drivers/block/zram/zram_drv.c

   1 /*
   2  * Compressed RAM block device
   3  *
   4  * Copyright (C) 2008, 2009, 2010  Nitin Gupta
   5  *               2012, 2013 Minchan Kim
   6  *
   7  * This code is released using a dual license strategy: BSD/GPL
   8  * You can choose the licence that better fits your requirements.
   9  *
  10  * Released under the terms of 3-clause BSD License
  11  * Released under the terms of GNU General Public License Version 2.0
  12  *
  13  */
  14
  15 #define KMSG_COMPONENT "zram"
  16 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
  17
  18 #include <linux/module.h>
  19 #include <linux/kernel.h>
  20 #include <linux/bio.h>
  21 #include <linux/bitops.h>
  22 #include <linux/blkdev.h>
  23 #include <linux/buffer_head.h>
  24 #include <linux/device.h>
  25 #include <linux/genhd.h>
  26 #include <linux/highmem.h>
  27 #include <linux/slab.h>
  28 #include <linux/backing-dev.h>
  29 #include <linux/string.h>
  30 #include <linux/vmalloc.h>
  31 #include <linux/err.h>
  32 #include <linux/idr.h>
  33 #include <linux/sysfs.h>
  34 #include <linux/cpuhotplug.h>
  35
  36 #include "zram_drv.h"
  37
  38 static DEFINE_IDR(zram_index_idr);
  39 /* idr index must be protected */
  40 static DEFINE_MUTEX(zram_index_mutex);
  41
  42 static int zram_major;
  43 static const char *default_compressor = "lzo";
  44
  45 /* Module params (documentation at end) */
  46 static unsigned int num_devices = 1;
  47
  48 static void zram_free_page(struct zram *zram, size_t index);
  49
  50 static inline bool init_done(struct zram *zram)
  51 {
  52         return zram->disksize;
  53 }
  54
  55 static inline struct zram *dev_to_zram(struct device *dev)
  56 {
  57         return (struct zram *)dev_to_disk(dev)->private_data;
  58 }
  59
  60 static unsigned long zram_get_handle(struct zram *zram, u32 index)
  61 {
  62         return zram->table[index].handle;
  63 }
  64
  65 static void zram_set_handle(struct zram *zram, u32 index, unsigned long handle)
  66 {
  67         zram->table[index].handle = handle;
  68 }
  69
  70 /* flag operations require table entry bit_spin_lock() being held */
  71 static int zram_test_flag(struct zram *zram, u32 index,
  72                         enum zram_pageflags flag)
  73 {
  74         return zram->table[index].value & BIT(flag);
  75 }
  76
  77 static void zram_set_flag(struct zram *zram, u32 index,
  78                         enum zram_pageflags flag)
  79 {
  80         zram->table[index].value |= BIT(flag);
  81 }
  82
  83 static void zram_clear_flag(struct zram *zram, u32 index,
  84                         enum zram_pageflags flag)
  85 {
  86         zram->table[index].value &= ~BIT(flag);
  87 }
  88
  89 static inline void zram_set_element(struct zram *zram, u32 index,
  90                         unsigned long element)
  91 {
  92         zram->table[index].element = element;
  93 }
  94
  95 static unsigned long zram_get_element(struct zram *zram, u32 index)
  96 {
  97         return zram->table[index].element;
  98 }
  99
 100 static size_t zram_get_obj_size(struct zram *zram, u32 index)
 101 {
 102         return zram->table[index].value & (BIT(ZRAM_FLAG_SHIFT) - 1);
 103 }
 104
 105 static void zram_set_obj_size(struct zram *zram,
 106                                         u32 index, size_t size)
 107 {
 108         unsigned long flags = zram->table[index].value >> ZRAM_FLAG_SHIFT;
 109
 110         zram->table[index].value = (flags << ZRAM_FLAG_SHIFT) | size;
 111 }
 112
 113 #if PAGE_SIZE != 4096
 114 static inline bool is_partial_io(struct bio_vec *bvec)
 115 {
 116         return bvec->bv_len != PAGE_SIZE;
 117 }
 118 #else
 119 static inline bool is_partial_io(struct bio_vec *bvec)
 120 {
 121         return false;
 122 }
 123 #endif
 124
 125 static void zram_revalidate_disk(struct zram *zram)
 126 {
 127         revalidate_disk(zram->disk);
 128         /* revalidate_disk reset the BDI_CAP_STABLE_WRITES so set again */
 129         zram->disk->queue->backing_dev_info->capabilities |=
 130                 BDI_CAP_STABLE_WRITES;
 131 }
 132
 133 /*
 134  * Check if request is within bounds and aligned on zram logical blocks.
 135  */
 136 static inline bool valid_io_request(struct zram *zram,
 137                 sector_t start, unsigned int size)
 138 {
 139         u64 end, bound;
 140
 141         /* unaligned request */
 142         if (unlikely(start & (ZRAM_SECTOR_PER_LOGICAL_BLOCK - 1)))
 143                 return false;
 144         if (unlikely(size & (ZRAM_LOGICAL_BLOCK_SIZE - 1)))
 145                 return false;
 146
 147         end = start + (size >> SECTOR_SHIFT);
 148         bound = zram->disksize >> SECTOR_SHIFT;
 149         /* out of range range */
 150         if (unlikely(start >= bound || end > bound || start > end))
 151                 return false;
 152
 153         /* I/O request is valid */
 154         return true;
 155 }
 156
 157 static void update_position(u32 *index, int *offset, struct bio_vec *bvec)
 158 {
 159         *index  += (*offset + bvec->bv_len) / PAGE_SIZE;
 160         *offset = (*offset + bvec->bv_len) % PAGE_SIZE;
 161 }
 162
 163 static inline void update_used_max(struct zram *zram,
 164                                         const unsigned long pages)
 165 {
 166         unsigned long old_max, cur_max;
 167
 168         old_max = atomic_long_read(&zram->stats.max_used_pages);
 169
 170         do {
 171                 cur_max = old_max;
 172                 if (pages > cur_max)
 173                         old_max = atomic_long_cmpxchg(
 174                                 &zram->stats.max_used_pages, cur_max, pages);
 175         } while (old_max != cur_max);
 176 }
 177
 178 static inline void zram_fill_page(char *ptr, unsigned long len,
 179                                         unsigned long value)
 180 {
 181         int i;
 182         unsigned long *page = (unsigned long *)ptr;
 183
 184         WARN_ON_ONCE(!IS_ALIGNED(len, sizeof(unsigned long)));
 185
 186         if (likely(value == 0)) {
 187                 memset(ptr, 0, len);
 188         } else {
 189                 for (i = 0; i < len / sizeof(*page); i++)
 190                         page[i] = value;
 191         }
 192 }
 193
 194 static bool page_same_filled(void *ptr, unsigned long *element)
 195 {
 196         unsigned int pos;
 197         unsigned long *page;
 198         unsigned long val;
 199
 200         page = (unsigned long *)ptr;
 201         val = page[0];
 202
 203         for (pos = 1; pos < PAGE_SIZE / sizeof(*page); pos++) {
 204                 if (val != page[pos])
 205                         return false;
 206         }
 207
 208         *element = val;
 209
 210         return true;
 211 }
 212
 213 static ssize_t initstate_show(struct device *dev,
 214                 struct device_attribute *attr, char *buf)
 215 {
 216         u32 val;
 217         struct zram *zram = dev_to_zram(dev);
 218
 219         down_read(&zram->init_lock);
 220         val = init_done(zram);
 221         up_read(&zram->init_lock);
 222
 223         return scnprintf(buf, PAGE_SIZE, "%u\n", val);
 224 }
 225
 226 static ssize_t disksize_show(struct device *dev,
 227                 struct device_attribute *attr, char *buf)
 228 {
 229         struct zram *zram = dev_to_zram(dev);
 230
 231         return scnprintf(buf, PAGE_SIZE, "%llu\n", zram->disksize);
 232 }
 233
 234 static ssize_t mem_limit_store(struct device *dev,
 235                 struct device_attribute *attr, const char *buf, size_t len)
 236 {
 237         u64 limit;
 238         char *tmp;
 239         struct zram *zram = dev_to_zram(dev);
 240
 241         limit = memparse(buf, &tmp);
 242         if (buf == tmp) /* no chars parsed, invalid input */
 243                 return -EINVAL;
 244
 245         down_write(&zram->init_lock);
 246         zram->limit_pages = PAGE_ALIGN(limit) >> PAGE_SHIFT;
 247         up_write(&zram->init_lock);
 248
 249         return len;
 250 }
 251
 252 static ssize_t mem_used_max_store(struct device *dev,
 253                 struct device_attribute *attr, const char *buf, size_t len)
 254 {
 255         int err;
 256         unsigned long val;
 257         struct zram *zram = dev_to_zram(dev);
 258
 259         err = kstrtoul(buf, 10, &val);
 260         if (err || val != 0)
 261                 return -EINVAL;
 262
 263         down_read(&zram->init_lock);
 264         if (init_done(zram)) {
 265                 atomic_long_set(&zram->stats.max_used_pages,
 266                                 zs_get_total_pages(zram->mem_pool));
 267         }
 268         up_read(&zram->init_lock);
 269
 270         return len;
 271 }
 272
 273 #ifdef CONFIG_ZRAM_WRITEBACK
 274 static bool zram_wb_enabled(struct zram *zram)
 275 {
 276         return zram->backing_dev;
 277 }
 278
 279 static void reset_bdev(struct zram *zram)
 280 {
 281         struct block_device *bdev;
 282
 283         if (!zram_wb_enabled(zram))
 284                 return;
 285
 286         bdev = zram->bdev;
 287         if (zram->old_block_size)
 288                 set_blocksize(bdev, zram->old_block_size);
 289         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 290         /* hope filp_close flush all of IO */
 291         filp_close(zram->backing_dev, NULL);
 292         zram->backing_dev = NULL;
 293         zram->old_block_size = 0;
 294         zram->bdev = NULL;
 295
 296         kvfree(zram->bitmap);
 297         zram->bitmap = NULL;
 298 }
 299
 300 static ssize_t backing_dev_show(struct device *dev,
 301                 struct device_attribute *attr, char *buf)
 302 {
 303         struct zram *zram = dev_to_zram(dev);
 304         struct file *file = zram->backing_dev;
 305         char *p;
 306         ssize_t ret;
 307
 308         down_read(&zram->init_lock);
 309         if (!zram_wb_enabled(zram)) {
 310                 memcpy(buf, "none\n", 5);
 311                 up_read(&zram->init_lock);
 312                 return 5;
 313         }
 314
 315         p = file_path(file, buf, PAGE_SIZE - 1);
 316         if (IS_ERR(p)) {
 317                 ret = PTR_ERR(p);
 318                 goto out;
 319         }
 320
 321         ret = strlen(p);
 322         memmove(buf, p, ret);
 323         buf[ret++] = '\n';
 324 out:
 325         up_read(&zram->init_lock);
 326         return ret;
 327 }
 328
 329 static ssize_t backing_dev_store(struct device *dev,
 330                 struct device_attribute *attr, const char *buf, size_t len)
 331 {
 332         char *file_name;
 333         struct file *backing_dev = NULL;
 334         struct inode *inode;
 335         struct address_space *mapping;
 336         unsigned int bitmap_sz, old_block_size = 0;
 337         unsigned long nr_pages, *bitmap = NULL;
 338         struct block_device *bdev = NULL;
 339         int err;
 340         struct zram *zram = dev_to_zram(dev);
 341
 342         file_name = kmalloc(PATH_MAX, GFP_KERNEL);
 343         if (!file_name)
 344                 return -ENOMEM;
 345
 346         down_write(&zram->init_lock);
 347         if (init_done(zram)) {
 348                 pr_info("Can't setup backing device for initialized device\n");
 349                 err = -EBUSY;
 350                 goto out;
 351         }
 352
 353         strlcpy(file_name, buf, len);
 354
 355         backing_dev = filp_open(file_name, O_RDWR|O_LARGEFILE, 0);
 356         if (IS_ERR(backing_dev)) {
 357                 err = PTR_ERR(backing_dev);
 358                 backing_dev = NULL;
 359                 goto out;
 360         }
 361
 362         mapping = backing_dev->f_mapping;
 363         inode = mapping->host;
 364
 365         /* Support only block device in this moment */
 366         if (!S_ISBLK(inode->i_mode)) {
 367                 err = -ENOTBLK;
 368                 goto out;
 369         }
 370
 371         bdev = bdgrab(I_BDEV(inode));
 372         err = blkdev_get(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL, zram);
 373         if (err < 0)
 374                 goto out;
 375
 376         nr_pages = i_size_read(inode) >> PAGE_SHIFT;
 377         bitmap_sz = BITS_TO_LONGS(nr_pages) * sizeof(long);
 378         bitmap = kvzalloc(bitmap_sz, GFP_KERNEL);
 379         if (!bitmap) {
 380                 err = -ENOMEM;
 381                 goto out;
 382         }
 383
 384         old_block_size = block_size(bdev);
 385         err = set_blocksize(bdev, PAGE_SIZE);
 386         if (err)
 387                 goto out;
 388
 389         reset_bdev(zram);
 390         spin_lock_init(&zram->bitmap_lock);
 391
 392         zram->old_block_size = old_block_size;
 393         zram->bdev = bdev;
 394         zram->backing_dev = backing_dev;
 395         zram->bitmap = bitmap;
 396         zram->nr_pages = nr_pages;
 397         up_write(&zram->init_lock);
 398
 399         pr_info("setup backing device %s\n", file_name);
 400         kfree(file_name);
 401
 402         return len;
 403 out:
 404         if (bitmap)
 405                 kvfree(bitmap);
 406
 407         if (bdev)
 408                 blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
 409
 410         if (backing_dev)
 411                 filp_close(backing_dev, NULL);
 412
 413         up_write(&zram->init_lock);
 414
 415         kfree(file_name);
 416
 417         return err;
 418 }
 419
 420 static unsigned long get_entry_bdev(struct zram *zram)
 421 {
 422         unsigned long entry;
 423
 424         spin_lock(&zram->bitmap_lock);
 425         /* skip 0 bit to confuse zram.handle = 0 */
 426         entry = find_next_zero_bit(zram->bitmap, zram->nr_pages, 1);
 427         if (entry == zram->nr_pages) {
 428                 spin_unlock(&zram->bitmap_lock);
 429                 return 0;
 430         }
 431
 432         set_bit(entry, zram->bitmap);
 433         spin_unlock(&zram->bitmap_lock);
 434
 435         return entry;
 436 }
 437
 438 static void put_entry_bdev(struct zram *zram, unsigned long entry)
 439 {
 440         int was_set;
 441
 442         spin_lock(&zram->bitmap_lock);
 443         was_set = test_and_clear_bit(entry, zram->bitmap);
 444         spin_unlock(&zram->bitmap_lock);
 445         WARN_ON_ONCE(!was_set);
 446 }
 447
 448 void zram_page_end_io(struct bio *bio)
 449 {
 450         struct page *page = bio->bi_io_vec[0].bv_page;
 451
 452         page_endio(page, op_is_write(bio_op(bio)),
 453                         blk_status_to_errno(bio->bi_status));
 454         bio_put(bio);
 455 }
 456
 457 static int write_to_bdev(struct zram *zram, struct bio_vec *bvec,
 458                                         u32 index, struct bio *parent,
 459                                         unsigned long *pentry)
 460 {
 461         struct bio *bio;
 462         unsigned long entry;
 463
 464         bio = bio_alloc(GFP_ATOMIC, 1);
 465         if (!bio)
 466                 return -ENOMEM;
 467
 468         entry = get_entry_bdev(zram);
 469         if (!entry) {
 470                 bio_put(bio);
 471                 return -ENOSPC;
 472         }
 473
 474         bio->bi_iter.bi_sector = entry * (PAGE_SIZE >> 9);
 475         bio->bi_bdev = zram->bdev;
 476         if (!bio_add_page(bio, bvec->bv_page, bvec->bv_len,
 477                                         bvec->bv_offset)) {
 478                 bio_put(bio);
 479                 put_entry_bdev(zram, entry);
 480                 return -EIO;
 481         }
 482
 483         if (!parent) {
 484                 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC;
 485                 bio->bi_end_io = zram_page_end_io;
 486         } else {
 487                 bio->bi_opf = parent->bi_opf;
 488                 bio_chain(bio, parent);
 489         }
 490
 491         submit_bio(bio);
 492         *pentry = entry;
 493
 494         return 0;
 495 }
 496
 497 static void zram_wb_clear(struct zram *zram, u32 index)
 498 {
 499         unsigned long entry;
 500
 501         zram_clear_flag(zram, index, ZRAM_WB);
 502         entry = zram_get_element(zram, index);
 503         zram_set_element(zram, index, 0);
 504         put_entry_bdev(zram, entry);
 505 }
 506
 507 #else
 508 static bool zram_wb_enabled(struct zram *zram) { return false; }
 509 static inline void reset_bdev(struct zram *zram) {};
 510 static int write_to_bdev(struct zram *zram, struct bio_vec *bvec,
 511                                         u32 index, struct bio *parent,
 512                                         unsigned long *pentry)
 513
 514 {
 515         return -EIO;
 516 }
 517 static void zram_wb_clear(struct zram *zram, u32 index) {}
 518 #endif
 519
 520
 521 /*
 522  * We switched to per-cpu streams and this attr is not needed anymore.
 523  * However, we will keep it around for some time, because:
 524  * a) we may revert per-cpu streams in the future
 525  * b) it's visible to user space and we need to follow our 2 years
 526  *    retirement rule; but we already have a number of 'soon to be
 527  *    altered' attrs, so max_comp_streams need to wait for the next
 528  *    layoff cycle.
 529  */
 530 static ssize_t max_comp_streams_show(struct device *dev,
 531                 struct device_attribute *attr, char *buf)
 532 {
 533         return scnprintf(buf, PAGE_SIZE, "%d\n", num_online_cpus());
 534 }
 535
 536 static ssize_t max_comp_streams_store(struct device *dev,
 537                 struct device_attribute *attr, const char *buf, size_t len)
 538 {
 539         return len;
 540 }
 541
 542 static ssize_t comp_algorithm_show(struct device *dev,
 543                 struct device_attribute *attr, char *buf)
 544 {
 545         size_t sz;
 546         struct zram *zram = dev_to_zram(dev);
 547
 548         down_read(&zram->init_lock);
 549         sz = zcomp_available_show(zram->compressor, buf);
 550         up_read(&zram->init_lock);
 551
 552         return sz;
 553 }
 554
 555 static ssize_t comp_algorithm_store(struct device *dev,
 556                 struct device_attribute *attr, const char *buf, size_t len)
 557 {
 558         struct zram *zram = dev_to_zram(dev);
 559         char compressor[ARRAY_SIZE(zram->compressor)];
 560         size_t sz;
 561
 562         strlcpy(compressor, buf, sizeof(compressor));
 563         /* ignore trailing newline */
 564         sz = strlen(compressor);
 565         if (sz > 0 && compressor[sz - 1] == '\n')
 566                 compressor[sz - 1] = 0x00;
 567
 568         if (!zcomp_available_algorithm(compressor))
 569                 return -EINVAL;
 570
 571         down_write(&zram->init_lock);
 572         if (init_done(zram)) {
 573                 up_write(&zram->init_lock);
 574                 pr_info("Can't change algorithm for initialized device\n");
 575                 return -EBUSY;
 576         }
 577
 578         strcpy(zram->compressor, compressor);
 579         up_write(&zram->init_lock);
 580         return len;
 581 }
 582
 583 static ssize_t compact_store(struct device *dev,
 584                 struct device_attribute *attr, const char *buf, size_t len)
 585 {
 586         struct zram *zram = dev_to_zram(dev);
 587
 588         down_read(&zram->init_lock);
 589         if (!init_done(zram)) {
 590                 up_read(&zram->init_lock);
 591                 return -EINVAL;
 592         }
 593
 594         zs_compact(zram->mem_pool);
 595         up_read(&zram->init_lock);
 596
 597         return len;
 598 }
 599
 600 static ssize_t io_stat_show(struct device *dev,
 601                 struct device_attribute *attr, char *buf)
 602 {
 603         struct zram *zram = dev_to_zram(dev);
 604         ssize_t ret;
 605
 606         down_read(&zram->init_lock);
 607         ret = scnprintf(buf, PAGE_SIZE,
 608                         "%8llu %8llu %8llu %8llu\n",
 609                         (u64)atomic64_read(&zram->stats.failed_reads),
 610                         (u64)atomic64_read(&zram->stats.failed_writes),
 611                         (u64)atomic64_read(&zram->stats.invalid_io),
 612                         (u64)atomic64_read(&zram->stats.notify_free));
 613         up_read(&zram->init_lock);
 614
 615         return ret;
 616 }
 617
 618 static ssize_t mm_stat_show(struct device *dev,
 619                 struct device_attribute *attr, char *buf)
 620 {
 621         struct zram *zram = dev_to_zram(dev);
 622         struct zs_pool_stats pool_stats;
 623         u64 orig_size, mem_used = 0;
 624         long max_used;
 625         ssize_t ret;
 626
 627         memset(&pool_stats, 0x00, sizeof(struct zs_pool_stats));
 628
 629         down_read(&zram->init_lock);
 630         if (init_done(zram)) {
 631                 mem_used = zs_get_total_pages(zram->mem_pool);
 632                 zs_pool_stats(zram->mem_pool, &pool_stats);
 633         }
 634
 635         orig_size = atomic64_read(&zram->stats.pages_stored);
 636         max_used = atomic_long_read(&zram->stats.max_used_pages);
 637
 638         ret = scnprintf(buf, PAGE_SIZE,
 639                         "%8llu %8llu %8llu %8lu %8ld %8llu %8lu\n",
 640                         orig_size << PAGE_SHIFT,
 641                         (u64)atomic64_read(&zram->stats.compr_data_size),
 642                         mem_used << PAGE_SHIFT,
 643                         zram->limit_pages << PAGE_SHIFT,
 644                         max_used << PAGE_SHIFT,
 645                         (u64)atomic64_read(&zram->stats.same_pages),
 646                         pool_stats.pages_compacted);
 647         up_read(&zram->init_lock);
 648
 649         return ret;
 650 }
 651
 652 static ssize_t debug_stat_show(struct device *dev,
 653                 struct device_attribute *attr, char *buf)
 654 {
 655         int version = 1;
 656         struct zram *zram = dev_to_zram(dev);
 657         ssize_t ret;
 658
 659         down_read(&zram->init_lock);
 660         ret = scnprintf(buf, PAGE_SIZE,
 661                         "version: %d\n%8llu\n",
 662                         version,
 663                         (u64)atomic64_read(&zram->stats.writestall));
 664         up_read(&zram->init_lock);
 665
 666         return ret;
 667 }
 668
 669 static DEVICE_ATTR_RO(io_stat);
 670 static DEVICE_ATTR_RO(mm_stat);
 671 static DEVICE_ATTR_RO(debug_stat);
 672
 673 static void zram_slot_lock(struct zram *zram, u32 index)
 674 {
 675         bit_spin_lock(ZRAM_ACCESS, &zram->table[index].value);
 676 }
 677
 678 static void zram_slot_unlock(struct zram *zram, u32 index)
 679 {
 680         bit_spin_unlock(ZRAM_ACCESS, &zram->table[index].value);
 681 }
 682
 683 static bool zram_same_page_read(struct zram *zram, u32 index,
 684                                 struct page *page,
 685                                 unsigned int offset, unsigned int len)
 686 {
 687         zram_slot_lock(zram, index);
 688         if (unlikely(!zram_get_handle(zram, index) ||
 689                         zram_test_flag(zram, index, ZRAM_SAME))) {
 690                 void *mem;
 691
 692                 zram_slot_unlock(zram, index);
 693                 mem = kmap_atomic(page);
 694                 zram_fill_page(mem + offset, len,
 695                                         zram_get_element(zram, index));
 696                 kunmap_atomic(mem);
 697                 return true;
 698         }
 699         zram_slot_unlock(zram, index);
 700
 701         return false;
 702 }
 703
 704 static void zram_meta_free(struct zram *zram, u64 disksize)
 705 {
 706         size_t num_pages = disksize >> PAGE_SHIFT;
 707         size_t index;
 708
 709         /* Free all pages that are still in this zram device */
 710         for (index = 0; index < num_pages; index++)
 711                 zram_free_page(zram, index);
 712
 713         zs_destroy_pool(zram->mem_pool);
 714         vfree(zram->table);
 715 }
 716
 717 static bool zram_meta_alloc(struct zram *zram, u64 disksize)
 718 {
 719         size_t num_pages;
 720
 721         num_pages = disksize >> PAGE_SHIFT;
 722         zram->table = vzalloc(num_pages * sizeof(*zram->table));
 723         if (!zram->table)
 724                 return false;
 725
 726         zram->mem_pool = zs_create_pool(zram->disk->disk_name);
 727         if (!zram->mem_pool) {
 728                 vfree(zram->table);
 729                 return false;
 730         }
 731
 732         return true;
 733 }
 734
 735 /*
 736  * To protect concurrent access to the same index entry,
 737  * caller should hold this table index entry's bit_spinlock to
 738  * indicate this index entry is accessing.
 739  */
 740 static void zram_free_page(struct zram *zram, size_t index)
 741 {
 742         unsigned long handle;
 743
 744         if (zram_wb_enabled(zram) && zram_test_flag(zram, index, ZRAM_WB)) {
 745                 zram_wb_clear(zram, index);
 746                 atomic64_dec(&zram->stats.pages_stored);
 747                 return;
 748         }
 749
 750         /*
 751          * No memory is allocated for same element filled pages.
 752          * Simply clear same page flag.
 753          */
 754         if (zram_test_flag(zram, index, ZRAM_SAME)) {
 755                 zram_clear_flag(zram, index, ZRAM_SAME);
 756                 zram_set_element(zram, index, 0);
 757                 atomic64_dec(&zram->stats.same_pages);
 758                 atomic64_dec(&zram->stats.pages_stored);
 759                 return;
 760         }
 761
 762         handle = zram_get_handle(zram, index);
 763         if (!handle)
 764                 return;
 765
 766         zs_free(zram->mem_pool, handle);
 767
 768         atomic64_sub(zram_get_obj_size(zram, index),
 769                         &zram->stats.compr_data_size);
 770         atomic64_dec(&zram->stats.pages_stored);
 771
 772         zram_set_handle(zram, index, 0);
 773         zram_set_obj_size(zram, index, 0);
 774 }
 775
 776 static int __zram_bvec_read(struct zram *zram, struct page *page, u32 index)
 777 {
 778         int ret;
 779         unsigned long handle;
 780         unsigned int size;
 781         void *src, *dst;
 782
 783         if (zram_same_page_read(zram, index, page, 0, PAGE_SIZE))
 784                 return 0;
 785
 786         zram_slot_lock(zram, index);
 787         handle = zram_get_handle(zram, index);
 788         size = zram_get_obj_size(zram, index);
 789
 790         src = zs_map_object(zram->mem_pool, handle, ZS_MM_RO);
 791         if (size == PAGE_SIZE) {
 792                 dst = kmap_atomic(page);
 793                 memcpy(dst, src, PAGE_SIZE);
 794                 kunmap_atomic(dst);
 795                 ret = 0;
 796         } else {
 797                 struct zcomp_strm *zstrm = zcomp_stream_get(zram->comp);
 798
 799                 dst = kmap_atomic(page);
 800                 ret = zcomp_decompress(zstrm, src, size, dst);
 801                 kunmap_atomic(dst);
 802                 zcomp_stream_put(zram->comp);
 803         }
 804         zs_unmap_object(zram->mem_pool, handle);
 805         zram_slot_unlock(zram, index);
 806
 807         /* Should NEVER happen. Return bio error if it does. */
 808         if (unlikely(ret))
 809                 pr_err("Decompression failed! err=%d, page=%u\n", ret, index);
 810
 811         return ret;
 812 }
 813
 814 static int zram_bvec_read(struct zram *zram, struct bio_vec *bvec,
 815                                 u32 index, int offset)
 816 {
 817         int ret;
 818         struct page *page;
 819
 820         page = bvec->bv_page;
 821         if (is_partial_io(bvec)) {
 822                 /* Use a temporary buffer to decompress the page */
 823                 page = alloc_page(GFP_NOIO|__GFP_HIGHMEM);
 824                 if (!page)
 825                         return -ENOMEM;
 826         }
 827
 828         ret = __zram_bvec_read(zram, page, index);
 829         if (unlikely(ret))
 830                 goto out;
 831
 832         if (is_partial_io(bvec)) {
 833                 void *dst = kmap_atomic(bvec->bv_page);
 834                 void *src = kmap_atomic(page);
 835
 836                 memcpy(dst + bvec->bv_offset, src + offset, bvec->bv_len);
 837                 kunmap_atomic(src);
 838                 kunmap_atomic(dst);
 839         }
 840 out:
 841         if (is_partial_io(bvec))
 842                 __free_page(page);
 843
 844         return ret;
 845 }
 846
 847 static int __zram_bvec_write(struct zram *zram, struct bio_vec *bvec,
 848                                 u32 index, struct bio *bio)
 849 {
 850         int ret = 0;
 851         unsigned long alloced_pages;
 852         unsigned long handle = 0;
 853         unsigned int comp_len = 0;
 854         void *src, *dst, *mem;
 855         struct zcomp_strm *zstrm;
 856         struct page *page = bvec->bv_page;
 857         unsigned long element = 0;
 858         enum zram_pageflags flags = 0;
 859         bool allow_wb = true;
 860
 861         mem = kmap_atomic(page);
 862         if (page_same_filled(mem, &element)) {
 863                 kunmap_atomic(mem);
 864                 /* Free memory associated with this sector now. */
 865                 flags = ZRAM_SAME;
 866                 atomic64_inc(&zram->stats.same_pages);
 867                 goto out;
 868         }
 869         kunmap_atomic(mem);
 870
 871 compress_again:
 872         zstrm = zcomp_stream_get(zram->comp);
 873         src = kmap_atomic(page);
 874         ret = zcomp_compress(zstrm, src, &comp_len);
 875         kunmap_atomic(src);
 876
 877         if (unlikely(ret)) {
 878                 zcomp_stream_put(zram->comp);
 879                 pr_err("Compression failed! err=%d\n", ret);
 880                 zs_free(zram->mem_pool, handle);
 881                 return ret;
 882         }
 883
 884         if (unlikely(comp_len > max_zpage_size)) {
 885                 if (zram_wb_enabled(zram) && allow_wb) {
 886                         zcomp_stream_put(zram->comp);
 887                         ret = write_to_bdev(zram, bvec, index, bio, &element);
 888                         if (!ret) {
 889                                 flags = ZRAM_WB;
 890                                 ret = 1;
 891                                 goto out;
 892                         }
 893                         allow_wb = false;
 894                         goto compress_again;
 895                 }
 896                 comp_len = PAGE_SIZE;
 897         }
 898
 899         /*
 900          * handle allocation has 2 paths:
 901          * a) fast path is executed with preemption disabled (for
 902          *  per-cpu streams) and has __GFP_DIRECT_RECLAIM bit clear,
 903          *  since we can't sleep;
 904          * b) slow path enables preemption and attempts to allocate
 905          *  the page with __GFP_DIRECT_RECLAIM bit set. we have to
 906          *  put per-cpu compression stream and, thus, to re-do
 907          *  the compression once handle is allocated.
 908          *
 909          * if we have a 'non-null' handle here then we are coming
 910          * from the slow path and handle has already been allocated.
 911          */
 912         if (!handle)
 913                 handle = zs_malloc(zram->mem_pool, comp_len,
 914                                 __GFP_KSWAPD_RECLAIM |
 915                                 __GFP_NOWARN |
 916                                 __GFP_HIGHMEM |
 917                                 __GFP_MOVABLE);
 918         if (!handle) {
 919                 zcomp_stream_put(zram->comp);
 920                 atomic64_inc(&zram->stats.writestall);
 921                 handle = zs_malloc(zram->mem_pool, comp_len,
 922                                 GFP_NOIO | __GFP_HIGHMEM |
 923                                 __GFP_MOVABLE);
 924                 if (handle)
 925                         goto compress_again;
 926                 return -ENOMEM;
 927         }
 928
 929         alloced_pages = zs_get_total_pages(zram->mem_pool);
 930         update_used_max(zram, alloced_pages);
 931
 932         if (zram->limit_pages && alloced_pages > zram->limit_pages) {
 933                 zcomp_stream_put(zram->comp);
 934                 zs_free(zram->mem_pool, handle);
 935                 return -ENOMEM;
 936         }
 937
 938         dst = zs_map_object(zram->mem_pool, handle, ZS_MM_WO);
 939
 940         src = zstrm->buffer;
 941         if (comp_len == PAGE_SIZE)
 942                 src = kmap_atomic(page);
 943         memcpy(dst, src, comp_len);
 944         if (comp_len == PAGE_SIZE)
 945                 kunmap_atomic(src);
 946
 947         zcomp_stream_put(zram->comp);
 948         zs_unmap_object(zram->mem_pool, handle);
 949         atomic64_add(comp_len, &zram->stats.compr_data_size);
 950 out:
 951         /*
 952          * Free memory associated with this sector
 953          * before overwriting unused sectors.
 954          */
 955         zram_slot_lock(zram, index);
 956         zram_free_page(zram, index);
 957
 958         if (flags) {
 959                 zram_set_flag(zram, index, flags);
 960                 zram_set_element(zram, index, element);
 961         }  else {
 962                 zram_set_handle(zram, index, handle);
 963                 zram_set_obj_size(zram, index, comp_len);
 964         }
 965         zram_slot_unlock(zram, index);
 966
 967         /* Update stats */
 968         atomic64_inc(&zram->stats.pages_stored);
 969         return ret;
 970 }
 971
 972 static int zram_bvec_write(struct zram *zram, struct bio_vec *bvec,
 973                                 u32 index, int offset, struct bio *bio)
 974 {
 975         int ret;
 976         struct page *page = NULL;
 977         void *src;
 978         struct bio_vec vec;
 979
 980         vec = *bvec;
 981         if (is_partial_io(bvec)) {
 982                 void *dst;
 983                 /*
 984                  * This is a partial IO. We need to read the full page
 985                  * before to write the changes.
 986                  */
 987                 page = alloc_page(GFP_NOIO|__GFP_HIGHMEM);
 988                 if (!page)
 989                         return -ENOMEM;
 990
 991                 ret = __zram_bvec_read(zram, page, index);
 992                 if (ret)
 993                         goto out;
 994
 995                 src = kmap_atomic(bvec->bv_page);
 996                 dst = kmap_atomic(page);
 997                 memcpy(dst + offset, src + bvec->bv_offset, bvec->bv_len);
 998                 kunmap_atomic(dst);
 999                 kunmap_atomic(src);
1000
1001                 vec.bv_page = page;
1002                 vec.bv_len = PAGE_SIZE;
1003                 vec.bv_offset = 0;
1004         }
1005
1006         ret = __zram_bvec_write(zram, &vec, index, bio);
1007 out:
1008         if (is_partial_io(bvec))
1009                 __free_page(page);
1010         return ret;
1011 }
1012
1013 /*
1014  * zram_bio_discard - handler on discard request
1015  * @index: physical block index in PAGE_SIZE units
1016  * @offset: byte offset within physical block
1017  */
1018 static void zram_bio_discard(struct zram *zram, u32 index,
1019                              int offset, struct bio *bio)
1020 {
1021         size_t n = bio->bi_iter.bi_size;
1022
1023         /*
1024          * zram manages data in physical block size units. Because logical block
1025          * size isn't identical with physical block size on some arch, we
1026          * could get a discard request pointing to a specific offset within a
1027          * certain physical block.  Although we can handle this request by
1028          * reading that physiclal block and decompressing and partially zeroing
1029          * and re-compressing and then re-storing it, this isn't reasonable
1030          * because our intent with a discard request is to save memory.  So
1031          * skipping this logical block is appropriate here.
1032          */
1033         if (offset) {
1034                 if (n <= (PAGE_SIZE - offset))
1035                         return;
1036
1037                 n -= (PAGE_SIZE - offset);
1038                 index++;
1039         }
1040
1041         while (n >= PAGE_SIZE) {
1042                 zram_slot_lock(zram, index);
1043                 zram_free_page(zram, index);
1044                 zram_slot_unlock(zram, index);
1045                 atomic64_inc(&zram->stats.notify_free);
1046                 index++;
1047                 n -= PAGE_SIZE;
1048         }
1049 }
1050
1051 /*
1052  * Returns errno if it has some problem. Otherwise return 0 or 1.
1053  * Returns 0 if IO request was done synchronously
1054  * Returns 1 if IO request was successfully submitted.
1055  */
1056 static int zram_bvec_rw(struct zram *zram, struct bio_vec *bvec, u32 index,
1057                         int offset, bool is_write, struct bio *bio)
1058 {
1059         unsigned long start_time = jiffies;
1060         int rw_acct = is_write ? REQ_OP_WRITE : REQ_OP_READ;
1061         int ret;
1062
1063         generic_start_io_acct(rw_acct, bvec->bv_len >> SECTOR_SHIFT,
1064                         &zram->disk->part0);
1065
1066         if (!is_write) {
1067                 atomic64_inc(&zram->stats.num_reads);
1068                 ret = zram_bvec_read(zram, bvec, index, offset);
1069                 flush_dcache_page(bvec->bv_page);
1070         } else {
1071                 atomic64_inc(&zram->stats.num_writes);
1072                 ret = zram_bvec_write(zram, bvec, index, offset, bio);
1073         }
1074
1075         generic_end_io_acct(rw_acct, &zram->disk->part0, start_time);
1076
1077         if (unlikely(ret < 0)) {
1078                 if (!is_write)
1079                         atomic64_inc(&zram->stats.failed_reads);
1080                 else
1081                         atomic64_inc(&zram->stats.failed_writes);
1082         }
1083
1084         return ret;
1085 }
1086
1087 static void __zram_make_request(struct zram *zram, struct bio *bio)
1088 {
1089         int offset;
1090         u32 index;
1091         struct bio_vec bvec;
1092         struct bvec_iter iter;
1093
1094         index = bio->bi_iter.bi_sector >> SECTORS_PER_PAGE_SHIFT;
1095         offset = (bio->bi_iter.bi_sector &
1096                   (SECTORS_PER_PAGE - 1)) << SECTOR_SHIFT;
1097
1098         switch (bio_op(bio)) {
1099         case REQ_OP_DISCARD:
1100         case REQ_OP_WRITE_ZEROES:
1101                 zram_bio_discard(zram, index, offset, bio);
1102                 bio_endio(bio);
1103                 return;
1104         default:
1105                 break;
1106         }
1107
1108         bio_for_each_segment(bvec, bio, iter) {
1109                 struct bio_vec bv = bvec;
1110                 unsigned int unwritten = bvec.bv_len;
1111
1112                 do {
1113                         bv.bv_len = min_t(unsigned int, PAGE_SIZE - offset,
1114                                                         unwritten);
1115                         if (zram_bvec_rw(zram, &bv, index, offset,
1116                                         op_is_write(bio_op(bio)), bio) < 0)
1117                                 goto out;
1118
1119                         bv.bv_offset += bv.bv_len;
1120                         unwritten -= bv.bv_len;
1121
1122                         update_position(&index, &offset, &bv);
1123                 } while (unwritten);
1124         }
1125
1126         bio_endio(bio);
1127         return;
1128
1129 out:
1130         bio_io_error(bio);
1131 }
1132
1133 /*
1134  * Handler function for all zram I/O requests.
1135  */
1136 static blk_qc_t zram_make_request(struct request_queue *queue, struct bio *bio)
1137 {
1138         struct zram *zram = queue->queuedata;
1139
1140         if (!valid_io_request(zram, bio->bi_iter.bi_sector,
1141                                         bio->bi_iter.bi_size)) {
1142                 atomic64_inc(&zram->stats.invalid_io);
1143                 goto error;
1144         }
1145
1146         __zram_make_request(zram, bio);
1147         return BLK_QC_T_NONE;
1148
1149 error:
1150         bio_io_error(bio);
1151         return BLK_QC_T_NONE;
1152 }
1153
1154 static void zram_slot_free_notify(struct block_device *bdev,
1155                                 unsigned long index)
1156 {
1157         struct zram *zram;
1158
1159         zram = bdev->bd_disk->private_data;
1160
1161         zram_slot_lock(zram, index);
1162         zram_free_page(zram, index);
1163         zram_slot_unlock(zram, index);
1164         atomic64_inc(&zram->stats.notify_free);
1165 }
1166
1167 static int zram_rw_page(struct block_device *bdev, sector_t sector,
1168                        struct page *page, bool is_write)
1169 {
1170         int offset, ret;
1171         u32 index;
1172         struct zram *zram;
1173         struct bio_vec bv;
1174
1175         zram = bdev->bd_disk->private_data;
1176
1177         if (!valid_io_request(zram, sector, PAGE_SIZE)) {
1178                 atomic64_inc(&zram->stats.invalid_io);
1179                 ret = -EINVAL;
1180                 goto out;
1181         }
1182
1183         index = sector >> SECTORS_PER_PAGE_SHIFT;
1184         offset = (sector & (SECTORS_PER_PAGE - 1)) << SECTOR_SHIFT;
1185
1186         bv.bv_page = page;
1187         bv.bv_len = PAGE_SIZE;
1188         bv.bv_offset = 0;
1189
1190         ret = zram_bvec_rw(zram, &bv, index, offset, is_write, NULL);
1191 out:
1192         /*
1193          * If I/O fails, just return error(ie, non-zero) without
1194          * calling page_endio.
1195          * It causes resubmit the I/O with bio request by upper functions
1196          * of rw_page(e.g., swap_readpage, __swap_writepage) and
1197          * bio->bi_end_io does things to handle the error
1198          * (e.g., SetPageError, set_page_dirty and extra works).
1199          */
1200         if (unlikely(ret < 0))
1201                 return ret;
1202
1203         switch (ret) {
1204         case 0:
1205                 page_endio(page, is_write, 0);
1206                 break;
1207         case 1:
1208                 ret = 0;
1209                 break;
1210         default:
1211                 WARN_ON(1);
1212         }
1213         return ret;
1214 }
1215
1216 static void zram_reset_device(struct zram *zram)
1217 {
1218         struct zcomp *comp;
1219         u64 disksize;
1220
1221         down_write(&zram->init_lock);
1222
1223         zram->limit_pages = 0;
1224
1225         if (!init_done(zram)) {
1226                 up_write(&zram->init_lock);
1227                 return;
1228         }
1229
1230         comp = zram->comp;
1231         disksize = zram->disksize;
1232         zram->disksize = 0;
1233
1234         set_capacity(zram->disk, 0);
1235         part_stat_set_all(&zram->disk->part0, 0);
1236
1237         up_write(&zram->init_lock);
1238         /* I/O operation under all of CPU are done so let's free */
1239         zram_meta_free(zram, disksize);
1240         memset(&zram->stats, 0, sizeof(zram->stats));
1241         zcomp_destroy(comp);
1242         reset_bdev(zram);
1243 }
1244
1245 static ssize_t disksize_store(struct device *dev,
1246                 struct device_attribute *attr, const char *buf, size_t len)
1247 {
1248         u64 disksize;
1249         struct zcomp *comp;
1250         struct zram *zram = dev_to_zram(dev);
1251         int err;
1252
1253         disksize = memparse(buf, NULL);
1254         if (!disksize)
1255                 return -EINVAL;
1256
1257         down_write(&zram->init_lock);
1258         if (init_done(zram)) {
1259                 pr_info("Cannot change disksize for initialized device\n");
1260                 err = -EBUSY;
1261                 goto out_unlock;
1262         }
1263
1264         disksize = PAGE_ALIGN(disksize);
1265         if (!zram_meta_alloc(zram, disksize)) {
1266                 err = -ENOMEM;
1267                 goto out_unlock;
1268         }
1269
1270         comp = zcomp_create(zram->compressor);
1271         if (IS_ERR(comp)) {
1272                 pr_err("Cannot initialise %s compressing backend\n",
1273                                 zram->compressor);
1274                 err = PTR_ERR(comp);
1275                 goto out_free_meta;
1276         }
1277
1278         zram->comp = comp;
1279         zram->disksize = disksize;
1280         set_capacity(zram->disk, zram->disksize >> SECTOR_SHIFT);
1281         zram_revalidate_disk(zram);
1282         up_write(&zram->init_lock);
1283
1284         return len;
1285
1286 out_free_meta:
1287         zram_meta_free(zram, disksize);
1288 out_unlock:
1289         up_write(&zram->init_lock);
1290         return err;
1291 }
1292
1293 static ssize_t reset_store(struct device *dev,
1294                 struct device_attribute *attr, const char *buf, size_t len)
1295 {
1296         int ret;
1297         unsigned short do_reset;
1298         struct zram *zram;
1299         struct block_device *bdev;
1300
1301         ret = kstrtou16(buf, 10, &do_reset);
1302         if (ret)
1303                 return ret;
1304
1305         if (!do_reset)
1306                 return -EINVAL;
1307
1308         zram = dev_to_zram(dev);
1309         bdev = bdget_disk(zram->disk, 0);
1310         if (!bdev)
1311                 return -ENOMEM;
1312
1313         mutex_lock(&bdev->bd_mutex);
1314         /* Do not reset an active device or claimed device */
1315         if (bdev->bd_openers || zram->claim) {
1316                 mutex_unlock(&bdev->bd_mutex);
1317                 bdput(bdev);
1318                 return -EBUSY;
1319         }
1320
1321         /* From now on, anyone can't open /dev/zram[0-9] */
1322         zram->claim = true;
1323         mutex_unlock(&bdev->bd_mutex);
1324
1325         /* Make sure all the pending I/O are finished */
1326         fsync_bdev(bdev);
1327         zram_reset_device(zram);
1328         zram_revalidate_disk(zram);
1329         bdput(bdev);
1330
1331         mutex_lock(&bdev->bd_mutex);
1332         zram->claim = false;
1333         mutex_unlock(&bdev->bd_mutex);
1334
1335         return len;
1336 }
1337
1338 static int zram_open(struct block_device *bdev, fmode_t mode)
1339 {
1340         int ret = 0;
1341         struct zram *zram;
1342
1343         WARN_ON(!mutex_is_locked(&bdev->bd_mutex));
1344
1345         zram = bdev->bd_disk->private_data;
1346         /* zram was claimed to reset so open request fails */
1347         if (zram->claim)
1348                 ret = -EBUSY;
1349
1350         return ret;
1351 }
1352
1353 static const struct block_device_operations zram_devops = {
1354         .open = zram_open,
1355         .swap_slot_free_notify = zram_slot_free_notify,
1356         .rw_page = zram_rw_page,
1357         .owner = THIS_MODULE
1358 };
1359
1360 static DEVICE_ATTR_WO(compact);
1361 static DEVICE_ATTR_RW(disksize);
1362 static DEVICE_ATTR_RO(initstate);
1363 static DEVICE_ATTR_WO(reset);
1364 static DEVICE_ATTR_WO(mem_limit);
1365 static DEVICE_ATTR_WO(mem_used_max);
1366 static DEVICE_ATTR_RW(max_comp_streams);
1367 static DEVICE_ATTR_RW(comp_algorithm);
1368 #ifdef CONFIG_ZRAM_WRITEBACK
1369 static DEVICE_ATTR_RW(backing_dev);
1370 #endif
1371
1372 static struct attribute *zram_disk_attrs[] = {
1373         &dev_attr_disksize.attr,
1374         &dev_attr_initstate.attr,
1375         &dev_attr_reset.attr,
1376         &dev_attr_compact.attr,
1377         &dev_attr_mem_limit.attr,
1378         &dev_attr_mem_used_max.attr,
1379         &dev_attr_max_comp_streams.attr,
1380         &dev_attr_comp_algorithm.attr,
1381 #ifdef CONFIG_ZRAM_WRITEBACK
1382         &dev_attr_backing_dev.attr,
1383 #endif
1384         &dev_attr_io_stat.attr,
1385         &dev_attr_mm_stat.attr,
1386         &dev_attr_debug_stat.attr,
1387         NULL,
1388 };
1389
1390 static const struct attribute_group zram_disk_attr_group = {
1391         .attrs = zram_disk_attrs,
1392 };
1393
1394 /*
1395  * Allocate and initialize new zram device. the function returns
1396  * '>= 0' device_id upon success, and negative value otherwise.
1397  */
1398 static int zram_add(void)
1399 {
1400         struct zram *zram;
1401         struct request_queue *queue;
1402         int ret, device_id;
1403
1404         zram = kzalloc(sizeof(struct zram), GFP_KERNEL);
1405         if (!zram)
1406                 return -ENOMEM;
1407
1408         ret = idr_alloc(&zram_index_idr, zram, 0, 0, GFP_KERNEL);
1409         if (ret < 0)
1410                 goto out_free_dev;
1411         device_id = ret;
1412
1413         init_rwsem(&zram->init_lock);
1414
1415         queue = blk_alloc_queue(GFP_KERNEL);
1416         if (!queue) {
1417                 pr_err("Error allocating disk queue for device %d\n",
1418                         device_id);
1419                 ret = -ENOMEM;
1420                 goto out_free_idr;
1421         }
1422
1423         blk_queue_make_request(queue, zram_make_request);
1424
1425         /* gendisk structure */
1426         zram->disk = alloc_disk(1);
1427         if (!zram->disk) {
1428                 pr_err("Error allocating disk structure for device %d\n",
1429                         device_id);
1430                 ret = -ENOMEM;
1431                 goto out_free_queue;
1432         }
1433
1434         zram->disk->major = zram_major;
1435         zram->disk->first_minor = device_id;
1436         zram->disk->fops = &zram_devops;
1437         zram->disk->queue = queue;
1438         zram->disk->queue->queuedata = zram;
1439         zram->disk->private_data = zram;
1440         snprintf(zram->disk->disk_name, 16, "zram%d", device_id);
1441
1442         /* Actual capacity set using syfs (/sys/block/zram<id>/disksize */
1443         set_capacity(zram->disk, 0);
1444         /* zram devices sort of resembles non-rotational disks */
1445         queue_flag_set_unlocked(QUEUE_FLAG_NONROT, zram->disk->queue);
1446         queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, zram->disk->queue);
1447         /*
1448          * To ensure that we always get PAGE_SIZE aligned
1449          * and n*PAGE_SIZED sized I/O requests.
1450          */
1451         blk_queue_physical_block_size(zram->disk->queue, PAGE_SIZE);
1452         blk_queue_logical_block_size(zram->disk->queue,
1453                                         ZRAM_LOGICAL_BLOCK_SIZE);
1454         blk_queue_io_min(zram->disk->queue, PAGE_SIZE);
1455         blk_queue_io_opt(zram->disk->queue, PAGE_SIZE);
1456         zram->disk->queue->limits.discard_granularity = PAGE_SIZE;
1457         blk_queue_max_discard_sectors(zram->disk->queue, UINT_MAX);
1458         queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, zram->disk->queue);
1459
1460         /*
1461          * zram_bio_discard() will clear all logical blocks if logical block
1462          * size is identical with physical block size(PAGE_SIZE). But if it is
1463          * different, we will skip discarding some parts of logical blocks in
1464          * the part of the request range which isn't aligned to physical block
1465          * size.  So we can't ensure that all discarded logical blocks are
1466          * zeroed.
1467          */
1468         if (ZRAM_LOGICAL_BLOCK_SIZE == PAGE_SIZE)
1469                 blk_queue_max_write_zeroes_sectors(zram->disk->queue, UINT_MAX);
1470
1471         add_disk(zram->disk);
1472
1473         ret = sysfs_create_group(&disk_to_dev(zram->disk)->kobj,
1474                                 &zram_disk_attr_group);
1475         if (ret < 0) {
1476                 pr_err("Error creating sysfs group for device %d\n",
1477                                 device_id);
1478                 goto out_free_disk;
1479         }
1480         strlcpy(zram->compressor, default_compressor, sizeof(zram->compressor));
1481
1482         pr_info("Added device: %s\n", zram->disk->disk_name);
1483         return device_id;
1484
1485 out_free_disk:
1486         del_gendisk(zram->disk);
1487         put_disk(zram->disk);
1488 out_free_queue:
1489         blk_cleanup_queue(queue);
1490 out_free_idr:
1491         idr_remove(&zram_index_idr, device_id);
1492 out_free_dev:
1493         kfree(zram);
1494         return ret;
1495 }
1496
1497 static int zram_remove(struct zram *zram)
1498 {
1499         struct block_device *bdev;
1500
1501         bdev = bdget_disk(zram->disk, 0);
1502         if (!bdev)
1503                 return -ENOMEM;
1504
1505         mutex_lock(&bdev->bd_mutex);
1506         if (bdev->bd_openers || zram->claim) {
1507                 mutex_unlock(&bdev->bd_mutex);
1508                 bdput(bdev);
1509                 return -EBUSY;
1510         }
1511
1512         zram->claim = true;
1513         mutex_unlock(&bdev->bd_mutex);
1514
1515         /*
1516          * Remove sysfs first, so no one will perform a disksize
1517          * store while we destroy the devices. This also helps during
1518          * hot_remove -- zram_reset_device() is the last holder of
1519          * ->init_lock, no later/concurrent disksize_store() or any
1520          * other sysfs handlers are possible.
1521          */
1522         sysfs_remove_group(&disk_to_dev(zram->disk)->kobj,
1523                         &zram_disk_attr_group);
1524
1525         /* Make sure all the pending I/O are finished */
1526         fsync_bdev(bdev);
1527         zram_reset_device(zram);
1528         bdput(bdev);
1529
1530         pr_info("Removed device: %s\n", zram->disk->disk_name);
1531
1532         blk_cleanup_queue(zram->disk->queue);
1533         del_gendisk(zram->disk);
1534         put_disk(zram->disk);
1535         kfree(zram);
1536         return 0;
1537 }
1538
1539 /* zram-control sysfs attributes */
1540
1541 /*
1542  * NOTE: hot_add attribute is not the usual read-only sysfs attribute. In a
1543  * sense that reading from this file does alter the state of your system -- it
1544  * creates a new un-initialized zram device and returns back this device's
1545  * device_id (or an error code if it fails to create a new device).
1546  */
1547 static ssize_t hot_add_show(struct class *class,
1548                         struct class_attribute *attr,
1549                         char *buf)
1550 {
1551         int ret;
1552
1553         mutex_lock(&zram_index_mutex);
1554         ret = zram_add();
1555         mutex_unlock(&zram_index_mutex);
1556
1557         if (ret < 0)
1558                 return ret;
1559         return scnprintf(buf, PAGE_SIZE, "%d\n", ret);
1560 }
1561 static CLASS_ATTR_RO(hot_add);
1562
1563 static ssize_t hot_remove_store(struct class *class,
1564                         struct class_attribute *attr,
1565                         const char *buf,
1566                         size_t count)
1567 {
1568         struct zram *zram;
1569         int ret, dev_id;
1570
1571         /* dev_id is gendisk->first_minor, which is `int' */
1572         ret = kstrtoint(buf, 10, &dev_id);
1573         if (ret)
1574                 return ret;
1575         if (dev_id < 0)
1576                 return -EINVAL;
1577
1578         mutex_lock(&zram_index_mutex);
1579
1580         zram = idr_find(&zram_index_idr, dev_id);
1581         if (zram) {
1582                 ret = zram_remove(zram);
1583                 if (!ret)
1584                         idr_remove(&zram_index_idr, dev_id);
1585         } else {
1586                 ret = -ENODEV;
1587         }
1588
1589         mutex_unlock(&zram_index_mutex);
1590         return ret ? ret : count;
1591 }
1592 static CLASS_ATTR_WO(hot_remove);
1593
1594 static struct attribute *zram_control_class_attrs[] = {
1595         &class_attr_hot_add.attr,
1596         &class_attr_hot_remove.attr,
1597         NULL,
1598 };
1599 ATTRIBUTE_GROUPS(zram_control_class);
1600
1601 static struct class zram_control_class = {
1602         .name           = "zram-control",
1603         .owner          = THIS_MODULE,
1604         .class_groups   = zram_control_class_groups,
1605 };
1606
1607 static int zram_remove_cb(int id, void *ptr, void *data)
1608 {
1609         zram_remove(ptr);
1610         return 0;
1611 }
1612
1613 static void destroy_devices(void)
1614 {
1615         class_unregister(&zram_control_class);
1616         idr_for_each(&zram_index_idr, &zram_remove_cb, NULL);
1617         idr_destroy(&zram_index_idr);
1618         unregister_blkdev(zram_major, "zram");
1619         cpuhp_remove_multi_state(CPUHP_ZCOMP_PREPARE);
1620 }
1621
1622 static int __init zram_init(void)
1623 {
1624         int ret;
1625
1626         ret = cpuhp_setup_state_multi(CPUHP_ZCOMP_PREPARE, "block/zram:prepare",
1627                                       zcomp_cpu_up_prepare, zcomp_cpu_dead);
1628         if (ret < 0)
1629                 return ret;
1630
1631         ret = class_register(&zram_control_class);
1632         if (ret) {
1633                 pr_err("Unable to register zram-control class\n");
1634                 cpuhp_remove_multi_state(CPUHP_ZCOMP_PREPARE);
1635                 return ret;
1636         }
1637
1638         zram_major = register_blkdev(0, "zram");
1639         if (zram_major <= 0) {
1640                 pr_err("Unable to get major number\n");
1641                 class_unregister(&zram_control_class);
1642                 cpuhp_remove_multi_state(CPUHP_ZCOMP_PREPARE);
1643                 return -EBUSY;
1644         }
1645
1646         while (num_devices != 0) {
1647                 mutex_lock(&zram_index_mutex);
1648                 ret = zram_add();
1649                 mutex_unlock(&zram_index_mutex);
1650                 if (ret < 0)
1651                         goto out_error;
1652                 num_devices--;
1653         }
1654
1655         return 0;
1656
1657 out_error:
1658         destroy_devices();
1659         return ret;
1660 }
1661
1662 static void __exit zram_exit(void)
1663 {
1664         destroy_devices();
1665 }
1666
1667 module_init(zram_init);
1668 module_exit(zram_exit);
1669
1670 module_param(num_devices, uint, 0);
1671 MODULE_PARM_DESC(num_devices, "Number of pre-created zram devices");
1672
1673 MODULE_LICENSE("Dual BSD/GPL");
1674 MODULE_AUTHOR("Nitin Gupta <ngupta@vflare.org>");
1675 MODULE_DESCRIPTION("Compressed RAM Block Device");