kernel/drivers/block/zram/zram_drv.c

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