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/lzo.h>
  33 #include <linux/string.h>
  34 #include <linux/vmalloc.h>
  35
  36 #include "zram_drv.h"
  37
  38 /* Globals */
  39 static int zram_major;
  40 static struct zram *zram_devices;
  41
  42 /* Module params (documentation at end) */
  43 static unsigned int num_devices = 1;
  44
  45 static inline struct zram *dev_to_zram(struct device *dev)
  46 {
  47         return (struct zram *)dev_to_disk(dev)->private_data;
  48 }
  49
  50 static ssize_t disksize_show(struct device *dev,
  51                 struct device_attribute *attr, char *buf)
  52 {
  53         struct zram *zram = dev_to_zram(dev);
  54
  55         return sprintf(buf, "%llu\n", zram->disksize);
  56 }
  57
  58 static ssize_t initstate_show(struct device *dev,
  59                 struct device_attribute *attr, char *buf)
  60 {
  61         struct zram *zram = dev_to_zram(dev);
  62
  63         return sprintf(buf, "%u\n", zram->init_done);
  64 }
  65
  66 static ssize_t num_reads_show(struct device *dev,
  67                 struct device_attribute *attr, char *buf)
  68 {
  69         struct zram *zram = dev_to_zram(dev);
  70
  71         return sprintf(buf, "%llu\n",
  72                         (u64)atomic64_read(&zram->stats.num_reads));
  73 }
  74
  75 static ssize_t num_writes_show(struct device *dev,
  76                 struct device_attribute *attr, char *buf)
  77 {
  78         struct zram *zram = dev_to_zram(dev);
  79
  80         return sprintf(buf, "%llu\n",
  81                         (u64)atomic64_read(&zram->stats.num_writes));
  82 }
  83
  84 static ssize_t invalid_io_show(struct device *dev,
  85                 struct device_attribute *attr, char *buf)
  86 {
  87         struct zram *zram = dev_to_zram(dev);
  88
  89         return sprintf(buf, "%llu\n",
  90                         (u64)atomic64_read(&zram->stats.invalid_io));
  91 }
  92
  93 static ssize_t notify_free_show(struct device *dev,
  94                 struct device_attribute *attr, char *buf)
  95 {
  96         struct zram *zram = dev_to_zram(dev);
  97
  98         return sprintf(buf, "%llu\n",
  99                         (u64)atomic64_read(&zram->stats.notify_free));
 100 }
 101
 102 static ssize_t zero_pages_show(struct device *dev,
 103                 struct device_attribute *attr, char *buf)
 104 {
 105         struct zram *zram = dev_to_zram(dev);
 106
 107         return sprintf(buf, "%u\n", atomic_read(&zram->stats.pages_zero));
 108 }
 109
 110 static ssize_t orig_data_size_show(struct device *dev,
 111                 struct device_attribute *attr, char *buf)
 112 {
 113         struct zram *zram = dev_to_zram(dev);
 114
 115         return sprintf(buf, "%llu\n",
 116                 (u64)(atomic_read(&zram->stats.pages_stored)) << PAGE_SHIFT);
 117 }
 118
 119 static ssize_t compr_data_size_show(struct device *dev,
 120                 struct device_attribute *attr, char *buf)
 121 {
 122         struct zram *zram = dev_to_zram(dev);
 123
 124         return sprintf(buf, "%llu\n",
 125                         (u64)atomic64_read(&zram->stats.compr_size));
 126 }
 127
 128 static ssize_t mem_used_total_show(struct device *dev,
 129                 struct device_attribute *attr, char *buf)
 130 {
 131         u64 val = 0;
 132         struct zram *zram = dev_to_zram(dev);
 133         struct zram_meta *meta = zram->meta;
 134
 135         down_read(&zram->init_lock);
 136         if (zram->init_done)
 137                 val = zs_get_total_size_bytes(meta->mem_pool);
 138         up_read(&zram->init_lock);
 139
 140         return sprintf(buf, "%llu\n", val);
 141 }
 142
 143 /* flag operations needs meta->tb_lock */
 144 static int zram_test_flag(struct zram_meta *meta, u32 index,
 145                         enum zram_pageflags flag)
 146 {
 147         return meta->table[index].flags & BIT(flag);
 148 }
 149
 150 static void zram_set_flag(struct zram_meta *meta, u32 index,
 151                         enum zram_pageflags flag)
 152 {
 153         meta->table[index].flags |= BIT(flag);
 154 }
 155
 156 static void zram_clear_flag(struct zram_meta *meta, u32 index,
 157                         enum zram_pageflags flag)
 158 {
 159         meta->table[index].flags &= ~BIT(flag);
 160 }
 161
 162 static inline int is_partial_io(struct bio_vec *bvec)
 163 {
 164         return bvec->bv_len != PAGE_SIZE;
 165 }
 166
 167 /*
 168  * Check if request is within bounds and aligned on zram logical blocks.
 169  */
 170 static inline int valid_io_request(struct zram *zram, struct bio *bio)
 171 {
 172         u64 start, end, bound;
 173
 174         /* unaligned request */
 175         if (unlikely(bio->bi_iter.bi_sector &
 176                      (ZRAM_SECTOR_PER_LOGICAL_BLOCK - 1)))
 177                 return 0;
 178         if (unlikely(bio->bi_iter.bi_size & (ZRAM_LOGICAL_BLOCK_SIZE - 1)))
 179                 return 0;
 180
 181         start = bio->bi_iter.bi_sector;
 182         end = start + (bio->bi_iter.bi_size >> SECTOR_SHIFT);
 183         bound = zram->disksize >> SECTOR_SHIFT;
 184         /* out of range range */
 185         if (unlikely(start >= bound || end > bound || start > end))
 186                 return 0;
 187
 188         /* I/O request is valid */
 189         return 1;
 190 }
 191
 192 static void zram_meta_free(struct zram_meta *meta)
 193 {
 194         zs_destroy_pool(meta->mem_pool);
 195         kfree(meta->compress_workmem);
 196         free_pages((unsigned long)meta->compress_buffer, 1);
 197         vfree(meta->table);
 198         kfree(meta);
 199 }
 200
 201 static struct zram_meta *zram_meta_alloc(u64 disksize)
 202 {
 203         size_t num_pages;
 204         struct zram_meta *meta = kmalloc(sizeof(*meta), GFP_KERNEL);
 205         if (!meta)
 206                 goto out;
 207
 208         meta->compress_workmem = kzalloc(LZO1X_MEM_COMPRESS, GFP_KERNEL);
 209         if (!meta->compress_workmem)
 210                 goto free_meta;
 211
 212         meta->compress_buffer =
 213                 (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, 1);
 214         if (!meta->compress_buffer) {
 215                 pr_err("Error allocating compressor buffer space\n");
 216                 goto free_workmem;
 217         }
 218
 219         num_pages = disksize >> PAGE_SHIFT;
 220         meta->table = vzalloc(num_pages * sizeof(*meta->table));
 221         if (!meta->table) {
 222                 pr_err("Error allocating zram address table\n");
 223                 goto free_buffer;
 224         }
 225
 226         meta->mem_pool = zs_create_pool(GFP_NOIO | __GFP_HIGHMEM);
 227         if (!meta->mem_pool) {
 228                 pr_err("Error creating memory pool\n");
 229                 goto free_table;
 230         }
 231
 232         rwlock_init(&meta->tb_lock);
 233         return meta;
 234
 235 free_table:
 236         vfree(meta->table);
 237 free_buffer:
 238         free_pages((unsigned long)meta->compress_buffer, 1);
 239 free_workmem:
 240         kfree(meta->compress_workmem);
 241 free_meta:
 242         kfree(meta);
 243         meta = NULL;
 244 out:
 245         return meta;
 246 }
 247
 248 static void update_position(u32 *index, int *offset, struct bio_vec *bvec)
 249 {
 250         if (*offset + bvec->bv_len >= PAGE_SIZE)
 251                 (*index)++;
 252         *offset = (*offset + bvec->bv_len) % PAGE_SIZE;
 253 }
 254
 255 static int page_zero_filled(void *ptr)
 256 {
 257         unsigned int pos;
 258         unsigned long *page;
 259
 260         page = (unsigned long *)ptr;
 261
 262         for (pos = 0; pos != PAGE_SIZE / sizeof(*page); pos++) {
 263                 if (page[pos])
 264                         return 0;
 265         }
 266
 267         return 1;
 268 }
 269
 270 static void handle_zero_page(struct bio_vec *bvec)
 271 {
 272         struct page *page = bvec->bv_page;
 273         void *user_mem;
 274
 275         user_mem = kmap_atomic(page);
 276         if (is_partial_io(bvec))
 277                 memset(user_mem + bvec->bv_offset, 0, bvec->bv_len);
 278         else
 279                 clear_page(user_mem);
 280         kunmap_atomic(user_mem);
 281
 282         flush_dcache_page(page);
 283 }
 284
 285 /* NOTE: caller should hold meta->tb_lock with write-side */
 286 static void zram_free_page(struct zram *zram, size_t index)
 287 {
 288         struct zram_meta *meta = zram->meta;
 289         unsigned long handle = meta->table[index].handle;
 290         u16 size = meta->table[index].size;
 291
 292         if (unlikely(!handle)) {
 293                 /*
 294                  * No memory is allocated for zero filled pages.
 295                  * Simply clear zero page flag.
 296                  */
 297                 if (zram_test_flag(meta, index, ZRAM_ZERO)) {
 298                         zram_clear_flag(meta, index, ZRAM_ZERO);
 299                         atomic_dec(&zram->stats.pages_zero);
 300                 }
 301                 return;
 302         }
 303
 304         if (unlikely(size > max_zpage_size))
 305                 atomic_dec(&zram->stats.bad_compress);
 306
 307         zs_free(meta->mem_pool, handle);
 308
 309         if (size <= PAGE_SIZE / 2)
 310                 atomic_dec(&zram->stats.good_compress);
 311
 312         atomic64_sub(meta->table[index].size, &zram->stats.compr_size);
 313         atomic_dec(&zram->stats.pages_stored);
 314
 315         meta->table[index].handle = 0;
 316         meta->table[index].size = 0;
 317 }
 318
 319 static int zram_decompress_page(struct zram *zram, char *mem, u32 index)
 320 {
 321         int ret = LZO_E_OK;
 322         size_t clen = PAGE_SIZE;
 323         unsigned char *cmem;
 324         struct zram_meta *meta = zram->meta;
 325         unsigned long handle;
 326         u16 size;
 327
 328         read_lock(&meta->tb_lock);
 329         handle = meta->table[index].handle;
 330         size = meta->table[index].size;
 331
 332         if (!handle || zram_test_flag(meta, index, ZRAM_ZERO)) {
 333                 read_unlock(&meta->tb_lock);
 334                 clear_page(mem);
 335                 return 0;
 336         }
 337
 338         cmem = zs_map_object(meta->mem_pool, handle, ZS_MM_RO);
 339         if (size == PAGE_SIZE)
 340                 copy_page(mem, cmem);
 341         else
 342                 ret = lzo1x_decompress_safe(cmem, size, mem, &clen);
 343         zs_unmap_object(meta->mem_pool, handle);
 344         read_unlock(&meta->tb_lock);
 345
 346         /* Should NEVER happen. Return bio error if it does. */
 347         if (unlikely(ret != LZO_E_OK)) {
 348                 pr_err("Decompression failed! err=%d, page=%u\n", ret, index);
 349                 atomic64_inc(&zram->stats.failed_reads);
 350                 return ret;
 351         }
 352
 353         return 0;
 354 }
 355
 356 static int zram_bvec_read(struct zram *zram, struct bio_vec *bvec,
 357                           u32 index, int offset, struct bio *bio)
 358 {
 359         int ret;
 360         struct page *page;
 361         unsigned char *user_mem, *uncmem = NULL;
 362         struct zram_meta *meta = zram->meta;
 363         page = bvec->bv_page;
 364
 365         read_lock(&meta->tb_lock);
 366         if (unlikely(!meta->table[index].handle) ||
 367                         zram_test_flag(meta, index, ZRAM_ZERO)) {
 368                 read_unlock(&meta->tb_lock);
 369                 handle_zero_page(bvec);
 370                 return 0;
 371         }
 372         read_unlock(&meta->tb_lock);
 373
 374         if (is_partial_io(bvec))
 375                 /* Use  a temporary buffer to decompress the page */
 376                 uncmem = kmalloc(PAGE_SIZE, GFP_NOIO);
 377
 378         user_mem = kmap_atomic(page);
 379         if (!is_partial_io(bvec))
 380                 uncmem = user_mem;
 381
 382         if (!uncmem) {
 383                 pr_info("Unable to allocate temp memory\n");
 384                 ret = -ENOMEM;
 385                 goto out_cleanup;
 386         }
 387
 388         ret = zram_decompress_page(zram, uncmem, index);
 389         /* Should NEVER happen. Return bio error if it does. */
 390         if (unlikely(ret != LZO_E_OK))
 391                 goto out_cleanup;
 392
 393         if (is_partial_io(bvec))
 394                 memcpy(user_mem + bvec->bv_offset, uncmem + offset,
 395                                 bvec->bv_len);
 396
 397         flush_dcache_page(page);
 398         ret = 0;
 399 out_cleanup:
 400         kunmap_atomic(user_mem);
 401         if (is_partial_io(bvec))
 402                 kfree(uncmem);
 403         return ret;
 404 }
 405
 406 static int zram_bvec_write(struct zram *zram, struct bio_vec *bvec, u32 index,
 407                            int offset)
 408 {
 409         int ret = 0;
 410         size_t clen;
 411         unsigned long handle;
 412         struct page *page;
 413         unsigned char *user_mem, *cmem, *src, *uncmem = NULL;
 414         struct zram_meta *meta = zram->meta;
 415
 416         page = bvec->bv_page;
 417         src = meta->compress_buffer;
 418
 419         if (is_partial_io(bvec)) {
 420                 /*
 421                  * This is a partial IO. We need to read the full page
 422                  * before to write the changes.
 423                  */
 424                 uncmem = kmalloc(PAGE_SIZE, GFP_NOIO);
 425                 if (!uncmem) {
 426                         ret = -ENOMEM;
 427                         goto out;
 428                 }
 429                 ret = zram_decompress_page(zram, uncmem, index);
 430                 if (ret)
 431                         goto out;
 432         }
 433
 434         user_mem = kmap_atomic(page);
 435
 436         if (is_partial_io(bvec)) {
 437                 memcpy(uncmem + offset, user_mem + bvec->bv_offset,
 438                        bvec->bv_len);
 439                 kunmap_atomic(user_mem);
 440                 user_mem = NULL;
 441         } else {
 442                 uncmem = user_mem;
 443         }
 444
 445         if (page_zero_filled(uncmem)) {
 446                 kunmap_atomic(user_mem);
 447                 /* Free memory associated with this sector now. */
 448                 write_lock(&zram->meta->tb_lock);
 449                 zram_free_page(zram, index);
 450                 zram_set_flag(meta, index, ZRAM_ZERO);
 451                 write_unlock(&zram->meta->tb_lock);
 452
 453                 atomic_inc(&zram->stats.pages_zero);
 454                 ret = 0;
 455                 goto out;
 456         }
 457
 458         ret = lzo1x_1_compress(uncmem, PAGE_SIZE, src, &clen,
 459                                meta->compress_workmem);
 460
 461         if (!is_partial_io(bvec)) {
 462                 kunmap_atomic(user_mem);
 463                 user_mem = NULL;
 464                 uncmem = NULL;
 465         }
 466
 467         if (unlikely(ret != LZO_E_OK)) {
 468                 pr_err("Compression failed! err=%d\n", ret);
 469                 goto out;
 470         }
 471
 472         if (unlikely(clen > max_zpage_size)) {
 473                 atomic_inc(&zram->stats.bad_compress);
 474                 clen = PAGE_SIZE;
 475                 src = NULL;
 476                 if (is_partial_io(bvec))
 477                         src = uncmem;
 478         }
 479
 480         handle = zs_malloc(meta->mem_pool, clen);
 481         if (!handle) {
 482                 pr_info("Error allocating memory for compressed page: %u, size=%zu\n",
 483                         index, clen);
 484                 ret = -ENOMEM;
 485                 goto out;
 486         }
 487         cmem = zs_map_object(meta->mem_pool, handle, ZS_MM_WO);
 488
 489         if ((clen == PAGE_SIZE) && !is_partial_io(bvec)) {
 490                 src = kmap_atomic(page);
 491                 copy_page(cmem, src);
 492                 kunmap_atomic(src);
 493         } else {
 494                 memcpy(cmem, src, clen);
 495         }
 496
 497         zs_unmap_object(meta->mem_pool, handle);
 498
 499         /*
 500          * Free memory associated with this sector
 501          * before overwriting unused sectors.
 502          */
 503         write_lock(&zram->meta->tb_lock);
 504         zram_free_page(zram, index);
 505
 506         meta->table[index].handle = handle;
 507         meta->table[index].size = clen;
 508         write_unlock(&zram->meta->tb_lock);
 509
 510         /* Update stats */
 511         atomic64_add(clen, &zram->stats.compr_size);
 512         atomic_inc(&zram->stats.pages_stored);
 513         if (clen <= PAGE_SIZE / 2)
 514                 atomic_inc(&zram->stats.good_compress);
 515
 516 out:
 517         if (is_partial_io(bvec))
 518                 kfree(uncmem);
 519
 520         if (ret)
 521                 atomic64_inc(&zram->stats.failed_writes);
 522         return ret;
 523 }
 524
 525 static int zram_bvec_rw(struct zram *zram, struct bio_vec *bvec, u32 index,
 526                         int offset, struct bio *bio, int rw)
 527 {
 528         int ret;
 529
 530         if (rw == READ) {
 531                 down_read(&zram->lock);
 532                 ret = zram_bvec_read(zram, bvec, index, offset, bio);
 533                 up_read(&zram->lock);
 534         } else {
 535                 down_write(&zram->lock);
 536                 ret = zram_bvec_write(zram, bvec, index, offset);
 537                 up_write(&zram->lock);
 538         }
 539
 540         return ret;
 541 }
 542
 543 static void zram_reset_device(struct zram *zram, bool reset_capacity)
 544 {
 545         size_t index;
 546         struct zram_meta *meta;
 547
 548         down_write(&zram->init_lock);
 549         if (!zram->init_done) {
 550                 up_write(&zram->init_lock);
 551                 return;
 552         }
 553
 554         meta = zram->meta;
 555         zram->init_done = 0;
 556
 557         /* Free all pages that are still in this zram device */
 558         for (index = 0; index < zram->disksize >> PAGE_SHIFT; index++) {
 559                 unsigned long handle = meta->table[index].handle;
 560                 if (!handle)
 561                         continue;
 562
 563                 zs_free(meta->mem_pool, handle);
 564         }
 565
 566         zram_meta_free(zram->meta);
 567         zram->meta = NULL;
 568         /* Reset stats */
 569         memset(&zram->stats, 0, sizeof(zram->stats));
 570
 571         zram->disksize = 0;
 572         if (reset_capacity)
 573                 set_capacity(zram->disk, 0);
 574         up_write(&zram->init_lock);
 575 }
 576
 577 static void zram_init_device(struct zram *zram, struct zram_meta *meta)
 578 {
 579         if (zram->disksize > 2 * (totalram_pages << PAGE_SHIFT)) {
 580                 pr_info(
 581                 "There is little point creating a zram of greater than "
 582                 "twice the size of memory since we expect a 2:1 compression "
 583                 "ratio. Note that zram uses about 0.1%% of the size of "
 584                 "the disk when not in use so a huge zram is "
 585                 "wasteful.\n"
 586                 "\tMemory Size: %lu kB\n"
 587                 "\tSize you selected: %llu kB\n"
 588                 "Continuing anyway ...\n",
 589                 (totalram_pages << PAGE_SHIFT) >> 10, zram->disksize >> 10
 590                 );
 591         }
 592
 593         /* zram devices sort of resembles non-rotational disks */
 594         queue_flag_set_unlocked(QUEUE_FLAG_NONROT, zram->disk->queue);
 595
 596         zram->meta = meta;
 597         zram->init_done = 1;
 598
 599         pr_debug("Initialization done!\n");
 600 }
 601
 602 static ssize_t disksize_store(struct device *dev,
 603                 struct device_attribute *attr, const char *buf, size_t len)
 604 {
 605         u64 disksize;
 606         struct zram_meta *meta;
 607         struct zram *zram = dev_to_zram(dev);
 608
 609         disksize = memparse(buf, NULL);
 610         if (!disksize)
 611                 return -EINVAL;
 612
 613         disksize = PAGE_ALIGN(disksize);
 614         meta = zram_meta_alloc(disksize);
 615         down_write(&zram->init_lock);
 616         if (zram->init_done) {
 617                 up_write(&zram->init_lock);
 618                 zram_meta_free(meta);
 619                 pr_info("Cannot change disksize for initialized device\n");
 620                 return -EBUSY;
 621         }
 622
 623         zram->disksize = disksize;
 624         set_capacity(zram->disk, zram->disksize >> SECTOR_SHIFT);
 625         zram_init_device(zram, meta);
 626         up_write(&zram->init_lock);
 627
 628         return len;
 629 }
 630
 631 static ssize_t reset_store(struct device *dev,
 632                 struct device_attribute *attr, const char *buf, size_t len)
 633 {
 634         int ret;
 635         unsigned short do_reset;
 636         struct zram *zram;
 637         struct block_device *bdev;
 638
 639         zram = dev_to_zram(dev);
 640         bdev = bdget_disk(zram->disk, 0);
 641
 642         if (!bdev)
 643                 return -ENOMEM;
 644
 645         /* Do not reset an active device! */
 646         if (bdev->bd_holders) {
 647                 ret = -EBUSY;
 648                 goto out;
 649         }
 650
 651         ret = kstrtou16(buf, 10, &do_reset);
 652         if (ret)
 653                 goto out;
 654
 655         if (!do_reset) {
 656                 ret = -EINVAL;
 657                 goto out;
 658         }
 659
 660         /* Make sure all pending I/O is finished */
 661         fsync_bdev(bdev);
 662         bdput(bdev);
 663
 664         zram_reset_device(zram, true);
 665         return len;
 666
 667 out:
 668         bdput(bdev);
 669         return ret;
 670 }
 671
 672 static void __zram_make_request(struct zram *zram, struct bio *bio, int rw)
 673 {
 674         int offset;
 675         u32 index;
 676         struct bio_vec bvec;
 677         struct bvec_iter iter;
 678
 679         switch (rw) {
 680         case READ:
 681                 atomic64_inc(&zram->stats.num_reads);
 682                 break;
 683         case WRITE:
 684                 atomic64_inc(&zram->stats.num_writes);
 685                 break;
 686         }
 687
 688         index = bio->bi_iter.bi_sector >> SECTORS_PER_PAGE_SHIFT;
 689         offset = (bio->bi_iter.bi_sector &
 690                   (SECTORS_PER_PAGE - 1)) << SECTOR_SHIFT;
 691
 692         bio_for_each_segment(bvec, bio, iter) {
 693                 int max_transfer_size = PAGE_SIZE - offset;
 694
 695                 if (bvec.bv_len > max_transfer_size) {
 696                         /*
 697                          * zram_bvec_rw() can only make operation on a single
 698                          * zram page. Split the bio vector.
 699                          */
 700                         struct bio_vec bv;
 701
 702                         bv.bv_page = bvec.bv_page;
 703                         bv.bv_len = max_transfer_size;
 704                         bv.bv_offset = bvec.bv_offset;
 705
 706                         if (zram_bvec_rw(zram, &bv, index, offset, bio, rw) < 0)
 707                                 goto out;
 708
 709                         bv.bv_len = bvec.bv_len - max_transfer_size;
 710                         bv.bv_offset += max_transfer_size;
 711                         if (zram_bvec_rw(zram, &bv, index+1, 0, bio, rw) < 0)
 712                                 goto out;
 713                 } else
 714                         if (zram_bvec_rw(zram, &bvec, index, offset, bio, rw)
 715                             < 0)
 716                                 goto out;
 717
 718                 update_position(&index, &offset, &bvec);
 719         }
 720
 721         set_bit(BIO_UPTODATE, &bio->bi_flags);
 722         bio_endio(bio, 0);
 723         return;
 724
 725 out:
 726         bio_io_error(bio);
 727 }
 728
 729 /*
 730  * Handler function for all zram I/O requests.
 731  */
 732 static void zram_make_request(struct request_queue *queue, struct bio *bio)
 733 {
 734         struct zram *zram = queue->queuedata;
 735
 736         down_read(&zram->init_lock);
 737         if (unlikely(!zram->init_done))
 738                 goto error;
 739
 740         if (!valid_io_request(zram, bio)) {
 741                 atomic64_inc(&zram->stats.invalid_io);
 742                 goto error;
 743         }
 744
 745         __zram_make_request(zram, bio, bio_data_dir(bio));
 746         up_read(&zram->init_lock);
 747
 748         return;
 749
 750 error:
 751         up_read(&zram->init_lock);
 752         bio_io_error(bio);
 753 }
 754
 755 static void zram_slot_free_notify(struct block_device *bdev,
 756                                 unsigned long index)
 757 {
 758         struct zram *zram;
 759         struct zram_meta *meta;
 760
 761         zram = bdev->bd_disk->private_data;
 762         meta = zram->meta;
 763
 764         write_lock(&meta->tb_lock);
 765         zram_free_page(zram, index);
 766         write_unlock(&meta->tb_lock);
 767         atomic64_inc(&zram->stats.notify_free);
 768 }
 769
 770 static const struct block_device_operations zram_devops = {
 771         .swap_slot_free_notify = zram_slot_free_notify,
 772         .owner = THIS_MODULE
 773 };
 774
 775 static DEVICE_ATTR(disksize, S_IRUGO | S_IWUSR,
 776                 disksize_show, disksize_store);
 777 static DEVICE_ATTR(initstate, S_IRUGO, initstate_show, NULL);
 778 static DEVICE_ATTR(reset, S_IWUSR, NULL, reset_store);
 779 static DEVICE_ATTR(num_reads, S_IRUGO, num_reads_show, NULL);
 780 static DEVICE_ATTR(num_writes, S_IRUGO, num_writes_show, NULL);
 781 static DEVICE_ATTR(invalid_io, S_IRUGO, invalid_io_show, NULL);
 782 static DEVICE_ATTR(notify_free, S_IRUGO, notify_free_show, NULL);
 783 static DEVICE_ATTR(zero_pages, S_IRUGO, zero_pages_show, NULL);
 784 static DEVICE_ATTR(orig_data_size, S_IRUGO, orig_data_size_show, NULL);
 785 static DEVICE_ATTR(compr_data_size, S_IRUGO, compr_data_size_show, NULL);
 786 static DEVICE_ATTR(mem_used_total, S_IRUGO, mem_used_total_show, NULL);
 787
 788 static struct attribute *zram_disk_attrs[] = {
 789         &dev_attr_disksize.attr,
 790         &dev_attr_initstate.attr,
 791         &dev_attr_reset.attr,
 792         &dev_attr_num_reads.attr,
 793         &dev_attr_num_writes.attr,
 794         &dev_attr_invalid_io.attr,
 795         &dev_attr_notify_free.attr,
 796         &dev_attr_zero_pages.attr,
 797         &dev_attr_orig_data_size.attr,
 798         &dev_attr_compr_data_size.attr,
 799         &dev_attr_mem_used_total.attr,
 800         NULL,
 801 };
 802
 803 static struct attribute_group zram_disk_attr_group = {
 804         .attrs = zram_disk_attrs,
 805 };
 806
 807 static int create_device(struct zram *zram, int device_id)
 808 {
 809         int ret = -ENOMEM;
 810
 811         init_rwsem(&zram->lock);
 812         init_rwsem(&zram->init_lock);
 813
 814         zram->queue = blk_alloc_queue(GFP_KERNEL);
 815         if (!zram->queue) {
 816                 pr_err("Error allocating disk queue for device %d\n",
 817                         device_id);
 818                 goto out;
 819         }
 820
 821         blk_queue_make_request(zram->queue, zram_make_request);
 822         zram->queue->queuedata = zram;
 823
 824          /* gendisk structure */
 825         zram->disk = alloc_disk(1);
 826         if (!zram->disk) {
 827                 pr_warn("Error allocating disk structure for device %d\n",
 828                         device_id);
 829                 goto out_free_queue;
 830         }
 831
 832         zram->disk->major = zram_major;
 833         zram->disk->first_minor = device_id;
 834         zram->disk->fops = &zram_devops;
 835         zram->disk->queue = zram->queue;
 836         zram->disk->private_data = zram;
 837         snprintf(zram->disk->disk_name, 16, "zram%d", device_id);
 838
 839         /* Actual capacity set using syfs (/sys/block/zram<id>/disksize */
 840         set_capacity(zram->disk, 0);
 841
 842         /*
 843          * To ensure that we always get PAGE_SIZE aligned
 844          * and n*PAGE_SIZED sized I/O requests.
 845          */
 846         blk_queue_physical_block_size(zram->disk->queue, PAGE_SIZE);
 847         blk_queue_logical_block_size(zram->disk->queue,
 848                                         ZRAM_LOGICAL_BLOCK_SIZE);
 849         blk_queue_io_min(zram->disk->queue, PAGE_SIZE);
 850         blk_queue_io_opt(zram->disk->queue, PAGE_SIZE);
 851
 852         add_disk(zram->disk);
 853
 854         ret = sysfs_create_group(&disk_to_dev(zram->disk)->kobj,
 855                                 &zram_disk_attr_group);
 856         if (ret < 0) {
 857                 pr_warn("Error creating sysfs group");
 858                 goto out_free_disk;
 859         }
 860
 861         zram->init_done = 0;
 862         return 0;
 863
 864 out_free_disk:
 865         del_gendisk(zram->disk);
 866         put_disk(zram->disk);
 867 out_free_queue:
 868         blk_cleanup_queue(zram->queue);
 869 out:
 870         return ret;
 871 }
 872
 873 static void destroy_device(struct zram *zram)
 874 {
 875         sysfs_remove_group(&disk_to_dev(zram->disk)->kobj,
 876                         &zram_disk_attr_group);
 877
 878         del_gendisk(zram->disk);
 879         put_disk(zram->disk);
 880
 881         blk_cleanup_queue(zram->queue);
 882 }
 883
 884 static int __init zram_init(void)
 885 {
 886         int ret, dev_id;
 887
 888         if (num_devices > max_num_devices) {
 889                 pr_warn("Invalid value for num_devices: %u\n",
 890                                 num_devices);
 891                 ret = -EINVAL;
 892                 goto out;
 893         }
 894
 895         zram_major = register_blkdev(0, "zram");
 896         if (zram_major <= 0) {
 897                 pr_warn("Unable to get major number\n");
 898                 ret = -EBUSY;
 899                 goto out;
 900         }
 901
 902         /* Allocate the device array and initialize each one */
 903         zram_devices = kzalloc(num_devices * sizeof(struct zram), GFP_KERNEL);
 904         if (!zram_devices) {
 905                 ret = -ENOMEM;
 906                 goto unregister;
 907         }
 908
 909         for (dev_id = 0; dev_id < num_devices; dev_id++) {
 910                 ret = create_device(&zram_devices[dev_id], dev_id);
 911                 if (ret)
 912                         goto free_devices;
 913         }
 914
 915         pr_info("Created %u device(s) ...\n", num_devices);
 916
 917         return 0;
 918
 919 free_devices:
 920         while (dev_id)
 921                 destroy_device(&zram_devices[--dev_id]);
 922         kfree(zram_devices);
 923 unregister:
 924         unregister_blkdev(zram_major, "zram");
 925 out:
 926         return ret;
 927 }
 928
 929 static void __exit zram_exit(void)
 930 {
 931         int i;
 932         struct zram *zram;
 933
 934         for (i = 0; i < num_devices; i++) {
 935                 zram = &zram_devices[i];
 936
 937                 destroy_device(zram);
 938                 /*
 939                  * Shouldn't access zram->disk after destroy_device
 940                  * because destroy_device already released zram->disk.
 941                  */
 942                 zram_reset_device(zram, false);
 943         }
 944
 945         unregister_blkdev(zram_major, "zram");
 946
 947         kfree(zram_devices);
 948         pr_debug("Cleanup done!\n");
 949 }
 950
 951 module_init(zram_init);
 952 module_exit(zram_exit);
 953
 954 module_param(num_devices, uint, 0);
 955 MODULE_PARM_DESC(num_devices, "Number of zram devices");
 956
 957 MODULE_LICENSE("Dual BSD/GPL");
 958 MODULE_AUTHOR("Nitin Gupta <ngupta@vflare.org>");
 959 MODULE_DESCRIPTION("Compressed RAM Block Device");