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/highmem.h>
  26 #include <linux/slab.h>
  27 #include <linux/backing-dev.h>
  28 #include <linux/string.h>
  29 #include <linux/vmalloc.h>
  30 #include <linux/err.h>
  31 #include <linux/idr.h>
  32 #include <linux/sysfs.h>
  33 #include <linux/debugfs.h>
  34 #include <linux/cpuhotplug.h>
  35 #include <linux/part_stat.h>
  36
  37 #include "zram_drv.h"
  38
  39 static DEFINE_IDR(zram_index_idr);
  40 /* idr index must be protected */
  41 static DEFINE_MUTEX(zram_index_mutex);
  42
  43 static int zram_major;
  44 static const char *default_compressor = CONFIG_ZRAM_DEF_COMP;
  45
  46 /* Module params (documentation at end) */
  47 static unsigned int num_devices = 1;
  48 /*
  49  * Pages that compress to sizes equals or greater than this are stored
  50  * uncompressed in memory.
  51  */
  52 static size_t huge_class_size;
  53
  54 static const struct block_device_operations zram_devops;
  55 static const struct block_device_operations zram_wb_devops;
  56
  57 static void zram_free_page(struct zram *zram, size_t index);
  58 static int zram_bvec_read(struct zram *zram, struct bio_vec *bvec,
  59                                 u32 index, int offset, struct bio *bio);
  60
  61
  62 static int zram_slot_trylock(struct zram *zram, u32 index)
  63 {
  64         return bit_spin_trylock(ZRAM_LOCK, &zram->table[index].flags);
  65 }
  66
  67 static void zram_slot_lock(struct zram *zram, u32 index)
  68 {
  69         bit_spin_lock(ZRAM_LOCK, &zram->table[index].flags);
  70 }
  71
  72 static void zram_slot_unlock(struct zram *zram, u32 index)
  73 {
  74         bit_spin_unlock(ZRAM_LOCK, &zram->table[index].flags);
  75 }
  76
  77 static inline bool init_done(struct zram *zram)
  78 {
  79         return zram->disksize;
  80 }
  81
  82 static inline struct zram *dev_to_zram(struct device *dev)
  83 {
  84         return (struct zram *)dev_to_disk(dev)->private_data;
  85 }
  86
  87 static unsigned long zram_get_handle(struct zram *zram, u32 index)
  88 {
  89         return zram->table[index].handle;
  90 }
  91
  92 static void zram_set_handle(struct zram *zram, u32 index, unsigned long handle)
  93 {
  94         zram->table[index].handle = handle;
  95 }
  96
  97 /* flag operations require table entry bit_spin_lock() being held */
  98 static bool zram_test_flag(struct zram *zram, u32 index,
  99                         enum zram_pageflags flag)
 100 {
 101         return zram->table[index].flags & BIT(flag);
 102 }
 103
 104 static void zram_set_flag(struct zram *zram, u32 index,
 105                         enum zram_pageflags flag)
 106 {
 107         zram->table[index].flags |= BIT(flag);
 108 }
 109
 110 static void zram_clear_flag(struct zram *zram, u32 index,
 111                         enum zram_pageflags flag)
 112 {
 113         zram->table[index].flags &= ~BIT(flag);
 114 }
 115
 116 static inline void zram_set_element(struct zram *zram, u32 index,
 117                         unsigned long element)
 118 {
 119         zram->table[index].element = element;
 120 }
 121
 122 static unsigned long zram_get_element(struct zram *zram, u32 index)
 123 {
 124         return zram->table[index].element;
 125 }
 126
 127 static size_t zram_get_obj_size(struct zram *zram, u32 index)
 128 {
 129         return zram->table[index].flags & (BIT(ZRAM_FLAG_SHIFT) - 1);
 130 }
 131
 132 static void zram_set_obj_size(struct zram *zram,
 133                                         u32 index, size_t size)
 134 {
 135         unsigned long flags = zram->table[index].flags >> ZRAM_FLAG_SHIFT;
 136
 137         zram->table[index].flags = (flags << ZRAM_FLAG_SHIFT) | size;
 138 }
 139
 140 static inline bool zram_allocated(struct zram *zram, u32 index)
 141 {
 142         return zram_get_obj_size(zram, index) ||
 143                         zram_test_flag(zram, index, ZRAM_SAME) ||
 144                         zram_test_flag(zram, index, ZRAM_WB);
 145 }
 146
 147 #if PAGE_SIZE != 4096
 148 static inline bool is_partial_io(struct bio_vec *bvec)
 149 {
 150         return bvec->bv_len != PAGE_SIZE;
 151 }
 152 #else
 153 static inline bool is_partial_io(struct bio_vec *bvec)
 154 {
 155         return false;
 156 }
 157 #endif
 158
 159 /*
 160  * Check if request is within bounds and aligned on zram logical blocks.
 161  */
 162 static inline bool valid_io_request(struct zram *zram,
 163                 sector_t start, unsigned int size)
 164 {
 165         u64 end, bound;
 166
 167         /* unaligned request */
 168         if (unlikely(start & (ZRAM_SECTOR_PER_LOGICAL_BLOCK - 1)))
 169                 return false;
 170         if (unlikely(size & (ZRAM_LOGICAL_BLOCK_SIZE - 1)))
 171                 return false;
 172
 173         end = start + (size >> SECTOR_SHIFT);
 174         bound = zram->disksize >> SECTOR_SHIFT;
 175         /* out of range range */
 176         if (unlikely(start >= bound || end > bound || start > end))
 177                 return false;
 178
 179         /* I/O request is valid */
 180         return true;
 181 }
 182
 183 static void update_position(u32 *index, int *offset, struct bio_vec *bvec)
 184 {
 185         *index  += (*offset + bvec->bv_len) / PAGE_SIZE;
 186         *offset = (*offset + bvec->bv_len) % PAGE_SIZE;
 187 }
 188
 189 static inline void update_used_max(struct zram *zram,
 190                                         const unsigned long pages)
 191 {
 192         unsigned long old_max, cur_max;
 193
 194         old_max = atomic_long_read(&zram->stats.max_used_pages);
 195
 196         do {
 197                 cur_max = old_max;
 198                 if (pages > cur_max)
 199                         old_max = atomic_long_cmpxchg(
 200                                 &zram->stats.max_used_pages, cur_max, pages);
 201         } while (old_max != cur_max);
 202 }
 203
 204 static inline void zram_fill_page(void *ptr, unsigned long len,
 205                                         unsigned long value)
 206 {
 207         WARN_ON_ONCE(!IS_ALIGNED(len, sizeof(unsigned long)));
 208         memset_l(ptr, value, len / sizeof(unsigned long));
 209 }
 210
 211 static bool page_same_filled(void *ptr, unsigned long *element)
 212 {
 213         unsigned long *page;
 214         unsigned long val;
 215         unsigned int pos, last_pos = PAGE_SIZE / sizeof(*page) - 1;
 216
 217         page = (unsigned long *)ptr;
 218         val = page[0];
 219
 220         if (val != page[last_pos])
 221                 return false;
 222
 223         for (pos = 1; pos < last_pos; pos++) {
 224                 if (val != page[pos])
 225                         return false;
 226         }
 227
 228         *element = val;
 229
 230         return true;
 231 }
 232
 233 static ssize_t initstate_show(struct device *dev,
 234                 struct device_attribute *attr, char *buf)
 235 {
 236         u32 val;
 237         struct zram *zram = dev_to_zram(dev);
 238
 239         down_read(&zram->init_lock);
 240         val = init_done(zram);
 241         up_read(&zram->init_lock);
 242
 243         return scnprintf(buf, PAGE_SIZE, "%u\n", val);
 244 }
 245
 246 static ssize_t disksize_show(struct device *dev,
 247                 struct device_attribute *attr, char *buf)
 248 {
 249         struct zram *zram = dev_to_zram(dev);
 250
 251         return scnprintf(buf, PAGE_SIZE, "%llu\n", zram->disksize);
 252 }
 253
 254 static ssize_t mem_limit_store(struct device *dev,
 255                 struct device_attribute *attr, const char *buf, size_t len)
 256 {
 257         u64 limit;
 258         char *tmp;
 259         struct zram *zram = dev_to_zram(dev);
 260
 261         limit = memparse(buf, &tmp);
 262         if (buf == tmp) /* no chars parsed, invalid input */
 263                 return -EINVAL;
 264
 265         down_write(&zram->init_lock);
 266         zram->limit_pages = PAGE_ALIGN(limit) >> PAGE_SHIFT;
 267         up_write(&zram->init_lock);
 268
 269         return len;
 270 }
 271
 272 static ssize_t mem_used_max_store(struct device *dev,
 273                 struct device_attribute *attr, const char *buf, size_t len)
 274 {
 275         int err;
 276         unsigned long val;
 277         struct zram *zram = dev_to_zram(dev);
 278
 279         err = kstrtoul(buf, 10, &val);
 280         if (err || val != 0)
 281                 return -EINVAL;
 282
 283         down_read(&zram->init_lock);
 284         if (init_done(zram)) {
 285                 atomic_long_set(&zram->stats.max_used_pages,
 286                                 zs_get_total_pages(zram->mem_pool));
 287         }
 288         up_read(&zram->init_lock);
 289
 290         return len;
 291 }
 292
 293 /*
 294  * Mark all pages which are older than or equal to cutoff as IDLE.
 295  * Callers should hold the zram init lock in read mode
 296  */
 297 static void mark_idle(struct zram *zram, ktime_t cutoff)
 298 {
 299         int is_idle = 1;
 300         unsigned long nr_pages = zram->disksize >> PAGE_SHIFT;
 301         int index;
 302
 303         for (index = 0; index < nr_pages; index++) {
 304                 /*
 305                  * Do not mark ZRAM_UNDER_WB slot as ZRAM_IDLE to close race.
 306                  * See the comment in writeback_store.
 307                  */
 308                 zram_slot_lock(zram, index);
 309                 if (zram_allocated(zram, index) &&
 310                                 !zram_test_flag(zram, index, ZRAM_UNDER_WB)) {
 311 #ifdef CONFIG_ZRAM_MEMORY_TRACKING
 312                         is_idle = !cutoff || ktime_after(cutoff, zram->table[index].ac_time);
 313 #endif
 314                         if (is_idle)
 315                                 zram_set_flag(zram, index, ZRAM_IDLE);
 316                 }
 317                 zram_slot_unlock(zram, index);
 318         }
 319 }
 320
 321 static ssize_t idle_store(struct device *dev,
 322                 struct device_attribute *attr, const char *buf, size_t len)
 323 {
 324         struct zram *zram = dev_to_zram(dev);
 325         ktime_t cutoff_time = 0;
 326         ssize_t rv = -EINVAL;
 327
 328         if (!sysfs_streq(buf, "all")) {
 329                 /*
 330                  * If it did not parse as 'all' try to treat it as an integer when
 331                  * we have memory tracking enabled.
 332                  */
 333                 u64 age_sec;
 334
 335                 if (IS_ENABLED(CONFIG_ZRAM_MEMORY_TRACKING) && !kstrtoull(buf, 0, &age_sec))
 336                         cutoff_time = ktime_sub(ktime_get_boottime(),
 337                                         ns_to_ktime(age_sec * NSEC_PER_SEC));
 338                 else
 339                         goto out;
 340         }
 341
 342         down_read(&zram->init_lock);
 343         if (!init_done(zram))
 344                 goto out_unlock;
 345
 346         /* A cutoff_time of 0 marks everything as idle, this is the "all" behavior */
 347         mark_idle(zram, cutoff_time);
 348         rv = len;
 349
 350 out_unlock:
 351         up_read(&zram->init_lock);
 352 out:
 353         return rv;
 354 }
 355
 356 #ifdef CONFIG_ZRAM_WRITEBACK
 357 static ssize_t writeback_limit_enable_store(struct device *dev,
 358                 struct device_attribute *attr, const char *buf, size_t len)
 359 {
 360         struct zram *zram = dev_to_zram(dev);
 361         u64 val;
 362         ssize_t ret = -EINVAL;
 363
 364         if (kstrtoull(buf, 10, &val))
 365                 return ret;
 366
 367         down_read(&zram->init_lock);
 368         spin_lock(&zram->wb_limit_lock);
 369         zram->wb_limit_enable = val;
 370         spin_unlock(&zram->wb_limit_lock);
 371         up_read(&zram->init_lock);
 372         ret = len;
 373
 374         return ret;
 375 }
 376
 377 static ssize_t writeback_limit_enable_show(struct device *dev,
 378                 struct device_attribute *attr, char *buf)
 379 {
 380         bool val;
 381         struct zram *zram = dev_to_zram(dev);
 382
 383         down_read(&zram->init_lock);
 384         spin_lock(&zram->wb_limit_lock);
 385         val = zram->wb_limit_enable;
 386         spin_unlock(&zram->wb_limit_lock);
 387         up_read(&zram->init_lock);
 388
 389         return scnprintf(buf, PAGE_SIZE, "%d\n", val);
 390 }
 391
 392 static ssize_t writeback_limit_store(struct device *dev,
 393                 struct device_attribute *attr, const char *buf, size_t len)
 394 {
 395         struct zram *zram = dev_to_zram(dev);
 396         u64 val;
 397         ssize_t ret = -EINVAL;
 398
 399         if (kstrtoull(buf, 10, &val))
 400                 return ret;
 401
 402         down_read(&zram->init_lock);
 403         spin_lock(&zram->wb_limit_lock);
 404         zram->bd_wb_limit = val;
 405         spin_unlock(&zram->wb_limit_lock);
 406         up_read(&zram->init_lock);
 407         ret = len;
 408
 409         return ret;
 410 }
 411
 412 static ssize_t writeback_limit_show(struct device *dev,
 413                 struct device_attribute *attr, char *buf)
 414 {
 415         u64 val;
 416         struct zram *zram = dev_to_zram(dev);
 417
 418         down_read(&zram->init_lock);
 419         spin_lock(&zram->wb_limit_lock);
 420         val = zram->bd_wb_limit;
 421         spin_unlock(&zram->wb_limit_lock);
 422         up_read(&zram->init_lock);
 423
 424         return scnprintf(buf, PAGE_SIZE, "%llu\n", val);
 425 }
 426
 427 static void reset_bdev(struct zram *zram)
 428 {
 429         struct block_device *bdev;
 430
 431         if (!zram->backing_dev)
 432                 return;
 433
 434         bdev = zram->bdev;
 435         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 436         /* hope filp_close flush all of IO */
 437         filp_close(zram->backing_dev, NULL);
 438         zram->backing_dev = NULL;
 439         zram->bdev = NULL;
 440         zram->disk->fops = &zram_devops;
 441         kvfree(zram->bitmap);
 442         zram->bitmap = NULL;
 443 }
 444
 445 static ssize_t backing_dev_show(struct device *dev,
 446                 struct device_attribute *attr, char *buf)
 447 {
 448         struct file *file;
 449         struct zram *zram = dev_to_zram(dev);
 450         char *p;
 451         ssize_t ret;
 452
 453         down_read(&zram->init_lock);
 454         file = zram->backing_dev;
 455         if (!file) {
 456                 memcpy(buf, "none\n", 5);
 457                 up_read(&zram->init_lock);
 458                 return 5;
 459         }
 460
 461         p = file_path(file, buf, PAGE_SIZE - 1);
 462         if (IS_ERR(p)) {
 463                 ret = PTR_ERR(p);
 464                 goto out;
 465         }
 466
 467         ret = strlen(p);
 468         memmove(buf, p, ret);
 469         buf[ret++] = '\n';
 470 out:
 471         up_read(&zram->init_lock);
 472         return ret;
 473 }
 474
 475 static ssize_t backing_dev_store(struct device *dev,
 476                 struct device_attribute *attr, const char *buf, size_t len)
 477 {
 478         char *file_name;
 479         size_t sz;
 480         struct file *backing_dev = NULL;
 481         struct inode *inode;
 482         struct address_space *mapping;
 483         unsigned int bitmap_sz;
 484         unsigned long nr_pages, *bitmap = NULL;
 485         struct block_device *bdev = NULL;
 486         int err;
 487         struct zram *zram = dev_to_zram(dev);
 488
 489         file_name = kmalloc(PATH_MAX, GFP_KERNEL);
 490         if (!file_name)
 491                 return -ENOMEM;
 492
 493         down_write(&zram->init_lock);
 494         if (init_done(zram)) {
 495                 pr_info("Can't setup backing device for initialized device\n");
 496                 err = -EBUSY;
 497                 goto out;
 498         }
 499
 500         strlcpy(file_name, buf, PATH_MAX);
 501         /* ignore trailing newline */
 502         sz = strlen(file_name);
 503         if (sz > 0 && file_name[sz - 1] == '\n')
 504                 file_name[sz - 1] = 0x00;
 505
 506         backing_dev = filp_open(file_name, O_RDWR|O_LARGEFILE, 0);
 507         if (IS_ERR(backing_dev)) {
 508                 err = PTR_ERR(backing_dev);
 509                 backing_dev = NULL;
 510                 goto out;
 511         }
 512
 513         mapping = backing_dev->f_mapping;
 514         inode = mapping->host;
 515
 516         /* Support only block device in this moment */
 517         if (!S_ISBLK(inode->i_mode)) {
 518                 err = -ENOTBLK;
 519                 goto out;
 520         }
 521
 522         bdev = blkdev_get_by_dev(inode->i_rdev,
 523                         FMODE_READ | FMODE_WRITE | FMODE_EXCL, zram);
 524         if (IS_ERR(bdev)) {
 525                 err = PTR_ERR(bdev);
 526                 bdev = NULL;
 527                 goto out;
 528         }
 529
 530         nr_pages = i_size_read(inode) >> PAGE_SHIFT;
 531         bitmap_sz = BITS_TO_LONGS(nr_pages) * sizeof(long);
 532         bitmap = kvzalloc(bitmap_sz, GFP_KERNEL);
 533         if (!bitmap) {
 534                 err = -ENOMEM;
 535                 goto out;
 536         }
 537
 538         reset_bdev(zram);
 539
 540         zram->bdev = bdev;
 541         zram->backing_dev = backing_dev;
 542         zram->bitmap = bitmap;
 543         zram->nr_pages = nr_pages;
 544         /*
 545          * With writeback feature, zram does asynchronous IO so it's no longer
 546          * synchronous device so let's remove synchronous io flag. Othewise,
 547          * upper layer(e.g., swap) could wait IO completion rather than
 548          * (submit and return), which will cause system sluggish.
 549          * Furthermore, when the IO function returns(e.g., swap_readpage),
 550          * upper layer expects IO was done so it could deallocate the page
 551          * freely but in fact, IO is going on so finally could cause
 552          * use-after-free when the IO is really done.
 553          */
 554         zram->disk->fops = &zram_wb_devops;
 555         up_write(&zram->init_lock);
 556
 557         pr_info("setup backing device %s\n", file_name);
 558         kfree(file_name);
 559
 560         return len;
 561 out:
 562         kvfree(bitmap);
 563
 564         if (bdev)
 565                 blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
 566
 567         if (backing_dev)
 568                 filp_close(backing_dev, NULL);
 569
 570         up_write(&zram->init_lock);
 571
 572         kfree(file_name);
 573
 574         return err;
 575 }
 576
 577 static unsigned long alloc_block_bdev(struct zram *zram)
 578 {
 579         unsigned long blk_idx = 1;
 580 retry:
 581         /* skip 0 bit to confuse zram.handle = 0 */
 582         blk_idx = find_next_zero_bit(zram->bitmap, zram->nr_pages, blk_idx);
 583         if (blk_idx == zram->nr_pages)
 584                 return 0;
 585
 586         if (test_and_set_bit(blk_idx, zram->bitmap))
 587                 goto retry;
 588
 589         atomic64_inc(&zram->stats.bd_count);
 590         return blk_idx;
 591 }
 592
 593 static void free_block_bdev(struct zram *zram, unsigned long blk_idx)
 594 {
 595         int was_set;
 596
 597         was_set = test_and_clear_bit(blk_idx, zram->bitmap);
 598         WARN_ON_ONCE(!was_set);
 599         atomic64_dec(&zram->stats.bd_count);
 600 }
 601
 602 static void zram_page_end_io(struct bio *bio)
 603 {
 604         struct page *page = bio_first_page_all(bio);
 605
 606         page_endio(page, op_is_write(bio_op(bio)),
 607                         blk_status_to_errno(bio->bi_status));
 608         bio_put(bio);
 609 }
 610
 611 /*
 612  * Returns 1 if the submission is successful.
 613  */
 614 static int read_from_bdev_async(struct zram *zram, struct bio_vec *bvec,
 615                         unsigned long entry, struct bio *parent)
 616 {
 617         struct bio *bio;
 618
 619         bio = bio_alloc(zram->bdev, 1, parent ? parent->bi_opf : REQ_OP_READ,
 620                         GFP_NOIO);
 621         if (!bio)
 622                 return -ENOMEM;
 623
 624         bio->bi_iter.bi_sector = entry * (PAGE_SIZE >> 9);
 625         if (!bio_add_page(bio, bvec->bv_page, bvec->bv_len, bvec->bv_offset)) {
 626                 bio_put(bio);
 627                 return -EIO;
 628         }
 629
 630         if (!parent)
 631                 bio->bi_end_io = zram_page_end_io;
 632         else
 633                 bio_chain(bio, parent);
 634
 635         submit_bio(bio);
 636         return 1;
 637 }
 638
 639 #define PAGE_WB_SIG "page_index="
 640
 641 #define PAGE_WRITEBACK 0
 642 #define HUGE_WRITEBACK (1<<0)
 643 #define IDLE_WRITEBACK (1<<1)
 644
 645
 646 static ssize_t writeback_store(struct device *dev,
 647                 struct device_attribute *attr, const char *buf, size_t len)
 648 {
 649         struct zram *zram = dev_to_zram(dev);
 650         unsigned long nr_pages = zram->disksize >> PAGE_SHIFT;
 651         unsigned long index = 0;
 652         struct bio bio;
 653         struct bio_vec bio_vec;
 654         struct page *page;
 655         ssize_t ret = len;
 656         int mode, err;
 657         unsigned long blk_idx = 0;
 658
 659         if (sysfs_streq(buf, "idle"))
 660                 mode = IDLE_WRITEBACK;
 661         else if (sysfs_streq(buf, "huge"))
 662                 mode = HUGE_WRITEBACK;
 663         else if (sysfs_streq(buf, "huge_idle"))
 664                 mode = IDLE_WRITEBACK | HUGE_WRITEBACK;
 665         else {
 666                 if (strncmp(buf, PAGE_WB_SIG, sizeof(PAGE_WB_SIG) - 1))
 667                         return -EINVAL;
 668
 669                 if (kstrtol(buf + sizeof(PAGE_WB_SIG) - 1, 10, &index) ||
 670                                 index >= nr_pages)
 671                         return -EINVAL;
 672
 673                 nr_pages = 1;
 674                 mode = PAGE_WRITEBACK;
 675         }
 676
 677         down_read(&zram->init_lock);
 678         if (!init_done(zram)) {
 679                 ret = -EINVAL;
 680                 goto release_init_lock;
 681         }
 682
 683         if (!zram->backing_dev) {
 684                 ret = -ENODEV;
 685                 goto release_init_lock;
 686         }
 687
 688         page = alloc_page(GFP_KERNEL);
 689         if (!page) {
 690                 ret = -ENOMEM;
 691                 goto release_init_lock;
 692         }
 693
 694         for (; nr_pages != 0; index++, nr_pages--) {
 695                 struct bio_vec bvec;
 696
 697                 bvec.bv_page = page;
 698                 bvec.bv_len = PAGE_SIZE;
 699                 bvec.bv_offset = 0;
 700
 701                 spin_lock(&zram->wb_limit_lock);
 702                 if (zram->wb_limit_enable && !zram->bd_wb_limit) {
 703                         spin_unlock(&zram->wb_limit_lock);
 704                         ret = -EIO;
 705                         break;
 706                 }
 707                 spin_unlock(&zram->wb_limit_lock);
 708
 709                 if (!blk_idx) {
 710                         blk_idx = alloc_block_bdev(zram);
 711                         if (!blk_idx) {
 712                                 ret = -ENOSPC;
 713                                 break;
 714                         }
 715                 }
 716
 717                 zram_slot_lock(zram, index);
 718                 if (!zram_allocated(zram, index))
 719                         goto next;
 720
 721                 if (zram_test_flag(zram, index, ZRAM_WB) ||
 722                                 zram_test_flag(zram, index, ZRAM_SAME) ||
 723                                 zram_test_flag(zram, index, ZRAM_UNDER_WB))
 724                         goto next;
 725
 726                 if (mode & IDLE_WRITEBACK &&
 727                           !zram_test_flag(zram, index, ZRAM_IDLE))
 728                         goto next;
 729                 if (mode & HUGE_WRITEBACK &&
 730                           !zram_test_flag(zram, index, ZRAM_HUGE))
 731                         goto next;
 732                 /*
 733                  * Clearing ZRAM_UNDER_WB is duty of caller.
 734                  * IOW, zram_free_page never clear it.
 735                  */
 736                 zram_set_flag(zram, index, ZRAM_UNDER_WB);
 737                 /* Need for hugepage writeback racing */
 738                 zram_set_flag(zram, index, ZRAM_IDLE);
 739                 zram_slot_unlock(zram, index);
 740                 if (zram_bvec_read(zram, &bvec, index, 0, NULL)) {
 741                         zram_slot_lock(zram, index);
 742                         zram_clear_flag(zram, index, ZRAM_UNDER_WB);
 743                         zram_clear_flag(zram, index, ZRAM_IDLE);
 744                         zram_slot_unlock(zram, index);
 745                         continue;
 746                 }
 747
 748                 bio_init(&bio, zram->bdev, &bio_vec, 1,
 749                          REQ_OP_WRITE | REQ_SYNC);
 750                 bio.bi_iter.bi_sector = blk_idx * (PAGE_SIZE >> 9);
 751
 752                 bio_add_page(&bio, bvec.bv_page, bvec.bv_len,
 753                                 bvec.bv_offset);
 754                 /*
 755                  * XXX: A single page IO would be inefficient for write
 756                  * but it would be not bad as starter.
 757                  */
 758                 err = submit_bio_wait(&bio);
 759                 if (err) {
 760                         zram_slot_lock(zram, index);
 761                         zram_clear_flag(zram, index, ZRAM_UNDER_WB);
 762                         zram_clear_flag(zram, index, ZRAM_IDLE);
 763                         zram_slot_unlock(zram, index);
 764                         /*
 765                          * Return last IO error unless every IO were
 766                          * not suceeded.
 767                          */
 768                         ret = err;
 769                         continue;
 770                 }
 771
 772                 atomic64_inc(&zram->stats.bd_writes);
 773                 /*
 774                  * We released zram_slot_lock so need to check if the slot was
 775                  * changed. If there is freeing for the slot, we can catch it
 776                  * easily by zram_allocated.
 777                  * A subtle case is the slot is freed/reallocated/marked as
 778                  * ZRAM_IDLE again. To close the race, idle_store doesn't
 779                  * mark ZRAM_IDLE once it found the slot was ZRAM_UNDER_WB.
 780                  * Thus, we could close the race by checking ZRAM_IDLE bit.
 781                  */
 782                 zram_slot_lock(zram, index);
 783                 if (!zram_allocated(zram, index) ||
 784                           !zram_test_flag(zram, index, ZRAM_IDLE)) {
 785                         zram_clear_flag(zram, index, ZRAM_UNDER_WB);
 786                         zram_clear_flag(zram, index, ZRAM_IDLE);
 787                         goto next;
 788                 }
 789
 790                 zram_free_page(zram, index);
 791                 zram_clear_flag(zram, index, ZRAM_UNDER_WB);
 792                 zram_set_flag(zram, index, ZRAM_WB);
 793                 zram_set_element(zram, index, blk_idx);
 794                 blk_idx = 0;
 795                 atomic64_inc(&zram->stats.pages_stored);
 796                 spin_lock(&zram->wb_limit_lock);
 797                 if (zram->wb_limit_enable && zram->bd_wb_limit > 0)
 798                         zram->bd_wb_limit -=  1UL << (PAGE_SHIFT - 12);
 799                 spin_unlock(&zram->wb_limit_lock);
 800 next:
 801                 zram_slot_unlock(zram, index);
 802         }
 803
 804         if (blk_idx)
 805                 free_block_bdev(zram, blk_idx);
 806         __free_page(page);
 807 release_init_lock:
 808         up_read(&zram->init_lock);
 809
 810         return ret;
 811 }
 812
 813 struct zram_work {
 814         struct work_struct work;
 815         struct zram *zram;
 816         unsigned long entry;
 817         struct bio *bio;
 818         struct bio_vec bvec;
 819 };
 820
 821 #if PAGE_SIZE != 4096
 822 static void zram_sync_read(struct work_struct *work)
 823 {
 824         struct zram_work *zw = container_of(work, struct zram_work, work);
 825         struct zram *zram = zw->zram;
 826         unsigned long entry = zw->entry;
 827         struct bio *bio = zw->bio;
 828
 829         read_from_bdev_async(zram, &zw->bvec, entry, bio);
 830 }
 831
 832 /*
 833  * Block layer want one ->submit_bio to be active at a time, so if we use
 834  * chained IO with parent IO in same context, it's a deadlock. To avoid that,
 835  * use a worker thread context.
 836  */
 837 static int read_from_bdev_sync(struct zram *zram, struct bio_vec *bvec,
 838                                 unsigned long entry, struct bio *bio)
 839 {
 840         struct zram_work work;
 841
 842         work.bvec = *bvec;
 843         work.zram = zram;
 844         work.entry = entry;
 845         work.bio = bio;
 846
 847         INIT_WORK_ONSTACK(&work.work, zram_sync_read);
 848         queue_work(system_unbound_wq, &work.work);
 849         flush_work(&work.work);
 850         destroy_work_on_stack(&work.work);
 851
 852         return 1;
 853 }
 854 #else
 855 static int read_from_bdev_sync(struct zram *zram, struct bio_vec *bvec,
 856                                 unsigned long entry, struct bio *bio)
 857 {
 858         WARN_ON(1);
 859         return -EIO;
 860 }
 861 #endif
 862
 863 static int read_from_bdev(struct zram *zram, struct bio_vec *bvec,
 864                         unsigned long entry, struct bio *parent, bool sync)
 865 {
 866         atomic64_inc(&zram->stats.bd_reads);
 867         if (sync)
 868                 return read_from_bdev_sync(zram, bvec, entry, parent);
 869         else
 870                 return read_from_bdev_async(zram, bvec, entry, parent);
 871 }
 872 #else
 873 static inline void reset_bdev(struct zram *zram) {};
 874 static int read_from_bdev(struct zram *zram, struct bio_vec *bvec,
 875                         unsigned long entry, struct bio *parent, bool sync)
 876 {
 877         return -EIO;
 878 }
 879
 880 static void free_block_bdev(struct zram *zram, unsigned long blk_idx) {};
 881 #endif
 882
 883 #ifdef CONFIG_ZRAM_MEMORY_TRACKING
 884
 885 static struct dentry *zram_debugfs_root;
 886
 887 static void zram_debugfs_create(void)
 888 {
 889         zram_debugfs_root = debugfs_create_dir("zram", NULL);
 890 }
 891
 892 static void zram_debugfs_destroy(void)
 893 {
 894         debugfs_remove_recursive(zram_debugfs_root);
 895 }
 896
 897 static void zram_accessed(struct zram *zram, u32 index)
 898 {
 899         zram_clear_flag(zram, index, ZRAM_IDLE);
 900         zram->table[index].ac_time = ktime_get_boottime();
 901 }
 902
 903 static ssize_t read_block_state(struct file *file, char __user *buf,
 904                                 size_t count, loff_t *ppos)
 905 {
 906         char *kbuf;
 907         ssize_t index, written = 0;
 908         struct zram *zram = file->private_data;
 909         unsigned long nr_pages = zram->disksize >> PAGE_SHIFT;
 910         struct timespec64 ts;
 911
 912         kbuf = kvmalloc(count, GFP_KERNEL);
 913         if (!kbuf)
 914                 return -ENOMEM;
 915
 916         down_read(&zram->init_lock);
 917         if (!init_done(zram)) {
 918                 up_read(&zram->init_lock);
 919                 kvfree(kbuf);
 920                 return -EINVAL;
 921         }
 922
 923         for (index = *ppos; index < nr_pages; index++) {
 924                 int copied;
 925
 926                 zram_slot_lock(zram, index);
 927                 if (!zram_allocated(zram, index))
 928                         goto next;
 929
 930                 ts = ktime_to_timespec64(zram->table[index].ac_time);
 931                 copied = snprintf(kbuf + written, count,
 932                         "%12zd %12lld.%06lu %c%c%c%c\n",
 933                         index, (s64)ts.tv_sec,
 934                         ts.tv_nsec / NSEC_PER_USEC,
 935                         zram_test_flag(zram, index, ZRAM_SAME) ? 's' : '.',
 936                         zram_test_flag(zram, index, ZRAM_WB) ? 'w' : '.',
 937                         zram_test_flag(zram, index, ZRAM_HUGE) ? 'h' : '.',
 938                         zram_test_flag(zram, index, ZRAM_IDLE) ? 'i' : '.');
 939
 940                 if (count <= copied) {
 941                         zram_slot_unlock(zram, index);
 942                         break;
 943                 }
 944                 written += copied;
 945                 count -= copied;
 946 next:
 947                 zram_slot_unlock(zram, index);
 948                 *ppos += 1;
 949         }
 950
 951         up_read(&zram->init_lock);
 952         if (copy_to_user(buf, kbuf, written))
 953                 written = -EFAULT;
 954         kvfree(kbuf);
 955
 956         return written;
 957 }
 958
 959 static const struct file_operations proc_zram_block_state_op = {
 960         .open = simple_open,
 961         .read = read_block_state,
 962         .llseek = default_llseek,
 963 };
 964
 965 static void zram_debugfs_register(struct zram *zram)
 966 {
 967         if (!zram_debugfs_root)
 968                 return;
 969
 970         zram->debugfs_dir = debugfs_create_dir(zram->disk->disk_name,
 971                                                 zram_debugfs_root);
 972         debugfs_create_file("block_state", 0400, zram->debugfs_dir,
 973                                 zram, &proc_zram_block_state_op);
 974 }
 975
 976 static void zram_debugfs_unregister(struct zram *zram)
 977 {
 978         debugfs_remove_recursive(zram->debugfs_dir);
 979 }
 980 #else
 981 static void zram_debugfs_create(void) {};
 982 static void zram_debugfs_destroy(void) {};
 983 static void zram_accessed(struct zram *zram, u32 index)
 984 {
 985         zram_clear_flag(zram, index, ZRAM_IDLE);
 986 };
 987 static void zram_debugfs_register(struct zram *zram) {};
 988 static void zram_debugfs_unregister(struct zram *zram) {};
 989 #endif
 990
 991 /*
 992  * We switched to per-cpu streams and this attr is not needed anymore.
 993  * However, we will keep it around for some time, because:
 994  * a) we may revert per-cpu streams in the future
 995  * b) it's visible to user space and we need to follow our 2 years
 996  *    retirement rule; but we already have a number of 'soon to be
 997  *    altered' attrs, so max_comp_streams need to wait for the next
 998  *    layoff cycle.
 999  */
1000 static ssize_t max_comp_streams_show(struct device *dev,
1001                 struct device_attribute *attr, char *buf)
1002 {
1003         return scnprintf(buf, PAGE_SIZE, "%d\n", num_online_cpus());
1004 }
1005
1006 static ssize_t max_comp_streams_store(struct device *dev,
1007                 struct device_attribute *attr, const char *buf, size_t len)
1008 {
1009         return len;
1010 }
1011
1012 static ssize_t comp_algorithm_show(struct device *dev,
1013                 struct device_attribute *attr, char *buf)
1014 {
1015         size_t sz;
1016         struct zram *zram = dev_to_zram(dev);
1017
1018         down_read(&zram->init_lock);
1019         sz = zcomp_available_show(zram->compressor, buf);
1020         up_read(&zram->init_lock);
1021
1022         return sz;
1023 }
1024
1025 static ssize_t comp_algorithm_store(struct device *dev,
1026                 struct device_attribute *attr, const char *buf, size_t len)
1027 {
1028         struct zram *zram = dev_to_zram(dev);
1029         char compressor[ARRAY_SIZE(zram->compressor)];
1030         size_t sz;
1031
1032         strlcpy(compressor, buf, sizeof(compressor));
1033         /* ignore trailing newline */
1034         sz = strlen(compressor);
1035         if (sz > 0 && compressor[sz - 1] == '\n')
1036                 compressor[sz - 1] = 0x00;
1037
1038         if (!zcomp_available_algorithm(compressor))
1039                 return -EINVAL;
1040
1041         down_write(&zram->init_lock);
1042         if (init_done(zram)) {
1043                 up_write(&zram->init_lock);
1044                 pr_info("Can't change algorithm for initialized device\n");
1045                 return -EBUSY;
1046         }
1047
1048         strcpy(zram->compressor, compressor);
1049         up_write(&zram->init_lock);
1050         return len;
1051 }
1052
1053 static ssize_t compact_store(struct device *dev,
1054                 struct device_attribute *attr, const char *buf, size_t len)
1055 {
1056         struct zram *zram = dev_to_zram(dev);
1057
1058         down_read(&zram->init_lock);
1059         if (!init_done(zram)) {
1060                 up_read(&zram->init_lock);
1061                 return -EINVAL;
1062         }
1063
1064         zs_compact(zram->mem_pool);
1065         up_read(&zram->init_lock);
1066
1067         return len;
1068 }
1069
1070 static ssize_t io_stat_show(struct device *dev,
1071                 struct device_attribute *attr, char *buf)
1072 {
1073         struct zram *zram = dev_to_zram(dev);
1074         ssize_t ret;
1075
1076         down_read(&zram->init_lock);
1077         ret = scnprintf(buf, PAGE_SIZE,
1078                         "%8llu %8llu %8llu %8llu\n",
1079                         (u64)atomic64_read(&zram->stats.failed_reads),
1080                         (u64)atomic64_read(&zram->stats.failed_writes),
1081                         (u64)atomic64_read(&zram->stats.invalid_io),
1082                         (u64)atomic64_read(&zram->stats.notify_free));
1083         up_read(&zram->init_lock);
1084
1085         return ret;
1086 }
1087
1088 static ssize_t mm_stat_show(struct device *dev,
1089                 struct device_attribute *attr, char *buf)
1090 {
1091         struct zram *zram = dev_to_zram(dev);
1092         struct zs_pool_stats pool_stats;
1093         u64 orig_size, mem_used = 0;
1094         long max_used;
1095         ssize_t ret;
1096
1097         memset(&pool_stats, 0x00, sizeof(struct zs_pool_stats));
1098
1099         down_read(&zram->init_lock);
1100         if (init_done(zram)) {
1101                 mem_used = zs_get_total_pages(zram->mem_pool);
1102                 zs_pool_stats(zram->mem_pool, &pool_stats);
1103         }
1104
1105         orig_size = atomic64_read(&zram->stats.pages_stored);
1106         max_used = atomic_long_read(&zram->stats.max_used_pages);
1107
1108         ret = scnprintf(buf, PAGE_SIZE,
1109                         "%8llu %8llu %8llu %8lu %8ld %8llu %8lu %8llu %8llu\n",
1110                         orig_size << PAGE_SHIFT,
1111                         (u64)atomic64_read(&zram->stats.compr_data_size),
1112                         mem_used << PAGE_SHIFT,
1113                         zram->limit_pages << PAGE_SHIFT,
1114                         max_used << PAGE_SHIFT,
1115                         (u64)atomic64_read(&zram->stats.same_pages),
1116                         atomic_long_read(&pool_stats.pages_compacted),
1117                         (u64)atomic64_read(&zram->stats.huge_pages),
1118                         (u64)atomic64_read(&zram->stats.huge_pages_since));
1119         up_read(&zram->init_lock);
1120
1121         return ret;
1122 }
1123
1124 #ifdef CONFIG_ZRAM_WRITEBACK
1125 #define FOUR_K(x) ((x) * (1 << (PAGE_SHIFT - 12)))
1126 static ssize_t bd_stat_show(struct device *dev,
1127                 struct device_attribute *attr, char *buf)
1128 {
1129         struct zram *zram = dev_to_zram(dev);
1130         ssize_t ret;
1131
1132         down_read(&zram->init_lock);
1133         ret = scnprintf(buf, PAGE_SIZE,
1134                 "%8llu %8llu %8llu\n",
1135                         FOUR_K((u64)atomic64_read(&zram->stats.bd_count)),
1136                         FOUR_K((u64)atomic64_read(&zram->stats.bd_reads)),
1137                         FOUR_K((u64)atomic64_read(&zram->stats.bd_writes)));
1138         up_read(&zram->init_lock);
1139
1140         return ret;
1141 }
1142 #endif
1143
1144 static ssize_t debug_stat_show(struct device *dev,
1145                 struct device_attribute *attr, char *buf)
1146 {
1147         int version = 2;
1148         struct zram *zram = dev_to_zram(dev);
1149         ssize_t ret;
1150
1151         down_read(&zram->init_lock);
1152         ret = scnprintf(buf, PAGE_SIZE,
1153                         "version: %d\n%8llu\n",
1154                         version,
1155                         (u64)atomic64_read(&zram->stats.miss_free));
1156         up_read(&zram->init_lock);
1157
1158         return ret;
1159 }
1160
1161 static DEVICE_ATTR_RO(io_stat);
1162 static DEVICE_ATTR_RO(mm_stat);
1163 #ifdef CONFIG_ZRAM_WRITEBACK
1164 static DEVICE_ATTR_RO(bd_stat);
1165 #endif
1166 static DEVICE_ATTR_RO(debug_stat);
1167
1168 static void zram_meta_free(struct zram *zram, u64 disksize)
1169 {
1170         size_t num_pages = disksize >> PAGE_SHIFT;
1171         size_t index;
1172
1173         /* Free all pages that are still in this zram device */
1174         for (index = 0; index < num_pages; index++)
1175                 zram_free_page(zram, index);
1176
1177         zs_destroy_pool(zram->mem_pool);
1178         vfree(zram->table);
1179 }
1180
1181 static bool zram_meta_alloc(struct zram *zram, u64 disksize)
1182 {
1183         size_t num_pages;
1184
1185         num_pages = disksize >> PAGE_SHIFT;
1186         zram->table = vzalloc(array_size(num_pages, sizeof(*zram->table)));
1187         if (!zram->table)
1188                 return false;
1189
1190         zram->mem_pool = zs_create_pool(zram->disk->disk_name);
1191         if (!zram->mem_pool) {
1192                 vfree(zram->table);
1193                 return false;
1194         }
1195
1196         if (!huge_class_size)
1197                 huge_class_size = zs_huge_class_size(zram->mem_pool);
1198         return true;
1199 }
1200
1201 /*
1202  * To protect concurrent access to the same index entry,
1203  * caller should hold this table index entry's bit_spinlock to
1204  * indicate this index entry is accessing.
1205  */
1206 static void zram_free_page(struct zram *zram, size_t index)
1207 {
1208         unsigned long handle;
1209
1210 #ifdef CONFIG_ZRAM_MEMORY_TRACKING
1211         zram->table[index].ac_time = 0;
1212 #endif
1213         if (zram_test_flag(zram, index, ZRAM_IDLE))
1214                 zram_clear_flag(zram, index, ZRAM_IDLE);
1215
1216         if (zram_test_flag(zram, index, ZRAM_HUGE)) {
1217                 zram_clear_flag(zram, index, ZRAM_HUGE);
1218                 atomic64_dec(&zram->stats.huge_pages);
1219         }
1220
1221         if (zram_test_flag(zram, index, ZRAM_WB)) {
1222                 zram_clear_flag(zram, index, ZRAM_WB);
1223                 free_block_bdev(zram, zram_get_element(zram, index));
1224                 goto out;
1225         }
1226
1227         /*
1228          * No memory is allocated for same element filled pages.
1229          * Simply clear same page flag.
1230          */
1231         if (zram_test_flag(zram, index, ZRAM_SAME)) {
1232                 zram_clear_flag(zram, index, ZRAM_SAME);
1233                 atomic64_dec(&zram->stats.same_pages);
1234                 goto out;
1235         }
1236
1237         handle = zram_get_handle(zram, index);
1238         if (!handle)
1239                 return;
1240
1241         zs_free(zram->mem_pool, handle);
1242
1243         atomic64_sub(zram_get_obj_size(zram, index),
1244                         &zram->stats.compr_data_size);
1245 out:
1246         atomic64_dec(&zram->stats.pages_stored);
1247         zram_set_handle(zram, index, 0);
1248         zram_set_obj_size(zram, index, 0);
1249         WARN_ON_ONCE(zram->table[index].flags &
1250                 ~(1UL << ZRAM_LOCK | 1UL << ZRAM_UNDER_WB));
1251 }
1252
1253 static int __zram_bvec_read(struct zram *zram, struct page *page, u32 index,
1254                                 struct bio *bio, bool partial_io)
1255 {
1256         struct zcomp_strm *zstrm;
1257         unsigned long handle;
1258         unsigned int size;
1259         void *src, *dst;
1260         int ret;
1261
1262         zram_slot_lock(zram, index);
1263         if (zram_test_flag(zram, index, ZRAM_WB)) {
1264                 struct bio_vec bvec;
1265
1266                 zram_slot_unlock(zram, index);
1267
1268                 bvec.bv_page = page;
1269                 bvec.bv_len = PAGE_SIZE;
1270                 bvec.bv_offset = 0;
1271                 return read_from_bdev(zram, &bvec,
1272                                 zram_get_element(zram, index),
1273                                 bio, partial_io);
1274         }
1275
1276         handle = zram_get_handle(zram, index);
1277         if (!handle || zram_test_flag(zram, index, ZRAM_SAME)) {
1278                 unsigned long value;
1279                 void *mem;
1280
1281                 value = handle ? zram_get_element(zram, index) : 0;
1282                 mem = kmap_atomic(page);
1283                 zram_fill_page(mem, PAGE_SIZE, value);
1284                 kunmap_atomic(mem);
1285                 zram_slot_unlock(zram, index);
1286                 return 0;
1287         }
1288
1289         size = zram_get_obj_size(zram, index);
1290
1291         if (size != PAGE_SIZE)
1292                 zstrm = zcomp_stream_get(zram->comp);
1293
1294         src = zs_map_object(zram->mem_pool, handle, ZS_MM_RO);
1295         if (size == PAGE_SIZE) {
1296                 dst = kmap_atomic(page);
1297                 memcpy(dst, src, PAGE_SIZE);
1298                 kunmap_atomic(dst);
1299                 ret = 0;
1300         } else {
1301                 dst = kmap_atomic(page);
1302                 ret = zcomp_decompress(zstrm, src, size, dst);
1303                 kunmap_atomic(dst);
1304                 zcomp_stream_put(zram->comp);
1305         }
1306         zs_unmap_object(zram->mem_pool, handle);
1307         zram_slot_unlock(zram, index);
1308
1309         /* Should NEVER happen. Return bio error if it does. */
1310         if (WARN_ON(ret))
1311                 pr_err("Decompression failed! err=%d, page=%u\n", ret, index);
1312
1313         return ret;
1314 }
1315
1316 static int zram_bvec_read(struct zram *zram, struct bio_vec *bvec,
1317                                 u32 index, int offset, struct bio *bio)
1318 {
1319         int ret;
1320         struct page *page;
1321
1322         page = bvec->bv_page;
1323         if (is_partial_io(bvec)) {
1324                 /* Use a temporary buffer to decompress the page */
1325                 page = alloc_page(GFP_NOIO|__GFP_HIGHMEM);
1326                 if (!page)
1327                         return -ENOMEM;
1328         }
1329
1330         ret = __zram_bvec_read(zram, page, index, bio, is_partial_io(bvec));
1331         if (unlikely(ret))
1332                 goto out;
1333
1334         if (is_partial_io(bvec)) {
1335                 void *src = kmap_atomic(page);
1336
1337                 memcpy_to_bvec(bvec, src + offset);
1338                 kunmap_atomic(src);
1339         }
1340 out:
1341         if (is_partial_io(bvec))
1342                 __free_page(page);
1343
1344         return ret;
1345 }
1346
1347 static int __zram_bvec_write(struct zram *zram, struct bio_vec *bvec,
1348                                 u32 index, struct bio *bio)
1349 {
1350         int ret = 0;
1351         unsigned long alloced_pages;
1352         unsigned long handle = 0;
1353         unsigned int comp_len = 0;
1354         void *src, *dst, *mem;
1355         struct zcomp_strm *zstrm;
1356         struct page *page = bvec->bv_page;
1357         unsigned long element = 0;
1358         enum zram_pageflags flags = 0;
1359
1360         mem = kmap_atomic(page);
1361         if (page_same_filled(mem, &element)) {
1362                 kunmap_atomic(mem);
1363                 /* Free memory associated with this sector now. */
1364                 flags = ZRAM_SAME;
1365                 atomic64_inc(&zram->stats.same_pages);
1366                 goto out;
1367         }
1368         kunmap_atomic(mem);
1369
1370         zstrm = zcomp_stream_get(zram->comp);
1371         src = kmap_atomic(page);
1372         ret = zcomp_compress(zstrm, src, &comp_len);
1373         kunmap_atomic(src);
1374
1375         if (unlikely(ret)) {
1376                 zcomp_stream_put(zram->comp);
1377                 pr_err("Compression failed! err=%d\n", ret);
1378                 return ret;
1379         }
1380
1381         if (comp_len >= huge_class_size)
1382                 comp_len = PAGE_SIZE;
1383
1384         handle = zs_malloc(zram->mem_pool, comp_len,
1385                         __GFP_KSWAPD_RECLAIM |
1386                         __GFP_NOWARN |
1387                         __GFP_HIGHMEM |
1388                         __GFP_MOVABLE);
1389
1390         if (unlikely(!handle)) {
1391                 zcomp_stream_put(zram->comp);
1392                 return -ENOMEM;
1393         }
1394
1395         alloced_pages = zs_get_total_pages(zram->mem_pool);
1396         update_used_max(zram, alloced_pages);
1397
1398         if (zram->limit_pages && alloced_pages > zram->limit_pages) {
1399                 zcomp_stream_put(zram->comp);
1400                 zs_free(zram->mem_pool, handle);
1401                 return -ENOMEM;
1402         }
1403
1404         dst = zs_map_object(zram->mem_pool, handle, ZS_MM_WO);
1405
1406         src = zstrm->buffer;
1407         if (comp_len == PAGE_SIZE)
1408                 src = kmap_atomic(page);
1409         memcpy(dst, src, comp_len);
1410         if (comp_len == PAGE_SIZE)
1411                 kunmap_atomic(src);
1412
1413         zcomp_stream_put(zram->comp);
1414         zs_unmap_object(zram->mem_pool, handle);
1415         atomic64_add(comp_len, &zram->stats.compr_data_size);
1416 out:
1417         /*
1418          * Free memory associated with this sector
1419          * before overwriting unused sectors.
1420          */
1421         zram_slot_lock(zram, index);
1422         zram_free_page(zram, index);
1423
1424         if (comp_len == PAGE_SIZE) {
1425                 zram_set_flag(zram, index, ZRAM_HUGE);
1426                 atomic64_inc(&zram->stats.huge_pages);
1427                 atomic64_inc(&zram->stats.huge_pages_since);
1428         }
1429
1430         if (flags) {
1431                 zram_set_flag(zram, index, flags);
1432                 zram_set_element(zram, index, element);
1433         }  else {
1434                 zram_set_handle(zram, index, handle);
1435                 zram_set_obj_size(zram, index, comp_len);
1436         }
1437         zram_slot_unlock(zram, index);
1438
1439         /* Update stats */
1440         atomic64_inc(&zram->stats.pages_stored);
1441         return ret;
1442 }
1443
1444 static int zram_bvec_write(struct zram *zram, struct bio_vec *bvec,
1445                                 u32 index, int offset, struct bio *bio)
1446 {
1447         int ret;
1448         struct page *page = NULL;
1449         struct bio_vec vec;
1450
1451         vec = *bvec;
1452         if (is_partial_io(bvec)) {
1453                 void *dst;
1454                 /*
1455                  * This is a partial IO. We need to read the full page
1456                  * before to write the changes.
1457                  */
1458                 page = alloc_page(GFP_NOIO|__GFP_HIGHMEM);
1459                 if (!page)
1460                         return -ENOMEM;
1461
1462                 ret = __zram_bvec_read(zram, page, index, bio, true);
1463                 if (ret)
1464                         goto out;
1465
1466                 dst = kmap_atomic(page);
1467                 memcpy_from_bvec(dst + offset, bvec);
1468                 kunmap_atomic(dst);
1469
1470                 vec.bv_page = page;
1471                 vec.bv_len = PAGE_SIZE;
1472                 vec.bv_offset = 0;
1473         }
1474
1475         ret = __zram_bvec_write(zram, &vec, index, bio);
1476 out:
1477         if (is_partial_io(bvec))
1478                 __free_page(page);
1479         return ret;
1480 }
1481
1482 /*
1483  * zram_bio_discard - handler on discard request
1484  * @index: physical block index in PAGE_SIZE units
1485  * @offset: byte offset within physical block
1486  */
1487 static void zram_bio_discard(struct zram *zram, u32 index,
1488                              int offset, struct bio *bio)
1489 {
1490         size_t n = bio->bi_iter.bi_size;
1491
1492         /*
1493          * zram manages data in physical block size units. Because logical block
1494          * size isn't identical with physical block size on some arch, we
1495          * could get a discard request pointing to a specific offset within a
1496          * certain physical block.  Although we can handle this request by
1497          * reading that physiclal block and decompressing and partially zeroing
1498          * and re-compressing and then re-storing it, this isn't reasonable
1499          * because our intent with a discard request is to save memory.  So
1500          * skipping this logical block is appropriate here.
1501          */
1502         if (offset) {
1503                 if (n <= (PAGE_SIZE - offset))
1504                         return;
1505
1506                 n -= (PAGE_SIZE - offset);
1507                 index++;
1508         }
1509
1510         while (n >= PAGE_SIZE) {
1511                 zram_slot_lock(zram, index);
1512                 zram_free_page(zram, index);
1513                 zram_slot_unlock(zram, index);
1514                 atomic64_inc(&zram->stats.notify_free);
1515                 index++;
1516                 n -= PAGE_SIZE;
1517         }
1518 }
1519
1520 /*
1521  * Returns errno if it has some problem. Otherwise return 0 or 1.
1522  * Returns 0 if IO request was done synchronously
1523  * Returns 1 if IO request was successfully submitted.
1524  */
1525 static int zram_bvec_rw(struct zram *zram, struct bio_vec *bvec, u32 index,
1526                         int offset, unsigned int op, struct bio *bio)
1527 {
1528         int ret;
1529
1530         if (!op_is_write(op)) {
1531                 atomic64_inc(&zram->stats.num_reads);
1532                 ret = zram_bvec_read(zram, bvec, index, offset, bio);
1533                 flush_dcache_page(bvec->bv_page);
1534         } else {
1535                 atomic64_inc(&zram->stats.num_writes);
1536                 ret = zram_bvec_write(zram, bvec, index, offset, bio);
1537         }
1538
1539         zram_slot_lock(zram, index);
1540         zram_accessed(zram, index);
1541         zram_slot_unlock(zram, index);
1542
1543         if (unlikely(ret < 0)) {
1544                 if (!op_is_write(op))
1545                         atomic64_inc(&zram->stats.failed_reads);
1546                 else
1547                         atomic64_inc(&zram->stats.failed_writes);
1548         }
1549
1550         return ret;
1551 }
1552
1553 static void __zram_make_request(struct zram *zram, struct bio *bio)
1554 {
1555         int offset;
1556         u32 index;
1557         struct bio_vec bvec;
1558         struct bvec_iter iter;
1559         unsigned long start_time;
1560
1561         index = bio->bi_iter.bi_sector >> SECTORS_PER_PAGE_SHIFT;
1562         offset = (bio->bi_iter.bi_sector &
1563                   (SECTORS_PER_PAGE - 1)) << SECTOR_SHIFT;
1564
1565         switch (bio_op(bio)) {
1566         case REQ_OP_DISCARD:
1567         case REQ_OP_WRITE_ZEROES:
1568                 zram_bio_discard(zram, index, offset, bio);
1569                 bio_endio(bio);
1570                 return;
1571         default:
1572                 break;
1573         }
1574
1575         start_time = bio_start_io_acct(bio);
1576         bio_for_each_segment(bvec, bio, iter) {
1577                 struct bio_vec bv = bvec;
1578                 unsigned int unwritten = bvec.bv_len;
1579
1580                 do {
1581                         bv.bv_len = min_t(unsigned int, PAGE_SIZE - offset,
1582                                                         unwritten);
1583                         if (zram_bvec_rw(zram, &bv, index, offset,
1584                                          bio_op(bio), bio) < 0) {
1585                                 bio->bi_status = BLK_STS_IOERR;
1586                                 break;
1587                         }
1588
1589                         bv.bv_offset += bv.bv_len;
1590                         unwritten -= bv.bv_len;
1591
1592                         update_position(&index, &offset, &bv);
1593                 } while (unwritten);
1594         }
1595         bio_end_io_acct(bio, start_time);
1596         bio_endio(bio);
1597 }
1598
1599 /*
1600  * Handler function for all zram I/O requests.
1601  */
1602 static void zram_submit_bio(struct bio *bio)
1603 {
1604         struct zram *zram = bio->bi_bdev->bd_disk->private_data;
1605
1606         if (!valid_io_request(zram, bio->bi_iter.bi_sector,
1607                                         bio->bi_iter.bi_size)) {
1608                 atomic64_inc(&zram->stats.invalid_io);
1609                 bio_io_error(bio);
1610                 return;
1611         }
1612
1613         __zram_make_request(zram, bio);
1614 }
1615
1616 static void zram_slot_free_notify(struct block_device *bdev,
1617                                 unsigned long index)
1618 {
1619         struct zram *zram;
1620
1621         zram = bdev->bd_disk->private_data;
1622
1623         atomic64_inc(&zram->stats.notify_free);
1624         if (!zram_slot_trylock(zram, index)) {
1625                 atomic64_inc(&zram->stats.miss_free);
1626                 return;
1627         }
1628
1629         zram_free_page(zram, index);
1630         zram_slot_unlock(zram, index);
1631 }
1632
1633 static int zram_rw_page(struct block_device *bdev, sector_t sector,
1634                        struct page *page, unsigned int op)
1635 {
1636         int offset, ret;
1637         u32 index;
1638         struct zram *zram;
1639         struct bio_vec bv;
1640         unsigned long start_time;
1641
1642         if (PageTransHuge(page))
1643                 return -ENOTSUPP;
1644         zram = bdev->bd_disk->private_data;
1645
1646         if (!valid_io_request(zram, sector, PAGE_SIZE)) {
1647                 atomic64_inc(&zram->stats.invalid_io);
1648                 ret = -EINVAL;
1649                 goto out;
1650         }
1651
1652         index = sector >> SECTORS_PER_PAGE_SHIFT;
1653         offset = (sector & (SECTORS_PER_PAGE - 1)) << SECTOR_SHIFT;
1654
1655         bv.bv_page = page;
1656         bv.bv_len = PAGE_SIZE;
1657         bv.bv_offset = 0;
1658
1659         start_time = bdev_start_io_acct(bdev->bd_disk->part0,
1660                         SECTORS_PER_PAGE, op, jiffies);
1661         ret = zram_bvec_rw(zram, &bv, index, offset, op, NULL);
1662         bdev_end_io_acct(bdev->bd_disk->part0, op, start_time);
1663 out:
1664         /*
1665          * If I/O fails, just return error(ie, non-zero) without
1666          * calling page_endio.
1667          * It causes resubmit the I/O with bio request by upper functions
1668          * of rw_page(e.g., swap_readpage, __swap_writepage) and
1669          * bio->bi_end_io does things to handle the error
1670          * (e.g., SetPageError, set_page_dirty and extra works).
1671          */
1672         if (unlikely(ret < 0))
1673                 return ret;
1674
1675         switch (ret) {
1676         case 0:
1677                 page_endio(page, op_is_write(op), 0);
1678                 break;
1679         case 1:
1680                 ret = 0;
1681                 break;
1682         default:
1683                 WARN_ON(1);
1684         }
1685         return ret;
1686 }
1687
1688 static void zram_reset_device(struct zram *zram)
1689 {
1690         struct zcomp *comp;
1691         u64 disksize;
1692
1693         down_write(&zram->init_lock);
1694
1695         zram->limit_pages = 0;
1696
1697         if (!init_done(zram)) {
1698                 up_write(&zram->init_lock);
1699                 return;
1700         }
1701
1702         comp = zram->comp;
1703         disksize = zram->disksize;
1704         zram->disksize = 0;
1705
1706         set_capacity_and_notify(zram->disk, 0);
1707         part_stat_set_all(zram->disk->part0, 0);
1708
1709         /* I/O operation under all of CPU are done so let's free */
1710         zram_meta_free(zram, disksize);
1711         memset(&zram->stats, 0, sizeof(zram->stats));
1712         zcomp_destroy(comp);
1713         reset_bdev(zram);
1714
1715         up_write(&zram->init_lock);
1716 }
1717
1718 static ssize_t disksize_store(struct device *dev,
1719                 struct device_attribute *attr, const char *buf, size_t len)
1720 {
1721         u64 disksize;
1722         struct zcomp *comp;
1723         struct zram *zram = dev_to_zram(dev);
1724         int err;
1725
1726         disksize = memparse(buf, NULL);
1727         if (!disksize)
1728                 return -EINVAL;
1729
1730         down_write(&zram->init_lock);
1731         if (init_done(zram)) {
1732                 pr_info("Cannot change disksize for initialized device\n");
1733                 err = -EBUSY;
1734                 goto out_unlock;
1735         }
1736
1737         disksize = PAGE_ALIGN(disksize);
1738         if (!zram_meta_alloc(zram, disksize)) {
1739                 err = -ENOMEM;
1740                 goto out_unlock;
1741         }
1742
1743         comp = zcomp_create(zram->compressor);
1744         if (IS_ERR(comp)) {
1745                 pr_err("Cannot initialise %s compressing backend\n",
1746                                 zram->compressor);
1747                 err = PTR_ERR(comp);
1748                 goto out_free_meta;
1749         }
1750
1751         zram->comp = comp;
1752         zram->disksize = disksize;
1753         set_capacity_and_notify(zram->disk, zram->disksize >> SECTOR_SHIFT);
1754         up_write(&zram->init_lock);
1755
1756         return len;
1757
1758 out_free_meta:
1759         zram_meta_free(zram, disksize);
1760 out_unlock:
1761         up_write(&zram->init_lock);
1762         return err;
1763 }
1764
1765 static ssize_t reset_store(struct device *dev,
1766                 struct device_attribute *attr, const char *buf, size_t len)
1767 {
1768         int ret;
1769         unsigned short do_reset;
1770         struct zram *zram;
1771         struct gendisk *disk;
1772
1773         ret = kstrtou16(buf, 10, &do_reset);
1774         if (ret)
1775                 return ret;
1776
1777         if (!do_reset)
1778                 return -EINVAL;
1779
1780         zram = dev_to_zram(dev);
1781         disk = zram->disk;
1782
1783         mutex_lock(&disk->open_mutex);
1784         /* Do not reset an active device or claimed device */
1785         if (disk_openers(disk) || zram->claim) {
1786                 mutex_unlock(&disk->open_mutex);
1787                 return -EBUSY;
1788         }
1789
1790         /* From now on, anyone can't open /dev/zram[0-9] */
1791         zram->claim = true;
1792         mutex_unlock(&disk->open_mutex);
1793
1794         /* Make sure all the pending I/O are finished */
1795         sync_blockdev(disk->part0);
1796         zram_reset_device(zram);
1797
1798         mutex_lock(&disk->open_mutex);
1799         zram->claim = false;
1800         mutex_unlock(&disk->open_mutex);
1801
1802         return len;
1803 }
1804
1805 static int zram_open(struct block_device *bdev, fmode_t mode)
1806 {
1807         int ret = 0;
1808         struct zram *zram;
1809
1810         WARN_ON(!mutex_is_locked(&bdev->bd_disk->open_mutex));
1811
1812         zram = bdev->bd_disk->private_data;
1813         /* zram was claimed to reset so open request fails */
1814         if (zram->claim)
1815                 ret = -EBUSY;
1816
1817         return ret;
1818 }
1819
1820 static const struct block_device_operations zram_devops = {
1821         .open = zram_open,
1822         .submit_bio = zram_submit_bio,
1823         .swap_slot_free_notify = zram_slot_free_notify,
1824         .rw_page = zram_rw_page,
1825         .owner = THIS_MODULE
1826 };
1827
1828 #ifdef CONFIG_ZRAM_WRITEBACK
1829 static const struct block_device_operations zram_wb_devops = {
1830         .open = zram_open,
1831         .submit_bio = zram_submit_bio,
1832         .swap_slot_free_notify = zram_slot_free_notify,
1833         .owner = THIS_MODULE
1834 };
1835 #endif
1836
1837 static DEVICE_ATTR_WO(compact);
1838 static DEVICE_ATTR_RW(disksize);
1839 static DEVICE_ATTR_RO(initstate);
1840 static DEVICE_ATTR_WO(reset);
1841 static DEVICE_ATTR_WO(mem_limit);
1842 static DEVICE_ATTR_WO(mem_used_max);
1843 static DEVICE_ATTR_WO(idle);
1844 static DEVICE_ATTR_RW(max_comp_streams);
1845 static DEVICE_ATTR_RW(comp_algorithm);
1846 #ifdef CONFIG_ZRAM_WRITEBACK
1847 static DEVICE_ATTR_RW(backing_dev);
1848 static DEVICE_ATTR_WO(writeback);
1849 static DEVICE_ATTR_RW(writeback_limit);
1850 static DEVICE_ATTR_RW(writeback_limit_enable);
1851 #endif
1852
1853 static struct attribute *zram_disk_attrs[] = {
1854         &dev_attr_disksize.attr,
1855         &dev_attr_initstate.attr,
1856         &dev_attr_reset.attr,
1857         &dev_attr_compact.attr,
1858         &dev_attr_mem_limit.attr,
1859         &dev_attr_mem_used_max.attr,
1860         &dev_attr_idle.attr,
1861         &dev_attr_max_comp_streams.attr,
1862         &dev_attr_comp_algorithm.attr,
1863 #ifdef CONFIG_ZRAM_WRITEBACK
1864         &dev_attr_backing_dev.attr,
1865         &dev_attr_writeback.attr,
1866         &dev_attr_writeback_limit.attr,
1867         &dev_attr_writeback_limit_enable.attr,
1868 #endif
1869         &dev_attr_io_stat.attr,
1870         &dev_attr_mm_stat.attr,
1871 #ifdef CONFIG_ZRAM_WRITEBACK
1872         &dev_attr_bd_stat.attr,
1873 #endif
1874         &dev_attr_debug_stat.attr,
1875         NULL,
1876 };
1877
1878 ATTRIBUTE_GROUPS(zram_disk);
1879
1880 /*
1881  * Allocate and initialize new zram device. the function returns
1882  * '>= 0' device_id upon success, and negative value otherwise.
1883  */
1884 static int zram_add(void)
1885 {
1886         struct zram *zram;
1887         int ret, device_id;
1888
1889         zram = kzalloc(sizeof(struct zram), GFP_KERNEL);
1890         if (!zram)
1891                 return -ENOMEM;
1892
1893         ret = idr_alloc(&zram_index_idr, zram, 0, 0, GFP_KERNEL);
1894         if (ret < 0)
1895                 goto out_free_dev;
1896         device_id = ret;
1897
1898         init_rwsem(&zram->init_lock);
1899 #ifdef CONFIG_ZRAM_WRITEBACK
1900         spin_lock_init(&zram->wb_limit_lock);
1901 #endif
1902
1903         /* gendisk structure */
1904         zram->disk = blk_alloc_disk(NUMA_NO_NODE);
1905         if (!zram->disk) {
1906                 pr_err("Error allocating disk structure for device %d\n",
1907                         device_id);
1908                 ret = -ENOMEM;
1909                 goto out_free_idr;
1910         }
1911
1912         zram->disk->major = zram_major;
1913         zram->disk->first_minor = device_id;
1914         zram->disk->minors = 1;
1915         zram->disk->flags |= GENHD_FL_NO_PART;
1916         zram->disk->fops = &zram_devops;
1917         zram->disk->private_data = zram;
1918         snprintf(zram->disk->disk_name, 16, "zram%d", device_id);
1919
1920         /* Actual capacity set using syfs (/sys/block/zram<id>/disksize */
1921         set_capacity(zram->disk, 0);
1922         /* zram devices sort of resembles non-rotational disks */
1923         blk_queue_flag_set(QUEUE_FLAG_NONROT, zram->disk->queue);
1924         blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, zram->disk->queue);
1925
1926         /*
1927          * To ensure that we always get PAGE_SIZE aligned
1928          * and n*PAGE_SIZED sized I/O requests.
1929          */
1930         blk_queue_physical_block_size(zram->disk->queue, PAGE_SIZE);
1931         blk_queue_logical_block_size(zram->disk->queue,
1932                                         ZRAM_LOGICAL_BLOCK_SIZE);
1933         blk_queue_io_min(zram->disk->queue, PAGE_SIZE);
1934         blk_queue_io_opt(zram->disk->queue, PAGE_SIZE);
1935         zram->disk->queue->limits.discard_granularity = PAGE_SIZE;
1936         blk_queue_max_discard_sectors(zram->disk->queue, UINT_MAX);
1937
1938         /*
1939          * zram_bio_discard() will clear all logical blocks if logical block
1940          * size is identical with physical block size(PAGE_SIZE). But if it is
1941          * different, we will skip discarding some parts of logical blocks in
1942          * the part of the request range which isn't aligned to physical block
1943          * size.  So we can't ensure that all discarded logical blocks are
1944          * zeroed.
1945          */
1946         if (ZRAM_LOGICAL_BLOCK_SIZE == PAGE_SIZE)
1947                 blk_queue_max_write_zeroes_sectors(zram->disk->queue, UINT_MAX);
1948
1949         ret = device_add_disk(NULL, zram->disk, zram_disk_groups);
1950         if (ret)
1951                 goto out_cleanup_disk;
1952
1953         strlcpy(zram->compressor, default_compressor, sizeof(zram->compressor));
1954
1955         zram_debugfs_register(zram);
1956         pr_info("Added device: %s\n", zram->disk->disk_name);
1957         return device_id;
1958
1959 out_cleanup_disk:
1960         blk_cleanup_disk(zram->disk);
1961 out_free_idr:
1962         idr_remove(&zram_index_idr, device_id);
1963 out_free_dev:
1964         kfree(zram);
1965         return ret;
1966 }
1967
1968 static int zram_remove(struct zram *zram)
1969 {
1970         bool claimed;
1971
1972         mutex_lock(&zram->disk->open_mutex);
1973         if (disk_openers(zram->disk)) {
1974                 mutex_unlock(&zram->disk->open_mutex);
1975                 return -EBUSY;
1976         }
1977
1978         claimed = zram->claim;
1979         if (!claimed)
1980                 zram->claim = true;
1981         mutex_unlock(&zram->disk->open_mutex);
1982
1983         zram_debugfs_unregister(zram);
1984
1985         if (claimed) {
1986                 /*
1987                  * If we were claimed by reset_store(), del_gendisk() will
1988                  * wait until reset_store() is done, so nothing need to do.
1989                  */
1990                 ;
1991         } else {
1992                 /* Make sure all the pending I/O are finished */
1993                 sync_blockdev(zram->disk->part0);
1994                 zram_reset_device(zram);
1995         }
1996
1997         pr_info("Removed device: %s\n", zram->disk->disk_name);
1998
1999         del_gendisk(zram->disk);
2000
2001         /* del_gendisk drains pending reset_store */
2002         WARN_ON_ONCE(claimed && zram->claim);
2003
2004         /*
2005          * disksize_store() may be called in between zram_reset_device()
2006          * and del_gendisk(), so run the last reset to avoid leaking
2007          * anything allocated with disksize_store()
2008          */
2009         zram_reset_device(zram);
2010
2011         blk_cleanup_disk(zram->disk);
2012         kfree(zram);
2013         return 0;
2014 }
2015
2016 /* zram-control sysfs attributes */
2017
2018 /*
2019  * NOTE: hot_add attribute is not the usual read-only sysfs attribute. In a
2020  * sense that reading from this file does alter the state of your system -- it
2021  * creates a new un-initialized zram device and returns back this device's
2022  * device_id (or an error code if it fails to create a new device).
2023  */
2024 static ssize_t hot_add_show(struct class *class,
2025                         struct class_attribute *attr,
2026                         char *buf)
2027 {
2028         int ret;
2029
2030         mutex_lock(&zram_index_mutex);
2031         ret = zram_add();
2032         mutex_unlock(&zram_index_mutex);
2033
2034         if (ret < 0)
2035                 return ret;
2036         return scnprintf(buf, PAGE_SIZE, "%d\n", ret);
2037 }
2038 static struct class_attribute class_attr_hot_add =
2039         __ATTR(hot_add, 0400, hot_add_show, NULL);
2040
2041 static ssize_t hot_remove_store(struct class *class,
2042                         struct class_attribute *attr,
2043                         const char *buf,
2044                         size_t count)
2045 {
2046         struct zram *zram;
2047         int ret, dev_id;
2048
2049         /* dev_id is gendisk->first_minor, which is `int' */
2050         ret = kstrtoint(buf, 10, &dev_id);
2051         if (ret)
2052                 return ret;
2053         if (dev_id < 0)
2054                 return -EINVAL;
2055
2056         mutex_lock(&zram_index_mutex);
2057
2058         zram = idr_find(&zram_index_idr, dev_id);
2059         if (zram) {
2060                 ret = zram_remove(zram);
2061                 if (!ret)
2062                         idr_remove(&zram_index_idr, dev_id);
2063         } else {
2064                 ret = -ENODEV;
2065         }
2066
2067         mutex_unlock(&zram_index_mutex);
2068         return ret ? ret : count;
2069 }
2070 static CLASS_ATTR_WO(hot_remove);
2071
2072 static struct attribute *zram_control_class_attrs[] = {
2073         &class_attr_hot_add.attr,
2074         &class_attr_hot_remove.attr,
2075         NULL,
2076 };
2077 ATTRIBUTE_GROUPS(zram_control_class);
2078
2079 static struct class zram_control_class = {
2080         .name           = "zram-control",
2081         .owner          = THIS_MODULE,
2082         .class_groups   = zram_control_class_groups,
2083 };
2084
2085 static int zram_remove_cb(int id, void *ptr, void *data)
2086 {
2087         WARN_ON_ONCE(zram_remove(ptr));
2088         return 0;
2089 }
2090
2091 static void destroy_devices(void)
2092 {
2093         class_unregister(&zram_control_class);
2094         idr_for_each(&zram_index_idr, &zram_remove_cb, NULL);
2095         zram_debugfs_destroy();
2096         idr_destroy(&zram_index_idr);
2097         unregister_blkdev(zram_major, "zram");
2098         cpuhp_remove_multi_state(CPUHP_ZCOMP_PREPARE);
2099 }
2100
2101 static int __init zram_init(void)
2102 {
2103         int ret;
2104
2105         ret = cpuhp_setup_state_multi(CPUHP_ZCOMP_PREPARE, "block/zram:prepare",
2106                                       zcomp_cpu_up_prepare, zcomp_cpu_dead);
2107         if (ret < 0)
2108                 return ret;
2109
2110         ret = class_register(&zram_control_class);
2111         if (ret) {
2112                 pr_err("Unable to register zram-control class\n");
2113                 cpuhp_remove_multi_state(CPUHP_ZCOMP_PREPARE);
2114                 return ret;
2115         }
2116
2117         zram_debugfs_create();
2118         zram_major = register_blkdev(0, "zram");
2119         if (zram_major <= 0) {
2120                 pr_err("Unable to get major number\n");
2121                 class_unregister(&zram_control_class);
2122                 cpuhp_remove_multi_state(CPUHP_ZCOMP_PREPARE);
2123                 return -EBUSY;
2124         }
2125
2126         while (num_devices != 0) {
2127                 mutex_lock(&zram_index_mutex);
2128                 ret = zram_add();
2129                 mutex_unlock(&zram_index_mutex);
2130                 if (ret < 0)
2131                         goto out_error;
2132                 num_devices--;
2133         }
2134
2135         return 0;
2136
2137 out_error:
2138         destroy_devices();
2139         return ret;
2140 }
2141
2142 static void __exit zram_exit(void)
2143 {
2144         destroy_devices();
2145 }
2146
2147 module_init(zram_init);
2148 module_exit(zram_exit);
2149
2150 module_param(num_devices, uint, 0);
2151 MODULE_PARM_DESC(num_devices, "Number of pre-created zram devices");
2152
2153 MODULE_LICENSE("Dual BSD/GPL");
2154 MODULE_AUTHOR("Nitin Gupta <ngupta@vflare.org>");
2155 MODULE_DESCRIPTION("Compressed RAM Block Device");