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 struct zram *devices;
  41
  42 /* Module params (documentation at end) */
  43 unsigned int num_devices;
  44
  45 static void zram_stat_inc(u32 *v)
  46 {
  47         *v = *v + 1;
  48 }
  49
  50 static void zram_stat_dec(u32 *v)
  51 {
  52         *v = *v - 1;
  53 }
  54
  55 static void zram_stat64_add(struct zram *zram, u64 *v, u64 inc)
  56 {
  57         spin_lock(&zram->stat64_lock);
  58         *v = *v + inc;
  59         spin_unlock(&zram->stat64_lock);
  60 }
  61
  62 static void zram_stat64_sub(struct zram *zram, u64 *v, u64 dec)
  63 {
  64         spin_lock(&zram->stat64_lock);
  65         *v = *v - dec;
  66         spin_unlock(&zram->stat64_lock);
  67 }
  68
  69 static void zram_stat64_inc(struct zram *zram, u64 *v)
  70 {
  71         zram_stat64_add(zram, v, 1);
  72 }
  73
  74 static int zram_test_flag(struct zram *zram, u32 index,
  75                         enum zram_pageflags flag)
  76 {
  77         return zram->table[index].flags & BIT(flag);
  78 }
  79
  80 static void zram_set_flag(struct zram *zram, u32 index,
  81                         enum zram_pageflags flag)
  82 {
  83         zram->table[index].flags |= BIT(flag);
  84 }
  85
  86 static void zram_clear_flag(struct zram *zram, u32 index,
  87                         enum zram_pageflags flag)
  88 {
  89         zram->table[index].flags &= ~BIT(flag);
  90 }
  91
  92 static int page_zero_filled(void *ptr)
  93 {
  94         unsigned int pos;
  95         unsigned long *page;
  96
  97         page = (unsigned long *)ptr;
  98
  99         for (pos = 0; pos != PAGE_SIZE / sizeof(*page); pos++) {
 100                 if (page[pos])
 101                         return 0;
 102         }
 103
 104         return 1;
 105 }
 106
 107 static void zram_set_disksize(struct zram *zram, size_t totalram_bytes)
 108 {
 109         if (!zram->disksize) {
 110                 pr_info(
 111                 "disk size not provided. You can use disksize_kb module "
 112                 "param to specify size.\nUsing default: (%u%% of RAM).\n",
 113                 default_disksize_perc_ram
 114                 );
 115                 zram->disksize = default_disksize_perc_ram *
 116                                         (totalram_bytes / 100);
 117         }
 118
 119         if (zram->disksize > 2 * (totalram_bytes)) {
 120                 pr_info(
 121                 "There is little point creating a zram of greater than "
 122                 "twice the size of memory since we expect a 2:1 compression "
 123                 "ratio. Note that zram uses about 0.1%% of the size of "
 124                 "the disk when not in use so a huge zram is "
 125                 "wasteful.\n"
 126                 "\tMemory Size: %zu kB\n"
 127                 "\tSize you selected: %llu kB\n"
 128                 "Continuing anyway ...\n",
 129                 totalram_bytes >> 10, zram->disksize
 130                 );
 131         }
 132
 133         zram->disksize &= PAGE_MASK;
 134 }
 135
 136 static void zram_free_page(struct zram *zram, size_t index)
 137 {
 138         u32 clen;
 139         void *obj;
 140
 141         struct page *page = zram->table[index].page;
 142         u32 offset = zram->table[index].offset;
 143
 144         if (unlikely(!page)) {
 145                 /*
 146                  * No memory is allocated for zero filled pages.
 147                  * Simply clear zero page flag.
 148                  */
 149                 if (zram_test_flag(zram, index, ZRAM_ZERO)) {
 150                         zram_clear_flag(zram, index, ZRAM_ZERO);
 151                         zram_stat_dec(&zram->stats.pages_zero);
 152                 }
 153                 return;
 154         }
 155
 156         if (unlikely(zram_test_flag(zram, index, ZRAM_UNCOMPRESSED))) {
 157                 clen = PAGE_SIZE;
 158                 __free_page(page);
 159                 zram_clear_flag(zram, index, ZRAM_UNCOMPRESSED);
 160                 zram_stat_dec(&zram->stats.pages_expand);
 161                 goto out;
 162         }
 163
 164         obj = kmap_atomic(page, KM_USER0) + offset;
 165         clen = xv_get_object_size(obj) - sizeof(struct zobj_header);
 166         kunmap_atomic(obj, KM_USER0);
 167
 168         xv_free(zram->mem_pool, page, offset);
 169         if (clen <= PAGE_SIZE / 2)
 170                 zram_stat_dec(&zram->stats.good_compress);
 171
 172 out:
 173         zram_stat64_sub(zram, &zram->stats.compr_size, clen);
 174         zram_stat_dec(&zram->stats.pages_stored);
 175
 176         zram->table[index].page = NULL;
 177         zram->table[index].offset = 0;
 178 }
 179
 180 static void handle_zero_page(struct bio_vec *bvec)
 181 {
 182         struct page *page = bvec->bv_page;
 183         void *user_mem;
 184
 185         user_mem = kmap_atomic(page, KM_USER0);
 186         memset(user_mem + bvec->bv_offset, 0, bvec->bv_len);
 187         kunmap_atomic(user_mem, KM_USER0);
 188
 189         flush_dcache_page(page);
 190 }
 191
 192 static void handle_uncompressed_page(struct zram *zram, struct bio_vec *bvec,
 193                                      u32 index, int offset)
 194 {
 195         struct page *page = bvec->bv_page;
 196         unsigned char *user_mem, *cmem;
 197
 198         user_mem = kmap_atomic(page, KM_USER0);
 199         cmem = kmap_atomic(zram->table[index].page, KM_USER1);
 200
 201         memcpy(user_mem + bvec->bv_offset, cmem + offset, bvec->bv_len);
 202         kunmap_atomic(cmem, KM_USER1);
 203         kunmap_atomic(user_mem, KM_USER0);
 204
 205         flush_dcache_page(page);
 206 }
 207
 208 static inline int is_partial_io(struct bio_vec *bvec)
 209 {
 210         return bvec->bv_len != PAGE_SIZE;
 211 }
 212
 213 static int zram_bvec_read(struct zram *zram, struct bio_vec *bvec,
 214                           u32 index, int offset, struct bio *bio)
 215 {
 216         int ret;
 217         size_t clen;
 218         struct page *page;
 219         struct zobj_header *zheader;
 220         unsigned char *user_mem, *cmem, *uncmem = NULL;
 221
 222         page = bvec->bv_page;
 223
 224         if (zram_test_flag(zram, index, ZRAM_ZERO)) {
 225                 handle_zero_page(bvec);
 226                 return 0;
 227         }
 228
 229         /* Requested page is not present in compressed area */
 230         if (unlikely(!zram->table[index].page)) {
 231                 pr_debug("Read before write: sector=%lu, size=%u",
 232                          (ulong)(bio->bi_sector), bio->bi_size);
 233                 handle_zero_page(bvec);
 234                 return 0;
 235         }
 236
 237         /* Page is stored uncompressed since it's incompressible */
 238         if (unlikely(zram_test_flag(zram, index, ZRAM_UNCOMPRESSED))) {
 239                 handle_uncompressed_page(zram, bvec, index, offset);
 240                 return 0;
 241         }
 242
 243         if (is_partial_io(bvec)) {
 244                 /* Use  a temporary buffer to decompress the page */
 245                 uncmem = kmalloc(PAGE_SIZE, GFP_KERNEL);
 246                 if (!uncmem) {
 247                         pr_info("Error allocating temp memory!\n");
 248                         return -ENOMEM;
 249                 }
 250         }
 251
 252         user_mem = kmap_atomic(page, KM_USER0);
 253         if (!is_partial_io(bvec))
 254                 uncmem = user_mem;
 255         clen = PAGE_SIZE;
 256
 257         cmem = kmap_atomic(zram->table[index].page, KM_USER1) +
 258                 zram->table[index].offset;
 259
 260         ret = lzo1x_decompress_safe(cmem + sizeof(*zheader),
 261                                     xv_get_object_size(cmem) - sizeof(*zheader),
 262                                     uncmem, &clen);
 263
 264         if (is_partial_io(bvec)) {
 265                 memcpy(user_mem + bvec->bv_offset, uncmem + offset,
 266                        bvec->bv_len);
 267                 kfree(uncmem);
 268         }
 269
 270         kunmap_atomic(cmem, KM_USER1);
 271         kunmap_atomic(user_mem, KM_USER0);
 272
 273         /* Should NEVER happen. Return bio error if it does. */
 274         if (unlikely(ret != LZO_E_OK)) {
 275                 pr_err("Decompression failed! err=%d, page=%u\n", ret, index);
 276                 zram_stat64_inc(zram, &zram->stats.failed_reads);
 277                 return ret;
 278         }
 279
 280         flush_dcache_page(page);
 281
 282         return 0;
 283 }
 284
 285 static int zram_read_before_write(struct zram *zram, char *mem, u32 index)
 286 {
 287         int ret;
 288         size_t clen = PAGE_SIZE;
 289         struct zobj_header *zheader;
 290         unsigned char *cmem;
 291
 292         if (zram_test_flag(zram, index, ZRAM_ZERO) ||
 293             !zram->table[index].page) {
 294                 memset(mem, 0, PAGE_SIZE);
 295                 return 0;
 296         }
 297
 298         cmem = kmap_atomic(zram->table[index].page, KM_USER0) +
 299                 zram->table[index].offset;
 300
 301         /* Page is stored uncompressed since it's incompressible */
 302         if (unlikely(zram_test_flag(zram, index, ZRAM_UNCOMPRESSED))) {
 303                 memcpy(mem, cmem, PAGE_SIZE);
 304                 kunmap_atomic(cmem, KM_USER0);
 305                 return 0;
 306         }
 307
 308         ret = lzo1x_decompress_safe(cmem + sizeof(*zheader),
 309                                     xv_get_object_size(cmem) - sizeof(*zheader),
 310                                     mem, &clen);
 311         kunmap_atomic(cmem, KM_USER0);
 312
 313         /* Should NEVER happen. Return bio error if it does. */
 314         if (unlikely(ret != LZO_E_OK)) {
 315                 pr_err("Decompression failed! err=%d, page=%u\n", ret, index);
 316                 zram_stat64_inc(zram, &zram->stats.failed_reads);
 317                 return ret;
 318         }
 319
 320         return 0;
 321 }
 322
 323 static int zram_bvec_write(struct zram *zram, struct bio_vec *bvec, u32 index,
 324                            int offset)
 325 {
 326         int ret;
 327         u32 store_offset;
 328         size_t clen;
 329         struct zobj_header *zheader;
 330         struct page *page, *page_store;
 331         unsigned char *user_mem, *cmem, *src, *uncmem = NULL;
 332
 333         page = bvec->bv_page;
 334         src = zram->compress_buffer;
 335
 336         if (is_partial_io(bvec)) {
 337                 /*
 338                  * This is a partial IO. We need to read the full page
 339                  * before to write the changes.
 340                  */
 341                 uncmem = kmalloc(PAGE_SIZE, GFP_KERNEL);
 342                 if (!uncmem) {
 343                         pr_info("Error allocating temp memory!\n");
 344                         ret = -ENOMEM;
 345                         goto out;
 346                 }
 347                 ret = zram_read_before_write(zram, uncmem, index);
 348                 if (ret) {
 349                         kfree(uncmem);
 350                         goto out;
 351                 }
 352         }
 353
 354         /*
 355          * System overwrites unused sectors. Free memory associated
 356          * with this sector now.
 357          */
 358         if (zram->table[index].page ||
 359             zram_test_flag(zram, index, ZRAM_ZERO))
 360                 zram_free_page(zram, index);
 361
 362         user_mem = kmap_atomic(page, KM_USER0);
 363
 364         if (is_partial_io(bvec))
 365                 memcpy(uncmem + offset, user_mem + bvec->bv_offset,
 366                        bvec->bv_len);
 367         else
 368                 uncmem = user_mem;
 369
 370         if (page_zero_filled(uncmem)) {
 371                 kunmap_atomic(user_mem, KM_USER0);
 372                 if (is_partial_io(bvec))
 373                         kfree(uncmem);
 374                 zram_stat_inc(&zram->stats.pages_zero);
 375                 zram_set_flag(zram, index, ZRAM_ZERO);
 376                 ret = 0;
 377                 goto out;
 378         }
 379
 380         ret = lzo1x_1_compress(uncmem, PAGE_SIZE, src, &clen,
 381                                zram->compress_workmem);
 382
 383         kunmap_atomic(user_mem, KM_USER0);
 384         if (is_partial_io(bvec))
 385                         kfree(uncmem);
 386
 387         if (unlikely(ret != LZO_E_OK)) {
 388                 pr_err("Compression failed! err=%d\n", ret);
 389                 goto out;
 390         }
 391
 392         /*
 393          * Page is incompressible. Store it as-is (uncompressed)
 394          * since we do not want to return too many disk write
 395          * errors which has side effect of hanging the system.
 396          */
 397         if (unlikely(clen > max_zpage_size)) {
 398                 clen = PAGE_SIZE;
 399                 page_store = alloc_page(GFP_NOIO | __GFP_HIGHMEM);
 400                 if (unlikely(!page_store)) {
 401                         pr_info("Error allocating memory for "
 402                                 "incompressible page: %u\n", index);
 403                         ret = -ENOMEM;
 404                         goto out;
 405                 }
 406
 407                 store_offset = 0;
 408                 zram_set_flag(zram, index, ZRAM_UNCOMPRESSED);
 409                 zram_stat_inc(&zram->stats.pages_expand);
 410                 zram->table[index].page = page_store;
 411                 src = kmap_atomic(page, KM_USER0);
 412                 goto memstore;
 413         }
 414
 415         if (xv_malloc(zram->mem_pool, clen + sizeof(*zheader),
 416                       &zram->table[index].page, &store_offset,
 417                       GFP_NOIO | __GFP_HIGHMEM)) {
 418                 pr_info("Error allocating memory for compressed "
 419                         "page: %u, size=%zu\n", index, clen);
 420                 ret = -ENOMEM;
 421                 goto out;
 422         }
 423
 424 memstore:
 425         zram->table[index].offset = store_offset;
 426
 427         cmem = kmap_atomic(zram->table[index].page, KM_USER1) +
 428                 zram->table[index].offset;
 429
 430 #if 0
 431         /* Back-reference needed for memory defragmentation */
 432         if (!zram_test_flag(zram, index, ZRAM_UNCOMPRESSED)) {
 433                 zheader = (struct zobj_header *)cmem;
 434                 zheader->table_idx = index;
 435                 cmem += sizeof(*zheader);
 436         }
 437 #endif
 438
 439         memcpy(cmem, src, clen);
 440
 441         kunmap_atomic(cmem, KM_USER1);
 442         if (unlikely(zram_test_flag(zram, index, ZRAM_UNCOMPRESSED)))
 443                 kunmap_atomic(src, KM_USER0);
 444
 445         /* Update stats */
 446         zram_stat64_add(zram, &zram->stats.compr_size, clen);
 447         zram_stat_inc(&zram->stats.pages_stored);
 448         if (clen <= PAGE_SIZE / 2)
 449                 zram_stat_inc(&zram->stats.good_compress);
 450
 451         return 0;
 452
 453 out:
 454         if (ret)
 455                 zram_stat64_inc(zram, &zram->stats.failed_writes);
 456         return ret;
 457 }
 458
 459 static int zram_bvec_rw(struct zram *zram, struct bio_vec *bvec, u32 index,
 460                         int offset, struct bio *bio, int rw)
 461 {
 462         int ret;
 463
 464         if (rw == READ) {
 465                 down_read(&zram->lock);
 466                 ret = zram_bvec_read(zram, bvec, index, offset, bio);
 467                 up_read(&zram->lock);
 468         } else {
 469                 down_write(&zram->lock);
 470                 ret = zram_bvec_write(zram, bvec, index, offset);
 471                 up_write(&zram->lock);
 472         }
 473
 474         return ret;
 475 }
 476
 477 static void update_position(u32 *index, int *offset, struct bio_vec *bvec)
 478 {
 479         if (*offset + bvec->bv_len >= PAGE_SIZE)
 480                 (*index)++;
 481         *offset = (*offset + bvec->bv_len) % PAGE_SIZE;
 482 }
 483
 484 static void __zram_make_request(struct zram *zram, struct bio *bio, int rw)
 485 {
 486         int i, offset;
 487         u32 index;
 488         struct bio_vec *bvec;
 489
 490         switch (rw) {
 491         case READ:
 492                 zram_stat64_inc(zram, &zram->stats.num_reads);
 493                 break;
 494         case WRITE:
 495                 zram_stat64_inc(zram, &zram->stats.num_writes);
 496                 break;
 497         }
 498
 499         index = bio->bi_sector >> SECTORS_PER_PAGE_SHIFT;
 500         offset = (bio->bi_sector & (SECTORS_PER_PAGE - 1)) << SECTOR_SHIFT;
 501
 502         bio_for_each_segment(bvec, bio, i) {
 503                 int max_transfer_size = PAGE_SIZE - offset;
 504
 505                 if (bvec->bv_len > max_transfer_size) {
 506                         /*
 507                          * zram_bvec_rw() can only make operation on a single
 508                          * zram page. Split the bio vector.
 509                          */
 510                         struct bio_vec bv;
 511
 512                         bv.bv_page = bvec->bv_page;
 513                         bv.bv_len = max_transfer_size;
 514                         bv.bv_offset = bvec->bv_offset;
 515
 516                         if (zram_bvec_rw(zram, &bv, index, offset, bio, rw) < 0)
 517                                 goto out;
 518
 519                         bv.bv_len = bvec->bv_len - max_transfer_size;
 520                         bv.bv_offset += max_transfer_size;
 521                         if (zram_bvec_rw(zram, &bv, index+1, 0, bio, rw) < 0)
 522                                 goto out;
 523                 } else
 524                         if (zram_bvec_rw(zram, bvec, index, offset, bio, rw)
 525                             < 0)
 526                                 goto out;
 527
 528                 update_position(&index, &offset, bvec);
 529         }
 530
 531         set_bit(BIO_UPTODATE, &bio->bi_flags);
 532         bio_endio(bio, 0);
 533         return;
 534
 535 out:
 536         bio_io_error(bio);
 537 }
 538
 539 /*
 540  * Check if request is within bounds and aligned on zram logical blocks.
 541  */
 542 static inline int valid_io_request(struct zram *zram, struct bio *bio)
 543 {
 544         if (unlikely(
 545                 (bio->bi_sector >= (zram->disksize >> SECTOR_SHIFT)) ||
 546                 (bio->bi_sector & (ZRAM_SECTOR_PER_LOGICAL_BLOCK - 1)) ||
 547                 (bio->bi_size & (ZRAM_LOGICAL_BLOCK_SIZE - 1)))) {
 548
 549                 return 0;
 550         }
 551
 552         /* I/O request is valid */
 553         return 1;
 554 }
 555
 556 /*
 557  * Handler function for all zram I/O requests.
 558  */
 559 static int zram_make_request(struct request_queue *queue, struct bio *bio)
 560 {
 561         struct zram *zram = queue->queuedata;
 562
 563         if (!valid_io_request(zram, bio)) {
 564                 zram_stat64_inc(zram, &zram->stats.invalid_io);
 565                 bio_io_error(bio);
 566                 return 0;
 567         }
 568
 569         if (unlikely(!zram->init_done) && zram_init_device(zram)) {
 570                 bio_io_error(bio);
 571                 return 0;
 572         }
 573
 574         __zram_make_request(zram, bio, bio_data_dir(bio));
 575
 576         return 0;
 577 }
 578
 579 void zram_reset_device(struct zram *zram)
 580 {
 581         size_t index;
 582
 583         mutex_lock(&zram->init_lock);
 584         zram->init_done = 0;
 585
 586         /* Free various per-device buffers */
 587         kfree(zram->compress_workmem);
 588         free_pages((unsigned long)zram->compress_buffer, 1);
 589
 590         zram->compress_workmem = NULL;
 591         zram->compress_buffer = NULL;
 592
 593         /* Free all pages that are still in this zram device */
 594         for (index = 0; index < zram->disksize >> PAGE_SHIFT; index++) {
 595                 struct page *page;
 596                 u16 offset;
 597
 598                 page = zram->table[index].page;
 599                 offset = zram->table[index].offset;
 600
 601                 if (!page)
 602                         continue;
 603
 604                 if (unlikely(zram_test_flag(zram, index, ZRAM_UNCOMPRESSED)))
 605                         __free_page(page);
 606                 else
 607                         xv_free(zram->mem_pool, page, offset);
 608         }
 609
 610         vfree(zram->table);
 611         zram->table = NULL;
 612
 613         xv_destroy_pool(zram->mem_pool);
 614         zram->mem_pool = NULL;
 615
 616         /* Reset stats */
 617         memset(&zram->stats, 0, sizeof(zram->stats));
 618
 619         zram->disksize = 0;
 620         mutex_unlock(&zram->init_lock);
 621 }
 622
 623 int zram_init_device(struct zram *zram)
 624 {
 625         int ret;
 626         size_t num_pages;
 627
 628         mutex_lock(&zram->init_lock);
 629
 630         if (zram->init_done) {
 631                 mutex_unlock(&zram->init_lock);
 632                 return 0;
 633         }
 634
 635         zram_set_disksize(zram, totalram_pages << PAGE_SHIFT);
 636
 637         zram->compress_workmem = kzalloc(LZO1X_MEM_COMPRESS, GFP_KERNEL);
 638         if (!zram->compress_workmem) {
 639                 pr_err("Error allocating compressor working memory!\n");
 640                 ret = -ENOMEM;
 641                 goto fail;
 642         }
 643
 644         zram->compress_buffer = (void *)__get_free_pages(__GFP_ZERO, 1);
 645         if (!zram->compress_buffer) {
 646                 pr_err("Error allocating compressor buffer space\n");
 647                 ret = -ENOMEM;
 648                 goto fail;
 649         }
 650
 651         num_pages = zram->disksize >> PAGE_SHIFT;
 652         zram->table = vzalloc(num_pages * sizeof(*zram->table));
 653         if (!zram->table) {
 654                 pr_err("Error allocating zram address table\n");
 655                 /* To prevent accessing table entries during cleanup */
 656                 zram->disksize = 0;
 657                 ret = -ENOMEM;
 658                 goto fail;
 659         }
 660
 661         set_capacity(zram->disk, zram->disksize >> SECTOR_SHIFT);
 662
 663         /* zram devices sort of resembles non-rotational disks */
 664         queue_flag_set_unlocked(QUEUE_FLAG_NONROT, zram->disk->queue);
 665
 666         zram->mem_pool = xv_create_pool();
 667         if (!zram->mem_pool) {
 668                 pr_err("Error creating memory pool\n");
 669                 ret = -ENOMEM;
 670                 goto fail;
 671         }
 672
 673         zram->init_done = 1;
 674         mutex_unlock(&zram->init_lock);
 675
 676         pr_debug("Initialization done!\n");
 677         return 0;
 678
 679 fail:
 680         mutex_unlock(&zram->init_lock);
 681         zram_reset_device(zram);
 682
 683         pr_err("Initialization failed: err=%d\n", ret);
 684         return ret;
 685 }
 686
 687 void zram_slot_free_notify(struct block_device *bdev, unsigned long index)
 688 {
 689         struct zram *zram;
 690
 691         zram = bdev->bd_disk->private_data;
 692         zram_free_page(zram, index);
 693         zram_stat64_inc(zram, &zram->stats.notify_free);
 694 }
 695
 696 static const struct block_device_operations zram_devops = {
 697         .swap_slot_free_notify = zram_slot_free_notify,
 698         .owner = THIS_MODULE
 699 };
 700
 701 static int create_device(struct zram *zram, int device_id)
 702 {
 703         int ret = 0;
 704
 705         init_rwsem(&zram->lock);
 706         mutex_init(&zram->init_lock);
 707         spin_lock_init(&zram->stat64_lock);
 708
 709         zram->queue = blk_alloc_queue(GFP_KERNEL);
 710         if (!zram->queue) {
 711                 pr_err("Error allocating disk queue for device %d\n",
 712                         device_id);
 713                 ret = -ENOMEM;
 714                 goto out;
 715         }
 716
 717         blk_queue_make_request(zram->queue, zram_make_request);
 718         zram->queue->queuedata = zram;
 719
 720          /* gendisk structure */
 721         zram->disk = alloc_disk(1);
 722         if (!zram->disk) {
 723                 blk_cleanup_queue(zram->queue);
 724                 pr_warning("Error allocating disk structure for device %d\n",
 725                         device_id);
 726                 ret = -ENOMEM;
 727                 goto out;
 728         }
 729
 730         zram->disk->major = zram_major;
 731         zram->disk->first_minor = device_id;
 732         zram->disk->fops = &zram_devops;
 733         zram->disk->queue = zram->queue;
 734         zram->disk->private_data = zram;
 735         snprintf(zram->disk->disk_name, 16, "zram%d", device_id);
 736
 737         /* Actual capacity set using syfs (/sys/block/zram<id>/disksize */
 738         set_capacity(zram->disk, 0);
 739
 740         /*
 741          * To ensure that we always get PAGE_SIZE aligned
 742          * and n*PAGE_SIZED sized I/O requests.
 743          */
 744         blk_queue_physical_block_size(zram->disk->queue, PAGE_SIZE);
 745         blk_queue_logical_block_size(zram->disk->queue,
 746                                         ZRAM_LOGICAL_BLOCK_SIZE);
 747         blk_queue_io_min(zram->disk->queue, PAGE_SIZE);
 748         blk_queue_io_opt(zram->disk->queue, PAGE_SIZE);
 749
 750         add_disk(zram->disk);
 751
 752         ret = sysfs_create_group(&disk_to_dev(zram->disk)->kobj,
 753                                 &zram_disk_attr_group);
 754         if (ret < 0) {
 755                 pr_warning("Error creating sysfs group");
 756                 goto out;
 757         }
 758
 759         zram->init_done = 0;
 760
 761 out:
 762         return ret;
 763 }
 764
 765 static void destroy_device(struct zram *zram)
 766 {
 767         sysfs_remove_group(&disk_to_dev(zram->disk)->kobj,
 768                         &zram_disk_attr_group);
 769
 770         if (zram->disk) {
 771                 del_gendisk(zram->disk);
 772                 put_disk(zram->disk);
 773         }
 774
 775         if (zram->queue)
 776                 blk_cleanup_queue(zram->queue);
 777 }
 778
 779 static int __init zram_init(void)
 780 {
 781         int ret, dev_id;
 782
 783         if (num_devices > max_num_devices) {
 784                 pr_warning("Invalid value for num_devices: %u\n",
 785                                 num_devices);
 786                 ret = -EINVAL;
 787                 goto out;
 788         }
 789
 790         zram_major = register_blkdev(0, "zram");
 791         if (zram_major <= 0) {
 792                 pr_warning("Unable to get major number\n");
 793                 ret = -EBUSY;
 794                 goto out;
 795         }
 796
 797         if (!num_devices) {
 798                 pr_info("num_devices not specified. Using default: 1\n");
 799                 num_devices = 1;
 800         }
 801
 802         /* Allocate the device array and initialize each one */
 803         pr_info("Creating %u devices ...\n", num_devices);
 804         devices = kzalloc(num_devices * sizeof(struct zram), GFP_KERNEL);
 805         if (!devices) {
 806                 ret = -ENOMEM;
 807                 goto unregister;
 808         }
 809
 810         for (dev_id = 0; dev_id < num_devices; dev_id++) {
 811                 ret = create_device(&devices[dev_id], dev_id);
 812                 if (ret)
 813                         goto free_devices;
 814         }
 815
 816         return 0;
 817
 818 free_devices:
 819         while (dev_id)
 820                 destroy_device(&devices[--dev_id]);
 821         kfree(devices);
 822 unregister:
 823         unregister_blkdev(zram_major, "zram");
 824 out:
 825         return ret;
 826 }
 827
 828 static void __exit zram_exit(void)
 829 {
 830         int i;
 831         struct zram *zram;
 832
 833         for (i = 0; i < num_devices; i++) {
 834                 zram = &devices[i];
 835
 836                 destroy_device(zram);
 837                 if (zram->init_done)
 838                         zram_reset_device(zram);
 839         }
 840
 841         unregister_blkdev(zram_major, "zram");
 842
 843         kfree(devices);
 844         pr_debug("Cleanup done!\n");
 845 }
 846
 847 module_param(num_devices, uint, 0);
 848 MODULE_PARM_DESC(num_devices, "Number of zram devices");
 849
 850 module_init(zram_init);
 851 module_exit(zram_exit);
 852
 853 MODULE_LICENSE("Dual BSD/GPL");
 854 MODULE_AUTHOR("Nitin Gupta <ngupta@vflare.org>");
 855 MODULE_DESCRIPTION("Compressed RAM Block Device");