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 bool init_done(struct zram *zram)
  57 {
  58         return zram->disksize;
  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 compact_store(struct device *dev,
  67                 struct device_attribute *attr, const char *buf, size_t len)
  68 {
  69         unsigned long nr_migrated;
  70         struct zram *zram = dev_to_zram(dev);
  71         struct zram_meta *meta;
  72
  73         down_read(&zram->init_lock);
  74         if (!init_done(zram)) {
  75                 up_read(&zram->init_lock);
  76                 return -EINVAL;
  77         }
  78
  79         meta = zram->meta;
  80         nr_migrated = zs_compact(meta->mem_pool);
  81         atomic64_add(nr_migrated, &zram->stats.num_migrated);
  82         up_read(&zram->init_lock);
  83
  84         return len;
  85 }
  86
  87 static ssize_t disksize_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", zram->disksize);
  93 }
  94
  95 static ssize_t initstate_show(struct device *dev,
  96                 struct device_attribute *attr, char *buf)
  97 {
  98         u32 val;
  99         struct zram *zram = dev_to_zram(dev);
 100
 101         down_read(&zram->init_lock);
 102         val = init_done(zram);
 103         up_read(&zram->init_lock);
 104
 105         return scnprintf(buf, PAGE_SIZE, "%u\n", val);
 106 }
 107
 108 static ssize_t orig_data_size_show(struct device *dev,
 109                 struct device_attribute *attr, char *buf)
 110 {
 111         struct zram *zram = dev_to_zram(dev);
 112
 113         return scnprintf(buf, PAGE_SIZE, "%llu\n",
 114                 (u64)(atomic64_read(&zram->stats.pages_stored)) << PAGE_SHIFT);
 115 }
 116
 117 static ssize_t mem_used_total_show(struct device *dev,
 118                 struct device_attribute *attr, char *buf)
 119 {
 120         u64 val = 0;
 121         struct zram *zram = dev_to_zram(dev);
 122
 123         down_read(&zram->init_lock);
 124         if (init_done(zram)) {
 125                 struct zram_meta *meta = zram->meta;
 126                 val = zs_get_total_pages(meta->mem_pool);
 127         }
 128         up_read(&zram->init_lock);
 129
 130         return scnprintf(buf, PAGE_SIZE, "%llu\n", val << PAGE_SHIFT);
 131 }
 132
 133 static ssize_t max_comp_streams_show(struct device *dev,
 134                 struct device_attribute *attr, char *buf)
 135 {
 136         int val;
 137         struct zram *zram = dev_to_zram(dev);
 138
 139         down_read(&zram->init_lock);
 140         val = zram->max_comp_streams;
 141         up_read(&zram->init_lock);
 142
 143         return scnprintf(buf, PAGE_SIZE, "%d\n", val);
 144 }
 145
 146 static ssize_t mem_limit_show(struct device *dev,
 147                 struct device_attribute *attr, char *buf)
 148 {
 149         u64 val;
 150         struct zram *zram = dev_to_zram(dev);
 151
 152         down_read(&zram->init_lock);
 153         val = zram->limit_pages;
 154         up_read(&zram->init_lock);
 155
 156         return scnprintf(buf, PAGE_SIZE, "%llu\n", val << PAGE_SHIFT);
 157 }
 158
 159 static ssize_t mem_limit_store(struct device *dev,
 160                 struct device_attribute *attr, const char *buf, size_t len)
 161 {
 162         u64 limit;
 163         char *tmp;
 164         struct zram *zram = dev_to_zram(dev);
 165
 166         limit = memparse(buf, &tmp);
 167         if (buf == tmp) /* no chars parsed, invalid input */
 168                 return -EINVAL;
 169
 170         down_write(&zram->init_lock);
 171         zram->limit_pages = PAGE_ALIGN(limit) >> PAGE_SHIFT;
 172         up_write(&zram->init_lock);
 173
 174         return len;
 175 }
 176
 177 static ssize_t mem_used_max_show(struct device *dev,
 178                 struct device_attribute *attr, char *buf)
 179 {
 180         u64 val = 0;
 181         struct zram *zram = dev_to_zram(dev);
 182
 183         down_read(&zram->init_lock);
 184         if (init_done(zram))
 185                 val = atomic_long_read(&zram->stats.max_used_pages);
 186         up_read(&zram->init_lock);
 187
 188         return scnprintf(buf, PAGE_SIZE, "%llu\n", val << PAGE_SHIFT);
 189 }
 190
 191 static ssize_t mem_used_max_store(struct device *dev,
 192                 struct device_attribute *attr, const char *buf, size_t len)
 193 {
 194         int err;
 195         unsigned long val;
 196         struct zram *zram = dev_to_zram(dev);
 197
 198         err = kstrtoul(buf, 10, &val);
 199         if (err || val != 0)
 200                 return -EINVAL;
 201
 202         down_read(&zram->init_lock);
 203         if (init_done(zram)) {
 204                 struct zram_meta *meta = zram->meta;
 205                 atomic_long_set(&zram->stats.max_used_pages,
 206                                 zs_get_total_pages(meta->mem_pool));
 207         }
 208         up_read(&zram->init_lock);
 209
 210         return len;
 211 }
 212
 213 static ssize_t max_comp_streams_store(struct device *dev,
 214                 struct device_attribute *attr, const char *buf, size_t len)
 215 {
 216         int num;
 217         struct zram *zram = dev_to_zram(dev);
 218         int ret;
 219
 220         ret = kstrtoint(buf, 0, &num);
 221         if (ret < 0)
 222                 return ret;
 223         if (num < 1)
 224                 return -EINVAL;
 225
 226         down_write(&zram->init_lock);
 227         if (init_done(zram)) {
 228                 if (!zcomp_set_max_streams(zram->comp, num)) {
 229                         pr_info("Cannot change max compression streams\n");
 230                         ret = -EINVAL;
 231                         goto out;
 232                 }
 233         }
 234
 235         zram->max_comp_streams = num;
 236         ret = len;
 237 out:
 238         up_write(&zram->init_lock);
 239         return ret;
 240 }
 241
 242 static ssize_t comp_algorithm_show(struct device *dev,
 243                 struct device_attribute *attr, char *buf)
 244 {
 245         size_t sz;
 246         struct zram *zram = dev_to_zram(dev);
 247
 248         down_read(&zram->init_lock);
 249         sz = zcomp_available_show(zram->compressor, buf);
 250         up_read(&zram->init_lock);
 251
 252         return sz;
 253 }
 254
 255 static ssize_t comp_algorithm_store(struct device *dev,
 256                 struct device_attribute *attr, const char *buf, size_t len)
 257 {
 258         struct zram *zram = dev_to_zram(dev);
 259         down_write(&zram->init_lock);
 260         if (init_done(zram)) {
 261                 up_write(&zram->init_lock);
 262                 pr_info("Can't change algorithm for initialized device\n");
 263                 return -EBUSY;
 264         }
 265         strlcpy(zram->compressor, buf, sizeof(zram->compressor));
 266         up_write(&zram->init_lock);
 267         return len;
 268 }
 269
 270 /* flag operations needs meta->tb_lock */
 271 static int zram_test_flag(struct zram_meta *meta, u32 index,
 272                         enum zram_pageflags flag)
 273 {
 274         return meta->table[index].value & BIT(flag);
 275 }
 276
 277 static void zram_set_flag(struct zram_meta *meta, u32 index,
 278                         enum zram_pageflags flag)
 279 {
 280         meta->table[index].value |= BIT(flag);
 281 }
 282
 283 static void zram_clear_flag(struct zram_meta *meta, u32 index,
 284                         enum zram_pageflags flag)
 285 {
 286         meta->table[index].value &= ~BIT(flag);
 287 }
 288
 289 static size_t zram_get_obj_size(struct zram_meta *meta, u32 index)
 290 {
 291         return meta->table[index].value & (BIT(ZRAM_FLAG_SHIFT) - 1);
 292 }
 293
 294 static void zram_set_obj_size(struct zram_meta *meta,
 295                                         u32 index, size_t size)
 296 {
 297         unsigned long flags = meta->table[index].value >> ZRAM_FLAG_SHIFT;
 298
 299         meta->table[index].value = (flags << ZRAM_FLAG_SHIFT) | size;
 300 }
 301
 302 static inline int is_partial_io(struct bio_vec *bvec)
 303 {
 304         return bvec->bv_len != PAGE_SIZE;
 305 }
 306
 307 /*
 308  * Check if request is within bounds and aligned on zram logical blocks.
 309  */
 310 static inline int valid_io_request(struct zram *zram,
 311                 sector_t start, unsigned int size)
 312 {
 313         u64 end, bound;
 314
 315         /* unaligned request */
 316         if (unlikely(start & (ZRAM_SECTOR_PER_LOGICAL_BLOCK - 1)))
 317                 return 0;
 318         if (unlikely(size & (ZRAM_LOGICAL_BLOCK_SIZE - 1)))
 319                 return 0;
 320
 321         end = start + (size >> SECTOR_SHIFT);
 322         bound = zram->disksize >> SECTOR_SHIFT;
 323         /* out of range range */
 324         if (unlikely(start >= bound || end > bound || start > end))
 325                 return 0;
 326
 327         /* I/O request is valid */
 328         return 1;
 329 }
 330
 331 static void zram_meta_free(struct zram_meta *meta, u64 disksize)
 332 {
 333         size_t num_pages = disksize >> PAGE_SHIFT;
 334         size_t index;
 335
 336         /* Free all pages that are still in this zram device */
 337         for (index = 0; index < num_pages; index++) {
 338                 unsigned long handle = meta->table[index].handle;
 339
 340                 if (!handle)
 341                         continue;
 342
 343                 zs_free(meta->mem_pool, handle);
 344         }
 345
 346         zs_destroy_pool(meta->mem_pool);
 347         vfree(meta->table);
 348         kfree(meta);
 349 }
 350
 351 static struct zram_meta *zram_meta_alloc(int device_id, u64 disksize)
 352 {
 353         size_t num_pages;
 354         char pool_name[8];
 355         struct zram_meta *meta = kmalloc(sizeof(*meta), GFP_KERNEL);
 356
 357         if (!meta)
 358                 return NULL;
 359
 360         num_pages = disksize >> PAGE_SHIFT;
 361         meta->table = vzalloc(num_pages * sizeof(*meta->table));
 362         if (!meta->table) {
 363                 pr_err("Error allocating zram address table\n");
 364                 goto out_error;
 365         }
 366
 367         snprintf(pool_name, sizeof(pool_name), "zram%d", device_id);
 368         meta->mem_pool = zs_create_pool(pool_name, GFP_NOIO | __GFP_HIGHMEM);
 369         if (!meta->mem_pool) {
 370                 pr_err("Error creating memory pool\n");
 371                 goto out_error;
 372         }
 373
 374         return meta;
 375
 376 out_error:
 377         vfree(meta->table);
 378         kfree(meta);
 379         return NULL;
 380 }
 381
 382 static inline bool zram_meta_get(struct zram *zram)
 383 {
 384         if (atomic_inc_not_zero(&zram->refcount))
 385                 return true;
 386         return false;
 387 }
 388
 389 static inline void zram_meta_put(struct zram *zram)
 390 {
 391         atomic_dec(&zram->refcount);
 392 }
 393
 394 static void update_position(u32 *index, int *offset, struct bio_vec *bvec)
 395 {
 396         if (*offset + bvec->bv_len >= PAGE_SIZE)
 397                 (*index)++;
 398         *offset = (*offset + bvec->bv_len) % PAGE_SIZE;
 399 }
 400
 401 static int page_zero_filled(void *ptr)
 402 {
 403         unsigned int pos;
 404         unsigned long *page;
 405
 406         page = (unsigned long *)ptr;
 407
 408         for (pos = 0; pos != PAGE_SIZE / sizeof(*page); pos++) {
 409                 if (page[pos])
 410                         return 0;
 411         }
 412
 413         return 1;
 414 }
 415
 416 static void handle_zero_page(struct bio_vec *bvec)
 417 {
 418         struct page *page = bvec->bv_page;
 419         void *user_mem;
 420
 421         user_mem = kmap_atomic(page);
 422         if (is_partial_io(bvec))
 423                 memset(user_mem + bvec->bv_offset, 0, bvec->bv_len);
 424         else
 425                 clear_page(user_mem);
 426         kunmap_atomic(user_mem);
 427
 428         flush_dcache_page(page);
 429 }
 430
 431
 432 /*
 433  * To protect concurrent access to the same index entry,
 434  * caller should hold this table index entry's bit_spinlock to
 435  * indicate this index entry is accessing.
 436  */
 437 static void zram_free_page(struct zram *zram, size_t index)
 438 {
 439         struct zram_meta *meta = zram->meta;
 440         unsigned long handle = meta->table[index].handle;
 441
 442         if (unlikely(!handle)) {
 443                 /*
 444                  * No memory is allocated for zero filled pages.
 445                  * Simply clear zero page flag.
 446                  */
 447                 if (zram_test_flag(meta, index, ZRAM_ZERO)) {
 448                         zram_clear_flag(meta, index, ZRAM_ZERO);
 449                         atomic64_dec(&zram->stats.zero_pages);
 450                 }
 451                 return;
 452         }
 453
 454         zs_free(meta->mem_pool, handle);
 455
 456         atomic64_sub(zram_get_obj_size(meta, index),
 457                         &zram->stats.compr_data_size);
 458         atomic64_dec(&zram->stats.pages_stored);
 459
 460         meta->table[index].handle = 0;
 461         zram_set_obj_size(meta, index, 0);
 462 }
 463
 464 static int zram_decompress_page(struct zram *zram, char *mem, u32 index)
 465 {
 466         int ret = 0;
 467         unsigned char *cmem;
 468         struct zram_meta *meta = zram->meta;
 469         unsigned long handle;
 470         size_t size;
 471
 472         bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
 473         handle = meta->table[index].handle;
 474         size = zram_get_obj_size(meta, index);
 475
 476         if (!handle || zram_test_flag(meta, index, ZRAM_ZERO)) {
 477                 bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 478                 clear_page(mem);
 479                 return 0;
 480         }
 481
 482         cmem = zs_map_object(meta->mem_pool, handle, ZS_MM_RO);
 483         if (size == PAGE_SIZE)
 484                 copy_page(mem, cmem);
 485         else
 486                 ret = zcomp_decompress(zram->comp, cmem, size, mem);
 487         zs_unmap_object(meta->mem_pool, handle);
 488         bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 489
 490         /* Should NEVER happen. Return bio error if it does. */
 491         if (unlikely(ret)) {
 492                 pr_err("Decompression failed! err=%d, page=%u\n", ret, index);
 493                 return ret;
 494         }
 495
 496         return 0;
 497 }
 498
 499 static int zram_bvec_read(struct zram *zram, struct bio_vec *bvec,
 500                           u32 index, int offset)
 501 {
 502         int ret;
 503         struct page *page;
 504         unsigned char *user_mem, *uncmem = NULL;
 505         struct zram_meta *meta = zram->meta;
 506         page = bvec->bv_page;
 507
 508         bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
 509         if (unlikely(!meta->table[index].handle) ||
 510                         zram_test_flag(meta, index, ZRAM_ZERO)) {
 511                 bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 512                 handle_zero_page(bvec);
 513                 return 0;
 514         }
 515         bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 516
 517         if (is_partial_io(bvec))
 518                 /* Use  a temporary buffer to decompress the page */
 519                 uncmem = kmalloc(PAGE_SIZE, GFP_NOIO);
 520
 521         user_mem = kmap_atomic(page);
 522         if (!is_partial_io(bvec))
 523                 uncmem = user_mem;
 524
 525         if (!uncmem) {
 526                 pr_info("Unable to allocate temp memory\n");
 527                 ret = -ENOMEM;
 528                 goto out_cleanup;
 529         }
 530
 531         ret = zram_decompress_page(zram, uncmem, index);
 532         /* Should NEVER happen. Return bio error if it does. */
 533         if (unlikely(ret))
 534                 goto out_cleanup;
 535
 536         if (is_partial_io(bvec))
 537                 memcpy(user_mem + bvec->bv_offset, uncmem + offset,
 538                                 bvec->bv_len);
 539
 540         flush_dcache_page(page);
 541         ret = 0;
 542 out_cleanup:
 543         kunmap_atomic(user_mem);
 544         if (is_partial_io(bvec))
 545                 kfree(uncmem);
 546         return ret;
 547 }
 548
 549 static inline void update_used_max(struct zram *zram,
 550                                         const unsigned long pages)
 551 {
 552         unsigned long old_max, cur_max;
 553
 554         old_max = atomic_long_read(&zram->stats.max_used_pages);
 555
 556         do {
 557                 cur_max = old_max;
 558                 if (pages > cur_max)
 559                         old_max = atomic_long_cmpxchg(
 560                                 &zram->stats.max_used_pages, cur_max, pages);
 561         } while (old_max != cur_max);
 562 }
 563
 564 static int zram_bvec_write(struct zram *zram, struct bio_vec *bvec, u32 index,
 565                            int offset)
 566 {
 567         int ret = 0;
 568         size_t clen;
 569         unsigned long handle;
 570         struct page *page;
 571         unsigned char *user_mem, *cmem, *src, *uncmem = NULL;
 572         struct zram_meta *meta = zram->meta;
 573         struct zcomp_strm *zstrm;
 574         bool locked = false;
 575         unsigned long alloced_pages;
 576
 577         page = bvec->bv_page;
 578         if (is_partial_io(bvec)) {
 579                 /*
 580                  * This is a partial IO. We need to read the full page
 581                  * before to write the changes.
 582                  */
 583                 uncmem = kmalloc(PAGE_SIZE, GFP_NOIO);
 584                 if (!uncmem) {
 585                         ret = -ENOMEM;
 586                         goto out;
 587                 }
 588                 ret = zram_decompress_page(zram, uncmem, index);
 589                 if (ret)
 590                         goto out;
 591         }
 592
 593         zstrm = zcomp_strm_find(zram->comp);
 594         locked = true;
 595         user_mem = kmap_atomic(page);
 596
 597         if (is_partial_io(bvec)) {
 598                 memcpy(uncmem + offset, user_mem + bvec->bv_offset,
 599                        bvec->bv_len);
 600                 kunmap_atomic(user_mem);
 601                 user_mem = NULL;
 602         } else {
 603                 uncmem = user_mem;
 604         }
 605
 606         if (page_zero_filled(uncmem)) {
 607                 if (user_mem)
 608                         kunmap_atomic(user_mem);
 609                 /* Free memory associated with this sector now. */
 610                 bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
 611                 zram_free_page(zram, index);
 612                 zram_set_flag(meta, index, ZRAM_ZERO);
 613                 bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 614
 615                 atomic64_inc(&zram->stats.zero_pages);
 616                 ret = 0;
 617                 goto out;
 618         }
 619
 620         ret = zcomp_compress(zram->comp, zstrm, uncmem, &clen);
 621         if (!is_partial_io(bvec)) {
 622                 kunmap_atomic(user_mem);
 623                 user_mem = NULL;
 624                 uncmem = NULL;
 625         }
 626
 627         if (unlikely(ret)) {
 628                 pr_err("Compression failed! err=%d\n", ret);
 629                 goto out;
 630         }
 631         src = zstrm->buffer;
 632         if (unlikely(clen > max_zpage_size)) {
 633                 clen = PAGE_SIZE;
 634                 if (is_partial_io(bvec))
 635                         src = uncmem;
 636         }
 637
 638         handle = zs_malloc(meta->mem_pool, clen);
 639         if (!handle) {
 640                 pr_info("Error allocating memory for compressed page: %u, size=%zu\n",
 641                         index, clen);
 642                 ret = -ENOMEM;
 643                 goto out;
 644         }
 645
 646         alloced_pages = zs_get_total_pages(meta->mem_pool);
 647         if (zram->limit_pages && alloced_pages > zram->limit_pages) {
 648                 zs_free(meta->mem_pool, handle);
 649                 ret = -ENOMEM;
 650                 goto out;
 651         }
 652
 653         update_used_max(zram, alloced_pages);
 654
 655         cmem = zs_map_object(meta->mem_pool, handle, ZS_MM_WO);
 656
 657         if ((clen == PAGE_SIZE) && !is_partial_io(bvec)) {
 658                 src = kmap_atomic(page);
 659                 copy_page(cmem, src);
 660                 kunmap_atomic(src);
 661         } else {
 662                 memcpy(cmem, src, clen);
 663         }
 664
 665         zcomp_strm_release(zram->comp, zstrm);
 666         locked = false;
 667         zs_unmap_object(meta->mem_pool, handle);
 668
 669         /*
 670          * Free memory associated with this sector
 671          * before overwriting unused sectors.
 672          */
 673         bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
 674         zram_free_page(zram, index);
 675
 676         meta->table[index].handle = handle;
 677         zram_set_obj_size(meta, index, clen);
 678         bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 679
 680         /* Update stats */
 681         atomic64_add(clen, &zram->stats.compr_data_size);
 682         atomic64_inc(&zram->stats.pages_stored);
 683 out:
 684         if (locked)
 685                 zcomp_strm_release(zram->comp, zstrm);
 686         if (is_partial_io(bvec))
 687                 kfree(uncmem);
 688         return ret;
 689 }
 690
 691 static int zram_bvec_rw(struct zram *zram, struct bio_vec *bvec, u32 index,
 692                         int offset, int rw)
 693 {
 694         int ret;
 695
 696         if (rw == READ) {
 697                 atomic64_inc(&zram->stats.num_reads);
 698                 ret = zram_bvec_read(zram, bvec, index, offset);
 699         } else {
 700                 atomic64_inc(&zram->stats.num_writes);
 701                 ret = zram_bvec_write(zram, bvec, index, offset);
 702         }
 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_WO(compact);
1042 static DEVICE_ATTR_RW(disksize);
1043 static DEVICE_ATTR_RO(initstate);
1044 static DEVICE_ATTR_WO(reset);
1045 static DEVICE_ATTR_RO(orig_data_size);
1046 static DEVICE_ATTR_RO(mem_used_total);
1047 static DEVICE_ATTR_RW(mem_limit);
1048 static DEVICE_ATTR_RW(mem_used_max);
1049 static DEVICE_ATTR_RW(max_comp_streams);
1050 static DEVICE_ATTR_RW(comp_algorithm);
1051
1052 ZRAM_ATTR_RO(num_reads);
1053 ZRAM_ATTR_RO(num_writes);
1054 ZRAM_ATTR_RO(failed_reads);
1055 ZRAM_ATTR_RO(failed_writes);
1056 ZRAM_ATTR_RO(invalid_io);
1057 ZRAM_ATTR_RO(notify_free);
1058 ZRAM_ATTR_RO(zero_pages);
1059 ZRAM_ATTR_RO(compr_data_size);
1060
1061 static struct attribute *zram_disk_attrs[] = {
1062         &dev_attr_disksize.attr,
1063         &dev_attr_initstate.attr,
1064         &dev_attr_reset.attr,
1065         &dev_attr_num_reads.attr,
1066         &dev_attr_num_writes.attr,
1067         &dev_attr_failed_reads.attr,
1068         &dev_attr_failed_writes.attr,
1069         &dev_attr_compact.attr,
1070         &dev_attr_invalid_io.attr,
1071         &dev_attr_notify_free.attr,
1072         &dev_attr_zero_pages.attr,
1073         &dev_attr_orig_data_size.attr,
1074         &dev_attr_compr_data_size.attr,
1075         &dev_attr_mem_used_total.attr,
1076         &dev_attr_mem_limit.attr,
1077         &dev_attr_mem_used_max.attr,
1078         &dev_attr_max_comp_streams.attr,
1079         &dev_attr_comp_algorithm.attr,
1080         NULL,
1081 };
1082
1083 static struct attribute_group zram_disk_attr_group = {
1084         .attrs = zram_disk_attrs,
1085 };
1086
1087 static int create_device(struct zram *zram, int device_id)
1088 {
1089         struct request_queue *queue;
1090         int ret = -ENOMEM;
1091
1092         init_rwsem(&zram->init_lock);
1093
1094         queue = blk_alloc_queue(GFP_KERNEL);
1095         if (!queue) {
1096                 pr_err("Error allocating disk queue for device %d\n",
1097                         device_id);
1098                 goto out;
1099         }
1100
1101         blk_queue_make_request(queue, zram_make_request);
1102
1103          /* gendisk structure */
1104         zram->disk = alloc_disk(1);
1105         if (!zram->disk) {
1106                 pr_warn("Error allocating disk structure for device %d\n",
1107                         device_id);
1108                 goto out_free_queue;
1109         }
1110
1111         zram->disk->major = zram_major;
1112         zram->disk->first_minor = device_id;
1113         zram->disk->fops = &zram_devops;
1114         zram->disk->queue = queue;
1115         zram->disk->queue->queuedata = zram;
1116         zram->disk->private_data = zram;
1117         snprintf(zram->disk->disk_name, 16, "zram%d", device_id);
1118
1119         /* Actual capacity set using syfs (/sys/block/zram<id>/disksize */
1120         set_capacity(zram->disk, 0);
1121         /* zram devices sort of resembles non-rotational disks */
1122         queue_flag_set_unlocked(QUEUE_FLAG_NONROT, zram->disk->queue);
1123         queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, zram->disk->queue);
1124         /*
1125          * To ensure that we always get PAGE_SIZE aligned
1126          * and n*PAGE_SIZED sized I/O requests.
1127          */
1128         blk_queue_physical_block_size(zram->disk->queue, PAGE_SIZE);
1129         blk_queue_logical_block_size(zram->disk->queue,
1130                                         ZRAM_LOGICAL_BLOCK_SIZE);
1131         blk_queue_io_min(zram->disk->queue, PAGE_SIZE);
1132         blk_queue_io_opt(zram->disk->queue, PAGE_SIZE);
1133         zram->disk->queue->limits.discard_granularity = PAGE_SIZE;
1134         zram->disk->queue->limits.max_discard_sectors = UINT_MAX;
1135         /*
1136          * zram_bio_discard() will clear all logical blocks if logical block
1137          * size is identical with physical block size(PAGE_SIZE). But if it is
1138          * different, we will skip discarding some parts of logical blocks in
1139          * the part of the request range which isn't aligned to physical block
1140          * size.  So we can't ensure that all discarded logical blocks are
1141          * zeroed.
1142          */
1143         if (ZRAM_LOGICAL_BLOCK_SIZE == PAGE_SIZE)
1144                 zram->disk->queue->limits.discard_zeroes_data = 1;
1145         else
1146                 zram->disk->queue->limits.discard_zeroes_data = 0;
1147         queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, zram->disk->queue);
1148
1149         add_disk(zram->disk);
1150
1151         ret = sysfs_create_group(&disk_to_dev(zram->disk)->kobj,
1152                                 &zram_disk_attr_group);
1153         if (ret < 0) {
1154                 pr_warn("Error creating sysfs group");
1155                 goto out_free_disk;
1156         }
1157         strlcpy(zram->compressor, default_compressor, sizeof(zram->compressor));
1158         zram->meta = NULL;
1159         zram->max_comp_streams = 1;
1160         return 0;
1161
1162 out_free_disk:
1163         del_gendisk(zram->disk);
1164         put_disk(zram->disk);
1165 out_free_queue:
1166         blk_cleanup_queue(queue);
1167 out:
1168         return ret;
1169 }
1170
1171 static void destroy_devices(unsigned int nr)
1172 {
1173         struct zram *zram;
1174         unsigned int i;
1175
1176         for (i = 0; i < nr; i++) {
1177                 zram = &zram_devices[i];
1178                 /*
1179                  * Remove sysfs first, so no one will perform a disksize
1180                  * store while we destroy the devices
1181                  */
1182                 sysfs_remove_group(&disk_to_dev(zram->disk)->kobj,
1183                                 &zram_disk_attr_group);
1184
1185                 zram_reset_device(zram);
1186
1187                 blk_cleanup_queue(zram->disk->queue);
1188                 del_gendisk(zram->disk);
1189                 put_disk(zram->disk);
1190         }
1191
1192         kfree(zram_devices);
1193         unregister_blkdev(zram_major, "zram");
1194         pr_info("Destroyed %u device(s)\n", nr);
1195 }
1196
1197 static int __init zram_init(void)
1198 {
1199         int ret, dev_id;
1200
1201         if (num_devices > max_num_devices) {
1202                 pr_warn("Invalid value for num_devices: %u\n",
1203                                 num_devices);
1204                 return -EINVAL;
1205         }
1206
1207         zram_major = register_blkdev(0, "zram");
1208         if (zram_major <= 0) {
1209                 pr_warn("Unable to get major number\n");
1210                 return -EBUSY;
1211         }
1212
1213         /* Allocate the device array and initialize each one */
1214         zram_devices = kzalloc(num_devices * sizeof(struct zram), GFP_KERNEL);
1215         if (!zram_devices) {
1216                 unregister_blkdev(zram_major, "zram");
1217                 return -ENOMEM;
1218         }
1219
1220         for (dev_id = 0; dev_id < num_devices; dev_id++) {
1221                 ret = create_device(&zram_devices[dev_id], dev_id);
1222                 if (ret)
1223                         goto out_error;
1224         }
1225
1226         pr_info("Created %u device(s)\n", num_devices);
1227         return 0;
1228
1229 out_error:
1230         destroy_devices(dev_id);
1231         return ret;
1232 }
1233
1234 static void __exit zram_exit(void)
1235 {
1236         destroy_devices(num_devices);
1237 }
1238
1239 module_init(zram_init);
1240 module_exit(zram_exit);
1241
1242 module_param(num_devices, uint, 0);
1243 MODULE_PARM_DESC(num_devices, "Number of zram devices");
1244
1245 MODULE_LICENSE("Dual BSD/GPL");
1246 MODULE_AUTHOR("Nitin Gupta <ngupta@vflare.org>");
1247 MODULE_DESCRIPTION("Compressed RAM Block Device");