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_sector & (ZRAM_SECTOR_PER_LOGICAL_BLOCK - 1)))
 176                 return 0;
 177         if (unlikely(bio->bi_size & (ZRAM_LOGICAL_BLOCK_SIZE - 1)))
 178                 return 0;
 179
 180         start = bio->bi_sector;
 181         end = start + (bio->bi_size >> SECTOR_SHIFT);
 182         bound = zram->disksize >> SECTOR_SHIFT;
 183         /* out of range range */
 184         if (unlikely(start >= bound || end >= bound || start > end))
 185                 return 0;
 186
 187         /* I/O request is valid */
 188         return 1;
 189 }
 190
 191 static void zram_meta_free(struct zram_meta *meta)
 192 {
 193         zs_destroy_pool(meta->mem_pool);
 194         kfree(meta->compress_workmem);
 195         free_pages((unsigned long)meta->compress_buffer, 1);
 196         vfree(meta->table);
 197         kfree(meta);
 198 }
 199
 200 static struct zram_meta *zram_meta_alloc(u64 disksize)
 201 {
 202         size_t num_pages;
 203         struct zram_meta *meta = kmalloc(sizeof(*meta), GFP_KERNEL);
 204         if (!meta)
 205                 goto out;
 206
 207         meta->compress_workmem = kzalloc(LZO1X_MEM_COMPRESS, GFP_KERNEL);
 208         if (!meta->compress_workmem)
 209                 goto free_meta;
 210
 211         meta->compress_buffer =
 212                 (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, 1);
 213         if (!meta->compress_buffer) {
 214                 pr_err("Error allocating compressor buffer space\n");
 215                 goto free_workmem;
 216         }
 217
 218         num_pages = disksize >> PAGE_SHIFT;
 219         meta->table = vzalloc(num_pages * sizeof(*meta->table));
 220         if (!meta->table) {
 221                 pr_err("Error allocating zram address table\n");
 222                 goto free_buffer;
 223         }
 224
 225         meta->mem_pool = zs_create_pool(GFP_NOIO | __GFP_HIGHMEM);
 226         if (!meta->mem_pool) {
 227                 pr_err("Error creating memory pool\n");
 228                 goto free_table;
 229         }
 230
 231         rwlock_init(&meta->tb_lock);
 232         return meta;
 233
 234 free_table:
 235         vfree(meta->table);
 236 free_buffer:
 237         free_pages((unsigned long)meta->compress_buffer, 1);
 238 free_workmem:
 239         kfree(meta->compress_workmem);
 240 free_meta:
 241         kfree(meta);
 242         meta = NULL;
 243 out:
 244         return meta;
 245 }
 246
 247 static void update_position(u32 *index, int *offset, struct bio_vec *bvec)
 248 {
 249         if (*offset + bvec->bv_len >= PAGE_SIZE)
 250                 (*index)++;
 251         *offset = (*offset + bvec->bv_len) % PAGE_SIZE;
 252 }
 253
 254 static int page_zero_filled(void *ptr)
 255 {
 256         unsigned int pos;
 257         unsigned long *page;
 258
 259         page = (unsigned long *)ptr;
 260
 261         for (pos = 0; pos != PAGE_SIZE / sizeof(*page); pos++) {
 262                 if (page[pos])
 263                         return 0;
 264         }
 265
 266         return 1;
 267 }
 268
 269 static void handle_zero_page(struct bio_vec *bvec)
 270 {
 271         struct page *page = bvec->bv_page;
 272         void *user_mem;
 273
 274         user_mem = kmap_atomic(page);
 275         if (is_partial_io(bvec))
 276                 memset(user_mem + bvec->bv_offset, 0, bvec->bv_len);
 277         else
 278                 clear_page(user_mem);
 279         kunmap_atomic(user_mem);
 280
 281         flush_dcache_page(page);
 282 }
 283
 284 /* NOTE: caller should hold meta->tb_lock with write-side */
 285 static void zram_free_page(struct zram *zram, size_t index)
 286 {
 287         struct zram_meta *meta = zram->meta;
 288         unsigned long handle = meta->table[index].handle;
 289         u16 size = meta->table[index].size;
 290
 291         if (unlikely(!handle)) {
 292                 /*
 293                  * No memory is allocated for zero filled pages.
 294                  * Simply clear zero page flag.
 295                  */
 296                 if (zram_test_flag(meta, index, ZRAM_ZERO)) {
 297                         zram_clear_flag(meta, index, ZRAM_ZERO);
 298                         atomic_dec(&zram->stats.pages_zero);
 299                 }
 300                 return;
 301         }
 302
 303         if (unlikely(size > max_zpage_size))
 304                 atomic_dec(&zram->stats.bad_compress);
 305
 306         zs_free(meta->mem_pool, handle);
 307
 308         if (size <= PAGE_SIZE / 2)
 309                 atomic_dec(&zram->stats.good_compress);
 310
 311         atomic64_sub(meta->table[index].size, &zram->stats.compr_size);
 312         atomic_dec(&zram->stats.pages_stored);
 313
 314         meta->table[index].handle = 0;
 315         meta->table[index].size = 0;
 316 }
 317
 318 static int zram_decompress_page(struct zram *zram, char *mem, u32 index)
 319 {
 320         int ret = LZO_E_OK;
 321         size_t clen = PAGE_SIZE;
 322         unsigned char *cmem;
 323         struct zram_meta *meta = zram->meta;
 324         unsigned long handle;
 325         u16 size;
 326
 327         read_lock(&meta->tb_lock);
 328         handle = meta->table[index].handle;
 329         size = meta->table[index].size;
 330
 331         if (!handle || zram_test_flag(meta, index, ZRAM_ZERO)) {
 332                 read_unlock(&meta->tb_lock);
 333                 clear_page(mem);
 334                 return 0;
 335         }
 336
 337         cmem = zs_map_object(meta->mem_pool, handle, ZS_MM_RO);
 338         if (size == PAGE_SIZE)
 339                 copy_page(mem, cmem);
 340         else
 341                 ret = lzo1x_decompress_safe(cmem, size, mem, &clen);
 342         zs_unmap_object(meta->mem_pool, handle);
 343         read_unlock(&meta->tb_lock);
 344
 345         /* Should NEVER happen. Return bio error if it does. */
 346         if (unlikely(ret != LZO_E_OK)) {
 347                 pr_err("Decompression failed! err=%d, page=%u\n", ret, index);
 348                 atomic64_inc(&zram->stats.failed_reads);
 349                 return ret;
 350         }
 351
 352         return 0;
 353 }
 354
 355 static int zram_bvec_read(struct zram *zram, struct bio_vec *bvec,
 356                           u32 index, int offset, struct bio *bio)
 357 {
 358         int ret;
 359         struct page *page;
 360         unsigned char *user_mem, *uncmem = NULL;
 361         struct zram_meta *meta = zram->meta;
 362         page = bvec->bv_page;
 363
 364         read_lock(&meta->tb_lock);
 365         if (unlikely(!meta->table[index].handle) ||
 366                         zram_test_flag(meta, index, ZRAM_ZERO)) {
 367                 read_unlock(&meta->tb_lock);
 368                 handle_zero_page(bvec);
 369                 return 0;
 370         }
 371         read_unlock(&meta->tb_lock);
 372
 373         if (is_partial_io(bvec))
 374                 /* Use  a temporary buffer to decompress the page */
 375                 uncmem = kmalloc(PAGE_SIZE, GFP_NOIO);
 376
 377         user_mem = kmap_atomic(page);
 378         if (!is_partial_io(bvec))
 379                 uncmem = user_mem;
 380
 381         if (!uncmem) {
 382                 pr_info("Unable to allocate temp memory\n");
 383                 ret = -ENOMEM;
 384                 goto out_cleanup;
 385         }
 386
 387         ret = zram_decompress_page(zram, uncmem, index);
 388         /* Should NEVER happen. Return bio error if it does. */
 389         if (unlikely(ret != LZO_E_OK))
 390                 goto out_cleanup;
 391
 392         if (is_partial_io(bvec))
 393                 memcpy(user_mem + bvec->bv_offset, uncmem + offset,
 394                                 bvec->bv_len);
 395
 396         flush_dcache_page(page);
 397         ret = 0;
 398 out_cleanup:
 399         kunmap_atomic(user_mem);
 400         if (is_partial_io(bvec))
 401                 kfree(uncmem);
 402         return ret;
 403 }
 404
 405 static int zram_bvec_write(struct zram *zram, struct bio_vec *bvec, u32 index,
 406                            int offset)
 407 {
 408         int ret = 0;
 409         size_t clen;
 410         unsigned long handle;
 411         struct page *page;
 412         unsigned char *user_mem, *cmem, *src, *uncmem = NULL;
 413         struct zram_meta *meta = zram->meta;
 414
 415         page = bvec->bv_page;
 416         src = meta->compress_buffer;
 417
 418         if (is_partial_io(bvec)) {
 419                 /*
 420                  * This is a partial IO. We need to read the full page
 421                  * before to write the changes.
 422                  */
 423                 uncmem = kmalloc(PAGE_SIZE, GFP_NOIO);
 424                 if (!uncmem) {
 425                         ret = -ENOMEM;
 426                         goto out;
 427                 }
 428                 ret = zram_decompress_page(zram, uncmem, index);
 429                 if (ret)
 430                         goto out;
 431         }
 432
 433         user_mem = kmap_atomic(page);
 434
 435         if (is_partial_io(bvec)) {
 436                 memcpy(uncmem + offset, user_mem + bvec->bv_offset,
 437                        bvec->bv_len);
 438                 kunmap_atomic(user_mem);
 439                 user_mem = NULL;
 440         } else {
 441                 uncmem = user_mem;
 442         }
 443
 444         if (page_zero_filled(uncmem)) {
 445                 kunmap_atomic(user_mem);
 446                 /* Free memory associated with this sector now. */
 447                 write_lock(&zram->meta->tb_lock);
 448                 zram_free_page(zram, index);
 449                 zram_set_flag(meta, index, ZRAM_ZERO);
 450                 write_unlock(&zram->meta->tb_lock);
 451
 452                 atomic_inc(&zram->stats.pages_zero);
 453                 ret = 0;
 454                 goto out;
 455         }
 456
 457         ret = lzo1x_1_compress(uncmem, PAGE_SIZE, src, &clen,
 458                                meta->compress_workmem);
 459
 460         if (!is_partial_io(bvec)) {
 461                 kunmap_atomic(user_mem);
 462                 user_mem = NULL;
 463                 uncmem = NULL;
 464         }
 465
 466         if (unlikely(ret != LZO_E_OK)) {
 467                 pr_err("Compression failed! err=%d\n", ret);
 468                 goto out;
 469         }
 470
 471         if (unlikely(clen > max_zpage_size)) {
 472                 atomic_inc(&zram->stats.bad_compress);
 473                 clen = PAGE_SIZE;
 474                 src = NULL;
 475                 if (is_partial_io(bvec))
 476                         src = uncmem;
 477         }
 478
 479         handle = zs_malloc(meta->mem_pool, clen);
 480         if (!handle) {
 481                 pr_info("Error allocating memory for compressed page: %u, size=%zu\n",
 482                         index, clen);
 483                 ret = -ENOMEM;
 484                 goto out;
 485         }
 486         cmem = zs_map_object(meta->mem_pool, handle, ZS_MM_WO);
 487
 488         if ((clen == PAGE_SIZE) && !is_partial_io(bvec)) {
 489                 src = kmap_atomic(page);
 490                 copy_page(cmem, src);
 491                 kunmap_atomic(src);
 492         } else {
 493                 memcpy(cmem, src, clen);
 494         }
 495
 496         zs_unmap_object(meta->mem_pool, handle);
 497
 498         /*
 499          * Free memory associated with this sector
 500          * before overwriting unused sectors.
 501          */
 502         write_lock(&zram->meta->tb_lock);
 503         zram_free_page(zram, index);
 504
 505         meta->table[index].handle = handle;
 506         meta->table[index].size = clen;
 507         write_unlock(&zram->meta->tb_lock);
 508
 509         /* Update stats */
 510         atomic64_add(clen, &zram->stats.compr_size);
 511         atomic_inc(&zram->stats.pages_stored);
 512         if (clen <= PAGE_SIZE / 2)
 513                 atomic_inc(&zram->stats.good_compress);
 514
 515 out:
 516         if (is_partial_io(bvec))
 517                 kfree(uncmem);
 518
 519         if (ret)
 520                 atomic64_inc(&zram->stats.failed_writes);
 521         return ret;
 522 }
 523
 524 static void handle_pending_slot_free(struct zram *zram)
 525 {
 526         struct zram_slot_free *free_rq;
 527
 528         spin_lock(&zram->slot_free_lock);
 529         while (zram->slot_free_rq) {
 530                 free_rq = zram->slot_free_rq;
 531                 zram->slot_free_rq = free_rq->next;
 532                 zram_free_page(zram, free_rq->index);
 533                 kfree(free_rq);
 534         }
 535         spin_unlock(&zram->slot_free_lock);
 536 }
 537
 538 static int zram_bvec_rw(struct zram *zram, struct bio_vec *bvec, u32 index,
 539                         int offset, struct bio *bio, int rw)
 540 {
 541         int ret;
 542
 543         if (rw == READ) {
 544                 down_read(&zram->lock);
 545                 ret = zram_bvec_read(zram, bvec, index, offset, bio);
 546                 up_read(&zram->lock);
 547         } else {
 548                 down_write(&zram->lock);
 549                 handle_pending_slot_free(zram);
 550                 ret = zram_bvec_write(zram, bvec, index, offset);
 551                 up_write(&zram->lock);
 552         }
 553
 554         return ret;
 555 }
 556
 557 static void zram_reset_device(struct zram *zram, bool reset_capacity)
 558 {
 559         size_t index;
 560         struct zram_meta *meta;
 561
 562         down_write(&zram->init_lock);
 563         if (!zram->init_done) {
 564                 up_write(&zram->init_lock);
 565                 return;
 566         }
 567
 568         flush_work(&zram->free_work);
 569
 570         meta = zram->meta;
 571         zram->init_done = 0;
 572
 573         /* Free all pages that are still in this zram device */
 574         for (index = 0; index < zram->disksize >> PAGE_SHIFT; index++) {
 575                 unsigned long handle = meta->table[index].handle;
 576                 if (!handle)
 577                         continue;
 578
 579                 zs_free(meta->mem_pool, handle);
 580         }
 581
 582         zram_meta_free(zram->meta);
 583         zram->meta = NULL;
 584         /* Reset stats */
 585         memset(&zram->stats, 0, sizeof(zram->stats));
 586
 587         zram->disksize = 0;
 588         if (reset_capacity)
 589                 set_capacity(zram->disk, 0);
 590         up_write(&zram->init_lock);
 591 }
 592
 593 static void zram_init_device(struct zram *zram, struct zram_meta *meta)
 594 {
 595         if (zram->disksize > 2 * (totalram_pages << PAGE_SHIFT)) {
 596                 pr_info(
 597                 "There is little point creating a zram of greater than "
 598                 "twice the size of memory since we expect a 2:1 compression "
 599                 "ratio. Note that zram uses about 0.1%% of the size of "
 600                 "the disk when not in use so a huge zram is "
 601                 "wasteful.\n"
 602                 "\tMemory Size: %lu kB\n"
 603                 "\tSize you selected: %llu kB\n"
 604                 "Continuing anyway ...\n",
 605                 (totalram_pages << PAGE_SHIFT) >> 10, zram->disksize >> 10
 606                 );
 607         }
 608
 609         /* zram devices sort of resembles non-rotational disks */
 610         queue_flag_set_unlocked(QUEUE_FLAG_NONROT, zram->disk->queue);
 611
 612         zram->meta = meta;
 613         zram->init_done = 1;
 614
 615         pr_debug("Initialization done!\n");
 616 }
 617
 618 static ssize_t disksize_store(struct device *dev,
 619                 struct device_attribute *attr, const char *buf, size_t len)
 620 {
 621         u64 disksize;
 622         struct zram_meta *meta;
 623         struct zram *zram = dev_to_zram(dev);
 624
 625         disksize = memparse(buf, NULL);
 626         if (!disksize)
 627                 return -EINVAL;
 628
 629         disksize = PAGE_ALIGN(disksize);
 630         meta = zram_meta_alloc(disksize);
 631         down_write(&zram->init_lock);
 632         if (zram->init_done) {
 633                 up_write(&zram->init_lock);
 634                 zram_meta_free(meta);
 635                 pr_info("Cannot change disksize for initialized device\n");
 636                 return -EBUSY;
 637         }
 638
 639         zram->disksize = disksize;
 640         set_capacity(zram->disk, zram->disksize >> SECTOR_SHIFT);
 641         zram_init_device(zram, meta);
 642         up_write(&zram->init_lock);
 643
 644         return len;
 645 }
 646
 647 static ssize_t reset_store(struct device *dev,
 648                 struct device_attribute *attr, const char *buf, size_t len)
 649 {
 650         int ret;
 651         unsigned short do_reset;
 652         struct zram *zram;
 653         struct block_device *bdev;
 654
 655         zram = dev_to_zram(dev);
 656         bdev = bdget_disk(zram->disk, 0);
 657
 658         if (!bdev)
 659                 return -ENOMEM;
 660
 661         /* Do not reset an active device! */
 662         if (bdev->bd_holders) {
 663                 ret = -EBUSY;
 664                 goto out;
 665         }
 666
 667         ret = kstrtou16(buf, 10, &do_reset);
 668         if (ret)
 669                 goto out;
 670
 671         if (!do_reset) {
 672                 ret = -EINVAL;
 673                 goto out;
 674         }
 675
 676         /* Make sure all pending I/O is finished */
 677         fsync_bdev(bdev);
 678         bdput(bdev);
 679
 680         zram_reset_device(zram, true);
 681         return len;
 682
 683 out:
 684         bdput(bdev);
 685         return ret;
 686 }
 687
 688 static void __zram_make_request(struct zram *zram, struct bio *bio, int rw)
 689 {
 690         int i, offset;
 691         u32 index;
 692         struct bio_vec *bvec;
 693
 694         switch (rw) {
 695         case READ:
 696                 atomic64_inc(&zram->stats.num_reads);
 697                 break;
 698         case WRITE:
 699                 atomic64_inc(&zram->stats.num_writes);
 700                 break;
 701         }
 702
 703         index = bio->bi_sector >> SECTORS_PER_PAGE_SHIFT;
 704         offset = (bio->bi_sector & (SECTORS_PER_PAGE - 1)) << SECTOR_SHIFT;
 705
 706         bio_for_each_segment(bvec, bio, i) {
 707                 int max_transfer_size = PAGE_SIZE - offset;
 708
 709                 if (bvec->bv_len > max_transfer_size) {
 710                         /*
 711                          * zram_bvec_rw() can only make operation on a single
 712                          * zram page. Split the bio vector.
 713                          */
 714                         struct bio_vec bv;
 715
 716                         bv.bv_page = bvec->bv_page;
 717                         bv.bv_len = max_transfer_size;
 718                         bv.bv_offset = bvec->bv_offset;
 719
 720                         if (zram_bvec_rw(zram, &bv, index, offset, bio, rw) < 0)
 721                                 goto out;
 722
 723                         bv.bv_len = bvec->bv_len - max_transfer_size;
 724                         bv.bv_offset += max_transfer_size;
 725                         if (zram_bvec_rw(zram, &bv, index+1, 0, bio, rw) < 0)
 726                                 goto out;
 727                 } else
 728                         if (zram_bvec_rw(zram, bvec, index, offset, bio, rw)
 729                             < 0)
 730                                 goto out;
 731
 732                 update_position(&index, &offset, bvec);
 733         }
 734
 735         set_bit(BIO_UPTODATE, &bio->bi_flags);
 736         bio_endio(bio, 0);
 737         return;
 738
 739 out:
 740         bio_io_error(bio);
 741 }
 742
 743 /*
 744  * Handler function for all zram I/O requests.
 745  */
 746 static void zram_make_request(struct request_queue *queue, struct bio *bio)
 747 {
 748         struct zram *zram = queue->queuedata;
 749
 750         down_read(&zram->init_lock);
 751         if (unlikely(!zram->init_done))
 752                 goto error;
 753
 754         if (!valid_io_request(zram, bio)) {
 755                 atomic64_inc(&zram->stats.invalid_io);
 756                 goto error;
 757         }
 758
 759         __zram_make_request(zram, bio, bio_data_dir(bio));
 760         up_read(&zram->init_lock);
 761
 762         return;
 763
 764 error:
 765         up_read(&zram->init_lock);
 766         bio_io_error(bio);
 767 }
 768
 769 static void zram_slot_free(struct work_struct *work)
 770 {
 771         struct zram *zram;
 772
 773         zram = container_of(work, struct zram, free_work);
 774         down_write(&zram->lock);
 775         handle_pending_slot_free(zram);
 776         up_write(&zram->lock);
 777 }
 778
 779 static void add_slot_free(struct zram *zram, struct zram_slot_free *free_rq)
 780 {
 781         spin_lock(&zram->slot_free_lock);
 782         free_rq->next = zram->slot_free_rq;
 783         zram->slot_free_rq = free_rq;
 784         spin_unlock(&zram->slot_free_lock);
 785 }
 786
 787 static void zram_slot_free_notify(struct block_device *bdev,
 788                                 unsigned long index)
 789 {
 790         struct zram *zram;
 791         struct zram_slot_free *free_rq;
 792
 793         zram = bdev->bd_disk->private_data;
 794         atomic64_inc(&zram->stats.notify_free);
 795
 796         free_rq = kmalloc(sizeof(struct zram_slot_free), GFP_ATOMIC);
 797         if (!free_rq)
 798                 return;
 799
 800         free_rq->index = index;
 801         add_slot_free(zram, free_rq);
 802         schedule_work(&zram->free_work);
 803 }
 804
 805 static const struct block_device_operations zram_devops = {
 806         .swap_slot_free_notify = zram_slot_free_notify,
 807         .owner = THIS_MODULE
 808 };
 809
 810 static DEVICE_ATTR(disksize, S_IRUGO | S_IWUSR,
 811                 disksize_show, disksize_store);
 812 static DEVICE_ATTR(initstate, S_IRUGO, initstate_show, NULL);
 813 static DEVICE_ATTR(reset, S_IWUSR, NULL, reset_store);
 814 static DEVICE_ATTR(num_reads, S_IRUGO, num_reads_show, NULL);
 815 static DEVICE_ATTR(num_writes, S_IRUGO, num_writes_show, NULL);
 816 static DEVICE_ATTR(invalid_io, S_IRUGO, invalid_io_show, NULL);
 817 static DEVICE_ATTR(notify_free, S_IRUGO, notify_free_show, NULL);
 818 static DEVICE_ATTR(zero_pages, S_IRUGO, zero_pages_show, NULL);
 819 static DEVICE_ATTR(orig_data_size, S_IRUGO, orig_data_size_show, NULL);
 820 static DEVICE_ATTR(compr_data_size, S_IRUGO, compr_data_size_show, NULL);
 821 static DEVICE_ATTR(mem_used_total, S_IRUGO, mem_used_total_show, NULL);
 822
 823 static struct attribute *zram_disk_attrs[] = {
 824         &dev_attr_disksize.attr,
 825         &dev_attr_initstate.attr,
 826         &dev_attr_reset.attr,
 827         &dev_attr_num_reads.attr,
 828         &dev_attr_num_writes.attr,
 829         &dev_attr_invalid_io.attr,
 830         &dev_attr_notify_free.attr,
 831         &dev_attr_zero_pages.attr,
 832         &dev_attr_orig_data_size.attr,
 833         &dev_attr_compr_data_size.attr,
 834         &dev_attr_mem_used_total.attr,
 835         NULL,
 836 };
 837
 838 static struct attribute_group zram_disk_attr_group = {
 839         .attrs = zram_disk_attrs,
 840 };
 841
 842 static int create_device(struct zram *zram, int device_id)
 843 {
 844         int ret = -ENOMEM;
 845
 846         init_rwsem(&zram->lock);
 847         init_rwsem(&zram->init_lock);
 848
 849         INIT_WORK(&zram->free_work, zram_slot_free);
 850         spin_lock_init(&zram->slot_free_lock);
 851         zram->slot_free_rq = NULL;
 852
 853         zram->queue = blk_alloc_queue(GFP_KERNEL);
 854         if (!zram->queue) {
 855                 pr_err("Error allocating disk queue for device %d\n",
 856                         device_id);
 857                 goto out;
 858         }
 859
 860         blk_queue_make_request(zram->queue, zram_make_request);
 861         zram->queue->queuedata = zram;
 862
 863          /* gendisk structure */
 864         zram->disk = alloc_disk(1);
 865         if (!zram->disk) {
 866                 pr_warn("Error allocating disk structure for device %d\n",
 867                         device_id);
 868                 goto out_free_queue;
 869         }
 870
 871         zram->disk->major = zram_major;
 872         zram->disk->first_minor = device_id;
 873         zram->disk->fops = &zram_devops;
 874         zram->disk->queue = zram->queue;
 875         zram->disk->private_data = zram;
 876         snprintf(zram->disk->disk_name, 16, "zram%d", device_id);
 877
 878         /* Actual capacity set using syfs (/sys/block/zram<id>/disksize */
 879         set_capacity(zram->disk, 0);
 880
 881         /*
 882          * To ensure that we always get PAGE_SIZE aligned
 883          * and n*PAGE_SIZED sized I/O requests.
 884          */
 885         blk_queue_physical_block_size(zram->disk->queue, PAGE_SIZE);
 886         blk_queue_logical_block_size(zram->disk->queue,
 887                                         ZRAM_LOGICAL_BLOCK_SIZE);
 888         blk_queue_io_min(zram->disk->queue, PAGE_SIZE);
 889         blk_queue_io_opt(zram->disk->queue, PAGE_SIZE);
 890
 891         add_disk(zram->disk);
 892
 893         ret = sysfs_create_group(&disk_to_dev(zram->disk)->kobj,
 894                                 &zram_disk_attr_group);
 895         if (ret < 0) {
 896                 pr_warn("Error creating sysfs group");
 897                 goto out_free_disk;
 898         }
 899
 900         zram->init_done = 0;
 901         return 0;
 902
 903 out_free_disk:
 904         del_gendisk(zram->disk);
 905         put_disk(zram->disk);
 906 out_free_queue:
 907         blk_cleanup_queue(zram->queue);
 908 out:
 909         return ret;
 910 }
 911
 912 static void destroy_device(struct zram *zram)
 913 {
 914         sysfs_remove_group(&disk_to_dev(zram->disk)->kobj,
 915                         &zram_disk_attr_group);
 916
 917         del_gendisk(zram->disk);
 918         put_disk(zram->disk);
 919
 920         blk_cleanup_queue(zram->queue);
 921 }
 922
 923 static int __init zram_init(void)
 924 {
 925         int ret, dev_id;
 926
 927         if (num_devices > max_num_devices) {
 928                 pr_warn("Invalid value for num_devices: %u\n",
 929                                 num_devices);
 930                 ret = -EINVAL;
 931                 goto out;
 932         }
 933
 934         zram_major = register_blkdev(0, "zram");
 935         if (zram_major <= 0) {
 936                 pr_warn("Unable to get major number\n");
 937                 ret = -EBUSY;
 938                 goto out;
 939         }
 940
 941         /* Allocate the device array and initialize each one */
 942         zram_devices = kzalloc(num_devices * sizeof(struct zram), GFP_KERNEL);
 943         if (!zram_devices) {
 944                 ret = -ENOMEM;
 945                 goto unregister;
 946         }
 947
 948         for (dev_id = 0; dev_id < num_devices; dev_id++) {
 949                 ret = create_device(&zram_devices[dev_id], dev_id);
 950                 if (ret)
 951                         goto free_devices;
 952         }
 953
 954         pr_info("Created %u device(s) ...\n", num_devices);
 955
 956         return 0;
 957
 958 free_devices:
 959         while (dev_id)
 960                 destroy_device(&zram_devices[--dev_id]);
 961         kfree(zram_devices);
 962 unregister:
 963         unregister_blkdev(zram_major, "zram");
 964 out:
 965         return ret;
 966 }
 967
 968 static void __exit zram_exit(void)
 969 {
 970         int i;
 971         struct zram *zram;
 972
 973         for (i = 0; i < num_devices; i++) {
 974                 zram = &zram_devices[i];
 975
 976                 destroy_device(zram);
 977                 /*
 978                  * Shouldn't access zram->disk after destroy_device
 979                  * because destroy_device already released zram->disk.
 980                  */
 981                 zram_reset_device(zram, false);
 982         }
 983
 984         unregister_blkdev(zram_major, "zram");
 985
 986         kfree(zram_devices);
 987         pr_debug("Cleanup done!\n");
 988 }
 989
 990 module_init(zram_init);
 991 module_exit(zram_exit);
 992
 993 module_param(num_devices, uint, 0);
 994 MODULE_PARM_DESC(num_devices, "Number of zram devices");
 995
 996 MODULE_LICENSE("Dual BSD/GPL");
 997 MODULE_AUTHOR("Nitin Gupta <ngupta@vflare.org>");
 998 MODULE_DESCRIPTION("Compressed RAM Block Device");