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 zram_attr_##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 struct device_attribute dev_attr_##name =                        \
  55         __ATTR(name, S_IRUGO, zram_attr_##name##_show, NULL);
  56
  57 static inline int init_done(struct zram *zram)
  58 {
  59         return zram->meta != NULL;
  60 }
  61
  62 static inline struct zram *dev_to_zram(struct device *dev)
  63 {
  64         return (struct zram *)dev_to_disk(dev)->private_data;
  65 }
  66
  67 static ssize_t disksize_show(struct device *dev,
  68                 struct device_attribute *attr, char *buf)
  69 {
  70         struct zram *zram = dev_to_zram(dev);
  71
  72         return scnprintf(buf, PAGE_SIZE, "%llu\n", zram->disksize);
  73 }
  74
  75 static ssize_t initstate_show(struct device *dev,
  76                 struct device_attribute *attr, char *buf)
  77 {
  78         u32 val;
  79         struct zram *zram = dev_to_zram(dev);
  80
  81         down_read(&zram->init_lock);
  82         val = init_done(zram);
  83         up_read(&zram->init_lock);
  84
  85         return scnprintf(buf, PAGE_SIZE, "%u\n", val);
  86 }
  87
  88 static ssize_t orig_data_size_show(struct device *dev,
  89                 struct device_attribute *attr, char *buf)
  90 {
  91         struct zram *zram = dev_to_zram(dev);
  92
  93         return scnprintf(buf, PAGE_SIZE, "%llu\n",
  94                 (u64)(atomic64_read(&zram->stats.pages_stored)) << PAGE_SHIFT);
  95 }
  96
  97 static ssize_t mem_used_total_show(struct device *dev,
  98                 struct device_attribute *attr, char *buf)
  99 {
 100         u64 val = 0;
 101         struct zram *zram = dev_to_zram(dev);
 102         struct zram_meta *meta = zram->meta;
 103
 104         down_read(&zram->init_lock);
 105         if (init_done(zram))
 106                 val = zs_get_total_size_bytes(meta->mem_pool);
 107         up_read(&zram->init_lock);
 108
 109         return scnprintf(buf, PAGE_SIZE, "%llu\n", val);
 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 max_comp_streams_store(struct device *dev,
 126                 struct device_attribute *attr, const char *buf, size_t len)
 127 {
 128         int num;
 129         struct zram *zram = dev_to_zram(dev);
 130         int ret;
 131
 132         ret = kstrtoint(buf, 0, &num);
 133         if (ret < 0)
 134                 return ret;
 135         if (num < 1)
 136                 return -EINVAL;
 137
 138         down_write(&zram->init_lock);
 139         if (init_done(zram)) {
 140                 if (!zcomp_set_max_streams(zram->comp, num)) {
 141                         pr_info("Cannot change max compression streams\n");
 142                         ret = -EINVAL;
 143                         goto out;
 144                 }
 145         }
 146
 147         zram->max_comp_streams = num;
 148         ret = len;
 149 out:
 150         up_write(&zram->init_lock);
 151         return ret;
 152 }
 153
 154 static ssize_t comp_algorithm_show(struct device *dev,
 155                 struct device_attribute *attr, char *buf)
 156 {
 157         size_t sz;
 158         struct zram *zram = dev_to_zram(dev);
 159
 160         down_read(&zram->init_lock);
 161         sz = zcomp_available_show(zram->compressor, buf);
 162         up_read(&zram->init_lock);
 163
 164         return sz;
 165 }
 166
 167 static ssize_t comp_algorithm_store(struct device *dev,
 168                 struct device_attribute *attr, const char *buf, size_t len)
 169 {
 170         struct zram *zram = dev_to_zram(dev);
 171         down_write(&zram->init_lock);
 172         if (init_done(zram)) {
 173                 up_write(&zram->init_lock);
 174                 pr_info("Can't change algorithm for initialized device\n");
 175                 return -EBUSY;
 176         }
 177         strlcpy(zram->compressor, buf, sizeof(zram->compressor));
 178         up_write(&zram->init_lock);
 179         return len;
 180 }
 181
 182 /* flag operations needs meta->tb_lock */
 183 static int zram_test_flag(struct zram_meta *meta, u32 index,
 184                         enum zram_pageflags flag)
 185 {
 186         return meta->table[index].flags & BIT(flag);
 187 }
 188
 189 static void zram_set_flag(struct zram_meta *meta, u32 index,
 190                         enum zram_pageflags flag)
 191 {
 192         meta->table[index].flags |= BIT(flag);
 193 }
 194
 195 static void zram_clear_flag(struct zram_meta *meta, u32 index,
 196                         enum zram_pageflags flag)
 197 {
 198         meta->table[index].flags &= ~BIT(flag);
 199 }
 200
 201 static inline int is_partial_io(struct bio_vec *bvec)
 202 {
 203         return bvec->bv_len != PAGE_SIZE;
 204 }
 205
 206 /*
 207  * Check if request is within bounds and aligned on zram logical blocks.
 208  */
 209 static inline int valid_io_request(struct zram *zram, struct bio *bio)
 210 {
 211         u64 start, end, bound;
 212
 213         /* unaligned request */
 214         if (unlikely(bio->bi_iter.bi_sector &
 215                      (ZRAM_SECTOR_PER_LOGICAL_BLOCK - 1)))
 216                 return 0;
 217         if (unlikely(bio->bi_iter.bi_size & (ZRAM_LOGICAL_BLOCK_SIZE - 1)))
 218                 return 0;
 219
 220         start = bio->bi_iter.bi_sector;
 221         end = start + (bio->bi_iter.bi_size >> SECTOR_SHIFT);
 222         bound = zram->disksize >> SECTOR_SHIFT;
 223         /* out of range range */
 224         if (unlikely(start >= bound || end > bound || start > end))
 225                 return 0;
 226
 227         /* I/O request is valid */
 228         return 1;
 229 }
 230
 231 static void zram_meta_free(struct zram_meta *meta)
 232 {
 233         zs_destroy_pool(meta->mem_pool);
 234         vfree(meta->table);
 235         kfree(meta);
 236 }
 237
 238 static struct zram_meta *zram_meta_alloc(u64 disksize)
 239 {
 240         size_t num_pages;
 241         struct zram_meta *meta = kmalloc(sizeof(*meta), GFP_KERNEL);
 242         if (!meta)
 243                 goto out;
 244
 245         num_pages = disksize >> PAGE_SHIFT;
 246         meta->table = vzalloc(num_pages * sizeof(*meta->table));
 247         if (!meta->table) {
 248                 pr_err("Error allocating zram address table\n");
 249                 goto free_meta;
 250         }
 251
 252         meta->mem_pool = zs_create_pool(GFP_NOIO | __GFP_HIGHMEM);
 253         if (!meta->mem_pool) {
 254                 pr_err("Error creating memory pool\n");
 255                 goto free_table;
 256         }
 257
 258         rwlock_init(&meta->tb_lock);
 259         return meta;
 260
 261 free_table:
 262         vfree(meta->table);
 263 free_meta:
 264         kfree(meta);
 265         meta = NULL;
 266 out:
 267         return meta;
 268 }
 269
 270 static void update_position(u32 *index, int *offset, struct bio_vec *bvec)
 271 {
 272         if (*offset + bvec->bv_len >= PAGE_SIZE)
 273                 (*index)++;
 274         *offset = (*offset + bvec->bv_len) % PAGE_SIZE;
 275 }
 276
 277 static int page_zero_filled(void *ptr)
 278 {
 279         unsigned int pos;
 280         unsigned long *page;
 281
 282         page = (unsigned long *)ptr;
 283
 284         for (pos = 0; pos != PAGE_SIZE / sizeof(*page); pos++) {
 285                 if (page[pos])
 286                         return 0;
 287         }
 288
 289         return 1;
 290 }
 291
 292 static void handle_zero_page(struct bio_vec *bvec)
 293 {
 294         struct page *page = bvec->bv_page;
 295         void *user_mem;
 296
 297         user_mem = kmap_atomic(page);
 298         if (is_partial_io(bvec))
 299                 memset(user_mem + bvec->bv_offset, 0, bvec->bv_len);
 300         else
 301                 clear_page(user_mem);
 302         kunmap_atomic(user_mem);
 303
 304         flush_dcache_page(page);
 305 }
 306
 307 /* NOTE: caller should hold meta->tb_lock with write-side */
 308 static void zram_free_page(struct zram *zram, size_t index)
 309 {
 310         struct zram_meta *meta = zram->meta;
 311         unsigned long handle = meta->table[index].handle;
 312
 313         if (unlikely(!handle)) {
 314                 /*
 315                  * No memory is allocated for zero filled pages.
 316                  * Simply clear zero page flag.
 317                  */
 318                 if (zram_test_flag(meta, index, ZRAM_ZERO)) {
 319                         zram_clear_flag(meta, index, ZRAM_ZERO);
 320                         atomic64_dec(&zram->stats.zero_pages);
 321                 }
 322                 return;
 323         }
 324
 325         zs_free(meta->mem_pool, handle);
 326
 327         atomic64_sub(meta->table[index].size, &zram->stats.compr_data_size);
 328         atomic64_dec(&zram->stats.pages_stored);
 329
 330         meta->table[index].handle = 0;
 331         meta->table[index].size = 0;
 332 }
 333
 334 static int zram_decompress_page(struct zram *zram, char *mem, u32 index)
 335 {
 336         int ret = 0;
 337         unsigned char *cmem;
 338         struct zram_meta *meta = zram->meta;
 339         unsigned long handle;
 340         u16 size;
 341
 342         read_lock(&meta->tb_lock);
 343         handle = meta->table[index].handle;
 344         size = meta->table[index].size;
 345
 346         if (!handle || zram_test_flag(meta, index, ZRAM_ZERO)) {
 347                 read_unlock(&meta->tb_lock);
 348                 clear_page(mem);
 349                 return 0;
 350         }
 351
 352         cmem = zs_map_object(meta->mem_pool, handle, ZS_MM_RO);
 353         if (size == PAGE_SIZE)
 354                 copy_page(mem, cmem);
 355         else
 356                 ret = zcomp_decompress(zram->comp, cmem, size, mem);
 357         zs_unmap_object(meta->mem_pool, handle);
 358         read_unlock(&meta->tb_lock);
 359
 360         /* Should NEVER happen. Return bio error if it does. */
 361         if (unlikely(ret)) {
 362                 pr_err("Decompression failed! err=%d, page=%u\n", ret, index);
 363                 atomic64_inc(&zram->stats.failed_reads);
 364                 return ret;
 365         }
 366
 367         return 0;
 368 }
 369
 370 static int zram_bvec_read(struct zram *zram, struct bio_vec *bvec,
 371                           u32 index, int offset, struct bio *bio)
 372 {
 373         int ret;
 374         struct page *page;
 375         unsigned char *user_mem, *uncmem = NULL;
 376         struct zram_meta *meta = zram->meta;
 377         page = bvec->bv_page;
 378
 379         read_lock(&meta->tb_lock);
 380         if (unlikely(!meta->table[index].handle) ||
 381                         zram_test_flag(meta, index, ZRAM_ZERO)) {
 382                 read_unlock(&meta->tb_lock);
 383                 handle_zero_page(bvec);
 384                 return 0;
 385         }
 386         read_unlock(&meta->tb_lock);
 387
 388         if (is_partial_io(bvec))
 389                 /* Use  a temporary buffer to decompress the page */
 390                 uncmem = kmalloc(PAGE_SIZE, GFP_NOIO);
 391
 392         user_mem = kmap_atomic(page);
 393         if (!is_partial_io(bvec))
 394                 uncmem = user_mem;
 395
 396         if (!uncmem) {
 397                 pr_info("Unable to allocate temp memory\n");
 398                 ret = -ENOMEM;
 399                 goto out_cleanup;
 400         }
 401
 402         ret = zram_decompress_page(zram, uncmem, index);
 403         /* Should NEVER happen. Return bio error if it does. */
 404         if (unlikely(ret))
 405                 goto out_cleanup;
 406
 407         if (is_partial_io(bvec))
 408                 memcpy(user_mem + bvec->bv_offset, uncmem + offset,
 409                                 bvec->bv_len);
 410
 411         flush_dcache_page(page);
 412         ret = 0;
 413 out_cleanup:
 414         kunmap_atomic(user_mem);
 415         if (is_partial_io(bvec))
 416                 kfree(uncmem);
 417         return ret;
 418 }
 419
 420 static int zram_bvec_write(struct zram *zram, struct bio_vec *bvec, u32 index,
 421                            int offset)
 422 {
 423         int ret = 0;
 424         size_t clen;
 425         unsigned long handle;
 426         struct page *page;
 427         unsigned char *user_mem, *cmem, *src, *uncmem = NULL;
 428         struct zram_meta *meta = zram->meta;
 429         struct zcomp_strm *zstrm;
 430         bool locked = false;
 431
 432         page = bvec->bv_page;
 433         if (is_partial_io(bvec)) {
 434                 /*
 435                  * This is a partial IO. We need to read the full page
 436                  * before to write the changes.
 437                  */
 438                 uncmem = kmalloc(PAGE_SIZE, GFP_NOIO);
 439                 if (!uncmem) {
 440                         ret = -ENOMEM;
 441                         goto out;
 442                 }
 443                 ret = zram_decompress_page(zram, uncmem, index);
 444                 if (ret)
 445                         goto out;
 446         }
 447
 448         zstrm = zcomp_strm_find(zram->comp);
 449         locked = true;
 450         user_mem = kmap_atomic(page);
 451
 452         if (is_partial_io(bvec)) {
 453                 memcpy(uncmem + offset, user_mem + bvec->bv_offset,
 454                        bvec->bv_len);
 455                 kunmap_atomic(user_mem);
 456                 user_mem = NULL;
 457         } else {
 458                 uncmem = user_mem;
 459         }
 460
 461         if (page_zero_filled(uncmem)) {
 462                 kunmap_atomic(user_mem);
 463                 /* Free memory associated with this sector now. */
 464                 write_lock(&zram->meta->tb_lock);
 465                 zram_free_page(zram, index);
 466                 zram_set_flag(meta, index, ZRAM_ZERO);
 467                 write_unlock(&zram->meta->tb_lock);
 468
 469                 atomic64_inc(&zram->stats.zero_pages);
 470                 ret = 0;
 471                 goto out;
 472         }
 473
 474         ret = zcomp_compress(zram->comp, zstrm, uncmem, &clen);
 475         if (!is_partial_io(bvec)) {
 476                 kunmap_atomic(user_mem);
 477                 user_mem = NULL;
 478                 uncmem = NULL;
 479         }
 480
 481         if (unlikely(ret)) {
 482                 pr_err("Compression failed! err=%d\n", ret);
 483                 goto out;
 484         }
 485         src = zstrm->buffer;
 486         if (unlikely(clen > max_zpage_size)) {
 487                 clen = PAGE_SIZE;
 488                 if (is_partial_io(bvec))
 489                         src = uncmem;
 490         }
 491
 492         handle = zs_malloc(meta->mem_pool, clen);
 493         if (!handle) {
 494                 pr_info("Error allocating memory for compressed page: %u, size=%zu\n",
 495                         index, clen);
 496                 ret = -ENOMEM;
 497                 goto out;
 498         }
 499         cmem = zs_map_object(meta->mem_pool, handle, ZS_MM_WO);
 500
 501         if ((clen == PAGE_SIZE) && !is_partial_io(bvec)) {
 502                 src = kmap_atomic(page);
 503                 copy_page(cmem, src);
 504                 kunmap_atomic(src);
 505         } else {
 506                 memcpy(cmem, src, clen);
 507         }
 508
 509         zcomp_strm_release(zram->comp, zstrm);
 510         locked = false;
 511         zs_unmap_object(meta->mem_pool, handle);
 512
 513         /*
 514          * Free memory associated with this sector
 515          * before overwriting unused sectors.
 516          */
 517         write_lock(&zram->meta->tb_lock);
 518         zram_free_page(zram, index);
 519
 520         meta->table[index].handle = handle;
 521         meta->table[index].size = clen;
 522         write_unlock(&zram->meta->tb_lock);
 523
 524         /* Update stats */
 525         atomic64_add(clen, &zram->stats.compr_data_size);
 526         atomic64_inc(&zram->stats.pages_stored);
 527 out:
 528         if (locked)
 529                 zcomp_strm_release(zram->comp, zstrm);
 530         if (is_partial_io(bvec))
 531                 kfree(uncmem);
 532         if (ret)
 533                 atomic64_inc(&zram->stats.failed_writes);
 534         return ret;
 535 }
 536
 537 static int zram_bvec_rw(struct zram *zram, struct bio_vec *bvec, u32 index,
 538                         int offset, struct bio *bio)
 539 {
 540         int ret;
 541         int rw = bio_data_dir(bio);
 542
 543         if (rw == READ) {
 544                 atomic64_inc(&zram->stats.num_reads);
 545                 ret = zram_bvec_read(zram, bvec, index, offset, bio);
 546         } else {
 547                 atomic64_inc(&zram->stats.num_writes);
 548                 ret = zram_bvec_write(zram, bvec, index, offset);
 549         }
 550
 551         return ret;
 552 }
 553
 554 /*
 555  * zram_bio_discard - handler on discard request
 556  * @index: physical block index in PAGE_SIZE units
 557  * @offset: byte offset within physical block
 558  */
 559 static void zram_bio_discard(struct zram *zram, u32 index,
 560                              int offset, struct bio *bio)
 561 {
 562         size_t n = bio->bi_iter.bi_size;
 563
 564         /*
 565          * zram manages data in physical block size units. Because logical block
 566          * size isn't identical with physical block size on some arch, we
 567          * could get a discard request pointing to a specific offset within a
 568          * certain physical block.  Although we can handle this request by
 569          * reading that physiclal block and decompressing and partially zeroing
 570          * and re-compressing and then re-storing it, this isn't reasonable
 571          * because our intent with a discard request is to save memory.  So
 572          * skipping this logical block is appropriate here.
 573          */
 574         if (offset) {
 575                 if (n <= (PAGE_SIZE - offset))
 576                         return;
 577
 578                 n -= (PAGE_SIZE - offset);
 579                 index++;
 580         }
 581
 582         while (n >= PAGE_SIZE) {
 583                 /*
 584                  * Discard request can be large so the lock hold times could be
 585                  * lengthy.  So take the lock once per page.
 586                  */
 587                 write_lock(&zram->meta->tb_lock);
 588                 zram_free_page(zram, index);
 589                 write_unlock(&zram->meta->tb_lock);
 590                 index++;
 591                 n -= PAGE_SIZE;
 592         }
 593 }
 594
 595 static void zram_reset_device(struct zram *zram, bool reset_capacity)
 596 {
 597         size_t index;
 598         struct zram_meta *meta;
 599
 600         down_write(&zram->init_lock);
 601         if (!init_done(zram)) {
 602                 up_write(&zram->init_lock);
 603                 return;
 604         }
 605
 606         meta = zram->meta;
 607         /* Free all pages that are still in this zram device */
 608         for (index = 0; index < zram->disksize >> PAGE_SHIFT; index++) {
 609                 unsigned long handle = meta->table[index].handle;
 610                 if (!handle)
 611                         continue;
 612
 613                 zs_free(meta->mem_pool, handle);
 614         }
 615
 616         zcomp_destroy(zram->comp);
 617         zram->max_comp_streams = 1;
 618
 619         zram_meta_free(zram->meta);
 620         zram->meta = NULL;
 621         /* Reset stats */
 622         memset(&zram->stats, 0, sizeof(zram->stats));
 623
 624         zram->disksize = 0;
 625         if (reset_capacity) {
 626                 set_capacity(zram->disk, 0);
 627                 revalidate_disk(zram->disk);
 628         }
 629         up_write(&zram->init_lock);
 630 }
 631
 632 static ssize_t disksize_store(struct device *dev,
 633                 struct device_attribute *attr, const char *buf, size_t len)
 634 {
 635         u64 disksize;
 636         struct zcomp *comp;
 637         struct zram_meta *meta;
 638         struct zram *zram = dev_to_zram(dev);
 639         int err;
 640
 641         disksize = memparse(buf, NULL);
 642         if (!disksize)
 643                 return -EINVAL;
 644
 645         disksize = PAGE_ALIGN(disksize);
 646         meta = zram_meta_alloc(disksize);
 647         if (!meta)
 648                 return -ENOMEM;
 649
 650         comp = zcomp_create(zram->compressor, zram->max_comp_streams);
 651         if (IS_ERR(comp)) {
 652                 pr_info("Cannot initialise %s compressing backend\n",
 653                                 zram->compressor);
 654                 err = PTR_ERR(comp);
 655                 goto out_free_meta;
 656         }
 657
 658         down_write(&zram->init_lock);
 659         if (init_done(zram)) {
 660                 pr_info("Cannot change disksize for initialized device\n");
 661                 err = -EBUSY;
 662                 goto out_destroy_comp;
 663         }
 664
 665         zram->meta = meta;
 666         zram->comp = comp;
 667         zram->disksize = disksize;
 668         set_capacity(zram->disk, zram->disksize >> SECTOR_SHIFT);
 669         revalidate_disk(zram->disk);
 670         up_write(&zram->init_lock);
 671         return len;
 672
 673 out_destroy_comp:
 674         up_write(&zram->init_lock);
 675         zcomp_destroy(comp);
 676 out_free_meta:
 677         zram_meta_free(meta);
 678         return err;
 679 }
 680
 681 static ssize_t reset_store(struct device *dev,
 682                 struct device_attribute *attr, const char *buf, size_t len)
 683 {
 684         int ret;
 685         unsigned short do_reset;
 686         struct zram *zram;
 687         struct block_device *bdev;
 688
 689         zram = dev_to_zram(dev);
 690         bdev = bdget_disk(zram->disk, 0);
 691
 692         if (!bdev)
 693                 return -ENOMEM;
 694
 695         /* Do not reset an active device! */
 696         if (bdev->bd_holders) {
 697                 ret = -EBUSY;
 698                 goto out;
 699         }
 700
 701         ret = kstrtou16(buf, 10, &do_reset);
 702         if (ret)
 703                 goto out;
 704
 705         if (!do_reset) {
 706                 ret = -EINVAL;
 707                 goto out;
 708         }
 709
 710         /* Make sure all pending I/O is finished */
 711         fsync_bdev(bdev);
 712         bdput(bdev);
 713
 714         zram_reset_device(zram, true);
 715         return len;
 716
 717 out:
 718         bdput(bdev);
 719         return ret;
 720 }
 721
 722 static void __zram_make_request(struct zram *zram, struct bio *bio)
 723 {
 724         int offset;
 725         u32 index;
 726         struct bio_vec bvec;
 727         struct bvec_iter iter;
 728
 729         index = bio->bi_iter.bi_sector >> SECTORS_PER_PAGE_SHIFT;
 730         offset = (bio->bi_iter.bi_sector &
 731                   (SECTORS_PER_PAGE - 1)) << SECTOR_SHIFT;
 732
 733         if (unlikely(bio->bi_rw & REQ_DISCARD)) {
 734                 zram_bio_discard(zram, index, offset, bio);
 735                 bio_endio(bio, 0);
 736                 return;
 737         }
 738
 739         bio_for_each_segment(bvec, bio, iter) {
 740                 int max_transfer_size = PAGE_SIZE - offset;
 741
 742                 if (bvec.bv_len > max_transfer_size) {
 743                         /*
 744                          * zram_bvec_rw() can only make operation on a single
 745                          * zram page. Split the bio vector.
 746                          */
 747                         struct bio_vec bv;
 748
 749                         bv.bv_page = bvec.bv_page;
 750                         bv.bv_len = max_transfer_size;
 751                         bv.bv_offset = bvec.bv_offset;
 752
 753                         if (zram_bvec_rw(zram, &bv, index, offset, bio) < 0)
 754                                 goto out;
 755
 756                         bv.bv_len = bvec.bv_len - max_transfer_size;
 757                         bv.bv_offset += max_transfer_size;
 758                         if (zram_bvec_rw(zram, &bv, index + 1, 0, bio) < 0)
 759                                 goto out;
 760                 } else
 761                         if (zram_bvec_rw(zram, &bvec, index, offset, bio) < 0)
 762                                 goto out;
 763
 764                 update_position(&index, &offset, &bvec);
 765         }
 766
 767         set_bit(BIO_UPTODATE, &bio->bi_flags);
 768         bio_endio(bio, 0);
 769         return;
 770
 771 out:
 772         bio_io_error(bio);
 773 }
 774
 775 /*
 776  * Handler function for all zram I/O requests.
 777  */
 778 static void zram_make_request(struct request_queue *queue, struct bio *bio)
 779 {
 780         struct zram *zram = queue->queuedata;
 781
 782         down_read(&zram->init_lock);
 783         if (unlikely(!init_done(zram)))
 784                 goto error;
 785
 786         if (!valid_io_request(zram, bio)) {
 787                 atomic64_inc(&zram->stats.invalid_io);
 788                 goto error;
 789         }
 790
 791         __zram_make_request(zram, bio);
 792         up_read(&zram->init_lock);
 793
 794         return;
 795
 796 error:
 797         up_read(&zram->init_lock);
 798         bio_io_error(bio);
 799 }
 800
 801 static void zram_slot_free_notify(struct block_device *bdev,
 802                                 unsigned long index)
 803 {
 804         struct zram *zram;
 805         struct zram_meta *meta;
 806
 807         zram = bdev->bd_disk->private_data;
 808         meta = zram->meta;
 809
 810         write_lock(&meta->tb_lock);
 811         zram_free_page(zram, index);
 812         write_unlock(&meta->tb_lock);
 813         atomic64_inc(&zram->stats.notify_free);
 814 }
 815
 816 static const struct block_device_operations zram_devops = {
 817         .swap_slot_free_notify = zram_slot_free_notify,
 818         .owner = THIS_MODULE
 819 };
 820
 821 static DEVICE_ATTR(disksize, S_IRUGO | S_IWUSR,
 822                 disksize_show, disksize_store);
 823 static DEVICE_ATTR(initstate, S_IRUGO, initstate_show, NULL);
 824 static DEVICE_ATTR(reset, S_IWUSR, NULL, reset_store);
 825 static DEVICE_ATTR(orig_data_size, S_IRUGO, orig_data_size_show, NULL);
 826 static DEVICE_ATTR(mem_used_total, S_IRUGO, mem_used_total_show, NULL);
 827 static DEVICE_ATTR(max_comp_streams, S_IRUGO | S_IWUSR,
 828                 max_comp_streams_show, max_comp_streams_store);
 829 static DEVICE_ATTR(comp_algorithm, S_IRUGO | S_IWUSR,
 830                 comp_algorithm_show, comp_algorithm_store);
 831
 832 ZRAM_ATTR_RO(num_reads);
 833 ZRAM_ATTR_RO(num_writes);
 834 ZRAM_ATTR_RO(failed_reads);
 835 ZRAM_ATTR_RO(failed_writes);
 836 ZRAM_ATTR_RO(invalid_io);
 837 ZRAM_ATTR_RO(notify_free);
 838 ZRAM_ATTR_RO(zero_pages);
 839 ZRAM_ATTR_RO(compr_data_size);
 840
 841 static struct attribute *zram_disk_attrs[] = {
 842         &dev_attr_disksize.attr,
 843         &dev_attr_initstate.attr,
 844         &dev_attr_reset.attr,
 845         &dev_attr_num_reads.attr,
 846         &dev_attr_num_writes.attr,
 847         &dev_attr_failed_reads.attr,
 848         &dev_attr_failed_writes.attr,
 849         &dev_attr_invalid_io.attr,
 850         &dev_attr_notify_free.attr,
 851         &dev_attr_zero_pages.attr,
 852         &dev_attr_orig_data_size.attr,
 853         &dev_attr_compr_data_size.attr,
 854         &dev_attr_mem_used_total.attr,
 855         &dev_attr_max_comp_streams.attr,
 856         &dev_attr_comp_algorithm.attr,
 857         NULL,
 858 };
 859
 860 static struct attribute_group zram_disk_attr_group = {
 861         .attrs = zram_disk_attrs,
 862 };
 863
 864 static int create_device(struct zram *zram, int device_id)
 865 {
 866         int ret = -ENOMEM;
 867
 868         init_rwsem(&zram->init_lock);
 869
 870         zram->queue = blk_alloc_queue(GFP_KERNEL);
 871         if (!zram->queue) {
 872                 pr_err("Error allocating disk queue for device %d\n",
 873                         device_id);
 874                 goto out;
 875         }
 876
 877         blk_queue_make_request(zram->queue, zram_make_request);
 878         zram->queue->queuedata = zram;
 879
 880          /* gendisk structure */
 881         zram->disk = alloc_disk(1);
 882         if (!zram->disk) {
 883                 pr_warn("Error allocating disk structure for device %d\n",
 884                         device_id);
 885                 goto out_free_queue;
 886         }
 887
 888         zram->disk->major = zram_major;
 889         zram->disk->first_minor = device_id;
 890         zram->disk->fops = &zram_devops;
 891         zram->disk->queue = zram->queue;
 892         zram->disk->private_data = zram;
 893         snprintf(zram->disk->disk_name, 16, "zram%d", device_id);
 894
 895         /* Actual capacity set using syfs (/sys/block/zram<id>/disksize */
 896         set_capacity(zram->disk, 0);
 897         /* zram devices sort of resembles non-rotational disks */
 898         queue_flag_set_unlocked(QUEUE_FLAG_NONROT, zram->disk->queue);
 899         /*
 900          * To ensure that we always get PAGE_SIZE aligned
 901          * and n*PAGE_SIZED sized I/O requests.
 902          */
 903         blk_queue_physical_block_size(zram->disk->queue, PAGE_SIZE);
 904         blk_queue_logical_block_size(zram->disk->queue,
 905                                         ZRAM_LOGICAL_BLOCK_SIZE);
 906         blk_queue_io_min(zram->disk->queue, PAGE_SIZE);
 907         blk_queue_io_opt(zram->disk->queue, PAGE_SIZE);
 908         zram->disk->queue->limits.discard_granularity = PAGE_SIZE;
 909         zram->disk->queue->limits.max_discard_sectors = UINT_MAX;
 910         /*
 911          * zram_bio_discard() will clear all logical blocks if logical block
 912          * size is identical with physical block size(PAGE_SIZE). But if it is
 913          * different, we will skip discarding some parts of logical blocks in
 914          * the part of the request range which isn't aligned to physical block
 915          * size.  So we can't ensure that all discarded logical blocks are
 916          * zeroed.
 917          */
 918         if (ZRAM_LOGICAL_BLOCK_SIZE == PAGE_SIZE)
 919                 zram->disk->queue->limits.discard_zeroes_data = 1;
 920         else
 921                 zram->disk->queue->limits.discard_zeroes_data = 0;
 922         queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, zram->disk->queue);
 923
 924         add_disk(zram->disk);
 925
 926         ret = sysfs_create_group(&disk_to_dev(zram->disk)->kobj,
 927                                 &zram_disk_attr_group);
 928         if (ret < 0) {
 929                 pr_warn("Error creating sysfs group");
 930                 goto out_free_disk;
 931         }
 932         strlcpy(zram->compressor, default_compressor, sizeof(zram->compressor));
 933         zram->meta = NULL;
 934         zram->max_comp_streams = 1;
 935         return 0;
 936
 937 out_free_disk:
 938         del_gendisk(zram->disk);
 939         put_disk(zram->disk);
 940 out_free_queue:
 941         blk_cleanup_queue(zram->queue);
 942 out:
 943         return ret;
 944 }
 945
 946 static void destroy_device(struct zram *zram)
 947 {
 948         sysfs_remove_group(&disk_to_dev(zram->disk)->kobj,
 949                         &zram_disk_attr_group);
 950
 951         del_gendisk(zram->disk);
 952         put_disk(zram->disk);
 953
 954         blk_cleanup_queue(zram->queue);
 955 }
 956
 957 static int __init zram_init(void)
 958 {
 959         int ret, dev_id;
 960
 961         if (num_devices > max_num_devices) {
 962                 pr_warn("Invalid value for num_devices: %u\n",
 963                                 num_devices);
 964                 ret = -EINVAL;
 965                 goto out;
 966         }
 967
 968         zram_major = register_blkdev(0, "zram");
 969         if (zram_major <= 0) {
 970                 pr_warn("Unable to get major number\n");
 971                 ret = -EBUSY;
 972                 goto out;
 973         }
 974
 975         /* Allocate the device array and initialize each one */
 976         zram_devices = kzalloc(num_devices * sizeof(struct zram), GFP_KERNEL);
 977         if (!zram_devices) {
 978                 ret = -ENOMEM;
 979                 goto unregister;
 980         }
 981
 982         for (dev_id = 0; dev_id < num_devices; dev_id++) {
 983                 ret = create_device(&zram_devices[dev_id], dev_id);
 984                 if (ret)
 985                         goto free_devices;
 986         }
 987
 988         pr_info("Created %u device(s) ...\n", num_devices);
 989
 990         return 0;
 991
 992 free_devices:
 993         while (dev_id)
 994                 destroy_device(&zram_devices[--dev_id]);
 995         kfree(zram_devices);
 996 unregister:
 997         unregister_blkdev(zram_major, "zram");
 998 out:
 999         return ret;
1000 }
1001
1002 static void __exit zram_exit(void)
1003 {
1004         int i;
1005         struct zram *zram;
1006
1007         for (i = 0; i < num_devices; i++) {
1008                 zram = &zram_devices[i];
1009
1010                 destroy_device(zram);
1011                 /*
1012                  * Shouldn't access zram->disk after destroy_device
1013                  * because destroy_device already released zram->disk.
1014                  */
1015                 zram_reset_device(zram, false);
1016         }
1017
1018         unregister_blkdev(zram_major, "zram");
1019
1020         kfree(zram_devices);
1021         pr_debug("Cleanup done!\n");
1022 }
1023
1024 module_init(zram_init);
1025 module_exit(zram_exit);
1026
1027 module_param(num_devices, uint, 0);
1028 MODULE_PARM_DESC(num_devices, "Number of zram devices");
1029
1030 MODULE_LICENSE("Dual BSD/GPL");
1031 MODULE_AUTHOR("Nitin Gupta <ngupta@vflare.org>");
1032 MODULE_DESCRIPTION("Compressed RAM Block Device");