drivers/block/zram/zram_drv.c

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