Pileus Git - ~andy/linux/blob - 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", 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)(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 static int zram_test_flag(struct zram_meta *meta, u32 index,
 144                         enum zram_pageflags flag)
 145 {
 146         return meta->table[index].flags & BIT(flag);
 147 }
 148
 149 static void zram_set_flag(struct zram_meta *meta, u32 index,
 150                         enum zram_pageflags flag)
 151 {
 152         meta->table[index].flags |= BIT(flag);
 153 }
 154
 155 static void zram_clear_flag(struct zram_meta *meta, u32 index,
 156                         enum zram_pageflags flag)
 157 {
 158         meta->table[index].flags &= ~BIT(flag);
 159 }
 160
 161 static inline int is_partial_io(struct bio_vec *bvec)
 162 {
 163         return bvec->bv_len != PAGE_SIZE;
 164 }
 165
 166 /*
 167  * Check if request is within bounds and aligned on zram logical blocks.
 168  */
 169 static inline int valid_io_request(struct zram *zram, struct bio *bio)
 170 {
 171         u64 start, end, bound;
 172
 173         /* unaligned request */
 174         if (unlikely(bio->bi_iter.bi_sector &
 175                      (ZRAM_SECTOR_PER_LOGICAL_BLOCK - 1)))
 176                 return 0;
 177         if (unlikely(bio->bi_iter.bi_size & (ZRAM_LOGICAL_BLOCK_SIZE - 1)))
 178                 return 0;
 179
 180         start = bio->bi_iter.bi_sector;
 181         end = start + (bio->bi_iter.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         return meta;
 232
 233 free_table:
 234         vfree(meta->table);
 235 free_buffer:
 236         free_pages((unsigned long)meta->compress_buffer, 1);
 237 free_workmem:
 238         kfree(meta->compress_workmem);
 239 free_meta:
 240         kfree(meta);
 241         meta = NULL;
 242 out:
 243         return meta;
 244 }
 245
 246 static void update_position(u32 *index, int *offset, struct bio_vec *bvec)
 247 {
 248         if (*offset + bvec->bv_len >= PAGE_SIZE)
 249                 (*index)++;
 250         *offset = (*offset + bvec->bv_len) % PAGE_SIZE;
 251 }
 252
 253 static int page_zero_filled(void *ptr)
 254 {
 255         unsigned int pos;
 256         unsigned long *page;
 257
 258         page = (unsigned long *)ptr;
 259
 260         for (pos = 0; pos != PAGE_SIZE / sizeof(*page); pos++) {
 261                 if (page[pos])
 262                         return 0;
 263         }
 264
 265         return 1;
 266 }
 267
 268 static void handle_zero_page(struct bio_vec *bvec)
 269 {
 270         struct page *page = bvec->bv_page;
 271         void *user_mem;
 272
 273         user_mem = kmap_atomic(page);
 274         if (is_partial_io(bvec))
 275                 memset(user_mem + bvec->bv_offset, 0, bvec->bv_len);
 276         else
 277                 clear_page(user_mem);
 278         kunmap_atomic(user_mem);
 279
 280         flush_dcache_page(page);
 281 }
 282
 283 static void zram_free_page(struct zram *zram, size_t index)
 284 {
 285         struct zram_meta *meta = zram->meta;
 286         unsigned long handle = meta->table[index].handle;
 287         u16 size = meta->table[index].size;
 288
 289         if (unlikely(!handle)) {
 290                 /*
 291                  * No memory is allocated for zero filled pages.
 292                  * Simply clear zero page flag.
 293                  */
 294                 if (zram_test_flag(meta, index, ZRAM_ZERO)) {
 295                         zram_clear_flag(meta, index, ZRAM_ZERO);
 296                         zram->stats.pages_zero--;
 297                 }
 298                 return;
 299         }
 300
 301         if (unlikely(size > max_zpage_size))
 302                 zram->stats.bad_compress--;
 303
 304         zs_free(meta->mem_pool, handle);
 305
 306         if (size <= PAGE_SIZE / 2)
 307                 zram->stats.good_compress--;
 308
 309         atomic64_sub(meta->table[index].size, &zram->stats.compr_size);
 310         zram->stats.pages_stored--;
 311
 312         meta->table[index].handle = 0;
 313         meta->table[index].size = 0;
 314 }
 315
 316 static int zram_decompress_page(struct zram *zram, char *mem, u32 index)
 317 {
 318         int ret = LZO_E_OK;
 319         size_t clen = PAGE_SIZE;
 320         unsigned char *cmem;
 321         struct zram_meta *meta = zram->meta;
 322         unsigned long handle = meta->table[index].handle;
 323
 324         if (!handle || zram_test_flag(meta, index, ZRAM_ZERO)) {
 325                 clear_page(mem);
 326                 return 0;
 327         }
 328
 329         cmem = zs_map_object(meta->mem_pool, handle, ZS_MM_RO);
 330         if (meta->table[index].size == PAGE_SIZE)
 331                 copy_page(mem, cmem);
 332         else
 333                 ret = lzo1x_decompress_safe(cmem, meta->table[index].size,
 334                                                 mem, &clen);
 335         zs_unmap_object(meta->mem_pool, handle);
 336
 337         /* Should NEVER happen. Return bio error if it does. */
 338         if (unlikely(ret != LZO_E_OK)) {
 339                 pr_err("Decompression failed! err=%d, page=%u\n", ret, index);
 340                 atomic64_inc(&zram->stats.failed_reads);
 341                 return ret;
 342         }
 343
 344         return 0;
 345 }
 346
 347 static int zram_bvec_read(struct zram *zram, struct bio_vec *bvec,
 348                           u32 index, int offset, struct bio *bio)
 349 {
 350         int ret;
 351         struct page *page;
 352         unsigned char *user_mem, *uncmem = NULL;
 353         struct zram_meta *meta = zram->meta;
 354         page = bvec->bv_page;
 355
 356         if (unlikely(!meta->table[index].handle) ||
 357                         zram_test_flag(meta, index, ZRAM_ZERO)) {
 358                 handle_zero_page(bvec);
 359                 return 0;
 360         }
 361
 362         if (is_partial_io(bvec))
 363                 /* Use  a temporary buffer to decompress the page */
 364                 uncmem = kmalloc(PAGE_SIZE, GFP_NOIO);
 365
 366         user_mem = kmap_atomic(page);
 367         if (!is_partial_io(bvec))
 368                 uncmem = user_mem;
 369
 370         if (!uncmem) {
 371                 pr_info("Unable to allocate temp memory\n");
 372                 ret = -ENOMEM;
 373                 goto out_cleanup;
 374         }
 375
 376         ret = zram_decompress_page(zram, uncmem, index);
 377         /* Should NEVER happen. Return bio error if it does. */
 378         if (unlikely(ret != LZO_E_OK))
 379                 goto out_cleanup;
 380
 381         if (is_partial_io(bvec))
 382                 memcpy(user_mem + bvec->bv_offset, uncmem + offset,
 383                                 bvec->bv_len);
 384
 385         flush_dcache_page(page);
 386         ret = 0;
 387 out_cleanup:
 388         kunmap_atomic(user_mem);
 389         if (is_partial_io(bvec))
 390                 kfree(uncmem);
 391         return ret;
 392 }
 393
 394 static int zram_bvec_write(struct zram *zram, struct bio_vec *bvec, u32 index,
 395                            int offset)
 396 {
 397         int ret = 0;
 398         size_t clen;
 399         unsigned long handle;
 400         struct page *page;
 401         unsigned char *user_mem, *cmem, *src, *uncmem = NULL;
 402         struct zram_meta *meta = zram->meta;
 403
 404         page = bvec->bv_page;
 405         src = meta->compress_buffer;
 406
 407         if (is_partial_io(bvec)) {
 408                 /*
 409                  * This is a partial IO. We need to read the full page
 410                  * before to write the changes.
 411                  */
 412                 uncmem = kmalloc(PAGE_SIZE, GFP_NOIO);
 413                 if (!uncmem) {
 414                         ret = -ENOMEM;
 415                         goto out;
 416                 }
 417                 ret = zram_decompress_page(zram, uncmem, index);
 418                 if (ret)
 419                         goto out;
 420         }
 421
 422         user_mem = kmap_atomic(page);
 423
 424         if (is_partial_io(bvec)) {
 425                 memcpy(uncmem + offset, user_mem + bvec->bv_offset,
 426                        bvec->bv_len);
 427                 kunmap_atomic(user_mem);
 428                 user_mem = NULL;
 429         } else {
 430                 uncmem = user_mem;
 431         }
 432
 433         if (page_zero_filled(uncmem)) {
 434                 kunmap_atomic(user_mem);
 435                 /* Free memory associated with this sector now. */
 436                 zram_free_page(zram, index);
 437
 438                 zram->stats.pages_zero++;
 439                 zram_set_flag(meta, index, ZRAM_ZERO);
 440                 ret = 0;
 441                 goto out;
 442         }
 443
 444         ret = lzo1x_1_compress(uncmem, PAGE_SIZE, src, &clen,
 445                                meta->compress_workmem);
 446
 447         if (!is_partial_io(bvec)) {
 448                 kunmap_atomic(user_mem);
 449                 user_mem = NULL;
 450                 uncmem = NULL;
 451         }
 452
 453         if (unlikely(ret != LZO_E_OK)) {
 454                 pr_err("Compression failed! err=%d\n", ret);
 455                 goto out;
 456         }
 457
 458         if (unlikely(clen > max_zpage_size)) {
 459                 zram->stats.bad_compress++;
 460                 clen = PAGE_SIZE;
 461                 src = NULL;
 462                 if (is_partial_io(bvec))
 463                         src = uncmem;
 464         }
 465
 466         handle = zs_malloc(meta->mem_pool, clen);
 467         if (!handle) {
 468                 pr_info("Error allocating memory for compressed page: %u, size=%zu\n",
 469                         index, clen);
 470                 ret = -ENOMEM;
 471                 goto out;
 472         }
 473         cmem = zs_map_object(meta->mem_pool, handle, ZS_MM_WO);
 474
 475         if ((clen == PAGE_SIZE) && !is_partial_io(bvec)) {
 476                 src = kmap_atomic(page);
 477                 copy_page(cmem, src);
 478                 kunmap_atomic(src);
 479         } else {
 480                 memcpy(cmem, src, clen);
 481         }
 482
 483         zs_unmap_object(meta->mem_pool, handle);
 484
 485         /*
 486          * Free memory associated with this sector
 487          * before overwriting unused sectors.
 488          */
 489         zram_free_page(zram, index);
 490
 491         meta->table[index].handle = handle;
 492         meta->table[index].size = clen;
 493
 494         /* Update stats */
 495         atomic64_add(clen, &zram->stats.compr_size);
 496         zram->stats.pages_stored++;
 497         if (clen <= PAGE_SIZE / 2)
 498                 zram->stats.good_compress++;
 499
 500 out:
 501         if (is_partial_io(bvec))
 502                 kfree(uncmem);
 503
 504         if (ret)
 505                 atomic64_inc(&zram->stats.failed_writes);
 506         return ret;
 507 }
 508
 509 static void handle_pending_slot_free(struct zram *zram)
 510 {
 511         struct zram_slot_free *free_rq;
 512
 513         spin_lock(&zram->slot_free_lock);
 514         while (zram->slot_free_rq) {
 515                 free_rq = zram->slot_free_rq;
 516                 zram->slot_free_rq = free_rq->next;
 517                 zram_free_page(zram, free_rq->index);
 518                 kfree(free_rq);
 519         }
 520         spin_unlock(&zram->slot_free_lock);
 521 }
 522
 523 static int zram_bvec_rw(struct zram *zram, struct bio_vec *bvec, u32 index,
 524                         int offset, struct bio *bio, int rw)
 525 {
 526         int ret;
 527
 528         if (rw == READ) {
 529                 down_read(&zram->lock);
 530                 ret = zram_bvec_read(zram, bvec, index, offset, bio);
 531                 up_read(&zram->lock);
 532         } else {
 533                 down_write(&zram->lock);
 534                 handle_pending_slot_free(zram);
 535                 ret = zram_bvec_write(zram, bvec, index, offset);
 536                 up_write(&zram->lock);
 537         }
 538
 539         return ret;
 540 }
 541
 542 static void zram_reset_device(struct zram *zram, bool reset_capacity)
 543 {
 544         size_t index;
 545         struct zram_meta *meta;
 546
 547         down_write(&zram->init_lock);
 548         if (!zram->init_done) {
 549                 up_write(&zram->init_lock);
 550                 return;
 551         }
 552
 553         flush_work(&zram->free_work);
 554
 555         meta = zram->meta;
 556         zram->init_done = 0;
 557
 558         /* Free all pages that are still in this zram device */
 559         for (index = 0; index < zram->disksize >> PAGE_SHIFT; index++) {
 560                 unsigned long handle = meta->table[index].handle;
 561                 if (!handle)
 562                         continue;
 563
 564                 zs_free(meta->mem_pool, handle);
 565         }
 566
 567         zram_meta_free(zram->meta);
 568         zram->meta = NULL;
 569         /* Reset stats */
 570         memset(&zram->stats, 0, sizeof(zram->stats));
 571
 572         zram->disksize = 0;
 573         if (reset_capacity)
 574                 set_capacity(zram->disk, 0);
 575         up_write(&zram->init_lock);
 576 }
 577
 578 static void zram_init_device(struct zram *zram, struct zram_meta *meta)
 579 {
 580         if (zram->disksize > 2 * (totalram_pages << PAGE_SHIFT)) {
 581                 pr_info(
 582                 "There is little point creating a zram of greater than "
 583                 "twice the size of memory since we expect a 2:1 compression "
 584                 "ratio. Note that zram uses about 0.1%% of the size of "
 585                 "the disk when not in use so a huge zram is "
 586                 "wasteful.\n"
 587                 "\tMemory Size: %lu kB\n"
 588                 "\tSize you selected: %llu kB\n"
 589                 "Continuing anyway ...\n",
 590                 (totalram_pages << PAGE_SHIFT) >> 10, zram->disksize >> 10
 591                 );
 592         }
 593
 594         /* zram devices sort of resembles non-rotational disks */
 595         queue_flag_set_unlocked(QUEUE_FLAG_NONROT, zram->disk->queue);
 596
 597         zram->meta = meta;
 598         zram->init_done = 1;
 599
 600         pr_debug("Initialization done!\n");
 601 }
 602
 603 static ssize_t disksize_store(struct device *dev,
 604                 struct device_attribute *attr, const char *buf, size_t len)
 605 {
 606         u64 disksize;
 607         struct zram_meta *meta;
 608         struct zram *zram = dev_to_zram(dev);
 609
 610         disksize = memparse(buf, NULL);
 611         if (!disksize)
 612                 return -EINVAL;
 613
 614         disksize = PAGE_ALIGN(disksize);
 615         meta = zram_meta_alloc(disksize);
 616         down_write(&zram->init_lock);
 617         if (zram->init_done) {
 618                 up_write(&zram->init_lock);
 619                 zram_meta_free(meta);
 620                 pr_info("Cannot change disksize for initialized device\n");
 621                 return -EBUSY;
 622         }
 623
 624         zram->disksize = disksize;
 625         set_capacity(zram->disk, zram->disksize >> SECTOR_SHIFT);
 626         zram_init_device(zram, meta);
 627         up_write(&zram->init_lock);
 628
 629         return len;
 630 }
 631
 632 static ssize_t reset_store(struct device *dev,
 633                 struct device_attribute *attr, const char *buf, size_t len)
 634 {
 635         int ret;
 636         unsigned short do_reset;
 637         struct zram *zram;
 638         struct block_device *bdev;
 639
 640         zram = dev_to_zram(dev);
 641         bdev = bdget_disk(zram->disk, 0);
 642
 643         if (!bdev)
 644                 return -ENOMEM;
 645
 646         /* Do not reset an active device! */
 647         if (bdev->bd_holders) {
 648                 ret = -EBUSY;
 649                 goto out;
 650         }
 651
 652         ret = kstrtou16(buf, 10, &do_reset);
 653         if (ret)
 654                 goto out;
 655
 656         if (!do_reset) {
 657                 ret = -EINVAL;
 658                 goto out;
 659         }
 660
 661         /* Make sure all pending I/O is finished */
 662         fsync_bdev(bdev);
 663         bdput(bdev);
 664
 665         zram_reset_device(zram, true);
 666         return len;
 667
 668 out:
 669         bdput(bdev);
 670         return ret;
 671 }
 672
 673 static void __zram_make_request(struct zram *zram, struct bio *bio, int rw)
 674 {
 675         int offset;
 676         u32 index;
 677         struct bio_vec bvec;
 678         struct bvec_iter iter;
 679
 680         switch (rw) {
 681         case READ:
 682                 atomic64_inc(&zram->stats.num_reads);
 683                 break;
 684         case WRITE:
 685                 atomic64_inc(&zram->stats.num_writes);
 686                 break;
 687         }
 688
 689         index = bio->bi_iter.bi_sector >> SECTORS_PER_PAGE_SHIFT;
 690         offset = (bio->bi_iter.bi_sector &
 691                   (SECTORS_PER_PAGE - 1)) << SECTOR_SHIFT;
 692
 693         bio_for_each_segment(bvec, bio, iter) {
 694                 int max_transfer_size = PAGE_SIZE - offset;
 695
 696                 if (bvec.bv_len > max_transfer_size) {
 697                         /*
 698                          * zram_bvec_rw() can only make operation on a single
 699                          * zram page. Split the bio vector.
 700                          */
 701                         struct bio_vec bv;
 702
 703                         bv.bv_page = bvec.bv_page;
 704                         bv.bv_len = max_transfer_size;
 705                         bv.bv_offset = bvec.bv_offset;
 706
 707                         if (zram_bvec_rw(zram, &bv, index, offset, bio, rw) < 0)
 708                                 goto out;
 709
 710                         bv.bv_len = bvec.bv_len - max_transfer_size;
 711                         bv.bv_offset += max_transfer_size;
 712                         if (zram_bvec_rw(zram, &bv, index+1, 0, bio, rw) < 0)
 713                                 goto out;
 714                 } else
 715                         if (zram_bvec_rw(zram, &bvec, index, offset, bio, rw)
 716                             < 0)
 717                                 goto out;
 718
 719                 update_position(&index, &offset, &bvec);
 720         }
 721
 722         set_bit(BIO_UPTODATE, &bio->bi_flags);
 723         bio_endio(bio, 0);
 724         return;
 725
 726 out:
 727         bio_io_error(bio);
 728 }
 729
 730 /*
 731  * Handler function for all zram I/O requests.
 732  */
 733 static void zram_make_request(struct request_queue *queue, struct bio *bio)
 734 {
 735         struct zram *zram = queue->queuedata;
 736
 737         down_read(&zram->init_lock);
 738         if (unlikely(!zram->init_done))
 739                 goto error;
 740
 741         if (!valid_io_request(zram, bio)) {
 742                 atomic64_inc(&zram->stats.invalid_io);
 743                 goto error;
 744         }
 745
 746         __zram_make_request(zram, bio, bio_data_dir(bio));
 747         up_read(&zram->init_lock);
 748
 749         return;
 750
 751 error:
 752         up_read(&zram->init_lock);
 753         bio_io_error(bio);
 754 }
 755
 756 static void zram_slot_free(struct work_struct *work)
 757 {
 758         struct zram *zram;
 759
 760         zram = container_of(work, struct zram, free_work);
 761         down_write(&zram->lock);
 762         handle_pending_slot_free(zram);
 763         up_write(&zram->lock);
 764 }
 765
 766 static void add_slot_free(struct zram *zram, struct zram_slot_free *free_rq)
 767 {
 768         spin_lock(&zram->slot_free_lock);
 769         free_rq->next = zram->slot_free_rq;
 770         zram->slot_free_rq = free_rq;
 771         spin_unlock(&zram->slot_free_lock);
 772 }
 773
 774 static void zram_slot_free_notify(struct block_device *bdev,
 775                                 unsigned long index)
 776 {
 777         struct zram *zram;
 778         struct zram_slot_free *free_rq;
 779
 780         zram = bdev->bd_disk->private_data;
 781         atomic64_inc(&zram->stats.notify_free);
 782
 783         free_rq = kmalloc(sizeof(struct zram_slot_free), GFP_ATOMIC);
 784         if (!free_rq)
 785                 return;
 786
 787         free_rq->index = index;
 788         add_slot_free(zram, free_rq);
 789         schedule_work(&zram->free_work);
 790 }
 791
 792 static const struct block_device_operations zram_devops = {
 793         .swap_slot_free_notify = zram_slot_free_notify,
 794         .owner = THIS_MODULE
 795 };
 796
 797 static DEVICE_ATTR(disksize, S_IRUGO | S_IWUSR,
 798                 disksize_show, disksize_store);
 799 static DEVICE_ATTR(initstate, S_IRUGO, initstate_show, NULL);
 800 static DEVICE_ATTR(reset, S_IWUSR, NULL, reset_store);
 801 static DEVICE_ATTR(num_reads, S_IRUGO, num_reads_show, NULL);
 802 static DEVICE_ATTR(num_writes, S_IRUGO, num_writes_show, NULL);
 803 static DEVICE_ATTR(invalid_io, S_IRUGO, invalid_io_show, NULL);
 804 static DEVICE_ATTR(notify_free, S_IRUGO, notify_free_show, NULL);
 805 static DEVICE_ATTR(zero_pages, S_IRUGO, zero_pages_show, NULL);
 806 static DEVICE_ATTR(orig_data_size, S_IRUGO, orig_data_size_show, NULL);
 807 static DEVICE_ATTR(compr_data_size, S_IRUGO, compr_data_size_show, NULL);
 808 static DEVICE_ATTR(mem_used_total, S_IRUGO, mem_used_total_show, NULL);
 809
 810 static struct attribute *zram_disk_attrs[] = {
 811         &dev_attr_disksize.attr,
 812         &dev_attr_initstate.attr,
 813         &dev_attr_reset.attr,
 814         &dev_attr_num_reads.attr,
 815         &dev_attr_num_writes.attr,
 816         &dev_attr_invalid_io.attr,
 817         &dev_attr_notify_free.attr,
 818         &dev_attr_zero_pages.attr,
 819         &dev_attr_orig_data_size.attr,
 820         &dev_attr_compr_data_size.attr,
 821         &dev_attr_mem_used_total.attr,
 822         NULL,
 823 };
 824
 825 static struct attribute_group zram_disk_attr_group = {
 826         .attrs = zram_disk_attrs,
 827 };
 828
 829 static int create_device(struct zram *zram, int device_id)
 830 {
 831         int ret = -ENOMEM;
 832
 833         init_rwsem(&zram->lock);
 834         init_rwsem(&zram->init_lock);
 835
 836         INIT_WORK(&zram->free_work, zram_slot_free);
 837         spin_lock_init(&zram->slot_free_lock);
 838         zram->slot_free_rq = NULL;
 839
 840         zram->queue = blk_alloc_queue(GFP_KERNEL);
 841         if (!zram->queue) {
 842                 pr_err("Error allocating disk queue for device %d\n",
 843                         device_id);
 844                 goto out;
 845         }
 846
 847         blk_queue_make_request(zram->queue, zram_make_request);
 848         zram->queue->queuedata = zram;
 849
 850          /* gendisk structure */
 851         zram->disk = alloc_disk(1);
 852         if (!zram->disk) {
 853                 pr_warn("Error allocating disk structure for device %d\n",
 854                         device_id);
 855                 goto out_free_queue;
 856         }
 857
 858         zram->disk->major = zram_major;
 859         zram->disk->first_minor = device_id;
 860         zram->disk->fops = &zram_devops;
 861         zram->disk->queue = zram->queue;
 862         zram->disk->private_data = zram;
 863         snprintf(zram->disk->disk_name, 16, "zram%d", device_id);
 864
 865         /* Actual capacity set using syfs (/sys/block/zram<id>/disksize */
 866         set_capacity(zram->disk, 0);
 867
 868         /*
 869          * To ensure that we always get PAGE_SIZE aligned
 870          * and n*PAGE_SIZED sized I/O requests.
 871          */
 872         blk_queue_physical_block_size(zram->disk->queue, PAGE_SIZE);
 873         blk_queue_logical_block_size(zram->disk->queue,
 874                                         ZRAM_LOGICAL_BLOCK_SIZE);
 875         blk_queue_io_min(zram->disk->queue, PAGE_SIZE);
 876         blk_queue_io_opt(zram->disk->queue, PAGE_SIZE);
 877
 878         add_disk(zram->disk);
 879
 880         ret = sysfs_create_group(&disk_to_dev(zram->disk)->kobj,
 881                                 &zram_disk_attr_group);
 882         if (ret < 0) {
 883                 pr_warn("Error creating sysfs group");
 884                 goto out_free_disk;
 885         }
 886
 887         zram->init_done = 0;
 888         return 0;
 889
 890 out_free_disk:
 891         del_gendisk(zram->disk);
 892         put_disk(zram->disk);
 893 out_free_queue:
 894         blk_cleanup_queue(zram->queue);
 895 out:
 896         return ret;
 897 }
 898
 899 static void destroy_device(struct zram *zram)
 900 {
 901         sysfs_remove_group(&disk_to_dev(zram->disk)->kobj,
 902                         &zram_disk_attr_group);
 903
 904         del_gendisk(zram->disk);
 905         put_disk(zram->disk);
 906
 907         blk_cleanup_queue(zram->queue);
 908 }
 909
 910 static int __init zram_init(void)
 911 {
 912         int ret, dev_id;
 913
 914         if (num_devices > max_num_devices) {
 915                 pr_warn("Invalid value for num_devices: %u\n",
 916                                 num_devices);
 917                 ret = -EINVAL;
 918                 goto out;
 919         }
 920
 921         zram_major = register_blkdev(0, "zram");
 922         if (zram_major <= 0) {
 923                 pr_warn("Unable to get major number\n");
 924                 ret = -EBUSY;
 925                 goto out;
 926         }
 927
 928         /* Allocate the device array and initialize each one */
 929         zram_devices = kzalloc(num_devices * sizeof(struct zram), GFP_KERNEL);
 930         if (!zram_devices) {
 931                 ret = -ENOMEM;
 932                 goto unregister;
 933         }
 934
 935         for (dev_id = 0; dev_id < num_devices; dev_id++) {
 936                 ret = create_device(&zram_devices[dev_id], dev_id);
 937                 if (ret)
 938                         goto free_devices;
 939         }
 940
 941         pr_info("Created %u device(s) ...\n", num_devices);
 942
 943         return 0;
 944
 945 free_devices:
 946         while (dev_id)
 947                 destroy_device(&zram_devices[--dev_id]);
 948         kfree(zram_devices);
 949 unregister:
 950         unregister_blkdev(zram_major, "zram");
 951 out:
 952         return ret;
 953 }
 954
 955 static void __exit zram_exit(void)
 956 {
 957         int i;
 958         struct zram *zram;
 959
 960         for (i = 0; i < num_devices; i++) {
 961                 zram = &zram_devices[i];
 962
 963                 destroy_device(zram);
 964                 /*
 965                  * Shouldn't access zram->disk after destroy_device
 966                  * because destroy_device already released zram->disk.
 967                  */
 968                 zram_reset_device(zram, false);
 969         }
 970
 971         unregister_blkdev(zram_major, "zram");
 972
 973         kfree(zram_devices);
 974         pr_debug("Cleanup done!\n");
 975 }
 976
 977 module_init(zram_init);
 978 module_exit(zram_exit);
 979
 980 module_param(num_devices, uint, 0);
 981 MODULE_PARM_DESC(num_devices, "Number of zram devices");
 982
 983 MODULE_LICENSE("Dual BSD/GPL");
 984 MODULE_AUTHOR("Nitin Gupta <ngupta@vflare.org>");
 985 MODULE_DESCRIPTION("Compressed RAM Block Device");