto put_zram;
 910         }
 911
 912         __zram_make_request(zram, bio);
 913         zram_meta_put(zram);
 914         return BLK_QC_T_NONE;
 915 put_zram:
 916         zram_meta_put(zram);
 917 error:
 918         bio_io_error(bio);
 919         return BLK_QC_T_NONE;
 920 }
 921
 922 static void zram_slot_free_notify(struct block_device *bdev,
 923                                 unsigned long index)
 924 {
 925         struct zram *zram;
 926         struct zram_meta *meta;
 927
 928         zram = bdev->bd_disk->private_data;
 929         meta = zram->meta;
 930
 931         bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
 932         zram_free_page(zram, index);
 933         bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 934         atomic64_inc(&zram->stats.notify_free);
 935 }
 936
 937 static int zram_rw_page(struct block_device *bdev, sector_t sector,
 938                        struct page *page, int rw)
 939 {
 940         int offset, err = -EIO;
 941         u32 index;
 942         struct zram *zram;
 943         struct bio_vec bv;
 944
 945         zram = bdev->bd_disk->private_data;
 946         if (unlikely(!zram_meta_get(zram)))
 947                 goto out;
 948
 949         if (!valid_io_request(zram, sector, PAGE_SIZE)) {
 950                 atomic64_inc(&zram->stats.invalid_io);
 951                 err = -EINVAL;
 952                 goto put_zram;
 953         }
 954
 955         index = sector >> SECTORS_PER_PAGE_SHIFT;
 956         offset = sector & (SECTORS_PER_PAGE - 1) << SECTOR_SHIFT;
 957
 958         bv.bv_page = page;
 959         bv.bv_len = PAGE_SIZE;
 960         bv.bv_offset = 0;
 961
 962         err = zram_bvec_rw(zram, &bv, index, offset, rw);
 963 put_zram:
 964         zram_meta_put(zram);
 965 out:
 966         /*
 967          * If I/O fails, just return error(ie, non-zero) without
 968          * calling page_endio.
 969          * It causes resubmit the I/O with bio request by upper functions
 970          * of rw_page(e.g., swap_readpage, __swap_writepage) and
 971          * bio->bi_end_io does things to handle the error
 972          * (e.g., SetPageError, set_page_dirty and extra works).
 973          */
 974         if (err == 0)
 975                 page_endio(page, rw, 0);
 976         return err;
 977 }
 978
 979 static void zram_reset_device(struct zram *zram)
 980 {
 981         struct zram_meta *meta;
 982         struct zcomp *comp;
 983         u64 disksize;
 984
 985         down_write(&zram->init_lock);
 986
 987         zram->limit_pages = 0;
 988
 989         if (!init_done(zram)) {
 990                 up_write(&zram->init_lock);
 991                 return;
 992         }
 993
 994         meta = zram->meta;
 995         comp = zram->comp;
 996         disksize = zram->disksize;
 997         /*
 998          * Refcount will go down to 0 eventually and r/w handler
 999          * cannot handle further I/O so it will bail out by
1000          * check zram_meta_get.
1001          */
1002         zram_meta_put(zram);
1003         /*
1004          * We want to free zram_meta in process context to avoid
1005          * deadlock between reclaim path and any other locks.
1006          */
1007         wait_event(zram->io_done, atomic_read(&zram->refcount) == 0);
1008
1009         /* Reset stats */
1010         memset(&zram->stats, 0, sizeof(zram->stats));
1011         zram->disksize = 0;
1012         zram->max_comp_streams = 1;
1013
1014         set_capacity(zram->disk, 0);
1015         part_stat_set_all(&zram->disk->part0, 0);
1016
1017         up_write(&zram->init_lock);
1018         /* I/O operation under all of CPU are done so let's free */
1019         zram_meta_free(meta, disksize);
1020         zcomp_destroy(comp);
1021 }
1022
1023 static ssize_t disksize_store(struct device *dev,
1024                 struct device_attribute *attr, const char *buf, size_t len)
1025 {
1026         u64 disksize;
1027         struct zcomp *comp;
1028         struct zram_meta *meta;
1029         struct zram *zram = dev_to_zram(dev);
1030         int err;
1031
1032         disksize = memparse(buf, NULL);
1033         if (!disksize)
1034                 return -EINVAL;
1035
1036         disksize = PAGE_ALIGN(disksize);
1037         meta = zram_meta_alloc(zram->disk->disk_name, disksize);
1038         if (!meta)
1039                 return -ENOMEM;
1040
1041         comp = zcomp_create(zram->compressor, zram->max_comp_streams);
1042         if (IS_ERR(comp)) {
1043                 pr_err("Cannot initialise %s compressing backend\n",
1044                                 zram->compressor);
1045                 err = PTR_ERR(comp);
1046                 goto out_free_meta;
1047         }
1048
1049         down_write(&zram->init_lock);
1050         if (init_done(zram)) {
1051                 pr_info("Cannot change disksize for initialized device\n");
1052                 err = -EBUSY;
1053                 goto out_destroy_comp;
1054         }
1055
1056         init_waitqueue_head(&zram->io_done);
1057         atomic_set(&zram->refcount, 1);
1058         zram->meta = meta;
1059         zram->comp = comp;
1060         zram->disksize = disksize;
1061         set_capacity(zram->disk, zram->disksize >> SECTOR_SHIFT);
1062         up_write(&zram->init_lock);
1063
1064         /*
1065          * Revalidate disk out of the init_lock to avoid lockdep splat.
1066          * It's okay because disk's capacity is protected by init_lock
1067          * so that revalidate_disk always sees up-to-date capacity.
1068          */
1069         revalidate_disk(zram->disk);
1070
1071         return len;
1072
1073 out_destroy_comp:
1074         up_write(&zram->init_lock);
1075         zcomp_destroy(comp);
1076 out_free_meta:
1077         zram_meta_free(meta, disksize);
1078         return err;
1079 }
1080
1081 static ssize_t reset_store(struct device *dev,
1082                 struct device_attribute *attr, const char *buf, size_t len)
1083 {
1084         int ret;
1085         unsigned short do_reset;
1086         struct zram *zram;
1087         struct block_device *bdev;
1088
1089         ret = kstrtou16(buf, 10, &do_reset);
1090         if (ret)
1091                 return ret;
1092
1093         if (!do_reset)
1094                 return -EINVAL;
1095
1096         zram = dev_to_zram(dev);
1097         bdev = bdget_disk(zram->disk, 0);
1098         if (!bdev)
1099                 return -ENOMEM;
1100
1101         mutex_lock(&bdev->bd_mutex);
1102         /* Do not reset an active device or claimed device */
1103         if (bdev->bd_openers || zram->claim) {
1104                 mutex_unlock(&bdev->bd_mutex);
1105                 bdput(bdev);
1106                 return -EBUSY;
1107         }
1108
1109         /* From now on, anyone can't open /dev/zram[0-9] */
1110         zram->claim = true;
1111         mutex_unlock(&bdev->bd_mutex);
1112
1113         /* Make sure all the pending I/O are finished */
1114         fsync_bdev(bdev);
1115         zram_reset_device(zram);
1116         revalidate_disk(zram->disk);
1117         bdput(bdev);
1118
1119         mutex_lock(&bdev->bd_mutex);
1120         zram->claim = false;
1121         mutex_unlock(&bdev->bd_mutex);
1122
1123         return len;
1124 }
1125
1126 static int zram_open(struct block_device *bdev, fmode_t mode)
1127 {
1128         int ret = 0;
1129         struct zram *zram;
1130
1131         WARN_ON(!mutex_is_locked(&bdev->bd_mutex));
1132
1133         zram = bdev->bd_disk->private_data;
1134         /* zram was claimed to reset so open request fails */
1135         if (zram->claim)
1136                 ret = -EBUSY;
1137
1138         return ret;
1139 }
1140
1141 static const struct block_device_operations zram_devops = {
1142         .open = zram_open,
1143         .swap_slot_free_notify = zram_slot_free_notify,
1144         .rw_page = zram_rw_page,
1145         .owner = THIS_MODULE
1146 };
1147
1148 static DEVICE_ATTR_WO(compact);
1149 static DEVICE_ATTR_RW(disksize);
1150 static DEVICE_ATTR_RO(initstate);
1151 static DEVICE_ATTR_WO(reset);
1152 static DEVICE_ATTR_RO(orig_data_size);
1153 static DEVICE_ATTR_RO(mem_used_total);
1154 static DEVICE_ATTR_RW(mem_limit);
1155 static DEVICE_ATTR_RW(mem_used_max);
1156 static DEVICE_ATTR_RW(max_comp_streams);
1157 static DEVICE_ATTR_RW(comp_algorithm);
1158
1159 static struct attribute *zram_disk_attrs[] = {
1160         &dev_attr_disksize.attr,
1161         &dev_attr_initstate.attr,
1162         &dev_attr_reset.attr,
1163         &dev_attr_num_reads.attr,
1164         &dev_attr_num_writes.attr,
1165         &dev_attr_failed_reads.attr,
1166         &dev_attr_failed_writes.attr,
1167         &dev_attr_compact.attr,
1168         &dev_attr_invalid_io.attr,
1169         &dev_attr_notify_free.attr,
1170         &dev_attr_zero_pages.attr,
1171         &dev_attr_orig_data_size.attr,
1172         &dev_attr_compr_data_size.attr,
1173         &dev_attr_mem_used_total.attr,
1174         &dev_attr_mem_limit.attr,
1175         &dev_attr_mem_used_max.attr,
1176         &dev_attr_max_comp_streams.attr,
1177         &dev_attr_comp_algorithm.attr,
1178         &dev_attr_io_stat.attr,
1179         &dev_attr_mm_stat.attr,
1180         NULL,
1181 };
1182
1183 static struct attribute_group zram_disk_attr_group = {
1184         .attrs = zram_disk_attrs,
1185 };
1186
1187 static const struct attribute_group *zram_disk_attr_groups[] = {
1188         &zram_disk_attr_group,
1189         NULL,
1190 };
1191
1192 /*
1193  * Allocate and initialize new zram device. the function returns
1194  * '>= 0' device_id upon success, and negative value otherwise.
1195  */
1196 static int zram_add(void)
1197 {
1198         struct zram *zram;
1199         struct request_queue *queue;
1200         int ret, device_id;
1201
1202         zram = kzalloc(sizeof(struct zram), GFP_KERNEL);
1203         if (!zram)
1204                 return -ENOMEM;
1205
1206         ret = idr_alloc(&zram_index_idr, zram, 0, 0, GFP_KERNEL);
1207         if (ret < 0)
1208                 goto out_free_dev;
1209         device_id = ret;
1210
1211         init_rwsem(&zram->init_lock);
1212
1213         queue = blk_alloc_queue(GFP_KERNEL);
1214         if (!queue) {
1215                 pr_err("Error allocating disk queue for device %d\n",
1216                         device_id);
1217                 ret = -ENOMEM;
1218                 goto out_free_idr;
1219         }
1220
1221         blk_queue_make_request(queue, zram_make_request);
1222
1223         /* gendisk structure */
1224         zram->disk = alloc_disk(1);
1225         if (!zram->disk) {
1226                 pr_err("Error allocating disk structure for device %d\n",
1227                         device_id);
1228                 ret = -ENOMEM;
1229                 goto out_free_queue;
1230         }
1231
1232         zram->disk->major = zram_major;
1233         zram->disk->first_minor = device_id;
1234         zram->disk->fops = &zram_devops;
1235         zram->disk->queue = queue;
1236         zram->disk->queue->queuedata = zram;
1237         zram->disk->private_data = zram;
1238         snprintf(zram->disk->disk_name, 16, "zram%d", device_id);
1239
1240         /* Actual capacity set using syfs (/sys/block/zram<id>/disksize */
1241         set_capacity(zram->disk, 0);
1242         /* zram devices sort of resembles non-rotational disks */
1243         queue_flag_set_unlocked(QUEUE_FLAG_NONROT, zram->disk->queue);
1244         queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, zram->disk->queue);
1245         /*
1246          * To ensure that we always get PAGE_SIZE aligned
1247          * and n*PAGE_SIZED sized I/O requests.
1248          */
1249         blk_queue_physical_block_size(zram->disk->queue, PAGE_SIZE);
1250         blk_queue_logical_block_size(zram->disk->queue,
1251                                         ZRAM_LOGICAL_BLOCK_SIZE);
1252         blk_queue_io_min(zram->disk->queue, PAGE_SIZE);
1253         blk_queue_io_opt(zram->disk->queue, PAGE_SIZE);
1254         zram->disk->queue->limits.discard_granularity = PAGE_SIZE;
1255         zram->disk->queue->limits.max_sectors = SECTORS_PER_PAGE;
1256         zram->disk->queue->limits.chunk_sectors = 0;
1257         blk_queue_max_discard_sectors(zram->disk->queue, UINT_MAX);
1258         /*
1259          * zram_bio_discard() will clear all logical blocks if logical block
1260          * size is identical with physical block size(PAGE_SIZE). But if it is
1261          * different, we will skip discarding some parts of logical blocks in
1262          * the part of the request range which isn't aligned to physical block
1263          * size.  So we can't ensure that all discarded logical blocks are
1264          * zeroed.
1265          */
1266         if (ZRAM_LOGICAL_BLOCK_SIZE == PAGE_SIZE)
1267                 zram->disk->queue->limits.discard_zeroes_data = 1;
1268         else
1269                 zram->disk->queue->limits.discard_zeroes_data = 0;
1270         queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, zram->disk->queue);
1271
1272         disk_to_dev(zram->disk)->groups = zram_disk_attr_groups;
1273         add_disk(zram->disk);
1274
1275         strlcpy(zram->compressor, default_compressor, sizeof(zram->compressor));
1276         zram->meta = NULL;
1277         zram->max_comp_streams = 1;
1278
1279         pr_info("Added device: %s\n", zram->disk->disk_name);
1280         return device_id;
1281
1282 out_free_queue:
1283         blk_cleanup_queue(queue);
1284 out_free_idr:
1285         idr_remove(&zram_index_idr, device_id);
1286 out_free_dev:
1287         kfree(zram);
1288         return ret;
1289 }
1290
1291 static int zram_remove(struct zram *zram)
1292 {
1293         struct block_device *bdev;
1294
1295         bdev = bdget_disk(zram->disk, 0);
1296         if (!bdev)
1297                 return -ENOMEM;
1298
1299         mutex_lock(&bdev->bd_mutex);
1300         if (bdev->bd_openers || zram->claim) {
1301                 mutex_unlock(&bdev->bd_mutex);
1302                 bdput(bdev);
1303                 return -EBUSY;
1304         }
1305
1306         zram->claim = true;
1307         mutex_unlock(&bdev->bd_mutex);
1308
1309         /* Make sure all the pending I/O are finished */
1310         fsync_bdev(bdev);
1311         zram_reset_device(zram);
1312         bdput(bdev);
1313
1314         pr_info("Removed device: %s\n", zram->disk->disk_name);
1315
1316         blk_cleanup_queue(zram->disk->queue);
1317         del_gendisk(zram->disk);
1318         put_disk(zram->disk);
1319         kfree(zram);
1320         return 0;
1321 }
1322
1323 /* zram-control sysfs attributes */
1324 static ssize_t hot_add_show(struct class *class,
1325                         struct class_attribute *attr,
1326                         char *buf)
1327 {
1328         int ret;
1329
1330         mutex_lock(&zram_index_mutex);
1331         ret = zram_add();
1332         mutex_unlock(&zram_index_mutex);
1333
1334         if (ret < 0)
1335                 return ret;
1336         return scnprintf(buf, PAGE_SIZE, "%d\n", ret);
1337 }
1338
1339 static ssize_t hot_remove_store(struct class *class,
1340                         struct class_attribute *attr,
1341                         const char *buf,
1342                         size_t count)
1343 {
1344         struct zram *zram;
1345         int ret, dev_id;
1346
1347         /* dev_id is gendisk->first_minor, which is `int' */
1348         ret = kstrtoint(buf, 10, &dev_id);
1349         if (ret)
1350                 return ret;
1351         if (dev_id < 0)
1352                 return -EINVAL;
1353
1354         mutex_lock(&zram_index_mutex);
1355
1356         zram = idr_find(&zram_index_idr, dev_id);
1357         if (zram) {
1358                 ret = zram_remove(zram);
1359                 if (!ret)
1360                         idr_remove(&zram_index_idr, dev_id);
1361         } else {
1362                 ret = -ENODEV;
1363         }
1364
1365         mutex_unlock(&zram_index_mutex);
1366         return ret ? ret : count;
1367 }
1368
1369 /*
1370  * NOTE: hot_add attribute is not the usual read-only sysfs attribute. In a
1371  * sense that reading from this file does alter the state of your system -- it
1372  * creates a new un-initialized zram device and returns back this device's
1373  * device_id (or an error code if it fails to create a new device).
1374  */
1375 static struct class_attribute zram_control_class_attrs[] = {
1376         __ATTR(hot_add, 0400, hot_add_show, NULL),
1377         __ATTR_WO(hot_remove),
1378         __ATTR_NULL,
1379 };
1380
1381 static struct class zram_control_class = {
1382         .name           = "zram-control",
1383         .owner          = THIS_MODULE,
1384         .class_attrs    = zram_control_class_attrs,
1385 };
1386
1387 static int zram_remove_cb(int id, void *ptr, void *data)
1388 {
1389         zram_remove(ptr);
1390         return 0;
1391 }
1392
1393 static void destroy_devices(void)
1394 {
1395         class_unregister(&zram_control_class);
1396         idr_for_each(&zram_index_idr, &zram_remove_cb, NULL);
1397         idr_destroy(&zram_index_idr);
1398         unregister_blkdev(zram_major, "zram");
1399 }
1400
1401 static int __init zram_init(void)
1402 {
1403         int ret;
1404
1405         ret = class_register(&zram_control_class);
1406         if (ret) {
1407                 pr_err("Unable to register zram-control class\n");
1408                 return ret;
1409         }
1410
1411         zram_major = register_blkdev(0, "zram");
1412         if (zram_major <= 0) {
1413                 pr_err("Unable to get major number\n");
1414                 class_unregister(&zram_control_class);
1415                 return -EBUSY;
1416         }
1417
1418         while (num_devices != 0) {
1419                 mutex_lock(&zram_index_mutex);
1420                 ret = zram_add();
1421                 mutex_unlock(&zram_index_mutex);
1422                 if (ret < 0)
1423                         goto out_error;
1424                 num_devices--;
1425         }
1426
1427         return 0;
1428
1429 out_error:
1430         destroy_devices();
1431         return ret;
1432 }
1433
1434 static void __exit zram_exit(void)
1435 {
1436         destroy_devices();
1437 }
1438
1439 module_init(zram_init);
1440 module_exit(zram_exit);
1441
1442 module_param(num_devices, uint, 0);
1443 MODULE_PARM_DESC(num_devices, "Number of pre-created zram devices");
1444
1445 MODULE_LICENSE("Dual BSD/GPL");
1446 MODULE_AUTHOR("Nitin Gupta <ngupta@vflare.org>");
1447 MODULE_DESCRIPTION("Compressed RAM Block Device");