drivers/staging/zram/zram_drv.c

   1 /*
   2  * Compressed RAM block device
   3  *
   4  * Copyright (C) 2008, 2009, 2010  Nitin Gupta
   5  *
   6  * This code is released using a dual license strategy: BSD/GPL
   7  * You can choose the licence that better fits your requirements.
   8  *
   9  * Released under the terms of 3-clause BSD License
  10  * Released under the terms of GNU General Public License Version 2.0
  11  *
  12  * Project home: http://compcache.googlecode.com
  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_sector & (ZRAM_SECTOR_PER_LOGICAL_BLOCK - 1)))
 175                 return 0;
 176         if (unlikely(bio->bi_size & (ZRAM_LOGICAL_BLOCK_SIZE - 1)))
 177                 return 0;
 178
 179         start = bio->bi_sector;
 180         end = start + (bio->bi_size >> SECTOR_SHIFT);
 181         bound = zram->disksize >> SECTOR_SHIFT;
 182         /* out of range range */
 183         if (unlikely(start >= bound || end > bound || start > end))
 184                 return 0;
 185
 186         /* I/O request is valid */
 187         return 1;
 188 }
 189
 190 static void zram_meta_free(struct zram_meta *meta)
 191 {
 192         zs_destroy_pool(meta->mem_pool);
 193         kfree(meta->compress_workmem);
 194         free_pages((unsigned long)meta->compress_buffer, 1);
 195         vfree(meta->table);
 196         kfree(meta);
 197 }
 198
 199 static struct zram_meta *zram_meta_alloc(u64 disksize)
 200 {
 201         size_t num_pages;
 202         struct zram_meta *meta = kmalloc(sizeof(*meta), GFP_KERNEL);
 203         if (!meta)
 204                 goto out;
 205
 206         meta->compress_workmem = kzalloc(LZO1X_MEM_COMPRESS, GFP_KERNEL);
 207         if (!meta->compress_workmem)
 208                 goto free_meta;
 209
 210         meta->compress_buffer =
 211                 (void *)__get_free_pages(GFP_KERNEL | __GFP_ZERO, 1);
 212         if (!meta->compress_buffer) {
 213                 pr_err("Error allocating compressor buffer space\n");
 214                 goto free_workmem;
 215         }
 216
 217         num_pages = disksize >> PAGE_SHIFT;
 218         meta->table = vzalloc(num_pages * sizeof(*meta->table));
 219         if (!meta->table) {
 220                 pr_err("Error allocating zram address table\n");
 221                 goto free_buffer;
 222         }
 223
 224         meta->mem_pool = zs_create_pool(GFP_NOIO | __GFP_HIGHMEM);
 225         if (!meta->mem_pool) {
 226                 pr_err("Error creating memory pool\n");
 227                 goto free_table;
 228         }
 229
 230         return meta;
 231
 232 free_table:
 233         vfree(meta->table);
 234 free_buffer:
 235         free_pages((unsigned long)meta->compress_buffer, 1);
 236 free_workmem:
 237         kfree(meta->compress_workmem);
 238 free_meta:
 239         kfree(meta);
 240         meta = NULL;
 241 out:
 242         return meta;
 243 }
 244
 245 static void update_position(u32 *index, int *offset, struct bio_vec *bvec)
 246 {
 247         if (*offset + bvec->bv_len >= PAGE_SIZE)
 248                 (*index)++;
 249         *offset = (*offset + bvec->bv_len) % PAGE_SIZE;
 250 }
 251
 252 static int page_zero_filled(void *ptr)
 253 {
 254         unsigned int pos;
 255         unsigned long *page;
 256
 257         page = (unsigned long *)ptr;
 258
 259         for (pos = 0; pos != PAGE_SIZE / sizeof(*page); pos++) {
 260                 if (page[pos])
 261                         return 0;
 262         }
 263
 264         return 1;
 265 }
 266
 267 static void handle_zero_page(struct bio_vec *bvec)
 268 {
 269         struct page *page = bvec->bv_page;
 270         void *user_mem;
 271
 272         user_mem = kmap_atomic(page);
 273         if (is_partial_io(bvec))
 274                 memset(user_mem + bvec->bv_offset, 0, bvec->bv_len);
 275         else
 276                 clear_page(user_mem);
 277         kunmap_atomic(user_mem);
 278
 279         flush_dcache_page(page);
 280 }
 281
 282 static void zram_free_page(struct zram *zram, size_t index)
 283 {
 284         struct zram_meta *meta = zram->meta;
 285         unsigned long handle = meta->table[index].handle;
 286         u16 size = meta->table[index].size;
 287
 288         if (unlikely(!handle)) {
 289                 /*
 290                  * No memory is allocated for zero filled pages.
 291                  * Simply clear zero page flag.
 292                  */
 293                 if (zram_test_flag(meta, index, ZRAM_ZERO)) {
 294                         zram_clear_flag(meta, index, ZRAM_ZERO);
 295                         zram->stats.pages_zero--;
 296                 }
 297                 return;
 298         }
 299
 300         if (unlikely(size > max_zpage_size))
 301                 zram->stats.bad_compress--;
 302
 303         zs_free(meta->mem_pool, handle);
 304
 305         if (size <= PAGE_SIZE / 2)
 306                 zram->stats.good_compress--;
 307
 308         atomic64_sub(meta->table[index].size, &zram->stats.compr_size);
 309         zram->stats.pages_stored--;
 310
 311         meta->table[index].handle = 0;
 312         meta->table[index].size = 0;
 313 }
 314
 315 static int zram_decompress_page(struct zram *zram, char *mem, u32 index)
 316 {
 317         int ret = LZO_E_OK;
 318         size_t clen = PAGE_SIZE;
 319         unsigned char *cmem;
 320         struct zram_meta *meta = zram->meta;
 321         unsigned long handle = meta->table[index].handle;
 322
 323         if (!handle || zram_test_flag(meta, index, ZRAM_ZERO)) {
 324                 clear_page(mem);
 325                 return 0;
 326         }
 327
 328         cmem = zs_map_object(meta->mem_pool, handle, ZS_MM_RO);
 329         if (meta->table[index].size == PAGE_SIZE)
 330                 copy_page(mem, cmem);
 331         else
 332                 ret = lzo1x_decompress_safe(cmem, meta->table[index].size,
 333                                                 mem, &clen);
 334         zs_unmap_object(meta->mem_pool, handle);
 335
 336         /* Should NEVER happen. Return bio error if it does. */
 337         if (unlikely(ret != LZO_E_OK)) {
 338                 pr_err("Decompression failed! err=%d, page=%u\n", ret, index);
 339                 atomic64_inc(&zram->stats.failed_reads);
 340                 return ret;
 341         }
 342
 343         return 0;
 344 }
 345
 346 static int zram_bvec_read(struct zram *zram, struct bio_vec *bvec,
 347                           u32 index, int offset, struct bio *bio)
 348 {
 349         int ret;
 350         struct page *page;
 351         unsigned char *user_mem, *uncmem = NULL;
 352         struct zram_meta *meta = zram->meta;
 353         page = bvec->bv_page;
 354
 355         if (unlikely(!meta->table[index].handle) ||
 356                         zram_test_flag(meta, index, ZRAM_ZERO)) {
 357                 handle_zero_page(bvec);
 358                 return 0;
 359         }
 360
 361         if (is_partial_io(bvec))
 362                 /* Use  a temporary buffer to decompress the page */
 363                 uncmem = kmalloc(PAGE_SIZE, GFP_NOIO);
 364
 365         user_mem = kmap_atomic(page);
 366         if (!is_partial_io(bvec))
 367                 uncmem = user_mem;
 368
 369         if (!uncmem) {
 370                 pr_info("Unable to allocate temp memory\n");
 371                 ret = -ENOMEM;
 372                 goto out_cleanup;
 373         }
 374
 375         ret = zram_decompress_page(zram, uncmem, index);
 376         /* Should NEVER happen. Return bio error if it does. */
 377         if (unlikely(ret != LZO_E_OK))
 378                 goto out_cleanup;
 379
 380         if (is_partial_io(bvec))
 381                 memcpy(user_mem + bvec->bv_offset, uncmem + offset,
 382                                 bvec->bv_len);
 383
 384         flush_dcache_page(page);
 385         ret = 0;
 386 out_cleanup:
 387         kunmap_atomic(user_mem);
 388         if (is_partial_io(bvec))
 389                 kfree(uncmem);
 390         return ret;
 391 }
 392
 393 static int zram_bvec_write(struct zram *zram, struct bio_vec *bvec, u32 index,
 394                            int offset)
 395 {
 396         int ret = 0;
 397         size_t clen;
 398         unsigned long handle;
 399         struct page *page;
 400         unsigned char *user_mem, *cmem, *src, *uncmem = NULL;
 401         struct zram_meta *meta = zram->meta;
 402
 403         page = bvec->bv_page;
 404         src = meta->compress_buffer;
 405
 406         if (is_partial_io(bvec)) {
 407                 /*
 408                  * This is a partial IO. We need to read the full page
 409                  * before to write the changes.
 410                  */
 411                 uncmem = kmalloc(PAGE_SIZE, GFP_NOIO);
 412                 if (!uncmem) {
 413                         ret = -ENOMEM;
 414                         goto out;
 415                 }
 416                 ret = zram_decompress_page(zram, uncmem, index);
 417                 if (ret)
 418                         goto out;
 419         }
 420
 421         /*
 422          * System overwrites unused sectors. Free memory associated
 423          * with this sector now.
 424          */
 425         if (meta->table[index].handle ||
 426             zram_test_flag(meta, index, ZRAM_ZERO))
 427                 zram_free_page(zram, index);
 428
 429         user_mem = kmap_atomic(page);
 430
 431         if (is_partial_io(bvec)) {
 432                 memcpy(uncmem + offset, user_mem + bvec->bv_offset,
 433                        bvec->bv_len);
 434                 kunmap_atomic(user_mem);
 435                 user_mem = NULL;
 436         } else {
 437                 uncmem = user_mem;
 438         }
 439
 440         if (page_zero_filled(uncmem)) {
 441                 kunmap_atomic(user_mem);
 442                 zram->stats.pages_zero++;
 443                 zram_set_flag(meta, index, ZRAM_ZERO);
 444                 ret = 0;
 445                 goto out;
 446         }
 447
 448         ret = lzo1x_1_compress(uncmem, PAGE_SIZE, src, &clen,
 449                                meta->compress_workmem);
 450
 451         if (!is_partial_io(bvec)) {
 452                 kunmap_atomic(user_mem);
 453                 user_mem = NULL;
 454                 uncmem = NULL;
 455         }
 456
 457         if (unlikely(ret != LZO_E_OK)) {
 458                 pr_err("Compression failed! err=%d\n", ret);
 459                 goto out;
 460         }
 461
 462         if (unlikely(clen > max_zpage_size)) {
 463                 zram->stats.bad_compress++;
 464                 clen = PAGE_SIZE;
 465                 src = NULL;
 466                 if (is_partial_io(bvec))
 467                         src = uncmem;
 468         }
 469
 470         handle = zs_malloc(meta->mem_pool, clen);
 471         if (!handle) {
 472                 pr_info("Error allocating memory for compressed page: %u, size=%zu\n",
 473                         index, clen);
 474                 ret = -ENOMEM;
 475                 goto out;
 476         }
 477         cmem = zs_map_object(meta->mem_pool, handle, ZS_MM_WO);
 478
 479         if ((clen == PAGE_SIZE) && !is_partial_io(bvec)) {
 480                 src = kmap_atomic(page);
 481                 copy_page(cmem, src);
 482                 kunmap_atomic(src);
 483         } else {
 484                 memcpy(cmem, src, clen);
 485         }
 486
 487         zs_unmap_object(meta->mem_pool, handle);
 488
 489         meta->table[index].handle = handle;
 490         meta->table[index].size = clen;
 491
 492         /* Update stats */
 493         atomic64_add(clen, &zram->stats.compr_size);
 494         zram->stats.pages_stored++;
 495         if (clen <= PAGE_SIZE / 2)
 496                 zram->stats.good_compress++;
 497
 498 out:
 499         if (is_partial_io(bvec))
 500                 kfree(uncmem);
 501
 502         if (ret)
 503                 atomic64_inc(&zram->stats.failed_writes);
 504         return ret;
 505 }
 506
 507 static int zram_bvec_rw(struct zram *zram, struct bio_vec *bvec, u32 index,
 508                         int offset, struct bio *bio, int rw)
 509 {
 510         int ret;
 511
 512         if (rw == READ) {
 513                 down_read(&zram->lock);
 514                 ret = zram_bvec_read(zram, bvec, index, offset, bio);
 515                 up_read(&zram->lock);
 516         } else {
 517                 down_write(&zram->lock);
 518                 ret = zram_bvec_write(zram, bvec, index, offset);
 519                 up_write(&zram->lock);
 520         }
 521
 522         return ret;
 523 }
 524
 525 static void zram_reset_device(struct zram *zram)
 526 {
 527         size_t index;
 528         struct zram_meta *meta;
 529
 530         down_write(&zram->init_lock);
 531         if (!zram->init_done) {
 532                 up_write(&zram->init_lock);
 533                 return;
 534         }
 535
 536         meta = zram->meta;
 537         zram->init_done = 0;
 538
 539         /* Free all pages that are still in this zram device */
 540         for (index = 0; index < zram->disksize >> PAGE_SHIFT; index++) {
 541                 unsigned long handle = meta->table[index].handle;
 542                 if (!handle)
 543                         continue;
 544
 545                 zs_free(meta->mem_pool, handle);
 546         }
 547
 548         zram_meta_free(zram->meta);
 549         zram->meta = NULL;
 550         /* Reset stats */
 551         memset(&zram->stats, 0, sizeof(zram->stats));
 552
 553         zram->disksize = 0;
 554         set_capacity(zram->disk, 0);
 555         up_write(&zram->init_lock);
 556 }
 557
 558 static void zram_init_device(struct zram *zram, struct zram_meta *meta)
 559 {
 560         if (zram->disksize > 2 * (totalram_pages << PAGE_SHIFT)) {
 561                 pr_info(
 562                 "There is little point creating a zram of greater than "
 563                 "twice the size of memory since we expect a 2:1 compression "
 564                 "ratio. Note that zram uses about 0.1%% of the size of "
 565                 "the disk when not in use so a huge zram is "
 566                 "wasteful.\n"
 567                 "\tMemory Size: %lu kB\n"
 568                 "\tSize you selected: %llu kB\n"
 569                 "Continuing anyway ...\n",
 570                 (totalram_pages << PAGE_SHIFT) >> 10, zram->disksize >> 10
 571                 );
 572         }
 573
 574         /* zram devices sort of resembles non-rotational disks */
 575         queue_flag_set_unlocked(QUEUE_FLAG_NONROT, zram->disk->queue);
 576
 577         zram->meta = meta;
 578         zram->init_done = 1;
 579
 580         pr_debug("Initialization done!\n");
 581 }
 582
 583 static ssize_t disksize_store(struct device *dev,
 584                 struct device_attribute *attr, const char *buf, size_t len)
 585 {
 586         u64 disksize;
 587         struct zram_meta *meta;
 588         struct zram *zram = dev_to_zram(dev);
 589
 590         disksize = memparse(buf, NULL);
 591         if (!disksize)
 592                 return -EINVAL;
 593
 594         disksize = PAGE_ALIGN(disksize);
 595         meta = zram_meta_alloc(disksize);
 596         down_write(&zram->init_lock);
 597         if (zram->init_done) {
 598                 up_write(&zram->init_lock);
 599                 zram_meta_free(meta);
 600                 pr_info("Cannot change disksize for initialized device\n");
 601                 return -EBUSY;
 602         }
 603
 604         zram->disksize = disksize;
 605         set_capacity(zram->disk, zram->disksize >> SECTOR_SHIFT);
 606         zram_init_device(zram, meta);
 607         up_write(&zram->init_lock);
 608
 609         return len;
 610 }
 611
 612 static ssize_t reset_store(struct device *dev,
 613                 struct device_attribute *attr, const char *buf, size_t len)
 614 {
 615         int ret;
 616         unsigned short do_reset;
 617         struct zram *zram;
 618         struct block_device *bdev;
 619
 620         zram = dev_to_zram(dev);
 621         bdev = bdget_disk(zram->disk, 0);
 622
 623         /* Do not reset an active device! */
 624         if (bdev->bd_holders)
 625                 return -EBUSY;
 626
 627         ret = kstrtou16(buf, 10, &do_reset);
 628         if (ret)
 629                 return ret;
 630
 631         if (!do_reset)
 632                 return -EINVAL;
 633
 634         /* Make sure all pending I/O is finished */
 635         if (bdev)
 636                 fsync_bdev(bdev);
 637
 638         zram_reset_device(zram);
 639         return len;
 640 }
 641
 642 static void __zram_make_request(struct zram *zram, struct bio *bio, int rw)
 643 {
 644         int i, offset;
 645         u32 index;
 646         struct bio_vec *bvec;
 647
 648         switch (rw) {
 649         case READ:
 650                 atomic64_inc(&zram->stats.num_reads);
 651                 break;
 652         case WRITE:
 653                 atomic64_inc(&zram->stats.num_writes);
 654                 break;
 655         }
 656
 657         index = bio->bi_sector >> SECTORS_PER_PAGE_SHIFT;
 658         offset = (bio->bi_sector & (SECTORS_PER_PAGE - 1)) << SECTOR_SHIFT;
 659
 660         bio_for_each_segment(bvec, bio, i) {
 661                 int max_transfer_size = PAGE_SIZE - offset;
 662
 663                 if (bvec->bv_len > max_transfer_size) {
 664                         /*
 665                          * zram_bvec_rw() can only make operation on a single
 666                          * zram page. Split the bio vector.
 667                          */
 668                         struct bio_vec bv;
 669
 670                         bv.bv_page = bvec->bv_page;
 671                         bv.bv_len = max_transfer_size;
 672                         bv.bv_offset = bvec->bv_offset;
 673
 674                         if (zram_bvec_rw(zram, &bv, index, offset, bio, rw) < 0)
 675                                 goto out;
 676
 677                         bv.bv_len = bvec->bv_len - max_transfer_size;
 678                         bv.bv_offset += max_transfer_size;
 679                         if (zram_bvec_rw(zram, &bv, index+1, 0, bio, rw) < 0)
 680                                 goto out;
 681                 } else
 682                         if (zram_bvec_rw(zram, bvec, index, offset, bio, rw)
 683                             < 0)
 684                                 goto out;
 685
 686                 update_position(&index, &offset, bvec);
 687         }
 688
 689         set_bit(BIO_UPTODATE, &bio->bi_flags);
 690         bio_endio(bio, 0);
 691         return;
 692
 693 out:
 694         bio_io_error(bio);
 695 }
 696
 697 /*
 698  * Handler function for all zram I/O requests.
 699  */
 700 static void zram_make_request(struct request_queue *queue, struct bio *bio)
 701 {
 702         struct zram *zram = queue->queuedata;
 703
 704         down_read(&zram->init_lock);
 705         if (unlikely(!zram->init_done))
 706                 goto error;
 707
 708         if (!valid_io_request(zram, bio)) {
 709                 atomic64_inc(&zram->stats.invalid_io);
 710                 goto error;
 711         }
 712
 713         __zram_make_request(zram, bio, bio_data_dir(bio));
 714         up_read(&zram->init_lock);
 715
 716         return;
 717
 718 error:
 719         up_read(&zram->init_lock);
 720         bio_io_error(bio);
 721 }
 722
 723 static void zram_slot_free_notify(struct block_device *bdev,
 724                                 unsigned long index)
 725 {
 726         struct zram *zram;
 727
 728         zram = bdev->bd_disk->private_data;
 729         down_write(&zram->lock);
 730         zram_free_page(zram, index);
 731         up_write(&zram->lock);
 732         atomic64_inc(&zram->stats.notify_free);
 733 }
 734
 735 static const struct block_device_operations zram_devops = {
 736         .swap_slot_free_notify = zram_slot_free_notify,
 737         .owner = THIS_MODULE
 738 };
 739
 740 static DEVICE_ATTR(disksize, S_IRUGO | S_IWUSR,
 741                 disksize_show, disksize_store);
 742 static DEVICE_ATTR(initstate, S_IRUGO, initstate_show, NULL);
 743 static DEVICE_ATTR(reset, S_IWUSR, NULL, reset_store);
 744 static DEVICE_ATTR(num_reads, S_IRUGO, num_reads_show, NULL);
 745 static DEVICE_ATTR(num_writes, S_IRUGO, num_writes_show, NULL);
 746 static DEVICE_ATTR(invalid_io, S_IRUGO, invalid_io_show, NULL);
 747 static DEVICE_ATTR(notify_free, S_IRUGO, notify_free_show, NULL);
 748 static DEVICE_ATTR(zero_pages, S_IRUGO, zero_pages_show, NULL);
 749 static DEVICE_ATTR(orig_data_size, S_IRUGO, orig_data_size_show, NULL);
 750 static DEVICE_ATTR(compr_data_size, S_IRUGO, compr_data_size_show, NULL);
 751 static DEVICE_ATTR(mem_used_total, S_IRUGO, mem_used_total_show, NULL);
 752
 753 static struct attribute *zram_disk_attrs[] = {
 754         &dev_attr_disksize.attr,
 755         &dev_attr_initstate.attr,
 756         &dev_attr_reset.attr,
 757         &dev_attr_num_reads.attr,
 758         &dev_attr_num_writes.attr,
 759         &dev_attr_invalid_io.attr,
 760         &dev_attr_notify_free.attr,
 761         &dev_attr_zero_pages.attr,
 762         &dev_attr_orig_data_size.attr,
 763         &dev_attr_compr_data_size.attr,
 764         &dev_attr_mem_used_total.attr,
 765         NULL,
 766 };
 767
 768 static struct attribute_group zram_disk_attr_group = {
 769         .attrs = zram_disk_attrs,
 770 };
 771
 772 static int create_device(struct zram *zram, int device_id)
 773 {
 774         int ret = -ENOMEM;
 775
 776         init_rwsem(&zram->lock);
 777         init_rwsem(&zram->init_lock);
 778
 779         zram->queue = blk_alloc_queue(GFP_KERNEL);
 780         if (!zram->queue) {
 781                 pr_err("Error allocating disk queue for device %d\n",
 782                         device_id);
 783                 goto out;
 784         }
 785
 786         blk_queue_make_request(zram->queue, zram_make_request);
 787         zram->queue->queuedata = zram;
 788
 789          /* gendisk structure */
 790         zram->disk = alloc_disk(1);
 791         if (!zram->disk) {
 792                 pr_warn("Error allocating disk structure for device %d\n",
 793                         device_id);
 794                 goto out_free_queue;
 795         }
 796
 797         zram->disk->major = zram_major;
 798         zram->disk->first_minor = device_id;
 799         zram->disk->fops = &zram_devops;
 800         zram->disk->queue = zram->queue;
 801         zram->disk->private_data = zram;
 802         snprintf(zram->disk->disk_name, 16, "zram%d", device_id);
 803
 804         /* Actual capacity set using syfs (/sys/block/zram<id>/disksize */
 805         set_capacity(zram->disk, 0);
 806
 807         /*
 808          * To ensure that we always get PAGE_SIZE aligned
 809          * and n*PAGE_SIZED sized I/O requests.
 810          */
 811         blk_queue_physical_block_size(zram->disk->queue, PAGE_SIZE);
 812         blk_queue_logical_block_size(zram->disk->queue,
 813                                         ZRAM_LOGICAL_BLOCK_SIZE);
 814         blk_queue_io_min(zram->disk->queue, PAGE_SIZE);
 815         blk_queue_io_opt(zram->disk->queue, PAGE_SIZE);
 816
 817         add_disk(zram->disk);
 818
 819         ret = sysfs_create_group(&disk_to_dev(zram->disk)->kobj,
 820                                 &zram_disk_attr_group);
 821         if (ret < 0) {
 822                 pr_warn("Error creating sysfs group");
 823                 goto out_free_disk;
 824         }
 825
 826         zram->init_done = 0;
 827         return 0;
 828
 829 out_free_disk:
 830         del_gendisk(zram->disk);
 831         put_disk(zram->disk);
 832 out_free_queue:
 833         blk_cleanup_queue(zram->queue);
 834 out:
 835         return ret;
 836 }
 837
 838 static void destroy_device(struct zram *zram)
 839 {
 840         sysfs_remove_group(&disk_to_dev(zram->disk)->kobj,
 841                         &zram_disk_attr_group);
 842
 843         if (zram->disk) {
 844                 del_gendisk(zram->disk);
 845                 put_disk(zram->disk);
 846         }
 847
 848         if (zram->queue)
 849                 blk_cleanup_queue(zram->queue);
 850 }
 851
 852 static int __init zram_init(void)
 853 {
 854         int ret, dev_id;
 855
 856         if (num_devices > max_num_devices) {
 857                 pr_warn("Invalid value for num_devices: %u\n",
 858                                 num_devices);
 859                 ret = -EINVAL;
 860                 goto out;
 861         }
 862
 863         zram_major = register_blkdev(0, "zram");
 864         if (zram_major <= 0) {
 865                 pr_warn("Unable to get major number\n");
 866                 ret = -EBUSY;
 867                 goto out;
 868         }
 869
 870         /* Allocate the device array and initialize each one */
 871         zram_devices = kzalloc(num_devices * sizeof(struct zram), GFP_KERNEL);
 872         if (!zram_devices) {
 873                 ret = -ENOMEM;
 874                 goto unregister;
 875         }
 876
 877         for (dev_id = 0; dev_id < num_devices; dev_id++) {
 878                 ret = create_device(&zram_devices[dev_id], dev_id);
 879                 if (ret)
 880                         goto free_devices;
 881         }
 882
 883         pr_info("Created %u device(s) ...\n", num_devices);
 884
 885         return 0;
 886
 887 free_devices:
 888         while (dev_id)
 889                 destroy_device(&zram_devices[--dev_id]);
 890         kfree(zram_devices);
 891 unregister:
 892         unregister_blkdev(zram_major, "zram");
 893 out:
 894         return ret;
 895 }
 896
 897 static void __exit zram_exit(void)
 898 {
 899         int i;
 900         struct zram *zram;
 901
 902         for (i = 0; i < num_devices; i++) {
 903                 zram = &zram_devices[i];
 904
 905                 get_disk(zram->disk);
 906                 destroy_device(zram);
 907                 zram_reset_device(zram);
 908                 put_disk(zram->disk);
 909         }
 910
 911         unregister_blkdev(zram_major, "zram");
 912
 913         kfree(zram_devices);
 914         pr_debug("Cleanup done!\n");
 915 }
 916
 917 module_init(zram_init);
 918 module_exit(zram_exit);
 919
 920 module_param(num_devices, uint, 0);
 921 MODULE_PARM_DESC(num_devices, "Number of zram devices");
 922
 923 MODULE_LICENSE("Dual BSD/GPL");
 924 MODULE_AUTHOR("Nitin Gupta <ngupta@vflare.org>");
 925 MODULE_DESCRIPTION("Compressed RAM Block Device");