drivers/block/zram/zram_drv.c

   1 /*
   2  * Compressed RAM block device
   3  *
   4  * Copyright (C) 2008, 2009, 2010  Nitin Gupta
   5  *               2012, 2013 Minchan Kim
   6  *
   7  * This code is released using a dual license strategy: BSD/GPL
   8  * You can choose the licence that better fits your requirements.
   9  *
  10  * Released under the terms of 3-clause BSD License
  11  * Released under the terms of GNU General Public License Version 2.0
  12  *
  13  */
  14
  15 #define KMSG_COMPONENT "zram"
  16 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
  17
  18 #include <linux/module.h>
  19 #include <linux/kernel.h>
  20 #include <linux/bio.h>
  21 #include <linux/bitops.h>
  22 #include <linux/blkdev.h>
  23 #include <linux/buffer_head.h>
  24 #include <linux/device.h>
  25 #include <linux/genhd.h>
  26 #include <linux/highmem.h>
  27 #include <linux/slab.h>
  28 #include <linux/string.h>
  29 #include <linux/vmalloc.h>
  30 #include <linux/err.h>
  31 #include <linux/idr.h>
  32
  33 #include "zram_drv.h"
  34
  35 static DEFINE_IDR(zram_index_idr);
  36 static int zram_major;
  37 static const char *default_compressor = "lzo";
  38
  39 /* Module params (documentation at end) */
  40 static unsigned int num_devices = 1;
  41
  42 static inline void deprecated_attr_warn(const char *name)
  43 {
  44         pr_warn_once("%d (%s) Attribute %s (and others) will be removed. %s\n",
  45                         task_pid_nr(current),
  46                         current->comm,
  47                         name,
  48                         "See zram documentation.");
  49 }
  50
  51 #define ZRAM_ATTR_RO(name)                                              \
  52 static ssize_t name##_show(struct device *d,                            \
  53                                 struct device_attribute *attr, char *b) \
  54 {                                                                       \
  55         struct zram *zram = dev_to_zram(d);                             \
  56                                                                         \
  57         deprecated_attr_warn(__stringify(name));                        \
  58         return scnprintf(b, PAGE_SIZE, "%llu\n",                        \
  59                 (u64)atomic64_read(&zram->stats.name));                 \
  60 }                                                                       \
  61 static DEVICE_ATTR_RO(name);
  62
  63 static inline bool init_done(struct zram *zram)
  64 {
  65         return zram->disksize;
  66 }
  67
  68 static inline struct zram *dev_to_zram(struct device *dev)
  69 {
  70         return (struct zram *)dev_to_disk(dev)->private_data;
  71 }
  72
  73 /* flag operations needs meta->tb_lock */
  74 static int zram_test_flag(struct zram_meta *meta, u32 index,
  75                         enum zram_pageflags flag)
  76 {
  77         return meta->table[index].value & BIT(flag);
  78 }
  79
  80 static void zram_set_flag(struct zram_meta *meta, u32 index,
  81                         enum zram_pageflags flag)
  82 {
  83         meta->table[index].value |= BIT(flag);
  84 }
  85
  86 static void zram_clear_flag(struct zram_meta *meta, u32 index,
  87                         enum zram_pageflags flag)
  88 {
  89         meta->table[index].value &= ~BIT(flag);
  90 }
  91
  92 static size_t zram_get_obj_size(struct zram_meta *meta, u32 index)
  93 {
  94         return meta->table[index].value & (BIT(ZRAM_FLAG_SHIFT) - 1);
  95 }
  96
  97 static void zram_set_obj_size(struct zram_meta *meta,
  98                                         u32 index, size_t size)
  99 {
 100         unsigned long flags = meta->table[index].value >> ZRAM_FLAG_SHIFT;
 101
 102         meta->table[index].value = (flags << ZRAM_FLAG_SHIFT) | size;
 103 }
 104
 105 static inline int is_partial_io(struct bio_vec *bvec)
 106 {
 107         return bvec->bv_len != PAGE_SIZE;
 108 }
 109
 110 /*
 111  * Check if request is within bounds and aligned on zram logical blocks.
 112  */
 113 static inline int valid_io_request(struct zram *zram,
 114                 sector_t start, unsigned int size)
 115 {
 116         u64 end, bound;
 117
 118         /* unaligned request */
 119         if (unlikely(start & (ZRAM_SECTOR_PER_LOGICAL_BLOCK - 1)))
 120                 return 0;
 121         if (unlikely(size & (ZRAM_LOGICAL_BLOCK_SIZE - 1)))
 122                 return 0;
 123
 124         end = start + (size >> SECTOR_SHIFT);
 125         bound = zram->disksize >> SECTOR_SHIFT;
 126         /* out of range range */
 127         if (unlikely(start >= bound || end > bound || start > end))
 128                 return 0;
 129
 130         /* I/O request is valid */
 131         return 1;
 132 }
 133
 134 static void update_position(u32 *index, int *offset, struct bio_vec *bvec)
 135 {
 136         if (*offset + bvec->bv_len >= PAGE_SIZE)
 137                 (*index)++;
 138         *offset = (*offset + bvec->bv_len) % PAGE_SIZE;
 139 }
 140
 141 static inline void update_used_max(struct zram *zram,
 142                                         const unsigned long pages)
 143 {
 144         unsigned long old_max, cur_max;
 145
 146         old_max = atomic_long_read(&zram->stats.max_used_pages);
 147
 148         do {
 149                 cur_max = old_max;
 150                 if (pages > cur_max)
 151                         old_max = atomic_long_cmpxchg(
 152                                 &zram->stats.max_used_pages, cur_max, pages);
 153         } while (old_max != cur_max);
 154 }
 155
 156 static int page_zero_filled(void *ptr)
 157 {
 158         unsigned int pos;
 159         unsigned long *page;
 160
 161         page = (unsigned long *)ptr;
 162
 163         for (pos = 0; pos != PAGE_SIZE / sizeof(*page); pos++) {
 164                 if (page[pos])
 165                         return 0;
 166         }
 167
 168         return 1;
 169 }
 170
 171 static void handle_zero_page(struct bio_vec *bvec)
 172 {
 173         struct page *page = bvec->bv_page;
 174         void *user_mem;
 175
 176         user_mem = kmap_atomic(page);
 177         if (is_partial_io(bvec))
 178                 memset(user_mem + bvec->bv_offset, 0, bvec->bv_len);
 179         else
 180                 clear_page(user_mem);
 181         kunmap_atomic(user_mem);
 182
 183         flush_dcache_page(page);
 184 }
 185
 186 static ssize_t initstate_show(struct device *dev,
 187                 struct device_attribute *attr, char *buf)
 188 {
 189         u32 val;
 190         struct zram *zram = dev_to_zram(dev);
 191
 192         down_read(&zram->init_lock);
 193         val = init_done(zram);
 194         up_read(&zram->init_lock);
 195
 196         return scnprintf(buf, PAGE_SIZE, "%u\n", val);
 197 }
 198
 199 static ssize_t disksize_show(struct device *dev,
 200                 struct device_attribute *attr, char *buf)
 201 {
 202         struct zram *zram = dev_to_zram(dev);
 203
 204         return scnprintf(buf, PAGE_SIZE, "%llu\n", zram->disksize);
 205 }
 206
 207 static ssize_t orig_data_size_show(struct device *dev,
 208                 struct device_attribute *attr, char *buf)
 209 {
 210         struct zram *zram = dev_to_zram(dev);
 211
 212         deprecated_attr_warn("orig_data_size");
 213         return scnprintf(buf, PAGE_SIZE, "%llu\n",
 214                 (u64)(atomic64_read(&zram->stats.pages_stored)) << PAGE_SHIFT);
 215 }
 216
 217 static ssize_t mem_used_total_show(struct device *dev,
 218                 struct device_attribute *attr, char *buf)
 219 {
 220         u64 val = 0;
 221         struct zram *zram = dev_to_zram(dev);
 222
 223         deprecated_attr_warn("mem_used_total");
 224         down_read(&zram->init_lock);
 225         if (init_done(zram)) {
 226                 struct zram_meta *meta = zram->meta;
 227                 val = zs_get_total_pages(meta->mem_pool);
 228         }
 229         up_read(&zram->init_lock);
 230
 231         return scnprintf(buf, PAGE_SIZE, "%llu\n", val << PAGE_SHIFT);
 232 }
 233
 234 static ssize_t mem_limit_show(struct device *dev,
 235                 struct device_attribute *attr, char *buf)
 236 {
 237         u64 val;
 238         struct zram *zram = dev_to_zram(dev);
 239
 240         deprecated_attr_warn("mem_limit");
 241         down_read(&zram->init_lock);
 242         val = zram->limit_pages;
 243         up_read(&zram->init_lock);
 244
 245         return scnprintf(buf, PAGE_SIZE, "%llu\n", val << PAGE_SHIFT);
 246 }
 247
 248 static ssize_t mem_limit_store(struct device *dev,
 249                 struct device_attribute *attr, const char *buf, size_t len)
 250 {
 251         u64 limit;
 252         char *tmp;
 253         struct zram *zram = dev_to_zram(dev);
 254
 255         limit = memparse(buf, &tmp);
 256         if (buf == tmp) /* no chars parsed, invalid input */
 257                 return -EINVAL;
 258
 259         down_write(&zram->init_lock);
 260         zram->limit_pages = PAGE_ALIGN(limit) >> PAGE_SHIFT;
 261         up_write(&zram->init_lock);
 262
 263         return len;
 264 }
 265
 266 static ssize_t mem_used_max_show(struct device *dev,
 267                 struct device_attribute *attr, char *buf)
 268 {
 269         u64 val = 0;
 270         struct zram *zram = dev_to_zram(dev);
 271
 272         deprecated_attr_warn("mem_used_max");
 273         down_read(&zram->init_lock);
 274         if (init_done(zram))
 275                 val = atomic_long_read(&zram->stats.max_used_pages);
 276         up_read(&zram->init_lock);
 277
 278         return scnprintf(buf, PAGE_SIZE, "%llu\n", val << PAGE_SHIFT);
 279 }
 280
 281 static ssize_t mem_used_max_store(struct device *dev,
 282                 struct device_attribute *attr, const char *buf, size_t len)
 283 {
 284         int err;
 285         unsigned long val;
 286         struct zram *zram = dev_to_zram(dev);
 287
 288         err = kstrtoul(buf, 10, &val);
 289         if (err || val != 0)
 290                 return -EINVAL;
 291
 292         down_read(&zram->init_lock);
 293         if (init_done(zram)) {
 294                 struct zram_meta *meta = zram->meta;
 295                 atomic_long_set(&zram->stats.max_used_pages,
 296                                 zs_get_total_pages(meta->mem_pool));
 297         }
 298         up_read(&zram->init_lock);
 299
 300         return len;
 301 }
 302
 303 static ssize_t max_comp_streams_show(struct device *dev,
 304                 struct device_attribute *attr, char *buf)
 305 {
 306         int val;
 307         struct zram *zram = dev_to_zram(dev);
 308
 309         down_read(&zram->init_lock);
 310         val = zram->max_comp_streams;
 311         up_read(&zram->init_lock);
 312
 313         return scnprintf(buf, PAGE_SIZE, "%d\n", val);
 314 }
 315
 316 static ssize_t max_comp_streams_store(struct device *dev,
 317                 struct device_attribute *attr, const char *buf, size_t len)
 318 {
 319         int num;
 320         struct zram *zram = dev_to_zram(dev);
 321         int ret;
 322
 323         ret = kstrtoint(buf, 0, &num);
 324         if (ret < 0)
 325                 return ret;
 326         if (num < 1)
 327                 return -EINVAL;
 328
 329         down_write(&zram->init_lock);
 330         if (init_done(zram)) {
 331                 if (!zcomp_set_max_streams(zram->comp, num)) {
 332                         pr_info("Cannot change max compression streams\n");
 333                         ret = -EINVAL;
 334                         goto out;
 335                 }
 336         }
 337
 338         zram->max_comp_streams = num;
 339         ret = len;
 340 out:
 341         up_write(&zram->init_lock);
 342         return ret;
 343 }
 344
 345 static ssize_t comp_algorithm_show(struct device *dev,
 346                 struct device_attribute *attr, char *buf)
 347 {
 348         size_t sz;
 349         struct zram *zram = dev_to_zram(dev);
 350
 351         down_read(&zram->init_lock);
 352         sz = zcomp_available_show(zram->compressor, buf);
 353         up_read(&zram->init_lock);
 354
 355         return sz;
 356 }
 357
 358 static ssize_t comp_algorithm_store(struct device *dev,
 359                 struct device_attribute *attr, const char *buf, size_t len)
 360 {
 361         struct zram *zram = dev_to_zram(dev);
 362         down_write(&zram->init_lock);
 363         if (init_done(zram)) {
 364                 up_write(&zram->init_lock);
 365                 pr_info("Can't change algorithm for initialized device\n");
 366                 return -EBUSY;
 367         }
 368         strlcpy(zram->compressor, buf, sizeof(zram->compressor));
 369         up_write(&zram->init_lock);
 370         return len;
 371 }
 372
 373 static ssize_t compact_store(struct device *dev,
 374                 struct device_attribute *attr, const char *buf, size_t len)
 375 {
 376         unsigned long nr_migrated;
 377         struct zram *zram = dev_to_zram(dev);
 378         struct zram_meta *meta;
 379
 380         down_read(&zram->init_lock);
 381         if (!init_done(zram)) {
 382                 up_read(&zram->init_lock);
 383                 return -EINVAL;
 384         }
 385
 386         meta = zram->meta;
 387         nr_migrated = zs_compact(meta->mem_pool);
 388         atomic64_add(nr_migrated, &zram->stats.num_migrated);
 389         up_read(&zram->init_lock);
 390
 391         return len;
 392 }
 393
 394 static ssize_t io_stat_show(struct device *dev,
 395                 struct device_attribute *attr, char *buf)
 396 {
 397         struct zram *zram = dev_to_zram(dev);
 398         ssize_t ret;
 399
 400         down_read(&zram->init_lock);
 401         ret = scnprintf(buf, PAGE_SIZE,
 402                         "%8llu %8llu %8llu %8llu\n",
 403                         (u64)atomic64_read(&zram->stats.failed_reads),
 404                         (u64)atomic64_read(&zram->stats.failed_writes),
 405                         (u64)atomic64_read(&zram->stats.invalid_io),
 406                         (u64)atomic64_read(&zram->stats.notify_free));
 407         up_read(&zram->init_lock);
 408
 409         return ret;
 410 }
 411
 412 static ssize_t mm_stat_show(struct device *dev,
 413                 struct device_attribute *attr, char *buf)
 414 {
 415         struct zram *zram = dev_to_zram(dev);
 416         u64 orig_size, mem_used = 0;
 417         long max_used;
 418         ssize_t ret;
 419
 420         down_read(&zram->init_lock);
 421         if (init_done(zram))
 422                 mem_used = zs_get_total_pages(zram->meta->mem_pool);
 423
 424         orig_size = atomic64_read(&zram->stats.pages_stored);
 425         max_used = atomic_long_read(&zram->stats.max_used_pages);
 426
 427         ret = scnprintf(buf, PAGE_SIZE,
 428                         "%8llu %8llu %8llu %8lu %8ld %8llu %8llu\n",
 429                         orig_size << PAGE_SHIFT,
 430                         (u64)atomic64_read(&zram->stats.compr_data_size),
 431                         mem_used << PAGE_SHIFT,
 432                         zram->limit_pages << PAGE_SHIFT,
 433                         max_used << PAGE_SHIFT,
 434                         (u64)atomic64_read(&zram->stats.zero_pages),
 435                         (u64)atomic64_read(&zram->stats.num_migrated));
 436         up_read(&zram->init_lock);
 437
 438         return ret;
 439 }
 440
 441 static DEVICE_ATTR_RO(io_stat);
 442 static DEVICE_ATTR_RO(mm_stat);
 443 ZRAM_ATTR_RO(num_reads);
 444 ZRAM_ATTR_RO(num_writes);
 445 ZRAM_ATTR_RO(failed_reads);
 446 ZRAM_ATTR_RO(failed_writes);
 447 ZRAM_ATTR_RO(invalid_io);
 448 ZRAM_ATTR_RO(notify_free);
 449 ZRAM_ATTR_RO(zero_pages);
 450 ZRAM_ATTR_RO(compr_data_size);
 451
 452 static inline bool zram_meta_get(struct zram *zram)
 453 {
 454         if (atomic_inc_not_zero(&zram->refcount))
 455                 return true;
 456         return false;
 457 }
 458
 459 static inline void zram_meta_put(struct zram *zram)
 460 {
 461         atomic_dec(&zram->refcount);
 462 }
 463
 464 static void zram_meta_free(struct zram_meta *meta, u64 disksize)
 465 {
 466         size_t num_pages = disksize >> PAGE_SHIFT;
 467         size_t index;
 468
 469         /* Free all pages that are still in this zram device */
 470         for (index = 0; index < num_pages; index++) {
 471                 unsigned long handle = meta->table[index].handle;
 472
 473                 if (!handle)
 474                         continue;
 475
 476                 zs_free(meta->mem_pool, handle);
 477         }
 478
 479         zs_destroy_pool(meta->mem_pool);
 480         vfree(meta->table);
 481         kfree(meta);
 482 }
 483
 484 static struct zram_meta *zram_meta_alloc(int device_id, u64 disksize)
 485 {
 486         size_t num_pages;
 487         char pool_name[8];
 488         struct zram_meta *meta = kmalloc(sizeof(*meta), GFP_KERNEL);
 489
 490         if (!meta)
 491                 return NULL;
 492
 493         num_pages = disksize >> PAGE_SHIFT;
 494         meta->table = vzalloc(num_pages * sizeof(*meta->table));
 495         if (!meta->table) {
 496                 pr_err("Error allocating zram address table\n");
 497                 goto out_error;
 498         }
 499
 500         snprintf(pool_name, sizeof(pool_name), "zram%d", device_id);
 501         meta->mem_pool = zs_create_pool(pool_name, GFP_NOIO | __GFP_HIGHMEM);
 502         if (!meta->mem_pool) {
 503                 pr_err("Error creating memory pool\n");
 504                 goto out_error;
 505         }
 506
 507         return meta;
 508
 509 out_error:
 510         vfree(meta->table);
 511         kfree(meta);
 512         return NULL;
 513 }
 514
 515 /*
 516  * To protect concurrent access to the same index entry,
 517  * caller should hold this table index entry's bit_spinlock to
 518  * indicate this index entry is accessing.
 519  */
 520 static void zram_free_page(struct zram *zram, size_t index)
 521 {
 522         struct zram_meta *meta = zram->meta;
 523         unsigned long handle = meta->table[index].handle;
 524
 525         if (unlikely(!handle)) {
 526                 /*
 527                  * No memory is allocated for zero filled pages.
 528                  * Simply clear zero page flag.
 529                  */
 530                 if (zram_test_flag(meta, index, ZRAM_ZERO)) {
 531                         zram_clear_flag(meta, index, ZRAM_ZERO);
 532                         atomic64_dec(&zram->stats.zero_pages);
 533                 }
 534                 return;
 535         }
 536
 537         zs_free(meta->mem_pool, handle);
 538
 539         atomic64_sub(zram_get_obj_size(meta, index),
 540                         &zram->stats.compr_data_size);
 541         atomic64_dec(&zram->stats.pages_stored);
 542
 543         meta->table[index].handle = 0;
 544         zram_set_obj_size(meta, index, 0);
 545 }
 546
 547 static int zram_decompress_page(struct zram *zram, char *mem, u32 index)
 548 {
 549         int ret = 0;
 550         unsigned char *cmem;
 551         struct zram_meta *meta = zram->meta;
 552         unsigned long handle;
 553         size_t size;
 554
 555         bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
 556         handle = meta->table[index].handle;
 557         size = zram_get_obj_size(meta, index);
 558
 559         if (!handle || zram_test_flag(meta, index, ZRAM_ZERO)) {
 560                 bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 561                 clear_page(mem);
 562                 return 0;
 563         }
 564
 565         cmem = zs_map_object(meta->mem_pool, handle, ZS_MM_RO);
 566         if (size == PAGE_SIZE)
 567                 copy_page(mem, cmem);
 568         else
 569                 ret = zcomp_decompress(zram->comp, cmem, size, mem);
 570         zs_unmap_object(meta->mem_pool, handle);
 571         bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 572
 573         /* Should NEVER happen. Return bio error if it does. */
 574         if (unlikely(ret)) {
 575                 pr_err("Decompression failed! err=%d, page=%u\n", ret, index);
 576                 return ret;
 577         }
 578
 579         return 0;
 580 }
 581
 582 static int zram_bvec_read(struct zram *zram, struct bio_vec *bvec,
 583                           u32 index, int offset)
 584 {
 585         int ret;
 586         struct page *page;
 587         unsigned char *user_mem, *uncmem = NULL;
 588         struct zram_meta *meta = zram->meta;
 589         page = bvec->bv_page;
 590
 591         bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
 592         if (unlikely(!meta->table[index].handle) ||
 593                         zram_test_flag(meta, index, ZRAM_ZERO)) {
 594                 bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 595                 handle_zero_page(bvec);
 596                 return 0;
 597         }
 598         bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 599
 600         if (is_partial_io(bvec))
 601                 /* Use  a temporary buffer to decompress the page */
 602                 uncmem = kmalloc(PAGE_SIZE, GFP_NOIO);
 603
 604         user_mem = kmap_atomic(page);
 605         if (!is_partial_io(bvec))
 606                 uncmem = user_mem;
 607
 608         if (!uncmem) {
 609                 pr_info("Unable to allocate temp memory\n");
 610                 ret = -ENOMEM;
 611                 goto out_cleanup;
 612         }
 613
 614         ret = zram_decompress_page(zram, uncmem, index);
 615         /* Should NEVER happen. Return bio error if it does. */
 616         if (unlikely(ret))
 617                 goto out_cleanup;
 618
 619         if (is_partial_io(bvec))
 620                 memcpy(user_mem + bvec->bv_offset, uncmem + offset,
 621                                 bvec->bv_len);
 622
 623         flush_dcache_page(page);
 624         ret = 0;
 625 out_cleanup:
 626         kunmap_atomic(user_mem);
 627         if (is_partial_io(bvec))
 628                 kfree(uncmem);
 629         return ret;
 630 }
 631
 632 static int zram_bvec_write(struct zram *zram, struct bio_vec *bvec, u32 index,
 633                            int offset)
 634 {
 635         int ret = 0;
 636         size_t clen;
 637         unsigned long handle;
 638         struct page *page;
 639         unsigned char *user_mem, *cmem, *src, *uncmem = NULL;
 640         struct zram_meta *meta = zram->meta;
 641         struct zcomp_strm *zstrm;
 642         bool locked = false;
 643         unsigned long alloced_pages;
 644
 645         page = bvec->bv_page;
 646         if (is_partial_io(bvec)) {
 647                 /*
 648                  * This is a partial IO. We need to read the full page
 649                  * before to write the changes.
 650                  */
 651                 uncmem = kmalloc(PAGE_SIZE, GFP_NOIO);
 652                 if (!uncmem) {
 653                         ret = -ENOMEM;
 654                         goto out;
 655                 }
 656                 ret = zram_decompress_page(zram, uncmem, index);
 657                 if (ret)
 658                         goto out;
 659         }
 660
 661         zstrm = zcomp_strm_find(zram->comp);
 662         locked = true;
 663         user_mem = kmap_atomic(page);
 664
 665         if (is_partial_io(bvec)) {
 666                 memcpy(uncmem + offset, user_mem + bvec->bv_offset,
 667                        bvec->bv_len);
 668                 kunmap_atomic(user_mem);
 669                 user_mem = NULL;
 670         } else {
 671                 uncmem = user_mem;
 672         }
 673
 674         if (page_zero_filled(uncmem)) {
 675                 if (user_mem)
 676                         kunmap_atomic(user_mem);
 677                 /* Free memory associated with this sector now. */
 678                 bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
 679                 zram_free_page(zram, index);
 680                 zram_set_flag(meta, index, ZRAM_ZERO);
 681                 bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 682
 683                 atomic64_inc(&zram->stats.zero_pages);
 684                 ret = 0;
 685                 goto out;
 686         }
 687
 688         ret = zcomp_compress(zram->comp, zstrm, uncmem, &clen);
 689         if (!is_partial_io(bvec)) {
 690                 kunmap_atomic(user_mem);
 691                 user_mem = NULL;
 692                 uncmem = NULL;
 693         }
 694
 695         if (unlikely(ret)) {
 696                 pr_err("Compression failed! err=%d\n", ret);
 697                 goto out;
 698         }
 699         src = zstrm->buffer;
 700         if (unlikely(clen > max_zpage_size)) {
 701                 clen = PAGE_SIZE;
 702                 if (is_partial_io(bvec))
 703                         src = uncmem;
 704         }
 705
 706         handle = zs_malloc(meta->mem_pool, clen);
 707         if (!handle) {
 708                 pr_info("Error allocating memory for compressed page: %u, size=%zu\n",
 709                         index, clen);
 710                 ret = -ENOMEM;
 711                 goto out;
 712         }
 713
 714         alloced_pages = zs_get_total_pages(meta->mem_pool);
 715         if (zram->limit_pages && alloced_pages > zram->limit_pages) {
 716                 zs_free(meta->mem_pool, handle);
 717                 ret = -ENOMEM;
 718                 goto out;
 719         }
 720
 721         update_used_max(zram, alloced_pages);
 722
 723         cmem = zs_map_object(meta->mem_pool, handle, ZS_MM_WO);
 724
 725         if ((clen == PAGE_SIZE) && !is_partial_io(bvec)) {
 726                 src = kmap_atomic(page);
 727                 copy_page(cmem, src);
 728                 kunmap_atomic(src);
 729         } else {
 730                 memcpy(cmem, src, clen);
 731         }
 732
 733         zcomp_strm_release(zram->comp, zstrm);
 734         locked = false;
 735         zs_unmap_object(meta->mem_pool, handle);
 736
 737         /*
 738          * Free memory associated with this sector
 739          * before overwriting unused sectors.
 740          */
 741         bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
 742         zram_free_page(zram, index);
 743
 744         meta->table[index].handle = handle;
 745         zram_set_obj_size(meta, index, clen);
 746         bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 747
 748         /* Update stats */
 749         atomic64_add(clen, &zram->stats.compr_data_size);
 750         atomic64_inc(&zram->stats.pages_stored);
 751 out:
 752         if (locked)
 753                 zcomp_strm_release(zram->comp, zstrm);
 754         if (is_partial_io(bvec))
 755                 kfree(uncmem);
 756         return ret;
 757 }
 758
 759 /*
 760  * zram_bio_discard - handler on discard request
 761  * @index: physical block index in PAGE_SIZE units
 762  * @offset: byte offset within physical block
 763  */
 764 static void zram_bio_discard(struct zram *zram, u32 index,
 765                              int offset, struct bio *bio)
 766 {
 767         size_t n = bio->bi_iter.bi_size;
 768         struct zram_meta *meta = zram->meta;
 769
 770         /*
 771          * zram manages data in physical block size units. Because logical block
 772          * size isn't identical with physical block size on some arch, we
 773          * could get a discard request pointing to a specific offset within a
 774          * certain physical block.  Although we can handle this request by
 775          * reading that physiclal block and decompressing and partially zeroing
 776          * and re-compressing and then re-storing it, this isn't reasonable
 777          * because our intent with a discard request is to save memory.  So
 778          * skipping this logical block is appropriate here.
 779          */
 780         if (offset) {
 781                 if (n <= (PAGE_SIZE - offset))
 782                         return;
 783
 784                 n -= (PAGE_SIZE - offset);
 785                 index++;
 786         }
 787
 788         while (n >= PAGE_SIZE) {
 789                 bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
 790                 zram_free_page(zram, index);
 791                 bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 792                 atomic64_inc(&zram->stats.notify_free);
 793                 index++;
 794                 n -= PAGE_SIZE;
 795         }
 796 }
 797
 798 static int zram_bvec_rw(struct zram *zram, struct bio_vec *bvec, u32 index,
 799                         int offset, int rw)
 800 {
 801         unsigned long start_time = jiffies;
 802         int ret;
 803
 804         generic_start_io_acct(rw, bvec->bv_len >> SECTOR_SHIFT,
 805                         &zram->disk->part0);
 806
 807         if (rw == READ) {
 808                 atomic64_inc(&zram->stats.num_reads);
 809                 ret = zram_bvec_read(zram, bvec, index, offset);
 810         } else {
 811                 atomic64_inc(&zram->stats.num_writes);
 812                 ret = zram_bvec_write(zram, bvec, index, offset);
 813         }
 814
 815         generic_end_io_acct(rw, &zram->disk->part0, start_time);
 816
 817         if (unlikely(ret)) {
 818                 if (rw == READ)
 819                         atomic64_inc(&zram->stats.failed_reads);
 820                 else
 821                         atomic64_inc(&zram->stats.failed_writes);
 822         }
 823
 824         return ret;
 825 }
 826
 827 static void __zram_make_request(struct zram *zram, struct bio *bio)
 828 {
 829         int offset, rw;
 830         u32 index;
 831         struct bio_vec bvec;
 832         struct bvec_iter iter;
 833
 834         index = bio->bi_iter.bi_sector >> SECTORS_PER_PAGE_SHIFT;
 835         offset = (bio->bi_iter.bi_sector &
 836                   (SECTORS_PER_PAGE - 1)) << SECTOR_SHIFT;
 837
 838         if (unlikely(bio->bi_rw & REQ_DISCARD)) {
 839                 zram_bio_discard(zram, index, offset, bio);
 840                 bio_endio(bio, 0);
 841                 return;
 842         }
 843
 844         rw = bio_data_dir(bio);
 845         bio_for_each_segment(bvec, bio, iter) {
 846                 int max_transfer_size = PAGE_SIZE - offset;
 847
 848                 if (bvec.bv_len > max_transfer_size) {
 849                         /*
 850                          * zram_bvec_rw() can only make operation on a single
 851                          * zram page. Split the bio vector.
 852                          */
 853                         struct bio_vec bv;
 854
 855                         bv.bv_page = bvec.bv_page;
 856                         bv.bv_len = max_transfer_size;
 857                         bv.bv_offset = bvec.bv_offset;
 858
 859                         if (zram_bvec_rw(zram, &bv, index, offset, rw) < 0)
 860                                 goto out;
 861
 862                         bv.bv_len = bvec.bv_len - max_transfer_size;
 863                         bv.bv_offset += max_transfer_size;
 864                         if (zram_bvec_rw(zram, &bv, index + 1, 0, rw) < 0)
 865                                 goto out;
 866                 } else
 867                         if (zram_bvec_rw(zram, &bvec, index, offset, rw) < 0)
 868                                 goto out;
 869
 870                 update_position(&index, &offset, &bvec);
 871         }
 872
 873         set_bit(BIO_UPTODATE, &bio->bi_flags);
 874         bio_endio(bio, 0);
 875         return;
 876
 877 out:
 878         bio_io_error(bio);
 879 }
 880
 881 /*
 882  * Handler function for all zram I/O requests.
 883  */
 884 static void zram_make_request(struct request_queue *queue, struct bio *bio)
 885 {
 886         struct zram *zram = queue->queuedata;
 887
 888         if (unlikely(!zram_meta_get(zram)))
 889                 goto error;
 890
 891         if (!valid_io_request(zram, bio->bi_iter.bi_sector,
 892                                         bio->bi_iter.bi_size)) {
 893                 atomic64_inc(&zram->stats.invalid_io);
 894                 goto put_zram;
 895         }
 896
 897         __zram_make_request(zram, bio);
 898         zram_meta_put(zram);
 899         return;
 900 put_zram:
 901         zram_meta_put(zram);
 902 error:
 903         bio_io_error(bio);
 904 }
 905
 906 static void zram_slot_free_notify(struct block_device *bdev,
 907                                 unsigned long index)
 908 {
 909         struct zram *zram;
 910         struct zram_meta *meta;
 911
 912         zram = bdev->bd_disk->private_data;
 913         meta = zram->meta;
 914
 915         bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
 916         zram_free_page(zram, index);
 917         bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 918         atomic64_inc(&zram->stats.notify_free);
 919 }
 920
 921 static int zram_rw_page(struct block_device *bdev, sector_t sector,
 922                        struct page *page, int rw)
 923 {
 924         int offset, err = -EIO;
 925         u32 index;
 926         struct zram *zram;
 927         struct bio_vec bv;
 928
 929         zram = bdev->bd_disk->private_data;
 930         if (unlikely(!zram_meta_get(zram)))
 931                 goto out;
 932
 933         if (!valid_io_request(zram, sector, PAGE_SIZE)) {
 934                 atomic64_inc(&zram->stats.invalid_io);
 935                 err = -EINVAL;
 936                 goto put_zram;
 937         }
 938
 939         index = sector >> SECTORS_PER_PAGE_SHIFT;
 940         offset = sector & (SECTORS_PER_PAGE - 1) << SECTOR_SHIFT;
 941
 942         bv.bv_page = page;
 943         bv.bv_len = PAGE_SIZE;
 944         bv.bv_offset = 0;
 945
 946         err = zram_bvec_rw(zram, &bv, index, offset, rw);
 947 put_zram:
 948         zram_meta_put(zram);
 949 out:
 950         /*
 951          * If I/O fails, just return error(ie, non-zero) without
 952          * calling page_endio.
 953          * It causes resubmit the I/O with bio request by upper functions
 954          * of rw_page(e.g., swap_readpage, __swap_writepage) and
 955          * bio->bi_end_io does things to handle the error
 956          * (e.g., SetPageError, set_page_dirty and extra works).
 957          */
 958         if (err == 0)
 959                 page_endio(page, rw, 0);
 960         return err;
 961 }
 962
 963 static void zram_reset_device(struct zram *zram)
 964 {
 965         struct zram_meta *meta;
 966         struct zcomp *comp;
 967         u64 disksize;
 968
 969         down_write(&zram->init_lock);
 970
 971         zram->limit_pages = 0;
 972
 973         if (!init_done(zram)) {
 974                 up_write(&zram->init_lock);
 975                 return;
 976         }
 977
 978         meta = zram->meta;
 979         comp = zram->comp;
 980         disksize = zram->disksize;
 981         /*
 982          * Refcount will go down to 0 eventually and r/w handler
 983          * cannot handle further I/O so it will bail out by
 984          * check zram_meta_get.
 985          */
 986         zram_meta_put(zram);
 987         /*
 988          * We want to free zram_meta in process context to avoid
 989          * deadlock between reclaim path and any other locks.
 990          */
 991         wait_event(zram->io_done, atomic_read(&zram->refcount) == 0);
 992
 993         /* Reset stats */
 994         memset(&zram->stats, 0, sizeof(zram->stats));
 995         zram->disksize = 0;
 996         zram->max_comp_streams = 1;
 997
 998         set_capacity(zram->disk, 0);
 999         part_stat_set_all(&zram->disk->part0, 0);
1000
1001         up_write(&zram->init_lock);
1002         /* I/O operation under all of CPU are done so let's free */
1003         zram_meta_free(meta, disksize);
1004         zcomp_destroy(comp);
1005 }
1006
1007 static ssize_t disksize_store(struct device *dev,
1008                 struct device_attribute *attr, const char *buf, size_t len)
1009 {
1010         u64 disksize;
1011         struct zcomp *comp;
1012         struct zram_meta *meta;
1013         struct zram *zram = dev_to_zram(dev);
1014         int err;
1015
1016         disksize = memparse(buf, NULL);
1017         if (!disksize)
1018                 return -EINVAL;
1019
1020         disksize = PAGE_ALIGN(disksize);
1021         meta = zram_meta_alloc(zram->disk->first_minor, disksize);
1022         if (!meta)
1023                 return -ENOMEM;
1024
1025         comp = zcomp_create(zram->compressor, zram->max_comp_streams);
1026         if (IS_ERR(comp)) {
1027                 pr_info("Cannot initialise %s compressing backend\n",
1028                                 zram->compressor);
1029                 err = PTR_ERR(comp);
1030                 goto out_free_meta;
1031         }
1032
1033         down_write(&zram->init_lock);
1034         if (init_done(zram)) {
1035                 pr_info("Cannot change disksize for initialized device\n");
1036                 err = -EBUSY;
1037                 goto out_destroy_comp;
1038         }
1039
1040         init_waitqueue_head(&zram->io_done);
1041         atomic_set(&zram->refcount, 1);
1042         zram->meta = meta;
1043         zram->comp = comp;
1044         zram->disksize = disksize;
1045         set_capacity(zram->disk, zram->disksize >> SECTOR_SHIFT);
1046         up_write(&zram->init_lock);
1047
1048         /*
1049          * Revalidate disk out of the init_lock to avoid lockdep splat.
1050          * It's okay because disk's capacity is protected by init_lock
1051          * so that revalidate_disk always sees up-to-date capacity.
1052          */
1053         revalidate_disk(zram->disk);
1054
1055         return len;
1056
1057 out_destroy_comp:
1058         up_write(&zram->init_lock);
1059         zcomp_destroy(comp);
1060 out_free_meta:
1061         zram_meta_free(meta, disksize);
1062         return err;
1063 }
1064
1065 static ssize_t reset_store(struct device *dev,
1066                 struct device_attribute *attr, const char *buf, size_t len)
1067 {
1068         int ret;
1069         unsigned short do_reset;
1070         struct zram *zram;
1071         struct block_device *bdev;
1072
1073         zram = dev_to_zram(dev);
1074         bdev = bdget_disk(zram->disk, 0);
1075
1076         if (!bdev)
1077                 return -ENOMEM;
1078
1079         mutex_lock(&bdev->bd_mutex);
1080         /* Do not reset an active device! */
1081         if (bdev->bd_openers) {
1082                 ret = -EBUSY;
1083                 goto out;
1084         }
1085
1086         ret = kstrtou16(buf, 10, &do_reset);
1087         if (ret)
1088                 goto out;
1089
1090         if (!do_reset) {
1091                 ret = -EINVAL;
1092                 goto out;
1093         }
1094
1095         /* Make sure all pending I/O is finished */
1096         fsync_bdev(bdev);
1097         zram_reset_device(zram);
1098
1099         mutex_unlock(&bdev->bd_mutex);
1100         revalidate_disk(zram->disk);
1101         bdput(bdev);
1102
1103         return len;
1104
1105 out:
1106         mutex_unlock(&bdev->bd_mutex);
1107         bdput(bdev);
1108         return ret;
1109 }
1110
1111 static const struct block_device_operations zram_devops = {
1112         .swap_slot_free_notify = zram_slot_free_notify,
1113         .rw_page = zram_rw_page,
1114         .owner = THIS_MODULE
1115 };
1116
1117 static DEVICE_ATTR_WO(compact);
1118 static DEVICE_ATTR_RW(disksize);
1119 static DEVICE_ATTR_RO(initstate);
1120 static DEVICE_ATTR_WO(reset);
1121 static DEVICE_ATTR_RO(orig_data_size);
1122 static DEVICE_ATTR_RO(mem_used_total);
1123 static DEVICE_ATTR_RW(mem_limit);
1124 static DEVICE_ATTR_RW(mem_used_max);
1125 static DEVICE_ATTR_RW(max_comp_streams);
1126 static DEVICE_ATTR_RW(comp_algorithm);
1127
1128 static struct attribute *zram_disk_attrs[] = {
1129         &dev_attr_disksize.attr,
1130         &dev_attr_initstate.attr,
1131         &dev_attr_reset.attr,
1132         &dev_attr_num_reads.attr,
1133         &dev_attr_num_writes.attr,
1134         &dev_attr_failed_reads.attr,
1135         &dev_attr_failed_writes.attr,
1136         &dev_attr_compact.attr,
1137         &dev_attr_invalid_io.attr,
1138         &dev_attr_notify_free.attr,
1139         &dev_attr_zero_pages.attr,
1140         &dev_attr_orig_data_size.attr,
1141         &dev_attr_compr_data_size.attr,
1142         &dev_attr_mem_used_total.attr,
1143         &dev_attr_mem_limit.attr,
1144         &dev_attr_mem_used_max.attr,
1145         &dev_attr_max_comp_streams.attr,
1146         &dev_attr_comp_algorithm.attr,
1147         &dev_attr_io_stat.attr,
1148         &dev_attr_mm_stat.attr,
1149         NULL,
1150 };
1151
1152 static struct attribute_group zram_disk_attr_group = {
1153         .attrs = zram_disk_attrs,
1154 };
1155
1156 static int zram_add(int device_id)
1157 {
1158         struct zram *zram;
1159         struct request_queue *queue;
1160         int ret;
1161
1162         zram = kzalloc(sizeof(struct zram), GFP_KERNEL);
1163         if (!zram)
1164                 return -ENOMEM;
1165
1166         ret = idr_alloc(&zram_index_idr, zram, device_id,
1167                         device_id + 1, GFP_KERNEL);
1168         if (ret < 0)
1169                 goto out_free_dev;
1170
1171         init_rwsem(&zram->init_lock);
1172
1173         queue = blk_alloc_queue(GFP_KERNEL);
1174         if (!queue) {
1175                 pr_err("Error allocating disk queue for device %d\n",
1176                         device_id);
1177                 ret = -ENOMEM;
1178                 goto out_free_idr;
1179         }
1180
1181         blk_queue_make_request(queue, zram_make_request);
1182
1183         /* gendisk structure */
1184         zram->disk = alloc_disk(1);
1185         if (!zram->disk) {
1186                 pr_warn("Error allocating disk structure for device %d\n",
1187                         device_id);
1188                 ret = -ENOMEM;
1189                 goto out_free_queue;
1190         }
1191
1192         zram->disk->major = zram_major;
1193         zram->disk->first_minor = device_id;
1194         zram->disk->fops = &zram_devops;
1195         zram->disk->queue = queue;
1196         zram->disk->queue->queuedata = zram;
1197         zram->disk->private_data = zram;
1198         snprintf(zram->disk->disk_name, 16, "zram%d", device_id);
1199
1200         /* Actual capacity set using syfs (/sys/block/zram<id>/disksize */
1201         set_capacity(zram->disk, 0);
1202         /* zram devices sort of resembles non-rotational disks */
1203         queue_flag_set_unlocked(QUEUE_FLAG_NONROT, zram->disk->queue);
1204         queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, zram->disk->queue);
1205         /*
1206          * To ensure that we always get PAGE_SIZE aligned
1207          * and n*PAGE_SIZED sized I/O requests.
1208          */
1209         blk_queue_physical_block_size(zram->disk->queue, PAGE_SIZE);
1210         blk_queue_logical_block_size(zram->disk->queue,
1211                                         ZRAM_LOGICAL_BLOCK_SIZE);
1212         blk_queue_io_min(zram->disk->queue, PAGE_SIZE);
1213         blk_queue_io_opt(zram->disk->queue, PAGE_SIZE);
1214         zram->disk->queue->limits.discard_granularity = PAGE_SIZE;
1215         zram->disk->queue->limits.max_discard_sectors = UINT_MAX;
1216         /*
1217          * zram_bio_discard() will clear all logical blocks if logical block
1218          * size is identical with physical block size(PAGE_SIZE). But if it is
1219          * different, we will skip discarding some parts of logical blocks in
1220          * the part of the request range which isn't aligned to physical block
1221          * size.  So we can't ensure that all discarded logical blocks are
1222          * zeroed.
1223          */
1224         if (ZRAM_LOGICAL_BLOCK_SIZE == PAGE_SIZE)
1225                 zram->disk->queue->limits.discard_zeroes_data = 1;
1226         else
1227                 zram->disk->queue->limits.discard_zeroes_data = 0;
1228         queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, zram->disk->queue);
1229
1230         add_disk(zram->disk);
1231
1232         ret = sysfs_create_group(&disk_to_dev(zram->disk)->kobj,
1233                                 &zram_disk_attr_group);
1234         if (ret < 0) {
1235                 pr_warn("Error creating sysfs group");
1236                 goto out_free_disk;
1237         }
1238         strlcpy(zram->compressor, default_compressor, sizeof(zram->compressor));
1239         zram->meta = NULL;
1240         zram->max_comp_streams = 1;
1241         return 0;
1242
1243 out_free_disk:
1244         del_gendisk(zram->disk);
1245         put_disk(zram->disk);
1246 out_free_queue:
1247         blk_cleanup_queue(queue);
1248 out_free_idr:
1249         idr_remove(&zram_index_idr, device_id);
1250 out_free_dev:
1251         kfree(zram);
1252         return ret;
1253 }
1254
1255 static void zram_remove(struct zram *zram)
1256 {
1257         /*
1258          * Remove sysfs first, so no one will perform a disksize
1259          * store while we destroy the devices
1260          */
1261         sysfs_remove_group(&disk_to_dev(zram->disk)->kobj,
1262                         &zram_disk_attr_group);
1263
1264         zram_reset_device(zram);
1265         idr_remove(&zram_index_idr, zram->disk->first_minor);
1266         blk_cleanup_queue(zram->disk->queue);
1267         del_gendisk(zram->disk);
1268         put_disk(zram->disk);
1269         kfree(zram);
1270 }
1271
1272 static int zram_remove_cb(int id, void *ptr, void *data)
1273 {
1274         zram_remove(ptr);
1275         return 0;
1276 }
1277
1278 static void destroy_devices(void)
1279 {
1280         idr_for_each(&zram_index_idr, &zram_remove_cb, NULL);
1281         idr_destroy(&zram_index_idr);
1282         unregister_blkdev(zram_major, "zram");
1283         pr_info("Destroyed device(s)\n");
1284 }
1285
1286 static int __init zram_init(void)
1287 {
1288         int ret, dev_id;
1289
1290         zram_major = register_blkdev(0, "zram");
1291         if (zram_major <= 0) {
1292                 pr_warn("Unable to get major number\n");
1293                 return -EBUSY;
1294         }
1295
1296         for (dev_id = 0; dev_id < num_devices; dev_id++) {
1297                 ret = zram_add(dev_id);
1298                 if (ret != 0)
1299                         goto out_error;
1300         }
1301
1302         pr_info("Created %u device(s)\n", num_devices);
1303         return 0;
1304
1305 out_error:
1306         destroy_devices();
1307         return ret;
1308 }
1309
1310 static void __exit zram_exit(void)
1311 {
1312         destroy_devices();
1313 }
1314
1315 module_init(zram_init);
1316 module_exit(zram_exit);
1317
1318 module_param(num_devices, uint, 0);
1319 MODULE_PARM_DESC(num_devices, "Number of pre-created zram devices");
1320
1321 MODULE_LICENSE("Dual BSD/GPL");
1322 MODULE_AUTHOR("Nitin Gupta <ngupta@vflare.org>");
1323 MODULE_DESCRIPTION("Compressed RAM Block Device");