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 #define ZRAM_ATTR_RO(name)                                              \
  47 static ssize_t name##_show(struct device *d,            \
  48                                 struct device_attribute *attr, char *b) \
  49 {                                                                       \
  50         struct zram *zram = dev_to_zram(d);                             \
  51         return scnprintf(b, PAGE_SIZE, "%llu\n",                        \
  52                 (u64)atomic64_read(&zram->stats.name));                 \
  53 }                                                                       \
  54 static DEVICE_ATTR_RO(name);
  55
  56 static inline int init_done(struct zram *zram)
  57 {
  58         return zram->meta != NULL;
  59 }
  60
  61 static inline struct zram *dev_to_zram(struct device *dev)
  62 {
  63         return (struct zram *)dev_to_disk(dev)->private_data;
  64 }
  65
  66 static ssize_t disksize_show(struct device *dev,
  67                 struct device_attribute *attr, char *buf)
  68 {
  69         struct zram *zram = dev_to_zram(dev);
  70
  71         return scnprintf(buf, PAGE_SIZE, "%llu\n", zram->disksize);
  72 }
  73
  74 static ssize_t initstate_show(struct device *dev,
  75                 struct device_attribute *attr, char *buf)
  76 {
  77         u32 val;
  78         struct zram *zram = dev_to_zram(dev);
  79
  80         down_read(&zram->init_lock);
  81         val = init_done(zram);
  82         up_read(&zram->init_lock);
  83
  84         return scnprintf(buf, PAGE_SIZE, "%u\n", val);
  85 }
  86
  87 static ssize_t orig_data_size_show(struct device *dev,
  88                 struct device_attribute *attr, char *buf)
  89 {
  90         struct zram *zram = dev_to_zram(dev);
  91
  92         return scnprintf(buf, PAGE_SIZE, "%llu\n",
  93                 (u64)(atomic64_read(&zram->stats.pages_stored)) << PAGE_SHIFT);
  94 }
  95
  96 static ssize_t mem_used_total_show(struct device *dev,
  97                 struct device_attribute *attr, char *buf)
  98 {
  99         u64 val = 0;
 100         struct zram *zram = dev_to_zram(dev);
 101
 102         down_read(&zram->init_lock);
 103         if (init_done(zram)) {
 104                 struct zram_meta *meta = zram->meta;
 105                 val = zs_get_total_pages(meta->mem_pool);
 106         }
 107         up_read(&zram->init_lock);
 108
 109         return scnprintf(buf, PAGE_SIZE, "%llu\n", val << PAGE_SHIFT);
 110 }
 111
 112 static ssize_t max_comp_streams_show(struct device *dev,
 113                 struct device_attribute *attr, char *buf)
 114 {
 115         int val;
 116         struct zram *zram = dev_to_zram(dev);
 117
 118         down_read(&zram->init_lock);
 119         val = zram->max_comp_streams;
 120         up_read(&zram->init_lock);
 121
 122         return scnprintf(buf, PAGE_SIZE, "%d\n", val);
 123 }
 124
 125 static ssize_t mem_limit_show(struct device *dev,
 126                 struct device_attribute *attr, char *buf)
 127 {
 128         u64 val;
 129         struct zram *zram = dev_to_zram(dev);
 130
 131         down_read(&zram->init_lock);
 132         val = zram->limit_pages;
 133         up_read(&zram->init_lock);
 134
 135         return scnprintf(buf, PAGE_SIZE, "%llu\n", val << PAGE_SHIFT);
 136 }
 137
 138 static ssize_t mem_limit_store(struct device *dev,
 139                 struct device_attribute *attr, const char *buf, size_t len)
 140 {
 141         u64 limit;
 142         char *tmp;
 143         struct zram *zram = dev_to_zram(dev);
 144
 145         limit = memparse(buf, &tmp);
 146         if (buf == tmp) /* no chars parsed, invalid input */
 147                 return -EINVAL;
 148
 149         down_write(&zram->init_lock);
 150         zram->limit_pages = PAGE_ALIGN(limit) >> PAGE_SHIFT;
 151         up_write(&zram->init_lock);
 152
 153         return len;
 154 }
 155
 156 static ssize_t mem_used_max_show(struct device *dev,
 157                 struct device_attribute *attr, char *buf)
 158 {
 159         u64 val = 0;
 160         struct zram *zram = dev_to_zram(dev);
 161
 162         down_read(&zram->init_lock);
 163         if (init_done(zram))
 164                 val = atomic_long_read(&zram->stats.max_used_pages);
 165         up_read(&zram->init_lock);
 166
 167         return scnprintf(buf, PAGE_SIZE, "%llu\n", val << PAGE_SHIFT);
 168 }
 169
 170 static ssize_t mem_used_max_store(struct device *dev,
 171                 struct device_attribute *attr, const char *buf, size_t len)
 172 {
 173         int err;
 174         unsigned long val;
 175         struct zram *zram = dev_to_zram(dev);
 176
 177         err = kstrtoul(buf, 10, &val);
 178         if (err || val != 0)
 179                 return -EINVAL;
 180
 181         down_read(&zram->init_lock);
 182         if (init_done(zram)) {
 183                 struct zram_meta *meta = zram->meta;
 184                 atomic_long_set(&zram->stats.max_used_pages,
 185                                 zs_get_total_pages(meta->mem_pool));
 186         }
 187         up_read(&zram->init_lock);
 188
 189         return len;
 190 }
 191
 192 static ssize_t max_comp_streams_store(struct device *dev,
 193                 struct device_attribute *attr, const char *buf, size_t len)
 194 {
 195         int num;
 196         struct zram *zram = dev_to_zram(dev);
 197         int ret;
 198
 199         ret = kstrtoint(buf, 0, &num);
 200         if (ret < 0)
 201                 return ret;
 202         if (num < 1)
 203                 return -EINVAL;
 204
 205         down_write(&zram->init_lock);
 206         if (init_done(zram)) {
 207                 if (!zcomp_set_max_streams(zram->comp, num)) {
 208                         pr_info("Cannot change max compression streams\n");
 209                         ret = -EINVAL;
 210                         goto out;
 211                 }
 212         }
 213
 214         zram->max_comp_streams = num;
 215         ret = len;
 216 out:
 217         up_write(&zram->init_lock);
 218         return ret;
 219 }
 220
 221 static ssize_t comp_algorithm_show(struct device *dev,
 222                 struct device_attribute *attr, char *buf)
 223 {
 224         size_t sz;
 225         struct zram *zram = dev_to_zram(dev);
 226
 227         down_read(&zram->init_lock);
 228         sz = zcomp_available_show(zram->compressor, buf);
 229         up_read(&zram->init_lock);
 230
 231         return sz;
 232 }
 233
 234 static ssize_t comp_algorithm_store(struct device *dev,
 235                 struct device_attribute *attr, const char *buf, size_t len)
 236 {
 237         struct zram *zram = dev_to_zram(dev);
 238         down_write(&zram->init_lock);
 239         if (init_done(zram)) {
 240                 up_write(&zram->init_lock);
 241                 pr_info("Can't change algorithm for initialized device\n");
 242                 return -EBUSY;
 243         }
 244         strlcpy(zram->compressor, buf, sizeof(zram->compressor));
 245         up_write(&zram->init_lock);
 246         return len;
 247 }
 248
 249 /* flag operations needs meta->tb_lock */
 250 static int zram_test_flag(struct zram_meta *meta, u32 index,
 251                         enum zram_pageflags flag)
 252 {
 253         return meta->table[index].value & BIT(flag);
 254 }
 255
 256 static void zram_set_flag(struct zram_meta *meta, u32 index,
 257                         enum zram_pageflags flag)
 258 {
 259         meta->table[index].value |= BIT(flag);
 260 }
 261
 262 static void zram_clear_flag(struct zram_meta *meta, u32 index,
 263                         enum zram_pageflags flag)
 264 {
 265         meta->table[index].value &= ~BIT(flag);
 266 }
 267
 268 static size_t zram_get_obj_size(struct zram_meta *meta, u32 index)
 269 {
 270         return meta->table[index].value & (BIT(ZRAM_FLAG_SHIFT) - 1);
 271 }
 272
 273 static void zram_set_obj_size(struct zram_meta *meta,
 274                                         u32 index, size_t size)
 275 {
 276         unsigned long flags = meta->table[index].value >> ZRAM_FLAG_SHIFT;
 277
 278         meta->table[index].value = (flags << ZRAM_FLAG_SHIFT) | size;
 279 }
 280
 281 static inline int is_partial_io(struct bio_vec *bvec)
 282 {
 283         return bvec->bv_len != PAGE_SIZE;
 284 }
 285
 286 /*
 287  * Check if request is within bounds and aligned on zram logical blocks.
 288  */
 289 static inline int valid_io_request(struct zram *zram,
 290                 sector_t start, unsigned int size)
 291 {
 292         u64 end, bound;
 293
 294         /* unaligned request */
 295         if (unlikely(start & (ZRAM_SECTOR_PER_LOGICAL_BLOCK - 1)))
 296                 return 0;
 297         if (unlikely(size & (ZRAM_LOGICAL_BLOCK_SIZE - 1)))
 298                 return 0;
 299
 300         end = start + (size >> SECTOR_SHIFT);
 301         bound = zram->disksize >> SECTOR_SHIFT;
 302         /* out of range range */
 303         if (unlikely(start >= bound || end > bound || start > end))
 304                 return 0;
 305
 306         /* I/O request is valid */
 307         return 1;
 308 }
 309
 310 static void zram_meta_free(struct zram_meta *meta)
 311 {
 312         zs_destroy_pool(meta->mem_pool);
 313         vfree(meta->table);
 314         kfree(meta);
 315 }
 316
 317 static struct zram_meta *zram_meta_alloc(u64 disksize)
 318 {
 319         size_t num_pages;
 320         struct zram_meta *meta = kmalloc(sizeof(*meta), GFP_KERNEL);
 321         if (!meta)
 322                 goto out;
 323
 324         num_pages = disksize >> PAGE_SHIFT;
 325         meta->table = vzalloc(num_pages * sizeof(*meta->table));
 326         if (!meta->table) {
 327                 pr_err("Error allocating zram address table\n");
 328                 goto free_meta;
 329         }
 330
 331         meta->mem_pool = zs_create_pool(GFP_NOIO | __GFP_HIGHMEM);
 332         if (!meta->mem_pool) {
 333                 pr_err("Error creating memory pool\n");
 334                 goto free_table;
 335         }
 336
 337         return meta;
 338
 339 free_table:
 340         vfree(meta->table);
 341 free_meta:
 342         kfree(meta);
 343         meta = NULL;
 344 out:
 345         return meta;
 346 }
 347
 348 static void update_position(u32 *index, int *offset, struct bio_vec *bvec)
 349 {
 350         if (*offset + bvec->bv_len >= PAGE_SIZE)
 351                 (*index)++;
 352         *offset = (*offset + bvec->bv_len) % PAGE_SIZE;
 353 }
 354
 355 static int page_zero_filled(void *ptr)
 356 {
 357         unsigned int pos;
 358         unsigned long *page;
 359
 360         page = (unsigned long *)ptr;
 361
 362         for (pos = 0; pos != PAGE_SIZE / sizeof(*page); pos++) {
 363                 if (page[pos])
 364                         return 0;
 365         }
 366
 367         return 1;
 368 }
 369
 370 static void handle_zero_page(struct bio_vec *bvec)
 371 {
 372         struct page *page = bvec->bv_page;
 373         void *user_mem;
 374
 375         user_mem = kmap_atomic(page);
 376         if (is_partial_io(bvec))
 377                 memset(user_mem + bvec->bv_offset, 0, bvec->bv_len);
 378         else
 379                 clear_page(user_mem);
 380         kunmap_atomic(user_mem);
 381
 382         flush_dcache_page(page);
 383 }
 384
 385
 386 /*
 387  * To protect concurrent access to the same index entry,
 388  * caller should hold this table index entry's bit_spinlock to
 389  * indicate this index entry is accessing.
 390  */
 391 static void zram_free_page(struct zram *zram, size_t index)
 392 {
 393         struct zram_meta *meta = zram->meta;
 394         unsigned long handle = meta->table[index].handle;
 395
 396         if (unlikely(!handle)) {
 397                 /*
 398                  * No memory is allocated for zero filled pages.
 399                  * Simply clear zero page flag.
 400                  */
 401                 if (zram_test_flag(meta, index, ZRAM_ZERO)) {
 402                         zram_clear_flag(meta, index, ZRAM_ZERO);
 403                         atomic64_dec(&zram->stats.zero_pages);
 404                 }
 405                 return;
 406         }
 407
 408         zs_free(meta->mem_pool, handle);
 409
 410         atomic64_sub(zram_get_obj_size(meta, index),
 411                         &zram->stats.compr_data_size);
 412         atomic64_dec(&zram->stats.pages_stored);
 413
 414         meta->table[index].handle = 0;
 415         zram_set_obj_size(meta, index, 0);
 416 }
 417
 418 static int zram_decompress_page(struct zram *zram, char *mem, u32 index)
 419 {
 420         int ret = 0;
 421         unsigned char *cmem;
 422         struct zram_meta *meta = zram->meta;
 423         unsigned long handle;
 424         size_t size;
 425
 426         bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
 427         handle = meta->table[index].handle;
 428         size = zram_get_obj_size(meta, index);
 429
 430         if (!handle || zram_test_flag(meta, index, ZRAM_ZERO)) {
 431                 bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 432                 clear_page(mem);
 433                 return 0;
 434         }
 435
 436         cmem = zs_map_object(meta->mem_pool, handle, ZS_MM_RO);
 437         if (size == PAGE_SIZE)
 438                 copy_page(mem, cmem);
 439         else
 440                 ret = zcomp_decompress(zram->comp, cmem, size, mem);
 441         zs_unmap_object(meta->mem_pool, handle);
 442         bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 443
 444         /* Should NEVER happen. Return bio error if it does. */
 445         if (unlikely(ret)) {
 446                 pr_err("Decompression failed! err=%d, page=%u\n", ret, index);
 447                 return ret;
 448         }
 449
 450         return 0;
 451 }
 452
 453 static int zram_bvec_read(struct zram *zram, struct bio_vec *bvec,
 454                           u32 index, int offset)
 455 {
 456         int ret;
 457         struct page *page;
 458         unsigned char *user_mem, *uncmem = NULL;
 459         struct zram_meta *meta = zram->meta;
 460         page = bvec->bv_page;
 461
 462         bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
 463         if (unlikely(!meta->table[index].handle) ||
 464                         zram_test_flag(meta, index, ZRAM_ZERO)) {
 465                 bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 466                 handle_zero_page(bvec);
 467                 return 0;
 468         }
 469         bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 470
 471         if (is_partial_io(bvec))
 472                 /* Use  a temporary buffer to decompress the page */
 473                 uncmem = kmalloc(PAGE_SIZE, GFP_NOIO);
 474
 475         user_mem = kmap_atomic(page);
 476         if (!is_partial_io(bvec))
 477                 uncmem = user_mem;
 478
 479         if (!uncmem) {
 480                 pr_info("Unable to allocate temp memory\n");
 481                 ret = -ENOMEM;
 482                 goto out_cleanup;
 483         }
 484
 485         ret = zram_decompress_page(zram, uncmem, index);
 486         /* Should NEVER happen. Return bio error if it does. */
 487         if (unlikely(ret))
 488                 goto out_cleanup;
 489
 490         if (is_partial_io(bvec))
 491                 memcpy(user_mem + bvec->bv_offset, uncmem + offset,
 492                                 bvec->bv_len);
 493
 494         flush_dcache_page(page);
 495         ret = 0;
 496 out_cleanup:
 497         kunmap_atomic(user_mem);
 498         if (is_partial_io(bvec))
 499                 kfree(uncmem);
 500         return ret;
 501 }
 502
 503 static inline void update_used_max(struct zram *zram,
 504                                         const unsigned long pages)
 505 {
 506         int old_max, cur_max;
 507
 508         old_max = atomic_long_read(&zram->stats.max_used_pages);
 509
 510         do {
 511                 cur_max = old_max;
 512                 if (pages > cur_max)
 513                         old_max = atomic_long_cmpxchg(
 514                                 &zram->stats.max_used_pages, cur_max, pages);
 515         } while (old_max != cur_max);
 516 }
 517
 518 static int zram_bvec_write(struct zram *zram, struct bio_vec *bvec, u32 index,
 519                            int offset)
 520 {
 521         int ret = 0;
 522         size_t clen;
 523         unsigned long handle;
 524         struct page *page;
 525         unsigned char *user_mem, *cmem, *src, *uncmem = NULL;
 526         struct zram_meta *meta = zram->meta;
 527         struct zcomp_strm *zstrm;
 528         bool locked = false;
 529         unsigned long alloced_pages;
 530
 531         page = bvec->bv_page;
 532         if (is_partial_io(bvec)) {
 533                 /*
 534                  * This is a partial IO. We need to read the full page
 535                  * before to write the changes.
 536                  */
 537                 uncmem = kmalloc(PAGE_SIZE, GFP_NOIO);
 538                 if (!uncmem) {
 539                         ret = -ENOMEM;
 540                         goto out;
 541                 }
 542                 ret = zram_decompress_page(zram, uncmem, index);
 543                 if (ret)
 544                         goto out;
 545         }
 546
 547         zstrm = zcomp_strm_find(zram->comp);
 548         locked = true;
 549         user_mem = kmap_atomic(page);
 550
 551         if (is_partial_io(bvec)) {
 552                 memcpy(uncmem + offset, user_mem + bvec->bv_offset,
 553                        bvec->bv_len);
 554                 kunmap_atomic(user_mem);
 555                 user_mem = NULL;
 556         } else {
 557                 uncmem = user_mem;
 558         }
 559
 560         if (page_zero_filled(uncmem)) {
 561                 if (user_mem)
 562                         kunmap_atomic(user_mem);
 563                 /* Free memory associated with this sector now. */
 564                 bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
 565                 zram_free_page(zram, index);
 566                 zram_set_flag(meta, index, ZRAM_ZERO);
 567                 bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 568
 569                 atomic64_inc(&zram->stats.zero_pages);
 570                 ret = 0;
 571                 goto out;
 572         }
 573
 574         ret = zcomp_compress(zram->comp, zstrm, uncmem, &clen);
 575         if (!is_partial_io(bvec)) {
 576                 kunmap_atomic(user_mem);
 577                 user_mem = NULL;
 578                 uncmem = NULL;
 579         }
 580
 581         if (unlikely(ret)) {
 582                 pr_err("Compression failed! err=%d\n", ret);
 583                 goto out;
 584         }
 585         src = zstrm->buffer;
 586         if (unlikely(clen > max_zpage_size)) {
 587                 clen = PAGE_SIZE;
 588                 if (is_partial_io(bvec))
 589                         src = uncmem;
 590         }
 591
 592         handle = zs_malloc(meta->mem_pool, clen);
 593         if (!handle) {
 594                 pr_info("Error allocating memory for compressed page: %u, size=%zu\n",
 595                         index, clen);
 596                 ret = -ENOMEM;
 597                 goto out;
 598         }
 599
 600         alloced_pages = zs_get_total_pages(meta->mem_pool);
 601         if (zram->limit_pages && alloced_pages > zram->limit_pages) {
 602                 zs_free(meta->mem_pool, handle);
 603                 ret = -ENOMEM;
 604                 goto out;
 605         }
 606
 607         update_used_max(zram, alloced_pages);
 608
 609         cmem = zs_map_object(meta->mem_pool, handle, ZS_MM_WO);
 610
 611         if ((clen == PAGE_SIZE) && !is_partial_io(bvec)) {
 612                 src = kmap_atomic(page);
 613                 copy_page(cmem, src);
 614                 kunmap_atomic(src);
 615         } else {
 616                 memcpy(cmem, src, clen);
 617         }
 618
 619         zcomp_strm_release(zram->comp, zstrm);
 620         locked = false;
 621         zs_unmap_object(meta->mem_pool, handle);
 622
 623         /*
 624          * Free memory associated with this sector
 625          * before overwriting unused sectors.
 626          */
 627         bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
 628         zram_free_page(zram, index);
 629
 630         meta->table[index].handle = handle;
 631         zram_set_obj_size(meta, index, clen);
 632         bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 633
 634         /* Update stats */
 635         atomic64_add(clen, &zram->stats.compr_data_size);
 636         atomic64_inc(&zram->stats.pages_stored);
 637 out:
 638         if (locked)
 639                 zcomp_strm_release(zram->comp, zstrm);
 640         if (is_partial_io(bvec))
 641                 kfree(uncmem);
 642         return ret;
 643 }
 644
 645 static int zram_bvec_rw(struct zram *zram, struct bio_vec *bvec, u32 index,
 646                         int offset, int rw)
 647 {
 648         int ret;
 649
 650         if (rw == READ) {
 651                 atomic64_inc(&zram->stats.num_reads);
 652                 ret = zram_bvec_read(zram, bvec, index, offset);
 653         } else {
 654                 atomic64_inc(&zram->stats.num_writes);
 655                 ret = zram_bvec_write(zram, bvec, index, offset);
 656         }
 657
 658         if (unlikely(ret)) {
 659                 if (rw == READ)
 660                         atomic64_inc(&zram->stats.failed_reads);
 661                 else
 662                         atomic64_inc(&zram->stats.failed_writes);
 663         }
 664
 665         return ret;
 666 }
 667
 668 /*
 669  * zram_bio_discard - handler on discard request
 670  * @index: physical block index in PAGE_SIZE units
 671  * @offset: byte offset within physical block
 672  */
 673 static void zram_bio_discard(struct zram *zram, u32 index,
 674                              int offset, struct bio *bio)
 675 {
 676         size_t n = bio->bi_iter.bi_size;
 677         struct zram_meta *meta = zram->meta;
 678
 679         /*
 680          * zram manages data in physical block size units. Because logical block
 681          * size isn't identical with physical block size on some arch, we
 682          * could get a discard request pointing to a specific offset within a
 683          * certain physical block.  Although we can handle this request by
 684          * reading that physiclal block and decompressing and partially zeroing
 685          * and re-compressing and then re-storing it, this isn't reasonable
 686          * because our intent with a discard request is to save memory.  So
 687          * skipping this logical block is appropriate here.
 688          */
 689         if (offset) {
 690                 if (n <= (PAGE_SIZE - offset))
 691                         return;
 692
 693                 n -= (PAGE_SIZE - offset);
 694                 index++;
 695         }
 696
 697         while (n >= PAGE_SIZE) {
 698                 bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
 699                 zram_free_page(zram, index);
 700                 bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 701                 atomic64_inc(&zram->stats.notify_free);
 702                 index++;
 703                 n -= PAGE_SIZE;
 704         }
 705 }
 706
 707 static void zram_reset_device(struct zram *zram, bool reset_capacity)
 708 {
 709         size_t index;
 710         struct zram_meta *meta;
 711
 712         down_write(&zram->init_lock);
 713
 714         zram->limit_pages = 0;
 715
 716         if (!init_done(zram)) {
 717                 up_write(&zram->init_lock);
 718                 return;
 719         }
 720
 721         meta = zram->meta;
 722         /* Free all pages that are still in this zram device */
 723         for (index = 0; index < zram->disksize >> PAGE_SHIFT; index++) {
 724                 unsigned long handle = meta->table[index].handle;
 725                 if (!handle)
 726                         continue;
 727
 728                 zs_free(meta->mem_pool, handle);
 729         }
 730
 731         zcomp_destroy(zram->comp);
 732         zram->max_comp_streams = 1;
 733
 734         zram_meta_free(zram->meta);
 735         zram->meta = NULL;
 736         /* Reset stats */
 737         memset(&zram->stats, 0, sizeof(zram->stats));
 738
 739         zram->disksize = 0;
 740         if (reset_capacity)
 741                 set_capacity(zram->disk, 0);
 742
 743         up_write(&zram->init_lock);
 744
 745         /*
 746          * Revalidate disk out of the init_lock to avoid lockdep splat.
 747          * It's okay because disk's capacity is protected by init_lock
 748          * so that revalidate_disk always sees up-to-date capacity.
 749          */
 750         if (reset_capacity)
 751                 revalidate_disk(zram->disk);
 752 }
 753
 754 static ssize_t disksize_store(struct device *dev,
 755                 struct device_attribute *attr, const char *buf, size_t len)
 756 {
 757         u64 disksize;
 758         struct zcomp *comp;
 759         struct zram_meta *meta;
 760         struct zram *zram = dev_to_zram(dev);
 761         int err;
 762
 763         disksize = memparse(buf, NULL);
 764         if (!disksize)
 765                 return -EINVAL;
 766
 767         disksize = PAGE_ALIGN(disksize);
 768         meta = zram_meta_alloc(disksize);
 769         if (!meta)
 770                 return -ENOMEM;
 771
 772         comp = zcomp_create(zram->compressor, zram->max_comp_streams);
 773         if (IS_ERR(comp)) {
 774                 pr_info("Cannot initialise %s compressing backend\n",
 775                                 zram->compressor);
 776                 err = PTR_ERR(comp);
 777                 goto out_free_meta;
 778         }
 779
 780         down_write(&zram->init_lock);
 781         if (init_done(zram)) {
 782                 pr_info("Cannot change disksize for initialized device\n");
 783                 err = -EBUSY;
 784                 goto out_destroy_comp;
 785         }
 786
 787         zram->meta = meta;
 788         zram->comp = comp;
 789         zram->disksize = disksize;
 790         set_capacity(zram->disk, zram->disksize >> SECTOR_SHIFT);
 791         up_write(&zram->init_lock);
 792
 793         /*
 794          * Revalidate disk out of the init_lock to avoid lockdep splat.
 795          * It's okay because disk's capacity is protected by init_lock
 796          * so that revalidate_disk always sees up-to-date capacity.
 797          */
 798         revalidate_disk(zram->disk);
 799
 800         return len;
 801
 802 out_destroy_comp:
 803         up_write(&zram->init_lock);
 804         zcomp_destroy(comp);
 805 out_free_meta:
 806         zram_meta_free(meta);
 807         return err;
 808 }
 809
 810 static ssize_t reset_store(struct device *dev,
 811                 struct device_attribute *attr, const char *buf, size_t len)
 812 {
 813         int ret;
 814         unsigned short do_reset;
 815         struct zram *zram;
 816         struct block_device *bdev;
 817
 818         zram = dev_to_zram(dev);
 819         bdev = bdget_disk(zram->disk, 0);
 820
 821         if (!bdev)
 822                 return -ENOMEM;
 823
 824         /* Do not reset an active device! */
 825         if (bdev->bd_holders) {
 826                 ret = -EBUSY;
 827                 goto out;
 828         }
 829
 830         ret = kstrtou16(buf, 10, &do_reset);
 831         if (ret)
 832                 goto out;
 833
 834         if (!do_reset) {
 835                 ret = -EINVAL;
 836                 goto out;
 837         }
 838
 839         /* Make sure all pending I/O is finished */
 840         fsync_bdev(bdev);
 841         bdput(bdev);
 842
 843         zram_reset_device(zram, true);
 844         return len;
 845
 846 out:
 847         bdput(bdev);
 848         return ret;
 849 }
 850
 851 static void __zram_make_request(struct zram *zram, struct bio *bio)
 852 {
 853         int offset, rw;
 854         u32 index;
 855         struct bio_vec bvec;
 856         struct bvec_iter iter;
 857
 858         index = bio->bi_iter.bi_sector >> SECTORS_PER_PAGE_SHIFT;
 859         offset = (bio->bi_iter.bi_sector &
 860                   (SECTORS_PER_PAGE - 1)) << SECTOR_SHIFT;
 861
 862         if (unlikely(bio->bi_rw & REQ_DISCARD)) {
 863                 zram_bio_discard(zram, index, offset, bio);
 864                 bio_endio(bio, 0);
 865                 return;
 866         }
 867
 868         rw = bio_data_dir(bio);
 869         bio_for_each_segment(bvec, bio, iter) {
 870                 int max_transfer_size = PAGE_SIZE - offset;
 871
 872                 if (bvec.bv_len > max_transfer_size) {
 873                         /*
 874                          * zram_bvec_rw() can only make operation on a single
 875                          * zram page. Split the bio vector.
 876                          */
 877                         struct bio_vec bv;
 878
 879                         bv.bv_page = bvec.bv_page;
 880                         bv.bv_len = max_transfer_size;
 881                         bv.bv_offset = bvec.bv_offset;
 882
 883                         if (zram_bvec_rw(zram, &bv, index, offset, rw) < 0)
 884                                 goto out;
 885
 886                         bv.bv_len = bvec.bv_len - max_transfer_size;
 887                         bv.bv_offset += max_transfer_size;
 888                         if (zram_bvec_rw(zram, &bv, index + 1, 0, rw) < 0)
 889                                 goto out;
 890                 } else
 891                         if (zram_bvec_rw(zram, &bvec, index, offset, rw) < 0)
 892                                 goto out;
 893
 894                 update_position(&index, &offset, &bvec);
 895         }
 896
 897         set_bit(BIO_UPTODATE, &bio->bi_flags);
 898         bio_endio(bio, 0);
 899         return;
 900
 901 out:
 902         bio_io_error(bio);
 903 }
 904
 905 /*
 906  * Handler function for all zram I/O requests.
 907  */
 908 static void zram_make_request(struct request_queue *queue, struct bio *bio)
 909 {
 910         struct zram *zram = queue->queuedata;
 911
 912         down_read(&zram->init_lock);
 913         if (unlikely(!init_done(zram)))
 914                 goto error;
 915
 916         if (!valid_io_request(zram, bio->bi_iter.bi_sector,
 917                                         bio->bi_iter.bi_size)) {
 918                 atomic64_inc(&zram->stats.invalid_io);
 919                 goto error;
 920         }
 921
 922         __zram_make_request(zram, bio);
 923         up_read(&zram->init_lock);
 924
 925         return;
 926
 927 error:
 928         up_read(&zram->init_lock);
 929         bio_io_error(bio);
 930 }
 931
 932 static void zram_slot_free_notify(struct block_device *bdev,
 933                                 unsigned long index)
 934 {
 935         struct zram *zram;
 936         struct zram_meta *meta;
 937
 938         zram = bdev->bd_disk->private_data;
 939         meta = zram->meta;
 940
 941         bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
 942         zram_free_page(zram, index);
 943         bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 944         atomic64_inc(&zram->stats.notify_free);
 945 }
 946
 947 static int zram_rw_page(struct block_device *bdev, sector_t sector,
 948                        struct page *page, int rw)
 949 {
 950         int offset, err;
 951         u32 index;
 952         struct zram *zram;
 953         struct bio_vec bv;
 954
 955         zram = bdev->bd_disk->private_data;
 956         if (!valid_io_request(zram, sector, PAGE_SIZE)) {
 957                 atomic64_inc(&zram->stats.invalid_io);
 958                 return -EINVAL;
 959         }
 960
 961         down_read(&zram->init_lock);
 962         if (unlikely(!init_done(zram))) {
 963                 err = -EIO;
 964                 goto out_unlock;
 965         }
 966
 967         index = sector >> SECTORS_PER_PAGE_SHIFT;
 968         offset = sector & (SECTORS_PER_PAGE - 1) << SECTOR_SHIFT;
 969
 970         bv.bv_page = page;
 971         bv.bv_len = PAGE_SIZE;
 972         bv.bv_offset = 0;
 973
 974         err = zram_bvec_rw(zram, &bv, index, offset, rw);
 975 out_unlock:
 976         up_read(&zram->init_lock);
 977         /*
 978          * If I/O fails, just return error(ie, non-zero) without
 979          * calling page_endio.
 980          * It causes resubmit the I/O with bio request by upper functions
 981          * of rw_page(e.g., swap_readpage, __swap_writepage) and
 982          * bio->bi_end_io does things to handle the error
 983          * (e.g., SetPageError, set_page_dirty and extra works).
 984          */
 985         if (err == 0)
 986                 page_endio(page, rw, 0);
 987         return err;
 988 }
 989
 990 static const struct block_device_operations zram_devops = {
 991         .swap_slot_free_notify = zram_slot_free_notify,
 992         .rw_page = zram_rw_page,
 993         .owner = THIS_MODULE
 994 };
 995
 996 static DEVICE_ATTR_RW(disksize);
 997 static DEVICE_ATTR_RO(initstate);
 998 static DEVICE_ATTR_WO(reset);
 999 static DEVICE_ATTR_RO(orig_data_size);
1000 static DEVICE_ATTR_RO(mem_used_total);
1001 static DEVICE_ATTR_RW(mem_limit);
1002 static DEVICE_ATTR_RW(mem_used_max);
1003 static DEVICE_ATTR_RW(max_comp_streams);
1004 static DEVICE_ATTR_RW(comp_algorithm);
1005
1006 ZRAM_ATTR_RO(num_reads);
1007 ZRAM_ATTR_RO(num_writes);
1008 ZRAM_ATTR_RO(failed_reads);
1009 ZRAM_ATTR_RO(failed_writes);
1010 ZRAM_ATTR_RO(invalid_io);
1011 ZRAM_ATTR_RO(notify_free);
1012 ZRAM_ATTR_RO(zero_pages);
1013 ZRAM_ATTR_RO(compr_data_size);
1014
1015 static struct attribute *zram_disk_attrs[] = {
1016         &dev_attr_disksize.attr,
1017         &dev_attr_initstate.attr,
1018         &dev_attr_reset.attr,
1019         &dev_attr_num_reads.attr,
1020         &dev_attr_num_writes.attr,
1021         &dev_attr_failed_reads.attr,
1022         &dev_attr_failed_writes.attr,
1023         &dev_attr_invalid_io.attr,
1024         &dev_attr_notify_free.attr,
1025         &dev_attr_zero_pages.attr,
1026         &dev_attr_orig_data_size.attr,
1027         &dev_attr_compr_data_size.attr,
1028         &dev_attr_mem_used_total.attr,
1029         &dev_attr_mem_limit.attr,
1030         &dev_attr_mem_used_max.attr,
1031         &dev_attr_max_comp_streams.attr,
1032         &dev_attr_comp_algorithm.attr,
1033         NULL,
1034 };
1035
1036 static struct attribute_group zram_disk_attr_group = {
1037         .attrs = zram_disk_attrs,
1038 };
1039
1040 static int create_device(struct zram *zram, int device_id)
1041 {
1042         int ret = -ENOMEM;
1043
1044         init_rwsem(&zram->init_lock);
1045
1046         zram->queue = blk_alloc_queue(GFP_KERNEL);
1047         if (!zram->queue) {
1048                 pr_err("Error allocating disk queue for device %d\n",
1049                         device_id);
1050                 goto out;
1051         }
1052
1053         blk_queue_make_request(zram->queue, zram_make_request);
1054         zram->queue->queuedata = zram;
1055
1056          /* gendisk structure */
1057         zram->disk = alloc_disk(1);
1058         if (!zram->disk) {
1059                 pr_warn("Error allocating disk structure for device %d\n",
1060                         device_id);
1061                 goto out_free_queue;
1062         }
1063
1064         zram->disk->major = zram_major;
1065         zram->disk->first_minor = device_id;
1066         zram->disk->fops = &zram_devops;
1067         zram->disk->queue = zram->queue;
1068         zram->disk->private_data = zram;
1069         snprintf(zram->disk->disk_name, 16, "zram%d", device_id);
1070
1071         /* Actual capacity set using syfs (/sys/block/zram<id>/disksize */
1072         set_capacity(zram->disk, 0);
1073         /* zram devices sort of resembles non-rotational disks */
1074         queue_flag_set_unlocked(QUEUE_FLAG_NONROT, zram->disk->queue);
1075         queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, zram->disk->queue);
1076         /*
1077          * To ensure that we always get PAGE_SIZE aligned
1078          * and n*PAGE_SIZED sized I/O requests.
1079          */
1080         blk_queue_physical_block_size(zram->disk->queue, PAGE_SIZE);
1081         blk_queue_logical_block_size(zram->disk->queue,
1082                                         ZRAM_LOGICAL_BLOCK_SIZE);
1083         blk_queue_io_min(zram->disk->queue, PAGE_SIZE);
1084         blk_queue_io_opt(zram->disk->queue, PAGE_SIZE);
1085         zram->disk->queue->limits.discard_granularity = PAGE_SIZE;
1086         zram->disk->queue->limits.max_discard_sectors = UINT_MAX;
1087         /*
1088          * zram_bio_discard() will clear all logical blocks if logical block
1089          * size is identical with physical block size(PAGE_SIZE). But if it is
1090          * different, we will skip discarding some parts of logical blocks in
1091          * the part of the request range which isn't aligned to physical block
1092          * size.  So we can't ensure that all discarded logical blocks are
1093          * zeroed.
1094          */
1095         if (ZRAM_LOGICAL_BLOCK_SIZE == PAGE_SIZE)
1096                 zram->disk->queue->limits.discard_zeroes_data = 1;
1097         else
1098                 zram->disk->queue->limits.discard_zeroes_data = 0;
1099         queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, zram->disk->queue);
1100
1101         add_disk(zram->disk);
1102
1103         ret = sysfs_create_group(&disk_to_dev(zram->disk)->kobj,
1104                                 &zram_disk_attr_group);
1105         if (ret < 0) {
1106                 pr_warn("Error creating sysfs group");
1107                 goto out_free_disk;
1108         }
1109         strlcpy(zram->compressor, default_compressor, sizeof(zram->compressor));
1110         zram->meta = NULL;
1111         zram->max_comp_streams = 1;
1112         return 0;
1113
1114 out_free_disk:
1115         del_gendisk(zram->disk);
1116         put_disk(zram->disk);
1117 out_free_queue:
1118         blk_cleanup_queue(zram->queue);
1119 out:
1120         return ret;
1121 }
1122
1123 static void destroy_device(struct zram *zram)
1124 {
1125         sysfs_remove_group(&disk_to_dev(zram->disk)->kobj,
1126                         &zram_disk_attr_group);
1127
1128         del_gendisk(zram->disk);
1129         put_disk(zram->disk);
1130
1131         blk_cleanup_queue(zram->queue);
1132 }
1133
1134 static int __init zram_init(void)
1135 {
1136         int ret, dev_id;
1137
1138         if (num_devices > max_num_devices) {
1139                 pr_warn("Invalid value for num_devices: %u\n",
1140                                 num_devices);
1141                 ret = -EINVAL;
1142                 goto out;
1143         }
1144
1145         zram_major = register_blkdev(0, "zram");
1146         if (zram_major <= 0) {
1147                 pr_warn("Unable to get major number\n");
1148                 ret = -EBUSY;
1149                 goto out;
1150         }
1151
1152         /* Allocate the device array and initialize each one */
1153         zram_devices = kzalloc(num_devices * sizeof(struct zram), GFP_KERNEL);
1154         if (!zram_devices) {
1155                 ret = -ENOMEM;
1156                 goto unregister;
1157         }
1158
1159         for (dev_id = 0; dev_id < num_devices; dev_id++) {
1160                 ret = create_device(&zram_devices[dev_id], dev_id);
1161                 if (ret)
1162                         goto free_devices;
1163         }
1164
1165         pr_info("Created %u device(s) ...\n", num_devices);
1166
1167         return 0;
1168
1169 free_devices:
1170         while (dev_id)
1171                 destroy_device(&zram_devices[--dev_id]);
1172         kfree(zram_devices);
1173 unregister:
1174         unregister_blkdev(zram_major, "zram");
1175 out:
1176         return ret;
1177 }
1178
1179 static void __exit zram_exit(void)
1180 {
1181         int i;
1182         struct zram *zram;
1183
1184         for (i = 0; i < num_devices; i++) {
1185                 zram = &zram_devices[i];
1186
1187                 destroy_device(zram);
1188                 /*
1189                  * Shouldn't access zram->disk after destroy_device
1190                  * because destroy_device already released zram->disk.
1191                  */
1192                 zram_reset_device(zram, false);
1193         }
1194
1195         unregister_blkdev(zram_major, "zram");
1196
1197         kfree(zram_devices);
1198         pr_debug("Cleanup done!\n");
1199 }
1200
1201 module_init(zram_init);
1202 module_exit(zram_exit);
1203
1204 module_param(num_devices, uint, 0);
1205 MODULE_PARM_DESC(num_devices, "Number of zram devices");
1206
1207 MODULE_LICENSE("Dual BSD/GPL");
1208 MODULE_AUTHOR("Nitin Gupta <ngupta@vflare.org>");
1209 MODULE_DESCRIPTION("Compressed RAM Block Device");