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 #ifdef CONFIG_ZRAM_DEBUG
  19 #define DEBUG
  20 #endif
  21
  22 #include <linux/module.h>
  23 #include <linux/kernel.h>
  24 #include <linux/bio.h>
  25 #include <linux/bitops.h>
  26 #include <linux/blkdev.h>
  27 #include <linux/buffer_head.h>
  28 #include <linux/device.h>
  29 #include <linux/genhd.h>
  30 #include <linux/highmem.h>
  31 #include <linux/slab.h>
  32 #include <linux/string.h>
  33 #include <linux/vmalloc.h>
  34 #include <linux/err.h>
  35
  36 #include "zram_drv.h"
  37
  38 /* Globals */
  39 static int zram_major;
  40 static struct zram *zram_devices;
  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 static ssize_t disksize_show(struct device *dev,
  78                 struct device_attribute *attr, char *buf)
  79 {
  80         struct zram *zram = dev_to_zram(dev);
  81
  82         return scnprintf(buf, PAGE_SIZE, "%llu\n", zram->disksize);
  83 }
  84
  85 static ssize_t initstate_show(struct device *dev,
  86                 struct device_attribute *attr, char *buf)
  87 {
  88         u32 val;
  89         struct zram *zram = dev_to_zram(dev);
  90
  91         down_read(&zram->init_lock);
  92         val = init_done(zram);
  93         up_read(&zram->init_lock);
  94
  95         return scnprintf(buf, PAGE_SIZE, "%u\n", val);
  96 }
  97
  98 static ssize_t orig_data_size_show(struct device *dev,
  99                 struct device_attribute *attr, char *buf)
 100 {
 101         struct zram *zram = dev_to_zram(dev);
 102
 103         deprecated_attr_warn("orig_data_size");
 104         return scnprintf(buf, PAGE_SIZE, "%llu\n",
 105                 (u64)(atomic64_read(&zram->stats.pages_stored)) << PAGE_SHIFT);
 106 }
 107
 108 static ssize_t mem_used_total_show(struct device *dev,
 109                 struct device_attribute *attr, char *buf)
 110 {
 111         u64 val = 0;
 112         struct zram *zram = dev_to_zram(dev);
 113
 114         deprecated_attr_warn("mem_used_total");
 115         down_read(&zram->init_lock);
 116         if (init_done(zram)) {
 117                 struct zram_meta *meta = zram->meta;
 118                 val = zs_get_total_pages(meta->mem_pool);
 119         }
 120         up_read(&zram->init_lock);
 121
 122         return scnprintf(buf, PAGE_SIZE, "%llu\n", val << PAGE_SHIFT);
 123 }
 124
 125 static ssize_t max_comp_streams_show(struct device *dev,
 126                 struct device_attribute *attr, char *buf)
 127 {
 128         int val;
 129         struct zram *zram = dev_to_zram(dev);
 130
 131         down_read(&zram->init_lock);
 132         val = zram->max_comp_streams;
 133         up_read(&zram->init_lock);
 134
 135         return scnprintf(buf, PAGE_SIZE, "%d\n", val);
 136 }
 137
 138 static ssize_t mem_limit_show(struct device *dev,
 139                 struct device_attribute *attr, char *buf)
 140 {
 141         u64 val;
 142         struct zram *zram = dev_to_zram(dev);
 143
 144         deprecated_attr_warn("mem_limit");
 145         down_read(&zram->init_lock);
 146         val = zram->limit_pages;
 147         up_read(&zram->init_lock);
 148
 149         return scnprintf(buf, PAGE_SIZE, "%llu\n", val << PAGE_SHIFT);
 150 }
 151
 152 static ssize_t mem_limit_store(struct device *dev,
 153                 struct device_attribute *attr, const char *buf, size_t len)
 154 {
 155         u64 limit;
 156         char *tmp;
 157         struct zram *zram = dev_to_zram(dev);
 158
 159         limit = memparse(buf, &tmp);
 160         if (buf == tmp) /* no chars parsed, invalid input */
 161                 return -EINVAL;
 162
 163         down_write(&zram->init_lock);
 164         zram->limit_pages = PAGE_ALIGN(limit) >> PAGE_SHIFT;
 165         up_write(&zram->init_lock);
 166
 167         return len;
 168 }
 169
 170 static ssize_t mem_used_max_show(struct device *dev,
 171                 struct device_attribute *attr, char *buf)
 172 {
 173         u64 val = 0;
 174         struct zram *zram = dev_to_zram(dev);
 175
 176         deprecated_attr_warn("mem_used_max");
 177         down_read(&zram->init_lock);
 178         if (init_done(zram))
 179                 val = atomic_long_read(&zram->stats.max_used_pages);
 180         up_read(&zram->init_lock);
 181
 182         return scnprintf(buf, PAGE_SIZE, "%llu\n", val << PAGE_SHIFT);
 183 }
 184
 185 static ssize_t mem_used_max_store(struct device *dev,
 186                 struct device_attribute *attr, const char *buf, size_t len)
 187 {
 188         int err;
 189         unsigned long val;
 190         struct zram *zram = dev_to_zram(dev);
 191
 192         err = kstrtoul(buf, 10, &val);
 193         if (err || val != 0)
 194                 return -EINVAL;
 195
 196         down_read(&zram->init_lock);
 197         if (init_done(zram)) {
 198                 struct zram_meta *meta = zram->meta;
 199                 atomic_long_set(&zram->stats.max_used_pages,
 200                                 zs_get_total_pages(meta->mem_pool));
 201         }
 202         up_read(&zram->init_lock);
 203
 204         return len;
 205 }
 206
 207 static ssize_t max_comp_streams_store(struct device *dev,
 208                 struct device_attribute *attr, const char *buf, size_t len)
 209 {
 210         int num;
 211         struct zram *zram = dev_to_zram(dev);
 212         int ret;
 213
 214         ret = kstrtoint(buf, 0, &num);
 215         if (ret < 0)
 216                 return ret;
 217         if (num < 1)
 218                 return -EINVAL;
 219
 220         down_write(&zram->init_lock);
 221         if (init_done(zram)) {
 222                 if (!zcomp_set_max_streams(zram->comp, num)) {
 223                         pr_info("Cannot change max compression streams\n");
 224                         ret = -EINVAL;
 225                         goto out;
 226                 }
 227         }
 228
 229         zram->max_comp_streams = num;
 230         ret = len;
 231 out:
 232         up_write(&zram->init_lock);
 233         return ret;
 234 }
 235
 236 static ssize_t comp_algorithm_show(struct device *dev,
 237                 struct device_attribute *attr, char *buf)
 238 {
 239         size_t sz;
 240         struct zram *zram = dev_to_zram(dev);
 241
 242         down_read(&zram->init_lock);
 243         sz = zcomp_available_show(zram->compressor, buf);
 244         up_read(&zram->init_lock);
 245
 246         return sz;
 247 }
 248
 249 static ssize_t comp_algorithm_store(struct device *dev,
 250                 struct device_attribute *attr, const char *buf, size_t len)
 251 {
 252         struct zram *zram = dev_to_zram(dev);
 253         down_write(&zram->init_lock);
 254         if (init_done(zram)) {
 255                 up_write(&zram->init_lock);
 256                 pr_info("Can't change algorithm for initialized device\n");
 257                 return -EBUSY;
 258         }
 259         strlcpy(zram->compressor, buf, sizeof(zram->compressor));
 260         up_write(&zram->init_lock);
 261         return len;
 262 }
 263
 264 /* flag operations needs meta->tb_lock */
 265 static int zram_test_flag(struct zram_meta *meta, u32 index,
 266                         enum zram_pageflags flag)
 267 {
 268         return meta->table[index].value & BIT(flag);
 269 }
 270
 271 static void zram_set_flag(struct zram_meta *meta, u32 index,
 272                         enum zram_pageflags flag)
 273 {
 274         meta->table[index].value |= BIT(flag);
 275 }
 276
 277 static void zram_clear_flag(struct zram_meta *meta, u32 index,
 278                         enum zram_pageflags flag)
 279 {
 280         meta->table[index].value &= ~BIT(flag);
 281 }
 282
 283 static size_t zram_get_obj_size(struct zram_meta *meta, u32 index)
 284 {
 285         return meta->table[index].value & (BIT(ZRAM_FLAG_SHIFT) - 1);
 286 }
 287
 288 static void zram_set_obj_size(struct zram_meta *meta,
 289                                         u32 index, size_t size)
 290 {
 291         unsigned long flags = meta->table[index].value >> ZRAM_FLAG_SHIFT;
 292
 293         meta->table[index].value = (flags << ZRAM_FLAG_SHIFT) | size;
 294 }
 295
 296 static inline int is_partial_io(struct bio_vec *bvec)
 297 {
 298         return bvec->bv_len != PAGE_SIZE;
 299 }
 300
 301 /*
 302  * Check if request is within bounds and aligned on zram logical blocks.
 303  */
 304 static inline int valid_io_request(struct zram *zram,
 305                 sector_t start, unsigned int size)
 306 {
 307         u64 end, bound;
 308
 309         /* unaligned request */
 310         if (unlikely(start & (ZRAM_SECTOR_PER_LOGICAL_BLOCK - 1)))
 311                 return 0;
 312         if (unlikely(size & (ZRAM_LOGICAL_BLOCK_SIZE - 1)))
 313                 return 0;
 314
 315         end = start + (size >> SECTOR_SHIFT);
 316         bound = zram->disksize >> SECTOR_SHIFT;
 317         /* out of range range */
 318         if (unlikely(start >= bound || end > bound || start > end))
 319                 return 0;
 320
 321         /* I/O request is valid */
 322         return 1;
 323 }
 324
 325 static void zram_meta_free(struct zram_meta *meta, u64 disksize)
 326 {
 327         size_t num_pages = disksize >> PAGE_SHIFT;
 328         size_t index;
 329
 330         /* Free all pages that are still in this zram device */
 331         for (index = 0; index < num_pages; index++) {
 332                 unsigned long handle = meta->table[index].handle;
 333
 334                 if (!handle)
 335                         continue;
 336
 337                 zs_free(meta->mem_pool, handle);
 338         }
 339
 340         zs_destroy_pool(meta->mem_pool);
 341         vfree(meta->table);
 342         kfree(meta);
 343 }
 344
 345 static struct zram_meta *zram_meta_alloc(int device_id, u64 disksize)
 346 {
 347         size_t num_pages;
 348         char pool_name[8];
 349         struct zram_meta *meta = kmalloc(sizeof(*meta), GFP_KERNEL);
 350
 351         if (!meta)
 352                 return NULL;
 353
 354         num_pages = disksize >> PAGE_SHIFT;
 355         meta->table = vzalloc(num_pages * sizeof(*meta->table));
 356         if (!meta->table) {
 357                 pr_err("Error allocating zram address table\n");
 358                 goto out_error;
 359         }
 360
 361         snprintf(pool_name, sizeof(pool_name), "zram%d", device_id);
 362         meta->mem_pool = zs_create_pool(pool_name, GFP_NOIO | __GFP_HIGHMEM);
 363         if (!meta->mem_pool) {
 364                 pr_err("Error creating memory pool\n");
 365                 goto out_error;
 366         }
 367
 368         return meta;
 369
 370 out_error:
 371         vfree(meta->table);
 372         kfree(meta);
 373         return NULL;
 374 }
 375
 376 static inline bool zram_meta_get(struct zram *zram)
 377 {
 378         if (atomic_inc_not_zero(&zram->refcount))
 379                 return true;
 380         return false;
 381 }
 382
 383 static inline void zram_meta_put(struct zram *zram)
 384 {
 385         atomic_dec(&zram->refcount);
 386 }
 387
 388 static void update_position(u32 *index, int *offset, struct bio_vec *bvec)
 389 {
 390         if (*offset + bvec->bv_len >= PAGE_SIZE)
 391                 (*index)++;
 392         *offset = (*offset + bvec->bv_len) % PAGE_SIZE;
 393 }
 394
 395 static int page_zero_filled(void *ptr)
 396 {
 397         unsigned int pos;
 398         unsigned long *page;
 399
 400         page = (unsigned long *)ptr;
 401
 402         for (pos = 0; pos != PAGE_SIZE / sizeof(*page); pos++) {
 403                 if (page[pos])
 404                         return 0;
 405         }
 406
 407         return 1;
 408 }
 409
 410 static void handle_zero_page(struct bio_vec *bvec)
 411 {
 412         struct page *page = bvec->bv_page;
 413         void *user_mem;
 414
 415         user_mem = kmap_atomic(page);
 416         if (is_partial_io(bvec))
 417                 memset(user_mem + bvec->bv_offset, 0, bvec->bv_len);
 418         else
 419                 clear_page(user_mem);
 420         kunmap_atomic(user_mem);
 421
 422         flush_dcache_page(page);
 423 }
 424
 425
 426 /*
 427  * To protect concurrent access to the same index entry,
 428  * caller should hold this table index entry's bit_spinlock to
 429  * indicate this index entry is accessing.
 430  */
 431 static void zram_free_page(struct zram *zram, size_t index)
 432 {
 433         struct zram_meta *meta = zram->meta;
 434         unsigned long handle = meta->table[index].handle;
 435
 436         if (unlikely(!handle)) {
 437                 /*
 438                  * No memory is allocated for zero filled pages.
 439                  * Simply clear zero page flag.
 440                  */
 441                 if (zram_test_flag(meta, index, ZRAM_ZERO)) {
 442                         zram_clear_flag(meta, index, ZRAM_ZERO);
 443                         atomic64_dec(&zram->stats.zero_pages);
 444                 }
 445                 return;
 446         }
 447
 448         zs_free(meta->mem_pool, handle);
 449
 450         atomic64_sub(zram_get_obj_size(meta, index),
 451                         &zram->stats.compr_data_size);
 452         atomic64_dec(&zram->stats.pages_stored);
 453
 454         meta->table[index].handle = 0;
 455         zram_set_obj_size(meta, index, 0);
 456 }
 457
 458 static int zram_decompress_page(struct zram *zram, char *mem, u32 index)
 459 {
 460         int ret = 0;
 461         unsigned char *cmem;
 462         struct zram_meta *meta = zram->meta;
 463         unsigned long handle;
 464         size_t size;
 465
 466         bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
 467         handle = meta->table[index].handle;
 468         size = zram_get_obj_size(meta, index);
 469
 470         if (!handle || zram_test_flag(meta, index, ZRAM_ZERO)) {
 471                 bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 472                 clear_page(mem);
 473                 return 0;
 474         }
 475
 476         cmem = zs_map_object(meta->mem_pool, handle, ZS_MM_RO);
 477         if (size == PAGE_SIZE)
 478                 copy_page(mem, cmem);
 479         else
 480                 ret = zcomp_decompress(zram->comp, cmem, size, mem);
 481         zs_unmap_object(meta->mem_pool, handle);
 482         bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 483
 484         /* Should NEVER happen. Return bio error if it does. */
 485         if (unlikely(ret)) {
 486                 pr_err("Decompression failed! err=%d, page=%u\n", ret, index);
 487                 return ret;
 488         }
 489
 490         return 0;
 491 }
 492
 493 static int zram_bvec_read(struct zram *zram, struct bio_vec *bvec,
 494                           u32 index, int offset)
 495 {
 496         int ret;
 497         struct page *page;
 498         unsigned char *user_mem, *uncmem = NULL;
 499         struct zram_meta *meta = zram->meta;
 500         page = bvec->bv_page;
 501
 502         bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
 503         if (unlikely(!meta->table[index].handle) ||
 504                         zram_test_flag(meta, index, ZRAM_ZERO)) {
 505                 bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 506                 handle_zero_page(bvec);
 507                 return 0;
 508         }
 509         bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 510
 511         if (is_partial_io(bvec))
 512                 /* Use  a temporary buffer to decompress the page */
 513                 uncmem = kmalloc(PAGE_SIZE, GFP_NOIO);
 514
 515         user_mem = kmap_atomic(page);
 516         if (!is_partial_io(bvec))
 517                 uncmem = user_mem;
 518
 519         if (!uncmem) {
 520                 pr_info("Unable to allocate temp memory\n");
 521                 ret = -ENOMEM;
 522                 goto out_cleanup;
 523         }
 524
 525         ret = zram_decompress_page(zram, uncmem, index);
 526         /* Should NEVER happen. Return bio error if it does. */
 527         if (unlikely(ret))
 528                 goto out_cleanup;
 529
 530         if (is_partial_io(bvec))
 531                 memcpy(user_mem + bvec->bv_offset, uncmem + offset,
 532                                 bvec->bv_len);
 533
 534         flush_dcache_page(page);
 535         ret = 0;
 536 out_cleanup:
 537         kunmap_atomic(user_mem);
 538         if (is_partial_io(bvec))
 539                 kfree(uncmem);
 540         return ret;
 541 }
 542
 543 static inline void update_used_max(struct zram *zram,
 544                                         const unsigned long pages)
 545 {
 546         unsigned long old_max, cur_max;
 547
 548         old_max = atomic_long_read(&zram->stats.max_used_pages);
 549
 550         do {
 551                 cur_max = old_max;
 552                 if (pages > cur_max)
 553                         old_max = atomic_long_cmpxchg(
 554                                 &zram->stats.max_used_pages, cur_max, pages);
 555         } while (old_max != cur_max);
 556 }
 557
 558 static int zram_bvec_write(struct zram *zram, struct bio_vec *bvec, u32 index,
 559                            int offset)
 560 {
 561         int ret = 0;
 562         size_t clen;
 563         unsigned long handle;
 564         struct page *page;
 565         unsigned char *user_mem, *cmem, *src, *uncmem = NULL;
 566         struct zram_meta *meta = zram->meta;
 567         struct zcomp_strm *zstrm;
 568         bool locked = false;
 569         unsigned long alloced_pages;
 570
 571         page = bvec->bv_page;
 572         if (is_partial_io(bvec)) {
 573                 /*
 574                  * This is a partial IO. We need to read the full page
 575                  * before to write the changes.
 576                  */
 577                 uncmem = kmalloc(PAGE_SIZE, GFP_NOIO);
 578                 if (!uncmem) {
 579                         ret = -ENOMEM;
 580                         goto out;
 581                 }
 582                 ret = zram_decompress_page(zram, uncmem, index);
 583                 if (ret)
 584                         goto out;
 585         }
 586
 587         zstrm = zcomp_strm_find(zram->comp);
 588         locked = true;
 589         user_mem = kmap_atomic(page);
 590
 591         if (is_partial_io(bvec)) {
 592                 memcpy(uncmem + offset, user_mem + bvec->bv_offset,
 593                        bvec->bv_len);
 594                 kunmap_atomic(user_mem);
 595                 user_mem = NULL;
 596         } else {
 597                 uncmem = user_mem;
 598         }
 599
 600         if (page_zero_filled(uncmem)) {
 601                 if (user_mem)
 602                         kunmap_atomic(user_mem);
 603                 /* Free memory associated with this sector now. */
 604                 bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
 605                 zram_free_page(zram, index);
 606                 zram_set_flag(meta, index, ZRAM_ZERO);
 607                 bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 608
 609                 atomic64_inc(&zram->stats.zero_pages);
 610                 ret = 0;
 611                 goto out;
 612         }
 613
 614         ret = zcomp_compress(zram->comp, zstrm, uncmem, &clen);
 615         if (!is_partial_io(bvec)) {
 616                 kunmap_atomic(user_mem);
 617                 user_mem = NULL;
 618                 uncmem = NULL;
 619         }
 620
 621         if (unlikely(ret)) {
 622                 pr_err("Compression failed! err=%d\n", ret);
 623                 goto out;
 624         }
 625         src = zstrm->buffer;
 626         if (unlikely(clen > max_zpage_size)) {
 627                 clen = PAGE_SIZE;
 628                 if (is_partial_io(bvec))
 629                         src = uncmem;
 630         }
 631
 632         handle = zs_malloc(meta->mem_pool, clen);
 633         if (!handle) {
 634                 pr_info("Error allocating memory for compressed page: %u, size=%zu\n",
 635                         index, clen);
 636                 ret = -ENOMEM;
 637                 goto out;
 638         }
 639
 640         alloced_pages = zs_get_total_pages(meta->mem_pool);
 641         if (zram->limit_pages && alloced_pages > zram->limit_pages) {
 642                 zs_free(meta->mem_pool, handle);
 643                 ret = -ENOMEM;
 644                 goto out;
 645         }
 646
 647         update_used_max(zram, alloced_pages);
 648
 649         cmem = zs_map_object(meta->mem_pool, handle, ZS_MM_WO);
 650
 651         if ((clen == PAGE_SIZE) && !is_partial_io(bvec)) {
 652                 src = kmap_atomic(page);
 653                 copy_page(cmem, src);
 654                 kunmap_atomic(src);
 655         } else {
 656                 memcpy(cmem, src, clen);
 657         }
 658
 659         zcomp_strm_release(zram->comp, zstrm);
 660         locked = false;
 661         zs_unmap_object(meta->mem_pool, handle);
 662
 663         /*
 664          * Free memory associated with this sector
 665          * before overwriting unused sectors.
 666          */
 667         bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
 668         zram_free_page(zram, index);
 669
 670         meta->table[index].handle = handle;
 671         zram_set_obj_size(meta, index, clen);
 672         bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 673
 674         /* Update stats */
 675         atomic64_add(clen, &zram->stats.compr_data_size);
 676         atomic64_inc(&zram->stats.pages_stored);
 677 out:
 678         if (locked)
 679                 zcomp_strm_release(zram->comp, zstrm);
 680         if (is_partial_io(bvec))
 681                 kfree(uncmem);
 682         return ret;
 683 }
 684
 685 static int zram_bvec_rw(struct zram *zram, struct bio_vec *bvec, u32 index,
 686                         int offset, int rw)
 687 {
 688         unsigned long start_time = jiffies;
 689         int ret;
 690
 691         generic_start_io_acct(rw, bvec->bv_len >> SECTOR_SHIFT,
 692                         &zram->disk->part0);
 693
 694         if (rw == READ) {
 695                 atomic64_inc(&zram->stats.num_reads);
 696                 ret = zram_bvec_read(zram, bvec, index, offset);
 697         } else {
 698                 atomic64_inc(&zram->stats.num_writes);
 699                 ret = zram_bvec_write(zram, bvec, index, offset);
 700         }
 701
 702         generic_end_io_acct(rw, &zram->disk->part0, start_time);
 703
 704         if (unlikely(ret)) {
 705                 if (rw == READ)
 706                         atomic64_inc(&zram->stats.failed_reads);
 707                 else
 708                         atomic64_inc(&zram->stats.failed_writes);
 709         }
 710
 711         return ret;
 712 }
 713
 714 /*
 715  * zram_bio_discard - handler on discard request
 716  * @index: physical block index in PAGE_SIZE units
 717  * @offset: byte offset within physical block
 718  */
 719 static void zram_bio_discard(struct zram *zram, u32 index,
 720                              int offset, struct bio *bio)
 721 {
 722         size_t n = bio->bi_iter.bi_size;
 723         struct zram_meta *meta = zram->meta;
 724
 725         /*
 726          * zram manages data in physical block size units. Because logical block
 727          * size isn't identical with physical block size on some arch, we
 728          * could get a discard request pointing to a specific offset within a
 729          * certain physical block.  Although we can handle this request by
 730          * reading that physiclal block and decompressing and partially zeroing
 731          * and re-compressing and then re-storing it, this isn't reasonable
 732          * because our intent with a discard request is to save memory.  So
 733          * skipping this logical block is appropriate here.
 734          */
 735         if (offset) {
 736                 if (n <= (PAGE_SIZE - offset))
 737                         return;
 738
 739                 n -= (PAGE_SIZE - offset);
 740                 index++;
 741         }
 742
 743         while (n >= PAGE_SIZE) {
 744                 bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
 745                 zram_free_page(zram, index);
 746                 bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 747                 atomic64_inc(&zram->stats.notify_free);
 748                 index++;
 749                 n -= PAGE_SIZE;
 750         }
 751 }
 752
 753 static void zram_reset_device(struct zram *zram)
 754 {
 755         struct zram_meta *meta;
 756         struct zcomp *comp;
 757         u64 disksize;
 758
 759         down_write(&zram->init_lock);
 760
 761         zram->limit_pages = 0;
 762
 763         if (!init_done(zram)) {
 764                 up_write(&zram->init_lock);
 765                 return;
 766         }
 767
 768         meta = zram->meta;
 769         comp = zram->comp;
 770         disksize = zram->disksize;
 771         /*
 772          * Refcount will go down to 0 eventually and r/w handler
 773          * cannot handle further I/O so it will bail out by
 774          * check zram_meta_get.
 775          */
 776         zram_meta_put(zram);
 777         /*
 778          * We want to free zram_meta in process context to avoid
 779          * deadlock between reclaim path and any other locks.
 780          */
 781         wait_event(zram->io_done, atomic_read(&zram->refcount) == 0);
 782
 783         /* Reset stats */
 784         memset(&zram->stats, 0, sizeof(zram->stats));
 785         zram->disksize = 0;
 786         zram->max_comp_streams = 1;
 787         set_capacity(zram->disk, 0);
 788
 789         up_write(&zram->init_lock);
 790         /* I/O operation under all of CPU are done so let's free */
 791         zram_meta_free(meta, disksize);
 792         zcomp_destroy(comp);
 793 }
 794
 795 static ssize_t disksize_store(struct device *dev,
 796                 struct device_attribute *attr, const char *buf, size_t len)
 797 {
 798         u64 disksize;
 799         struct zcomp *comp;
 800         struct zram_meta *meta;
 801         struct zram *zram = dev_to_zram(dev);
 802         int err;
 803
 804         disksize = memparse(buf, NULL);
 805         if (!disksize)
 806                 return -EINVAL;
 807
 808         disksize = PAGE_ALIGN(disksize);
 809         meta = zram_meta_alloc(zram->disk->first_minor, disksize);
 810         if (!meta)
 811                 return -ENOMEM;
 812
 813         comp = zcomp_create(zram->compressor, zram->max_comp_streams);
 814         if (IS_ERR(comp)) {
 815                 pr_info("Cannot initialise %s compressing backend\n",
 816                                 zram->compressor);
 817                 err = PTR_ERR(comp);
 818                 goto out_free_meta;
 819         }
 820
 821         down_write(&zram->init_lock);
 822         if (init_done(zram)) {
 823                 pr_info("Cannot change disksize for initialized device\n");
 824                 err = -EBUSY;
 825                 goto out_destroy_comp;
 826         }
 827
 828         init_waitqueue_head(&zram->io_done);
 829         atomic_set(&zram->refcount, 1);
 830         zram->meta = meta;
 831         zram->comp = comp;
 832         zram->disksize = disksize;
 833         set_capacity(zram->disk, zram->disksize >> SECTOR_SHIFT);
 834         up_write(&zram->init_lock);
 835
 836         /*
 837          * Revalidate disk out of the init_lock to avoid lockdep splat.
 838          * It's okay because disk's capacity is protected by init_lock
 839          * so that revalidate_disk always sees up-to-date capacity.
 840          */
 841         revalidate_disk(zram->disk);
 842
 843         return len;
 844
 845 out_destroy_comp:
 846         up_write(&zram->init_lock);
 847         zcomp_destroy(comp);
 848 out_free_meta:
 849         zram_meta_free(meta, disksize);
 850         return err;
 851 }
 852
 853 static ssize_t reset_store(struct device *dev,
 854                 struct device_attribute *attr, const char *buf, size_t len)
 855 {
 856         int ret;
 857         unsigned short do_reset;
 858         struct zram *zram;
 859         struct block_device *bdev;
 860
 861         zram = dev_to_zram(dev);
 862         bdev = bdget_disk(zram->disk, 0);
 863
 864         if (!bdev)
 865                 return -ENOMEM;
 866
 867         mutex_lock(&bdev->bd_mutex);
 868         /* Do not reset an active device! */
 869         if (bdev->bd_openers) {
 870                 ret = -EBUSY;
 871                 goto out;
 872         }
 873
 874         ret = kstrtou16(buf, 10, &do_reset);
 875         if (ret)
 876                 goto out;
 877
 878         if (!do_reset) {
 879                 ret = -EINVAL;
 880                 goto out;
 881         }
 882
 883         /* Make sure all pending I/O is finished */
 884         fsync_bdev(bdev);
 885         zram_reset_device(zram);
 886
 887         mutex_unlock(&bdev->bd_mutex);
 888         revalidate_disk(zram->disk);
 889         bdput(bdev);
 890
 891         return len;
 892
 893 out:
 894         mutex_unlock(&bdev->bd_mutex);
 895         bdput(bdev);
 896         return ret;
 897 }
 898
 899 static void __zram_make_request(struct zram *zram, struct bio *bio)
 900 {
 901         int offset, rw;
 902         u32 index;
 903         struct bio_vec bvec;
 904         struct bvec_iter iter;
 905
 906         index = bio->bi_iter.bi_sector >> SECTORS_PER_PAGE_SHIFT;
 907         offset = (bio->bi_iter.bi_sector &
 908                   (SECTORS_PER_PAGE - 1)) << SECTOR_SHIFT;
 909
 910         if (unlikely(bio->bi_rw & REQ_DISCARD)) {
 911                 zram_bio_discard(zram, index, offset, bio);
 912                 bio_endio(bio, 0);
 913                 return;
 914         }
 915
 916         rw = bio_data_dir(bio);
 917         bio_for_each_segment(bvec, bio, iter) {
 918                 int max_transfer_size = PAGE_SIZE - offset;
 919
 920                 if (bvec.bv_len > max_transfer_size) {
 921                         /*
 922                          * zram_bvec_rw() can only make operation on a single
 923                          * zram page. Split the bio vector.
 924                          */
 925                         struct bio_vec bv;
 926
 927                         bv.bv_page = bvec.bv_page;
 928                         bv.bv_len = max_transfer_size;
 929                         bv.bv_offset = bvec.bv_offset;
 930
 931                         if (zram_bvec_rw(zram, &bv, index, offset, rw) < 0)
 932                                 goto out;
 933
 934                         bv.bv_len = bvec.bv_len - max_transfer_size;
 935                         bv.bv_offset += max_transfer_size;
 936                         if (zram_bvec_rw(zram, &bv, index + 1, 0, rw) < 0)
 937                                 goto out;
 938                 } else
 939                         if (zram_bvec_rw(zram, &bvec, index, offset, rw) < 0)
 940                                 goto out;
 941
 942                 update_position(&index, &offset, &bvec);
 943         }
 944
 945         set_bit(BIO_UPTODATE, &bio->bi_flags);
 946         bio_endio(bio, 0);
 947         return;
 948
 949 out:
 950         bio_io_error(bio);
 951 }
 952
 953 /*
 954  * Handler function for all zram I/O requests.
 955  */
 956 static void zram_make_request(struct request_queue *queue, struct bio *bio)
 957 {
 958         struct zram *zram = queue->queuedata;
 959
 960         if (unlikely(!zram_meta_get(zram)))
 961                 goto error;
 962
 963         if (!valid_io_request(zram, bio->bi_iter.bi_sector,
 964                                         bio->bi_iter.bi_size)) {
 965                 atomic64_inc(&zram->stats.invalid_io);
 966                 goto put_zram;
 967         }
 968
 969         __zram_make_request(zram, bio);
 970         zram_meta_put(zram);
 971         return;
 972 put_zram:
 973         zram_meta_put(zram);
 974 error:
 975         bio_io_error(bio);
 976 }
 977
 978 static void zram_slot_free_notify(struct block_device *bdev,
 979                                 unsigned long index)
 980 {
 981         struct zram *zram;
 982         struct zram_meta *meta;
 983
 984         zram = bdev->bd_disk->private_data;
 985         meta = zram->meta;
 986
 987         bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
 988         zram_free_page(zram, index);
 989         bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 990         atomic64_inc(&zram->stats.notify_free);
 991 }
 992
 993 static int zram_rw_page(struct block_device *bdev, sector_t sector,
 994                        struct page *page, int rw)
 995 {
 996         int offset, err = -EIO;
 997         u32 index;
 998         struct zram *zram;
 999         struct bio_vec bv;
1000
1001         zram = bdev->bd_disk->private_data;
1002         if (unlikely(!zram_meta_get(zram)))
1003                 goto out;
1004
1005         if (!valid_io_request(zram, sector, PAGE_SIZE)) {
1006                 atomic64_inc(&zram->stats.invalid_io);
1007                 err = -EINVAL;
1008                 goto put_zram;
1009         }
1010
1011         index = sector >> SECTORS_PER_PAGE_SHIFT;
1012         offset = sector & (SECTORS_PER_PAGE - 1) << SECTOR_SHIFT;
1013
1014         bv.bv_page = page;
1015         bv.bv_len = PAGE_SIZE;
1016         bv.bv_offset = 0;
1017
1018         err = zram_bvec_rw(zram, &bv, index, offset, rw);
1019 put_zram:
1020         zram_meta_put(zram);
1021 out:
1022         /*
1023          * If I/O fails, just return error(ie, non-zero) without
1024          * calling page_endio.
1025          * It causes resubmit the I/O with bio request by upper functions
1026          * of rw_page(e.g., swap_readpage, __swap_writepage) and
1027          * bio->bi_end_io does things to handle the error
1028          * (e.g., SetPageError, set_page_dirty and extra works).
1029          */
1030         if (err == 0)
1031                 page_endio(page, rw, 0);
1032         return err;
1033 }
1034
1035 static const struct block_device_operations zram_devops = {
1036         .swap_slot_free_notify = zram_slot_free_notify,
1037         .rw_page = zram_rw_page,
1038         .owner = THIS_MODULE
1039 };
1040
1041 static DEVICE_ATTR_RW(disksize);
1042 static DEVICE_ATTR_RO(initstate);
1043 static DEVICE_ATTR_WO(reset);
1044 static DEVICE_ATTR_RO(orig_data_size);
1045 static DEVICE_ATTR_RO(mem_used_total);
1046 static DEVICE_ATTR_RW(mem_limit);
1047 static DEVICE_ATTR_RW(mem_used_max);
1048 static DEVICE_ATTR_RW(max_comp_streams);
1049 static DEVICE_ATTR_RW(comp_algorithm);
1050
1051 static ssize_t io_stat_show(struct device *dev,
1052                 struct device_attribute *attr, char *buf)
1053 {
1054         struct zram *zram = dev_to_zram(dev);
1055         ssize_t ret;
1056
1057         down_read(&zram->init_lock);
1058         ret = scnprintf(buf, PAGE_SIZE,
1059                         "%8llu %8llu %8llu %8llu\n",
1060                         (u64)atomic64_read(&zram->stats.failed_reads),
1061                         (u64)atomic64_read(&zram->stats.failed_writes),
1062                         (u64)atomic64_read(&zram->stats.invalid_io),
1063                         (u64)atomic64_read(&zram->stats.notify_free));
1064         up_read(&zram->init_lock);
1065
1066         return ret;
1067 }
1068
1069 static ssize_t mm_stat_show(struct device *dev,
1070                 struct device_attribute *attr, char *buf)
1071 {
1072         struct zram *zram = dev_to_zram(dev);
1073         u64 orig_size, mem_used = 0;
1074         long max_used;
1075         ssize_t ret;
1076
1077         down_read(&zram->init_lock);
1078         if (init_done(zram))
1079                 mem_used = zs_get_total_pages(zram->meta->mem_pool);
1080
1081         orig_size = atomic64_read(&zram->stats.pages_stored);
1082         max_used = atomic_long_read(&zram->stats.max_used_pages);
1083
1084         ret = scnprintf(buf, PAGE_SIZE,
1085                         "%8llu %8llu %8llu %8lu %8ld %8llu %8llu\n",
1086                         orig_size << PAGE_SHIFT,
1087                         (u64)atomic64_read(&zram->stats.compr_data_size),
1088                         mem_used << PAGE_SHIFT,
1089                         zram->limit_pages << PAGE_SHIFT,
1090                         max_used << PAGE_SHIFT,
1091                         (u64)atomic64_read(&zram->stats.zero_pages),
1092                         (u64)atomic64_read(&zram->stats.num_migrated));
1093         up_read(&zram->init_lock);
1094
1095         return ret;
1096 }
1097
1098 static DEVICE_ATTR_RO(io_stat);
1099 static DEVICE_ATTR_RO(mm_stat);
1100 ZRAM_ATTR_RO(num_reads);
1101 ZRAM_ATTR_RO(num_writes);
1102 ZRAM_ATTR_RO(failed_reads);
1103 ZRAM_ATTR_RO(failed_writes);
1104 ZRAM_ATTR_RO(invalid_io);
1105 ZRAM_ATTR_RO(notify_free);
1106 ZRAM_ATTR_RO(zero_pages);
1107 ZRAM_ATTR_RO(compr_data_size);
1108
1109 static struct attribute *zram_disk_attrs[] = {
1110         &dev_attr_disksize.attr,
1111         &dev_attr_initstate.attr,
1112         &dev_attr_reset.attr,
1113         &dev_attr_num_reads.attr,
1114         &dev_attr_num_writes.attr,
1115         &dev_attr_failed_reads.attr,
1116         &dev_attr_failed_writes.attr,
1117         &dev_attr_invalid_io.attr,
1118         &dev_attr_notify_free.attr,
1119         &dev_attr_zero_pages.attr,
1120         &dev_attr_orig_data_size.attr,
1121         &dev_attr_compr_data_size.attr,
1122         &dev_attr_mem_used_total.attr,
1123         &dev_attr_mem_limit.attr,
1124         &dev_attr_mem_used_max.attr,
1125         &dev_attr_max_comp_streams.attr,
1126         &dev_attr_comp_algorithm.attr,
1127         &dev_attr_io_stat.attr,
1128         &dev_attr_mm_stat.attr,
1129         NULL,
1130 };
1131
1132 static struct attribute_group zram_disk_attr_group = {
1133         .attrs = zram_disk_attrs,
1134 };
1135
1136 static int create_device(struct zram *zram, int device_id)
1137 {
1138         struct request_queue *queue;
1139         int ret = -ENOMEM;
1140
1141         init_rwsem(&zram->init_lock);
1142
1143         queue = blk_alloc_queue(GFP_KERNEL);
1144         if (!queue) {
1145                 pr_err("Error allocating disk queue for device %d\n",
1146                         device_id);
1147                 goto out;
1148         }
1149
1150         blk_queue_make_request(queue, zram_make_request);
1151
1152          /* gendisk structure */
1153         zram->disk = alloc_disk(1);
1154         if (!zram->disk) {
1155                 pr_warn("Error allocating disk structure for device %d\n",
1156                         device_id);
1157                 ret = -ENOMEM;
1158                 goto out_free_queue;
1159         }
1160
1161         zram->disk->major = zram_major;
1162         zram->disk->first_minor = device_id;
1163         zram->disk->fops = &zram_devops;
1164         zram->disk->queue = queue;
1165         zram->disk->queue->queuedata = zram;
1166         zram->disk->private_data = zram;
1167         snprintf(zram->disk->disk_name, 16, "zram%d", device_id);
1168
1169         /* Actual capacity set using syfs (/sys/block/zram<id>/disksize */
1170         set_capacity(zram->disk, 0);
1171         /* zram devices sort of resembles non-rotational disks */
1172         queue_flag_set_unlocked(QUEUE_FLAG_NONROT, zram->disk->queue);
1173         queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, zram->disk->queue);
1174         /*
1175          * To ensure that we always get PAGE_SIZE aligned
1176          * and n*PAGE_SIZED sized I/O requests.
1177          */
1178         blk_queue_physical_block_size(zram->disk->queue, PAGE_SIZE);
1179         blk_queue_logical_block_size(zram->disk->queue,
1180                                         ZRAM_LOGICAL_BLOCK_SIZE);
1181         blk_queue_io_min(zram->disk->queue, PAGE_SIZE);
1182         blk_queue_io_opt(zram->disk->queue, PAGE_SIZE);
1183         zram->disk->queue->limits.discard_granularity = PAGE_SIZE;
1184         zram->disk->queue->limits.max_discard_sectors = UINT_MAX;
1185         /*
1186          * zram_bio_discard() will clear all logical blocks if logical block
1187          * size is identical with physical block size(PAGE_SIZE). But if it is
1188          * different, we will skip discarding some parts of logical blocks in
1189          * the part of the request range which isn't aligned to physical block
1190          * size.  So we can't ensure that all discarded logical blocks are
1191          * zeroed.
1192          */
1193         if (ZRAM_LOGICAL_BLOCK_SIZE == PAGE_SIZE)
1194                 zram->disk->queue->limits.discard_zeroes_data = 1;
1195         else
1196                 zram->disk->queue->limits.discard_zeroes_data = 0;
1197         queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, zram->disk->queue);
1198
1199         add_disk(zram->disk);
1200
1201         ret = sysfs_create_group(&disk_to_dev(zram->disk)->kobj,
1202                                 &zram_disk_attr_group);
1203         if (ret < 0) {
1204                 pr_warn("Error creating sysfs group");
1205                 goto out_free_disk;
1206         }
1207         strlcpy(zram->compressor, default_compressor, sizeof(zram->compressor));
1208         zram->meta = NULL;
1209         zram->max_comp_streams = 1;
1210         return 0;
1211
1212 out_free_disk:
1213         del_gendisk(zram->disk);
1214         put_disk(zram->disk);
1215 out_free_queue:
1216         blk_cleanup_queue(queue);
1217 out:
1218         return ret;
1219 }
1220
1221 static void destroy_devices(unsigned int nr)
1222 {
1223         struct zram *zram;
1224         unsigned int i;
1225
1226         for (i = 0; i < nr; i++) {
1227                 zram = &zram_devices[i];
1228                 /*
1229                  * Remove sysfs first, so no one will perform a disksize
1230                  * store while we destroy the devices
1231                  */
1232                 sysfs_remove_group(&disk_to_dev(zram->disk)->kobj,
1233                                 &zram_disk_attr_group);
1234
1235                 zram_reset_device(zram);
1236
1237                 blk_cleanup_queue(zram->disk->queue);
1238                 del_gendisk(zram->disk);
1239                 put_disk(zram->disk);
1240         }
1241
1242         kfree(zram_devices);
1243         unregister_blkdev(zram_major, "zram");
1244         pr_info("Destroyed %u device(s)\n", nr);
1245 }
1246
1247 static int __init zram_init(void)
1248 {
1249         int ret, dev_id;
1250
1251         if (num_devices > max_num_devices) {
1252                 pr_warn("Invalid value for num_devices: %u\n",
1253                                 num_devices);
1254                 return -EINVAL;
1255         }
1256
1257         zram_major = register_blkdev(0, "zram");
1258         if (zram_major <= 0) {
1259                 pr_warn("Unable to get major number\n");
1260                 return -EBUSY;
1261         }
1262
1263         /* Allocate the device array and initialize each one */
1264         zram_devices = kzalloc(num_devices * sizeof(struct zram), GFP_KERNEL);
1265         if (!zram_devices) {
1266                 unregister_blkdev(zram_major, "zram");
1267                 return -ENOMEM;
1268         }
1269
1270         for (dev_id = 0; dev_id < num_devices; dev_id++) {
1271                 ret = create_device(&zram_devices[dev_id], dev_id);
1272                 if (ret)
1273                         goto out_error;
1274         }
1275
1276         pr_info("Created %u device(s)\n", num_devices);
1277         return 0;
1278
1279 out_error:
1280         destroy_devices(dev_id);
1281         return ret;
1282 }
1283
1284 static void __exit zram_exit(void)
1285 {
1286         destroy_devices(num_devices);
1287 }
1288
1289 module_init(zram_init);
1290 module_exit(zram_exit);
1291
1292 module_param(num_devices, uint, 0);
1293 MODULE_PARM_DESC(num_devices, "Number of zram devices");
1294
1295 MODULE_LICENSE("Dual BSD/GPL");
1296 MODULE_AUTHOR("Nitin Gupta <ngupta@vflare.org>");
1297 MODULE_DESCRIPTION("Compressed RAM Block Device");