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_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         /*
 445          * zram_slot_free_notify could miss free so that let's
 446          * double check.
 447          */
 448         if (unlikely(meta->table[index].handle ||
 449                         zram_test_flag(meta, index, ZRAM_ZERO)))
 450                 zram_free_page(zram, index);
 451
 452         ret = lzo1x_1_compress(uncmem, PAGE_SIZE, src, &clen,
 453                                meta->compress_workmem);
 454
 455         if (!is_partial_io(bvec)) {
 456                 kunmap_atomic(user_mem);
 457                 user_mem = NULL;
 458                 uncmem = NULL;
 459         }
 460
 461         if (unlikely(ret != LZO_E_OK)) {
 462                 pr_err("Compression failed! err=%d\n", ret);
 463                 goto out;
 464         }
 465
 466         if (unlikely(clen > max_zpage_size)) {
 467                 zram->stats.bad_compress++;
 468                 clen = PAGE_SIZE;
 469                 src = NULL;
 470                 if (is_partial_io(bvec))
 471                         src = uncmem;
 472         }
 473
 474         handle = zs_malloc(meta->mem_pool, clen);
 475         if (!handle) {
 476                 pr_info("Error allocating memory for compressed page: %u, size=%zu\n",
 477                         index, clen);
 478                 ret = -ENOMEM;
 479                 goto out;
 480         }
 481         cmem = zs_map_object(meta->mem_pool, handle, ZS_MM_WO);
 482
 483         if ((clen == PAGE_SIZE) && !is_partial_io(bvec)) {
 484                 src = kmap_atomic(page);
 485                 copy_page(cmem, src);
 486                 kunmap_atomic(src);
 487         } else {
 488                 memcpy(cmem, src, clen);
 489         }
 490
 491         zs_unmap_object(meta->mem_pool, handle);
 492
 493         /*
 494          * Free memory associated with this sector
 495          * before overwriting unused sectors.
 496          */
 497         zram_free_page(zram, index);
 498
 499         meta->table[index].handle = handle;
 500         meta->table[index].size = clen;
 501
 502         /* Update stats */
 503         atomic64_add(clen, &zram->stats.compr_size);
 504         zram->stats.pages_stored++;
 505         if (clen <= PAGE_SIZE / 2)
 506                 zram->stats.good_compress++;
 507
 508 out:
 509         if (is_partial_io(bvec))
 510                 kfree(uncmem);
 511
 512         if (ret)
 513                 atomic64_inc(&zram->stats.failed_writes);
 514         return ret;
 515 }
 516
 517 static void handle_pending_slot_free(struct zram *zram)
 518 {
 519         struct zram_slot_free *free_rq;
 520
 521         spin_lock(&zram->slot_free_lock);
 522         while (zram->slot_free_rq) {
 523                 free_rq = zram->slot_free_rq;
 524                 zram->slot_free_rq = free_rq->next;
 525                 zram_free_page(zram, free_rq->index);
 526                 kfree(free_rq);
 527         }
 528         spin_unlock(&zram->slot_free_lock);
 529 }
 530
 531 static int zram_bvec_rw(struct zram *zram, struct bio_vec *bvec, u32 index,
 532                         int offset, struct bio *bio, int rw)
 533 {
 534         int ret;
 535
 536         if (rw == READ) {
 537                 down_read(&zram->lock);
 538                 handle_pending_slot_free(zram);
 539                 ret = zram_bvec_read(zram, bvec, index, offset, bio);
 540                 up_read(&zram->lock);
 541         } else {
 542                 down_write(&zram->lock);
 543                 handle_pending_slot_free(zram);
 544                 ret = zram_bvec_write(zram, bvec, index, offset);
 545                 up_write(&zram->lock);
 546         }
 547
 548         return ret;
 549 }
 550
 551 static void zram_reset_device(struct zram *zram, bool reset_capacity)
 552 {
 553         size_t index;
 554         struct zram_meta *meta;
 555
 556         flush_work(&zram->free_work);
 557
 558         down_write(&zram->init_lock);
 559         if (!zram->init_done) {
 560                 up_write(&zram->init_lock);
 561                 return;
 562         }
 563
 564         meta = zram->meta;
 565         zram->init_done = 0;
 566
 567         /* Free all pages that are still in this zram device */
 568         for (index = 0; index < zram->disksize >> PAGE_SHIFT; index++) {
 569                 unsigned long handle = meta->table[index].handle;
 570                 if (!handle)
 571                         continue;
 572
 573                 zs_free(meta->mem_pool, handle);
 574         }
 575
 576         zram_meta_free(zram->meta);
 577         zram->meta = NULL;
 578         /* Reset stats */
 579         memset(&zram->stats, 0, sizeof(zram->stats));
 580
 581         zram->disksize = 0;
 582         if (reset_capacity)
 583                 set_capacity(zram->disk, 0);
 584         up_write(&zram->init_lock);
 585 }
 586
 587 static void zram_init_device(struct zram *zram, struct zram_meta *meta)
 588 {
 589         if (zram->disksize > 2 * (totalram_pages << PAGE_SHIFT)) {
 590                 pr_info(
 591                 "There is little point creating a zram of greater than "
 592                 "twice the size of memory since we expect a 2:1 compression "
 593                 "ratio. Note that zram uses about 0.1%% of the size of "
 594                 "the disk when not in use so a huge zram is "
 595                 "wasteful.\n"
 596                 "\tMemory Size: %lu kB\n"
 597                 "\tSize you selected: %llu kB\n"
 598                 "Continuing anyway ...\n",
 599                 (totalram_pages << PAGE_SHIFT) >> 10, zram->disksize >> 10
 600                 );
 601         }
 602
 603         /* zram devices sort of resembles non-rotational disks */
 604         queue_flag_set_unlocked(QUEUE_FLAG_NONROT, zram->disk->queue);
 605
 606         zram->meta = meta;
 607         zram->init_done = 1;
 608
 609         pr_debug("Initialization done!\n");
 610 }
 611
 612 static ssize_t disksize_store(struct device *dev,
 613                 struct device_attribute *attr, const char *buf, size_t len)
 614 {
 615         u64 disksize;
 616         struct zram_meta *meta;
 617         struct zram *zram = dev_to_zram(dev);
 618
 619         disksize = memparse(buf, NULL);
 620         if (!disksize)
 621                 return -EINVAL;
 622
 623         disksize = PAGE_ALIGN(disksize);
 624         meta = zram_meta_alloc(disksize);
 625         down_write(&zram->init_lock);
 626         if (zram->init_done) {
 627                 up_write(&zram->init_lock);
 628                 zram_meta_free(meta);
 629                 pr_info("Cannot change disksize for initialized device\n");
 630                 return -EBUSY;
 631         }
 632
 633         zram->disksize = disksize;
 634         set_capacity(zram->disk, zram->disksize >> SECTOR_SHIFT);
 635         zram_init_device(zram, meta);
 636         up_write(&zram->init_lock);
 637
 638         return len;
 639 }
 640
 641 static ssize_t reset_store(struct device *dev,
 642                 struct device_attribute *attr, const char *buf, size_t len)
 643 {
 644         int ret;
 645         unsigned short do_reset;
 646         struct zram *zram;
 647         struct block_device *bdev;
 648
 649         zram = dev_to_zram(dev);
 650         bdev = bdget_disk(zram->disk, 0);
 651
 652         if (!bdev)
 653                 return -ENOMEM;
 654
 655         /* Do not reset an active device! */
 656         if (bdev->bd_holders)
 657                 return -EBUSY;
 658
 659         ret = kstrtou16(buf, 10, &do_reset);
 660         if (ret)
 661                 return ret;
 662
 663         if (!do_reset)
 664                 return -EINVAL;
 665
 666         /* Make sure all pending I/O is finished */
 667         fsync_bdev(bdev);
 668
 669         zram_reset_device(zram, true);
 670         return len;
 671 }
 672
 673 static void __zram_make_request(struct zram *zram, struct bio *bio, int rw)
 674 {
 675         int i, offset;
 676         u32 index;
 677         struct bio_vec *bvec;
 678
 679         switch (rw) {
 680         case READ:
 681                 atomic64_inc(&zram->stats.num_reads);
 682                 break;
 683         case WRITE:
 684                 atomic64_inc(&zram->stats.num_writes);
 685                 break;
 686         }
 687
 688         index = bio->bi_iter.bi_sector >> SECTORS_PER_PAGE_SHIFT;
 689         offset = (bio->bi_iter.bi_sector &
 690                   (SECTORS_PER_PAGE - 1)) << SECTOR_SHIFT;
 691
 692         bio_for_each_segment(bvec, bio, i) {
 693                 int max_transfer_size = PAGE_SIZE - offset;
 694
 695                 if (bvec->bv_len > max_transfer_size) {
 696                         /*
 697                          * zram_bvec_rw() can only make operation on a single
 698                          * zram page. Split the bio vector.
 699                          */
 700                         struct bio_vec bv;
 701
 702                         bv.bv_page = bvec->bv_page;
 703                         bv.bv_len = max_transfer_size;
 704                         bv.bv_offset = bvec->bv_offset;
 705
 706                         if (zram_bvec_rw(zram, &bv, index, offset, bio, rw) < 0)
 707                                 goto out;
 708
 709                         bv.bv_len = bvec->bv_len - max_transfer_size;
 710                         bv.bv_offset += max_transfer_size;
 711                         if (zram_bvec_rw(zram, &bv, index+1, 0, bio, rw) < 0)
 712                                 goto out;
 713                 } else
 714                         if (zram_bvec_rw(zram, bvec, index, offset, bio, rw)
 715                             < 0)
 716                                 goto out;
 717
 718                 update_position(&index, &offset, bvec);
 719         }
 720
 721         set_bit(BIO_UPTODATE, &bio->bi_flags);
 722         bio_endio(bio, 0);
 723         return;
 724
 725 out:
 726         bio_io_error(bio);
 727 }
 728
 729 /*
 730  * Handler function for all zram I/O requests.
 731  */
 732 static void zram_make_request(struct request_queue *queue, struct bio *bio)
 733 {
 734         struct zram *zram = queue->queuedata;
 735
 736         down_read(&zram->init_lock);
 737         if (unlikely(!zram->init_done))
 738                 goto error;
 739
 740         if (!valid_io_request(zram, bio)) {
 741                 atomic64_inc(&zram->stats.invalid_io);
 742                 goto error;
 743         }
 744
 745         __zram_make_request(zram, bio, bio_data_dir(bio));
 746         up_read(&zram->init_lock);
 747
 748         return;
 749
 750 error:
 751         up_read(&zram->init_lock);
 752         bio_io_error(bio);
 753 }
 754
 755 static void zram_slot_free(struct work_struct *work)
 756 {
 757         struct zram *zram;
 758
 759         zram = container_of(work, struct zram, free_work);
 760         down_write(&zram->lock);
 761         handle_pending_slot_free(zram);
 762         up_write(&zram->lock);
 763 }
 764
 765 static void add_slot_free(struct zram *zram, struct zram_slot_free *free_rq)
 766 {
 767         spin_lock(&zram->slot_free_lock);
 768         free_rq->next = zram->slot_free_rq;
 769         zram->slot_free_rq = free_rq;
 770         spin_unlock(&zram->slot_free_lock);
 771 }
 772
 773 static void zram_slot_free_notify(struct block_device *bdev,
 774                                 unsigned long index)
 775 {
 776         struct zram *zram;
 777         struct zram_slot_free *free_rq;
 778
 779         zram = bdev->bd_disk->private_data;
 780         atomic64_inc(&zram->stats.notify_free);
 781
 782         free_rq = kmalloc(sizeof(struct zram_slot_free), GFP_ATOMIC);
 783         if (!free_rq)
 784                 return;
 785
 786         free_rq->index = index;
 787         add_slot_free(zram, free_rq);
 788         schedule_work(&zram->free_work);
 789 }
 790
 791 static const struct block_device_operations zram_devops = {
 792         .swap_slot_free_notify = zram_slot_free_notify,
 793         .owner = THIS_MODULE
 794 };
 795
 796 static DEVICE_ATTR(disksize, S_IRUGO | S_IWUSR,
 797                 disksize_show, disksize_store);
 798 static DEVICE_ATTR(initstate, S_IRUGO, initstate_show, NULL);
 799 static DEVICE_ATTR(reset, S_IWUSR, NULL, reset_store);
 800 static DEVICE_ATTR(num_reads, S_IRUGO, num_reads_show, NULL);
 801 static DEVICE_ATTR(num_writes, S_IRUGO, num_writes_show, NULL);
 802 static DEVICE_ATTR(invalid_io, S_IRUGO, invalid_io_show, NULL);
 803 static DEVICE_ATTR(notify_free, S_IRUGO, notify_free_show, NULL);
 804 static DEVICE_ATTR(zero_pages, S_IRUGO, zero_pages_show, NULL);
 805 static DEVICE_ATTR(orig_data_size, S_IRUGO, orig_data_size_show, NULL);
 806 static DEVICE_ATTR(compr_data_size, S_IRUGO, compr_data_size_show, NULL);
 807 static DEVICE_ATTR(mem_used_total, S_IRUGO, mem_used_total_show, NULL);
 808
 809 static struct attribute *zram_disk_attrs[] = {
 810         &dev_attr_disksize.attr,
 811         &dev_attr_initstate.attr,
 812         &dev_attr_reset.attr,
 813         &dev_attr_num_reads.attr,
 814         &dev_attr_num_writes.attr,
 815         &dev_attr_invalid_io.attr,
 816         &dev_attr_notify_free.attr,
 817         &dev_attr_zero_pages.attr,
 818         &dev_attr_orig_data_size.attr,
 819         &dev_attr_compr_data_size.attr,
 820         &dev_attr_mem_used_total.attr,
 821         NULL,
 822 };
 823
 824 static struct attribute_group zram_disk_attr_group = {
 825         .attrs = zram_disk_attrs,
 826 };
 827
 828 static int create_device(struct zram *zram, int device_id)
 829 {
 830         int ret = -ENOMEM;
 831
 832         init_rwsem(&zram->lock);
 833         init_rwsem(&zram->init_lock);
 834
 835         INIT_WORK(&zram->free_work, zram_slot_free);
 836         spin_lock_init(&zram->slot_free_lock);
 837         zram->slot_free_rq = NULL;
 838
 839         zram->queue = blk_alloc_queue(GFP_KERNEL);
 840         if (!zram->queue) {
 841                 pr_err("Error allocating disk queue for device %d\n",
 842                         device_id);
 843                 goto out;
 844         }
 845
 846         blk_queue_make_request(zram->queue, zram_make_request);
 847         zram->queue->queuedata = zram;
 848
 849          /* gendisk structure */
 850         zram->disk = alloc_disk(1);
 851         if (!zram->disk) {
 852                 pr_warn("Error allocating disk structure for device %d\n",
 853                         device_id);
 854                 goto out_free_queue;
 855         }
 856
 857         zram->disk->major = zram_major;
 858         zram->disk->first_minor = device_id;
 859         zram->disk->fops = &zram_devops;
 860         zram->disk->queue = zram->queue;
 861         zram->disk->private_data = zram;
 862         snprintf(zram->disk->disk_name, 16, "zram%d", device_id);
 863
 864         /* Actual capacity set using syfs (/sys/block/zram<id>/disksize */
 865         set_capacity(zram->disk, 0);
 866
 867         /*
 868          * To ensure that we always get PAGE_SIZE aligned
 869          * and n*PAGE_SIZED sized I/O requests.
 870          */
 871         blk_queue_physical_block_size(zram->disk->queue, PAGE_SIZE);
 872         blk_queue_logical_block_size(zram->disk->queue,
 873                                         ZRAM_LOGICAL_BLOCK_SIZE);
 874         blk_queue_io_min(zram->disk->queue, PAGE_SIZE);
 875         blk_queue_io_opt(zram->disk->queue, PAGE_SIZE);
 876
 877         add_disk(zram->disk);
 878
 879         ret = sysfs_create_group(&disk_to_dev(zram->disk)->kobj,
 880                                 &zram_disk_attr_group);
 881         if (ret < 0) {
 882                 pr_warn("Error creating sysfs group");
 883                 goto out_free_disk;
 884         }
 885
 886         zram->init_done = 0;
 887         return 0;
 888
 889 out_free_disk:
 890         del_gendisk(zram->disk);
 891         put_disk(zram->disk);
 892 out_free_queue:
 893         blk_cleanup_queue(zram->queue);
 894 out:
 895         return ret;
 896 }
 897
 898 static void destroy_device(struct zram *zram)
 899 {
 900         sysfs_remove_group(&disk_to_dev(zram->disk)->kobj,
 901                         &zram_disk_attr_group);
 902
 903         del_gendisk(zram->disk);
 904         put_disk(zram->disk);
 905
 906         blk_cleanup_queue(zram->queue);
 907 }
 908
 909 static int __init zram_init(void)
 910 {
 911         int ret, dev_id;
 912
 913         if (num_devices > max_num_devices) {
 914                 pr_warn("Invalid value for num_devices: %u\n",
 915                                 num_devices);
 916                 ret = -EINVAL;
 917                 goto out;
 918         }
 919
 920         zram_major = register_blkdev(0, "zram");
 921         if (zram_major <= 0) {
 922                 pr_warn("Unable to get major number\n");
 923                 ret = -EBUSY;
 924                 goto out;
 925         }
 926
 927         /* Allocate the device array and initialize each one */
 928         zram_devices = kzalloc(num_devices * sizeof(struct zram), GFP_KERNEL);
 929         if (!zram_devices) {
 930                 ret = -ENOMEM;
 931                 goto unregister;
 932         }
 933
 934         for (dev_id = 0; dev_id < num_devices; dev_id++) {
 935                 ret = create_device(&zram_devices[dev_id], dev_id);
 936                 if (ret)
 937                         goto free_devices;
 938         }
 939
 940         pr_info("Created %u device(s) ...\n", num_devices);
 941
 942         return 0;
 943
 944 free_devices:
 945         while (dev_id)
 946                 destroy_device(&zram_devices[--dev_id]);
 947         kfree(zram_devices);
 948 unregister:
 949         unregister_blkdev(zram_major, "zram");
 950 out:
 951         return ret;
 952 }
 953
 954 static void __exit zram_exit(void)
 955 {
 956         int i;
 957         struct zram *zram;
 958
 959         for (i = 0; i < num_devices; i++) {
 960                 zram = &zram_devices[i];
 961
 962                 destroy_device(zram);
 963                 /*
 964                  * Shouldn't access zram->disk after destroy_device
 965                  * because destroy_device already released zram->disk.
 966                  */
 967                 zram_reset_device(zram, false);
 968         }
 969
 970         unregister_blkdev(zram_major, "zram");
 971
 972         kfree(zram_devices);
 973         pr_debug("Cleanup done!\n");
 974 }
 975
 976 module_init(zram_init);
 977 module_exit(zram_exit);
 978
 979 module_param(num_devices, uint, 0);
 980 MODULE_PARM_DESC(num_devices, "Number of zram devices");
 981
 982 MODULE_LICENSE("Dual BSD/GPL");
 983 MODULE_AUTHOR("Nitin Gupta <ngupta@vflare.org>");
 984 MODULE_DESCRIPTION("Compressed RAM Block Device");