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 #ifdef CONFIG_ZRAM_WRITEBACK
  56 static const struct block_device_operations zram_wb_devops;
  57 #endif
  58
  59 static void zram_free_page(struct zram *zram, size_t index);
  60 static int zram_bvec_read(struct zram *zram, struct bio_vec *bvec,
  61                                 u32 index, int offset, struct bio *bio);
  62
  63
  64 static int zram_slot_trylock(struct zram *zram, u32 index)
  65 {
  66         return bit_spin_trylock(ZRAM_LOCK, &zram->table[index].flags);
  67 }
  68
  69 static void zram_slot_lock(struct zram *zram, u32 index)
  70 {
  71         bit_spin_lock(ZRAM_LOCK, &zram->table[index].flags);
  72 }
  73
  74 static void zram_slot_unlock(struct zram *zram, u32 index)
  75 {
  76         bit_spin_unlock(ZRAM_LOCK, &zram->table[index].flags);
  77 }
  78
  79 static inline bool init_done(struct zram *zram)
  80 {
  81         return zram->disksize;
  82 }
  83
  84 static inline struct zram *dev_to_zram(struct device *dev)
  85 {
  86         return (struct zram *)dev_to_disk(dev)->private_data;
  87 }
  88
  89 static unsigned long zram_get_handle(struct zram *zram, u32 index)
  90 {
  91         return zram->table[index].handle;
  92 }
  93
  94 static void zram_set_handle(struct zram *zram, u32 index, unsigned long handle)
  95 {
  96         zram->table[index].handle = handle;
  97 }
  98
  99 /* flag operations require table entry bit_spin_lock() being held */
 100 static bool zram_test_flag(struct zram *zram, u32 index,
 101                         enum zram_pageflags flag)
 102 {
 103         return zram->table[index].flags & BIT(flag);
 104 }
 105
 106 static void zram_set_flag(struct zram *zram, u32 index,
 107                         enum zram_pageflags flag)
 108 {
 109         zram->table[index].flags |= BIT(flag);
 110 }
 111
 112 static void zram_clear_flag(struct zram *zram, u32 index,
 113                         enum zram_pageflags flag)
 114 {
 115         zram->table[index].flags &= ~BIT(flag);
 116 }
 117
 118 static inline void zram_set_element(struct zram *zram, u32 index,
 119                         unsigned long element)
 120 {
 121         zram->table[index].element = element;
 122 }
 123
 124 static unsigned long zram_get_element(struct zram *zram, u32 index)
 125 {
 126         return zram->table[index].element;
 127 }
 128
 129 static size_t zram_get_obj_size(struct zram *zram, u32 index)
 130 {
 131         return zram->table[index].flags & (BIT(ZRAM_FLAG_SHIFT) - 1);
 132 }
 133
 134 static void zram_set_obj_size(struct zram *zram,
 135                                         u32 index, size_t size)
 136 {
 137         unsigned long flags = zram->table[index].flags >> ZRAM_FLAG_SHIFT;
 138
 139         zram->table[index].flags = (flags << ZRAM_FLAG_SHIFT) | size;
 140 }
 141
 142 static inline bool zram_allocated(struct zram *zram, u32 index)
 143 {
 144         return zram_get_obj_size(zram, index) ||
 145                         zram_test_flag(zram, index, ZRAM_SAME) ||
 146                         zram_test_flag(zram, index, ZRAM_WB);
 147 }
 148
 149 #if PAGE_SIZE != 4096
 150 static inline bool is_partial_io(struct bio_vec *bvec)
 151 {
 152         return bvec->bv_len != PAGE_SIZE;
 153 }
 154 #else
 155 static inline bool is_partial_io(struct bio_vec *bvec)
 156 {
 157         return false;
 158 }
 159 #endif
 160
 161 /*
 162  * Check if request is within bounds and aligned on zram logical blocks.
 163  */
 164 static inline bool valid_io_request(struct zram *zram,
 165                 sector_t start, unsigned int size)
 166 {
 167         u64 end, bound;
 168
 169         /* unaligned request */
 170         if (unlikely(start & (ZRAM_SECTOR_PER_LOGICAL_BLOCK - 1)))
 171                 return false;
 172         if (unlikely(size & (ZRAM_LOGICAL_BLOCK_SIZE - 1)))
 173                 return false;
 174
 175         end = start + (size >> SECTOR_SHIFT);
 176         bound = zram->disksize >> SECTOR_SHIFT;
 177         /* out of range range */
 178         if (unlikely(start >= bound || end > bound || start > end))
 179                 return false;
 180
 181         /* I/O request is valid */
 182         return true;
 183 }
 184
 185 static void update_position(u32 *index, int *offset, struct bio_vec *bvec)
 186 {
 187         *index  += (*offset + bvec->bv_len) / PAGE_SIZE;
 188         *offset = (*offset + bvec->bv_len) % PAGE_SIZE;
 189 }
 190
 191 static inline void update_used_max(struct zram *zram,
 192                                         const unsigned long pages)
 193 {
 194         unsigned long old_max, cur_max;
 195
 196         old_max = atomic_long_read(&zram->stats.max_used_pages);
 197
 198         do {
 199                 cur_max = old_max;
 200                 if (pages > cur_max)
 201                         old_max = atomic_long_cmpxchg(
 202                                 &zram->stats.max_used_pages, cur_max, pages);
 203         } while (old_max != cur_max);
 204 }
 205
 206 static inline void zram_fill_page(void *ptr, unsigned long len,
 207                                         unsigned long value)
 208 {
 209         WARN_ON_ONCE(!IS_ALIGNED(len, sizeof(unsigned long)));
 210         memset_l(ptr, value, len / sizeof(unsigned long));
 211 }
 212
 213 static bool page_same_filled(void *ptr, unsigned long *element)
 214 {
 215         unsigned long *page;
 216         unsigned long val;
 217         unsigned int pos, last_pos = PAGE_SIZE / sizeof(*page) - 1;
 218
 219         page = (unsigned long *)ptr;
 220         val = page[0];
 221
 222         if (val != page[last_pos])
 223                 return false;
 224
 225         for (pos = 1; pos < last_pos; pos++) {
 226                 if (val != page[pos])
 227                         return false;
 228         }
 229
 230         *element = val;
 231
 232         return true;
 233 }
 234
 235 static ssize_t initstate_show(struct device *dev,
 236                 struct device_attribute *attr, char *buf)
 237 {
 238         u32 val;
 239         struct zram *zram = dev_to_zram(dev);
 240
 241         down_read(&zram->init_lock);
 242         val = init_done(zram);
 243         up_read(&zram->init_lock);
 244
 245         return scnprintf(buf, PAGE_SIZE, "%u\n", val);
 246 }
 247
 248 static ssize_t disksize_show(struct device *dev,
 249                 struct device_attribute *attr, char *buf)
 250 {
 251         struct zram *zram = dev_to_zram(dev);
 252
 253         return scnprintf(buf, PAGE_SIZE, "%llu\n", zram->disksize);
 254 }
 255
 256 static ssize_t mem_limit_store(struct device *dev,
 257                 struct device_attribute *attr, const char *buf, size_t len)
 258 {
 259         u64 limit;
 260         char *tmp;
 261         struct zram *zram = dev_to_zram(dev);
 262
 263         limit = memparse(buf, &tmp);
 264         if (buf == tmp) /* no chars parsed, invalid input */
 265                 return -EINVAL;
 266
 267         down_write(&zram->init_lock);
 268         zram->limit_pages = PAGE_ALIGN(limit) >> PAGE_SHIFT;
 269         up_write(&zram->init_lock);
 270
 271         return len;
 272 }
 273
 274 static ssize_t mem_used_max_store(struct device *dev,
 275                 struct device_attribute *attr, const char *buf, size_t len)
 276 {
 277         int err;
 278         unsigned long val;
 279         struct zram *zram = dev_to_zram(dev);
 280
 281         err = kstrtoul(buf, 10, &val);
 282         if (err || val != 0)
 283                 return -EINVAL;
 284
 285         down_read(&zram->init_lock);
 286         if (init_done(zram)) {
 287                 atomic_long_set(&zram->stats.max_used_pages,
 288                                 zs_get_total_pages(zram->mem_pool));
 289         }
 290         up_read(&zram->init_lock);
 291
 292         return len;
 293 }
 294
 295 /*
 296  * Mark all pages which are older than or equal to cutoff as IDLE.
 297  * Callers should hold the zram init lock in read mode
 298  */
 299 static void mark_idle(struct zram *zram, ktime_t cutoff)
 300 {
 301         int is_idle = 1;
 302         unsigned long nr_pages = zram->disksize >> PAGE_SHIFT;
 303         int index;
 304
 305         for (index = 0; index < nr_pages; index++) {
 306                 /*
 307                  * Do not mark ZRAM_UNDER_WB slot as ZRAM_IDLE to close race.
 308                  * See the comment in writeback_store.
 309                  */
 310                 zram_slot_lock(zram, index);
 311                 if (zram_allocated(zram, index) &&
 312                                 !zram_test_flag(zram, index, ZRAM_UNDER_WB)) {
 313 #ifdef CONFIG_ZRAM_MEMORY_TRACKING
 314                         is_idle = !cutoff || ktime_after(cutoff, zram->table[index].ac_time);
 315 #endif
 316                         if (is_idle)
 317                                 zram_set_flag(zram, index, ZRAM_IDLE);
 318                 }
 319                 zram_slot_unlock(zram, index);
 320         }
 321 }
 322
 323 static ssize_t idle_store(struct device *dev,
 324                 struct device_attribute *attr, const char *buf, size_t len)
 325 {
 326         struct zram *zram = dev_to_zram(dev);
 327         ktime_t cutoff_time = 0;
 328         ssize_t rv = -EINVAL;
 329
 330         if (!sysfs_streq(buf, "all")) {
 331                 /*
 332                  * If it did not parse as 'all' try to treat it as an integer when
 333                  * we have memory tracking enabled.
 334                  */
 335                 u64 age_sec;
 336
 337                 if (IS_ENABLED(CONFIG_ZRAM_MEMORY_TRACKING) && !kstrtoull(buf, 0, &age_sec))
 338                         cutoff_time = ktime_sub(ktime_get_boottime(),
 339                                         ns_to_ktime(age_sec * NSEC_PER_SEC));
 340                 else
 341                         goto out;
 342         }
 343
 344         down_read(&zram->init_lock);
 345         if (!init_done(zram))
 346                 goto out_unlock;
 347
 348         /* A cutoff_time of 0 marks everything as idle, this is the "all" behavior */
 349         mark_idle(zram, cutoff_time);
 350         rv = len;
 351
 352 out_unlock:
 353         up_read(&zram->init_lock);
 354 out:
 355         return rv;
 356 }
 357
 358 #ifdef CONFIG_ZRAM_WRITEBACK
 359 static ssize_t writeback_limit_enable_store(struct device *dev,
 360                 struct device_attribute *attr, const char *buf, size_t len)
 361 {
 362         struct zram *zram = dev_to_zram(dev);
 363         u64 val;
 364         ssize_t ret = -EINVAL;
 365
 366         if (kstrtoull(buf, 10, &val))
 367                 return ret;
 368
 369         down_read(&zram->init_lock);
 370         spin_lock(&zram->wb_limit_lock);
 371         zram->wb_limit_enable = val;
 372         spin_unlock(&zram->wb_limit_lock);
 373         up_read(&zram->init_lock);
 374         ret = len;
 375
 376         return ret;
 377 }
 378
 379 static ssize_t writeback_limit_enable_show(struct device *dev,
 380                 struct device_attribute *attr, char *buf)
 381 {
 382         bool val;
 383         struct zram *zram = dev_to_zram(dev);
 384
 385         down_read(&zram->init_lock);
 386         spin_lock(&zram->wb_limit_lock);
 387         val = zram->wb_limit_enable;
 388         spin_unlock(&zram->wb_limit_lock);
 389         up_read(&zram->init_lock);
 390
 391         return scnprintf(buf, PAGE_SIZE, "%d\n", val);
 392 }
 393
 394 static ssize_t writeback_limit_store(struct device *dev,
 395                 struct device_attribute *attr, const char *buf, size_t len)
 396 {
 397         struct zram *zram = dev_to_zram(dev);
 398         u64 val;
 399         ssize_t ret = -EINVAL;
 400
 401         if (kstrtoull(buf, 10, &val))
 402                 return ret;
 403
 404         down_read(&zram->init_lock);
 405         spin_lock(&zram->wb_limit_lock);
 406         zram->bd_wb_limit = val;
 407         spin_unlock(&zram->wb_limit_lock);
 408         up_read(&zram->init_lock);
 409         ret = len;
 410
 411         return ret;
 412 }
 413
 414 static ssize_t writeback_limit_show(struct device *dev,
 415                 struct device_attribute *attr, char *buf)
 416 {
 417         u64 val;
 418         struct zram *zram = dev_to_zram(dev);
 419
 420         down_read(&zram->init_lock);
 421         spin_lock(&zram->wb_limit_lock);
 422         val = zram->bd_wb_limit;
 423         spin_unlock(&zram->wb_limit_lock);
 424         up_read(&zram->init_lock);
 425
 426         return scnprintf(buf, PAGE_SIZE, "%llu\n", val);
 427 }
 428
 429 static void reset_bdev(struct zram *zram)
 430 {
 431         struct block_device *bdev;
 432
 433         if (!zram->backing_dev)
 434                 return;
 435
 436         bdev = zram->bdev;
 437         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 438         /* hope filp_close flush all of IO */
 439         filp_close(zram->backing_dev, NULL);
 440         zram->backing_dev = NULL;
 441         zram->bdev = NULL;
 442         zram->disk->fops = &zram_devops;
 443         kvfree(zram->bitmap);
 444         zram->bitmap = NULL;
 445 }
 446
 447 static ssize_t backing_dev_show(struct device *dev,
 448                 struct device_attribute *attr, char *buf)
 449 {
 450         struct file *file;
 451         struct zram *zram = dev_to_zram(dev);
 452         char *p;
 453         ssize_t ret;
 454
 455         down_read(&zram->init_lock);
 456         file = zram->backing_dev;
 457         if (!file) {
 458                 memcpy(buf, "none\n", 5);
 459                 up_read(&zram->init_lock);
 460                 return 5;
 461         }
 462
 463         p = file_path(file, buf, PAGE_SIZE - 1);
 464         if (IS_ERR(p)) {
 465                 ret = PTR_ERR(p);
 466                 goto out;
 467         }
 468
 469         ret = strlen(p);
 470         memmove(buf, p, ret);
 471         buf[ret++] = '\n';
 472 out:
 473         up_read(&zram->init_lock);
 474         return ret;
 475 }
 476
 477 static ssize_t backing_dev_store(struct device *dev,
 478                 struct device_attribute *attr, const char *buf, size_t len)
 479 {
 480         char *file_name;
 481         size_t sz;
 482         struct file *backing_dev = NULL;
 483         struct inode *inode;
 484         struct address_space *mapping;
 485         unsigned int bitmap_sz;
 486         unsigned long nr_pages, *bitmap = NULL;
 487         struct block_device *bdev = NULL;
 488         int err;
 489         struct zram *zram = dev_to_zram(dev);
 490
 491         file_name = kmalloc(PATH_MAX, GFP_KERNEL);
 492         if (!file_name)
 493                 return -ENOMEM;
 494
 495         down_write(&zram->init_lock);
 496         if (init_done(zram)) {
 497                 pr_info("Can't setup backing device for initialized device\n");
 498                 err = -EBUSY;
 499                 goto out;
 500         }
 501
 502         strlcpy(file_name, buf, PATH_MAX);
 503         /* ignore trailing newline */
 504         sz = strlen(file_name);
 505         if (sz > 0 && file_name[sz - 1] == '\n')
 506                 file_name[sz - 1] = 0x00;
 507
 508         backing_dev = filp_open(file_name, O_RDWR|O_LARGEFILE, 0);
 509         if (IS_ERR(backing_dev)) {
 510                 err = PTR_ERR(backing_dev);
 511                 backing_dev = NULL;
 512                 goto out;
 513         }
 514
 515         mapping = backing_dev->f_mapping;
 516         inode = mapping->host;
 517
 518         /* Support only block device in this moment */
 519         if (!S_ISBLK(inode->i_mode)) {
 520                 err = -ENOTBLK;
 521                 goto out;
 522         }
 523
 524         bdev = blkdev_get_by_dev(inode->i_rdev,
 525                         FMODE_READ | FMODE_WRITE | FMODE_EXCL, zram);
 526         if (IS_ERR(bdev)) {
 527                 err = PTR_ERR(bdev);
 528                 bdev = NULL;
 529                 goto out;
 530         }
 531
 532         nr_pages = i_size_read(inode) >> PAGE_SHIFT;
 533         bitmap_sz = BITS_TO_LONGS(nr_pages) * sizeof(long);
 534         bitmap = kvzalloc(bitmap_sz, GFP_KERNEL);
 535         if (!bitmap) {
 536                 err = -ENOMEM;
 537                 goto out;
 538         }
 539
 540         reset_bdev(zram);
 541
 542         zram->bdev = bdev;
 543         zram->backing_dev = backing_dev;
 544         zram->bitmap = bitmap;
 545         zram->nr_pages = nr_pages;
 546         /*
 547          * With writeback feature, zram does asynchronous IO so it's no longer
 548          * synchronous device so let's remove synchronous io flag. Othewise,
 549          * upper layer(e.g., swap) could wait IO completion rather than
 550          * (submit and return), which will cause system sluggish.
 551          * Furthermore, when the IO function returns(e.g., swap_readpage),
 552          * upper layer expects IO was done so it could deallocate the page
 553          * freely but in fact, IO is going on so finally could cause
 554          * use-after-free when the IO is really done.
 555          */
 556         zram->disk->fops = &zram_wb_devops;
 557         up_write(&zram->init_lock);
 558
 559         pr_info("setup backing device %s\n", file_name);
 560         kfree(file_name);
 561
 562         return len;
 563 out:
 564         kvfree(bitmap);
 565
 566         if (bdev)
 567                 blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
 568
 569         if (backing_dev)
 570                 filp_close(backing_dev, NULL);
 571
 572         up_write(&zram->init_lock);
 573
 574         kfree(file_name);
 575
 576         return err;
 577 }
 578
 579 static unsigned long alloc_block_bdev(struct zram *zram)
 580 {
 581         unsigned long blk_idx = 1;
 582 retry:
 583         /* skip 0 bit to confuse zram.handle = 0 */
 584         blk_idx = find_next_zero_bit(zram->bitmap, zram->nr_pages, blk_idx);
 585         if (blk_idx == zram->nr_pages)
 586                 return 0;
 587
 588         if (test_and_set_bit(blk_idx, zram->bitmap))
 589                 goto retry;
 590
 591         atomic64_inc(&zram->stats.bd_count);
 592         return blk_idx;
 593 }
 594
 595 static void free_block_bdev(struct zram *zram, unsigned long blk_idx)
 596 {
 597         int was_set;
 598
 599         was_set = test_and_clear_bit(blk_idx, zram->bitmap);
 600         WARN_ON_ONCE(!was_set);
 601         atomic64_dec(&zram->stats.bd_count);
 602 }
 603
 604 static void zram_page_end_io(struct bio *bio)
 605 {
 606         struct page *page = bio_first_page_all(bio);
 607
 608         page_endio(page, op_is_write(bio_op(bio)),
 609                         blk_status_to_errno(bio->bi_status));
 610         bio_put(bio);
 611 }
 612
 613 /*
 614  * Returns 1 if the submission is successful.
 615  */
 616 static int read_from_bdev_async(struct zram *zram, struct bio_vec *bvec,
 617                         unsigned long entry, struct bio *parent)
 618 {
 619         struct bio *bio;
 620
 621         bio = bio_alloc(zram->bdev, 1, parent ? parent->bi_opf : REQ_OP_READ,
 622                         GFP_NOIO);
 623         if (!bio)
 624                 return -ENOMEM;
 625
 626         bio->bi_iter.bi_sector = entry * (PAGE_SIZE >> 9);
 627         if (!bio_add_page(bio, bvec->bv_page, bvec->bv_len, bvec->bv_offset)) {
 628                 bio_put(bio);
 629                 return -EIO;
 630         }
 631
 632         if (!parent)
 633                 bio->bi_end_io = zram_page_end_io;
 634         else
 635                 bio_chain(bio, parent);
 636
 637         submit_bio(bio);
 638         return 1;
 639 }
 640
 641 #define PAGE_WB_SIG "page_index="
 642
 643 #define PAGE_WRITEBACK 0
 644 #define HUGE_WRITEBACK (1<<0)
 645 #define IDLE_WRITEBACK (1<<1)
 646
 647
 648 static ssize_t writeback_store(struct device *dev,
 649                 struct device_attribute *attr, const char *buf, size_t len)
 650 {
 651         struct zram *zram = dev_to_zram(dev);
 652         unsigned long nr_pages = zram->disksize >> PAGE_SHIFT;
 653         unsigned long index = 0;
 654         struct bio bio;
 655         struct bio_vec bio_vec;
 656         struct page *page;
 657         ssize_t ret = len;
 658         int mode, err;
 659         unsigned long blk_idx = 0;
 660
 661         if (sysfs_streq(buf, "idle"))
 662                 mode = IDLE_WRITEBACK;
 663         else if (sysfs_streq(buf, "huge"))
 664                 mode = HUGE_WRITEBACK;
 665         else if (sysfs_streq(buf, "huge_idle"))
 666                 mode = IDLE_WRITEBACK | HUGE_WRITEBACK;
 667         else {
 668                 if (strncmp(buf, PAGE_WB_SIG, sizeof(PAGE_WB_SIG) - 1))
 669                         return -EINVAL;
 670
 671                 if (kstrtol(buf + sizeof(PAGE_WB_SIG) - 1, 10, &index) ||
 672                                 index >= nr_pages)
 673                         return -EINVAL;
 674
 675                 nr_pages = 1;
 676                 mode = PAGE_WRITEBACK;
 677         }
 678
 679         down_read(&zram->init_lock);
 680         if (!init_done(zram)) {
 681                 ret = -EINVAL;
 682                 goto release_init_lock;
 683         }
 684
 685         if (!zram->backing_dev) {
 686                 ret = -ENODEV;
 687                 goto release_init_lock;
 688         }
 689
 690         page = alloc_page(GFP_KERNEL);
 691         if (!page) {
 692                 ret = -ENOMEM;
 693                 goto release_init_lock;
 694         }
 695
 696         for (; nr_pages != 0; index++, nr_pages--) {
 697                 struct bio_vec bvec;
 698
 699                 bvec.bv_page = page;
 700                 bvec.bv_len = PAGE_SIZE;
 701                 bvec.bv_offset = 0;
 702
 703                 spin_lock(&zram->wb_limit_lock);
 704                 if (zram->wb_limit_enable && !zram->bd_wb_limit) {
 705                         spin_unlock(&zram->wb_limit_lock);
 706                         ret = -EIO;
 707                         break;
 708                 }
 709                 spin_unlock(&zram->wb_limit_lock);
 710
 711                 if (!blk_idx) {
 712                         blk_idx = alloc_block_bdev(zram);
 713                         if (!blk_idx) {
 714                                 ret = -ENOSPC;
 715                                 break;
 716                         }
 717                 }
 718
 719                 zram_slot_lock(zram, index);
 720                 if (!zram_allocated(zram, index))
 721                         goto next;
 722
 723                 if (zram_test_flag(zram, index, ZRAM_WB) ||
 724                                 zram_test_flag(zram, index, ZRAM_SAME) ||
 725                                 zram_test_flag(zram, index, ZRAM_UNDER_WB))
 726                         goto next;
 727
 728                 if (mode & IDLE_WRITEBACK &&
 729                           !zram_test_flag(zram, index, ZRAM_IDLE))
 730                         goto next;
 731                 if (mode & HUGE_WRITEBACK &&
 732                           !zram_test_flag(zram, index, ZRAM_HUGE))
 733                         goto next;
 734                 /*
 735                  * Clearing ZRAM_UNDER_WB is duty of caller.
 736                  * IOW, zram_free_page never clear it.
 737                  */
 738                 zram_set_flag(zram, index, ZRAM_UNDER_WB);
 739                 /* Need for hugepage writeback racing */
 740                 zram_set_flag(zram, index, ZRAM_IDLE);
 741                 zram_slot_unlock(zram, index);
 742                 if (zram_bvec_read(zram, &bvec, index, 0, NULL)) {
 743                         zram_slot_lock(zram, index);
 744                         zram_clear_flag(zram, index, ZRAM_UNDER_WB);
 745                         zram_clear_flag(zram, index, ZRAM_IDLE);
 746                         zram_slot_unlock(zram, index);
 747                         continue;
 748                 }
 749
 750                 bio_init(&bio, zram->bdev, &bio_vec, 1,
 751                          REQ_OP_WRITE | REQ_SYNC);
 752                 bio.bi_iter.bi_sector = blk_idx * (PAGE_SIZE >> 9);
 753
 754                 bio_add_page(&bio, bvec.bv_page, bvec.bv_len,
 755                                 bvec.bv_offset);
 756                 /*
 757                  * XXX: A single page IO would be inefficient for write
 758                  * but it would be not bad as starter.
 759                  */
 760                 err = submit_bio_wait(&bio);
 761                 if (err) {
 762                         zram_slot_lock(zram, index);
 763                         zram_clear_flag(zram, index, ZRAM_UNDER_WB);
 764                         zram_clear_flag(zram, index, ZRAM_IDLE);
 765                         zram_slot_unlock(zram, index);
 766                         /*
 767                          * Return last IO error unless every IO were
 768                          * not suceeded.
 769                          */
 770                         ret = err;
 771                         continue;
 772                 }
 773
 774                 atomic64_inc(&zram->stats.bd_writes);
 775                 /*
 776                  * We released zram_slot_lock so need to check if the slot was
 777                  * changed. If there is freeing for the slot, we can catch it
 778                  * easily by zram_allocated.
 779                  * A subtle case is the slot is freed/reallocated/marked as
 780                  * ZRAM_IDLE again. To close the race, idle_store doesn't
 781                  * mark ZRAM_IDLE once it found the slot was ZRAM_UNDER_WB.
 782                  * Thus, we could close the race by checking ZRAM_IDLE bit.
 783                  */
 784                 zram_slot_lock(zram, index);
 785                 if (!zram_allocated(zram, index) ||
 786                           !zram_test_flag(zram, index, ZRAM_IDLE)) {
 787                         zram_clear_flag(zram, index, ZRAM_UNDER_WB);
 788                         zram_clear_flag(zram, index, ZRAM_IDLE);
 789                         goto next;
 790                 }
 791
 792                 zram_free_page(zram, index);
 793                 zram_clear_flag(zram, index, ZRAM_UNDER_WB);
 794                 zram_set_flag(zram, index, ZRAM_WB);
 795                 zram_set_element(zram, index, blk_idx);
 796                 blk_idx = 0;
 797                 atomic64_inc(&zram->stats.pages_stored);
 798                 spin_lock(&zram->wb_limit_lock);
 799                 if (zram->wb_limit_enable && zram->bd_wb_limit > 0)
 800                         zram->bd_wb_limit -=  1UL << (PAGE_SHIFT - 12);
 801                 spin_unlock(&zram->wb_limit_lock);
 802 next:
 803                 zram_slot_unlock(zram, index);
 804         }
 805
 806         if (blk_idx)
 807                 free_block_bdev(zram, blk_idx);
 808         __free_page(page);
 809 release_init_lock:
 810         up_read(&zram->init_lock);
 811
 812         return ret;
 813 }
 814
 815 struct zram_work {
 816         struct work_struct work;
 817         struct zram *zram;
 818         unsigned long entry;
 819         struct bio *bio;
 820         struct bio_vec bvec;
 821 };
 822
 823 #if PAGE_SIZE != 4096
 824 static void zram_sync_read(struct work_struct *work)
 825 {
 826         struct zram_work *zw = container_of(work, struct zram_work, work);
 827         struct zram *zram = zw->zram;
 828         unsigned long entry = zw->entry;
 829         struct bio *bio = zw->bio;
 830
 831         read_from_bdev_async(zram, &zw->bvec, entry, bio);
 832 }
 833
 834 /*
 835  * Block layer want one ->submit_bio to be active at a time, so if we use
 836  * chained IO with parent IO in same context, it's a deadlock. To avoid that,
 837  * use a worker thread context.
 838  */
 839 static int read_from_bdev_sync(struct zram *zram, struct bio_vec *bvec,
 840                                 unsigned long entry, struct bio *bio)
 841 {
 842         struct zram_work work;
 843
 844         work.bvec = *bvec;
 845         work.zram = zram;
 846         work.entry = entry;
 847         work.bio = bio;
 848
 849         INIT_WORK_ONSTACK(&work.work, zram_sync_read);
 850         queue_work(system_unbound_wq, &work.work);
 851         flush_work(&work.work);
 852         destroy_work_on_stack(&work.work);
 853
 854         return 1;
 855 }
 856 #else
 857 static int read_from_bdev_sync(struct zram *zram, struct bio_vec *bvec,
 858                                 unsigned long entry, struct bio *bio)
 859 {
 860         WARN_ON(1);
 861         return -EIO;
 862 }
 863 #endif
 864
 865 static int read_from_bdev(struct zram *zram, struct bio_vec *bvec,
 866                         unsigned long entry, struct bio *parent, bool sync)
 867 {
 868         atomic64_inc(&zram->stats.bd_reads);
 869         if (sync)
 870                 return read_from_bdev_sync(zram, bvec, entry, parent);
 871         else
 872                 return read_from_bdev_async(zram, bvec, entry, parent);
 873 }
 874 #else
 875 static inline void reset_bdev(struct zram *zram) {};
 876 static int read_from_bdev(struct zram *zram, struct bio_vec *bvec,
 877                         unsigned long entry, struct bio *parent, bool sync)
 878 {
 879         return -EIO;
 880 }
 881
 882 static void free_block_bdev(struct zram *zram, unsigned long blk_idx) {};
 883 #endif
 884
 885 #ifdef CONFIG_ZRAM_MEMORY_TRACKING
 886
 887 static struct dentry *zram_debugfs_root;
 888
 889 static void zram_debugfs_create(void)
 890 {
 891         zram_debugfs_root = debugfs_create_dir("zram", NULL);
 892 }
 893
 894 static void zram_debugfs_destroy(void)
 895 {
 896         debugfs_remove_recursive(zram_debugfs_root);
 897 }
 898
 899 static void zram_accessed(struct zram *zram, u32 index)
 900 {
 901         zram_clear_flag(zram, index, ZRAM_IDLE);
 902         zram->table[index].ac_time = ktime_get_boottime();
 903 }
 904
 905 static ssize_t read_block_state(struct file *file, char __user *buf,
 906                                 size_t count, loff_t *ppos)
 907 {
 908         char *kbuf;
 909         ssize_t index, written = 0;
 910         struct zram *zram = file->private_data;
 911         unsigned long nr_pages = zram->disksize >> PAGE_SHIFT;
 912         struct timespec64 ts;
 913
 914         kbuf = kvmalloc(count, GFP_KERNEL);
 915         if (!kbuf)
 916                 return -ENOMEM;
 917
 918         down_read(&zram->init_lock);
 919         if (!init_done(zram)) {
 920                 up_read(&zram->init_lock);
 921                 kvfree(kbuf);
 922                 return -EINVAL;
 923         }
 924
 925         for (index = *ppos; index < nr_pages; index++) {
 926                 int copied;
 927
 928                 zram_slot_lock(zram, index);
 929                 if (!zram_allocated(zram, index))
 930                         goto next;
 931
 932                 ts = ktime_to_timespec64(zram->table[index].ac_time);
 933                 copied = snprintf(kbuf + written, count,
 934                         "%12zd %12lld.%06lu %c%c%c%c\n",
 935                         index, (s64)ts.tv_sec,
 936                         ts.tv_nsec / NSEC_PER_USEC,
 937                         zram_test_flag(zram, index, ZRAM_SAME) ? 's' : '.',
 938                         zram_test_flag(zram, index, ZRAM_WB) ? 'w' : '.',
 939                         zram_test_flag(zram, index, ZRAM_HUGE) ? 'h' : '.',
 940                         zram_test_flag(zram, index, ZRAM_IDLE) ? 'i' : '.');
 941
 942                 if (count <= copied) {
 943                         zram_slot_unlock(zram, index);
 944                         break;
 945                 }
 946                 written += copied;
 947                 count -= copied;
 948 next:
 949                 zram_slot_unlock(zram, index);
 950                 *ppos += 1;
 951         }
 952
 953         up_read(&zram->init_lock);
 954         if (copy_to_user(buf, kbuf, written))
 955                 written = -EFAULT;
 956         kvfree(kbuf);
 957
 958         return written;
 959 }
 960
 961 static const struct file_operations proc_zram_block_state_op = {
 962         .open = simple_open,
 963         .read = read_block_state,
 964         .llseek = default_llseek,
 965 };
 966
 967 static void zram_debugfs_register(struct zram *zram)
 968 {
 969         if (!zram_debugfs_root)
 970                 return;
 971
 972         zram->debugfs_dir = debugfs_create_dir(zram->disk->disk_name,
 973                                                 zram_debugfs_root);
 974         debugfs_create_file("block_state", 0400, zram->debugfs_dir,
 975                                 zram, &proc_zram_block_state_op);
 976 }
 977
 978 static void zram_debugfs_unregister(struct zram *zram)
 979 {
 980         debugfs_remove_recursive(zram->debugfs_dir);
 981 }
 982 #else
 983 static void zram_debugfs_create(void) {};
 984 static void zram_debugfs_destroy(void) {};
 985 static void zram_accessed(struct zram *zram, u32 index)
 986 {
 987         zram_clear_flag(zram, index, ZRAM_IDLE);
 988 };
 989 static void zram_debugfs_register(struct zram *zram) {};
 990 static void zram_debugfs_unregister(struct zram *zram) {};
 991 #endif
 992
 993 /*
 994  * We switched to per-cpu streams and this attr is not needed anymore.
 995  * However, we will keep it around for some time, because:
 996  * a) we may revert per-cpu streams in the future
 997  * b) it's visible to user space and we need to follow our 2 years
 998  *    retirement rule; but we already have a number of 'soon to be
 999  *    altered' attrs, so max_comp_streams need to wait for the next
1000  *    layoff cycle.
1001  */
1002 static ssize_t max_comp_streams_show(struct device *dev,
1003                 struct device_attribute *attr, char *buf)
1004 {
1005         return scnprintf(buf, PAGE_SIZE, "%d\n", num_online_cpus());
1006 }
1007
1008 static ssize_t max_comp_streams_store(struct device *dev,
1009                 struct device_attribute *attr, const char *buf, size_t len)
1010 {
1011         return len;
1012 }
1013
1014 static ssize_t comp_algorithm_show(struct device *dev,
1015                 struct device_attribute *attr, char *buf)
1016 {
1017         size_t sz;
1018         struct zram *zram = dev_to_zram(dev);
1019
1020         down_read(&zram->init_lock);
1021         sz = zcomp_available_show(zram->compressor, buf);
1022         up_read(&zram->init_lock);
1023
1024         return sz;
1025 }
1026
1027 static ssize_t comp_algorithm_store(struct device *dev,
1028                 struct device_attribute *attr, const char *buf, size_t len)
1029 {
1030         struct zram *zram = dev_to_zram(dev);
1031         char compressor[ARRAY_SIZE(zram->compressor)];
1032         size_t sz;
1033
1034         strlcpy(compressor, buf, sizeof(compressor));
1035         /* ignore trailing newline */
1036         sz = strlen(compressor);
1037         if (sz > 0 && compressor[sz - 1] == '\n')
1038                 compressor[sz - 1] = 0x00;
1039
1040         if (!zcomp_available_algorithm(compressor))
1041                 return -EINVAL;
1042
1043         down_write(&zram->init_lock);
1044         if (init_done(zram)) {
1045                 up_write(&zram->init_lock);
1046                 pr_info("Can't change algorithm for initialized device\n");
1047                 return -EBUSY;
1048         }
1049
1050         strcpy(zram->compressor, compressor);
1051         up_write(&zram->init_lock);
1052         return len;
1053 }
1054
1055 static ssize_t compact_store(struct device *dev,
1056                 struct device_attribute *attr, const char *buf, size_t len)
1057 {
1058         struct zram *zram = dev_to_zram(dev);
1059
1060         down_read(&zram->init_lock);
1061         if (!init_done(zram)) {
1062                 up_read(&zram->init_lock);
1063                 return -EINVAL;
1064         }
1065
1066         zs_compact(zram->mem_pool);
1067         up_read(&zram->init_lock);
1068
1069         return len;
1070 }
1071
1072 static ssize_t io_stat_show(struct device *dev,
1073                 struct device_attribute *attr, char *buf)
1074 {
1075         struct zram *zram = dev_to_zram(dev);
1076         ssize_t ret;
1077
1078         down_read(&zram->init_lock);
1079         ret = scnprintf(buf, PAGE_SIZE,
1080                         "%8llu %8llu %8llu %8llu\n",
1081                         (u64)atomic64_read(&zram->stats.failed_reads),
1082                         (u64)atomic64_read(&zram->stats.failed_writes),
1083                         (u64)atomic64_read(&zram->stats.invalid_io),
1084                         (u64)atomic64_read(&zram->stats.notify_free));
1085         up_read(&zram->init_lock);
1086
1087         return ret;
1088 }
1089
1090 static ssize_t mm_stat_show(struct device *dev,
1091                 struct device_attribute *attr, char *buf)
1092 {
1093         struct zram *zram = dev_to_zram(dev);
1094         struct zs_pool_stats pool_stats;
1095         u64 orig_size, mem_used = 0;
1096         long max_used;
1097         ssize_t ret;
1098
1099         memset(&pool_stats, 0x00, sizeof(struct zs_pool_stats));
1100
1101         down_read(&zram->init_lock);
1102         if (init_done(zram)) {
1103                 mem_used = zs_get_total_pages(zram->mem_pool);
1104                 zs_pool_stats(zram->mem_pool, &pool_stats);
1105         }
1106
1107         orig_size = atomic64_read(&zram->stats.pages_stored);
1108         max_used = atomic_long_read(&zram->stats.max_used_pages);
1109
1110         ret = scnprintf(buf, PAGE_SIZE,
1111                         "%8llu %8llu %8llu %8lu %8ld %8llu %8lu %8llu %8llu\n",
1112                         orig_size << PAGE_SHIFT,
1113                         (u64)atomic64_read(&zram->stats.compr_data_size),
1114                         mem_used << PAGE_SHIFT,
1115                         zram->limit_pages << PAGE_SHIFT,
1116                         max_used << PAGE_SHIFT,
1117                         (u64)atomic64_read(&zram->stats.same_pages),
1118                         atomic_long_read(&pool_stats.pages_compacted),
1119                         (u64)atomic64_read(&zram->stats.huge_pages),
1120                         (u64)atomic64_read(&zram->stats.huge_pages_since));
1121         up_read(&zram->init_lock);
1122
1123         return ret;
1124 }
1125
1126 #ifdef CONFIG_ZRAM_WRITEBACK
1127 #define FOUR_K(x) ((x) * (1 << (PAGE_SHIFT - 12)))
1128 static ssize_t bd_stat_show(struct device *dev,
1129                 struct device_attribute *attr, char *buf)
1130 {
1131         struct zram *zram = dev_to_zram(dev);
1132         ssize_t ret;
1133
1134         down_read(&zram->init_lock);
1135         ret = scnprintf(buf, PAGE_SIZE,
1136                 "%8llu %8llu %8llu\n",
1137                         FOUR_K((u64)atomic64_read(&zram->stats.bd_count)),
1138                         FOUR_K((u64)atomic64_read(&zram->stats.bd_reads)),
1139                         FOUR_K((u64)atomic64_read(&zram->stats.bd_writes)));
1140         up_read(&zram->init_lock);
1141
1142         return ret;
1143 }
1144 #endif
1145
1146 static ssize_t debug_stat_show(struct device *dev,
1147                 struct device_attribute *attr, char *buf)
1148 {
1149         int version = 2;
1150         struct zram *zram = dev_to_zram(dev);
1151         ssize_t ret;
1152
1153         down_read(&zram->init_lock);
1154         ret = scnprintf(buf, PAGE_SIZE,
1155                         "version: %d\n%8llu\n",
1156                         version,
1157                         (u64)atomic64_read(&zram->stats.miss_free));
1158         up_read(&zram->init_lock);
1159
1160         return ret;
1161 }
1162
1163 static DEVICE_ATTR_RO(io_stat);
1164 static DEVICE_ATTR_RO(mm_stat);
1165 #ifdef CONFIG_ZRAM_WRITEBACK
1166 static DEVICE_ATTR_RO(bd_stat);
1167 #endif
1168 static DEVICE_ATTR_RO(debug_stat);
1169
1170 static void zram_meta_free(struct zram *zram, u64 disksize)
1171 {
1172         size_t num_pages = disksize >> PAGE_SHIFT;
1173         size_t index;
1174
1175         /* Free all pages that are still in this zram device */
1176         for (index = 0; index < num_pages; index++)
1177                 zram_free_page(zram, index);
1178
1179         zs_destroy_pool(zram->mem_pool);
1180         vfree(zram->table);
1181 }
1182
1183 static bool zram_meta_alloc(struct zram *zram, u64 disksize)
1184 {
1185         size_t num_pages;
1186
1187         num_pages = disksize >> PAGE_SHIFT;
1188         zram->table = vzalloc(array_size(num_pages, sizeof(*zram->table)));
1189         if (!zram->table)
1190                 return false;
1191
1192         zram->mem_pool = zs_create_pool(zram->disk->disk_name);
1193         if (!zram->mem_pool) {
1194                 vfree(zram->table);
1195                 return false;
1196         }
1197
1198         if (!huge_class_size)
1199                 huge_class_size = zs_huge_class_size(zram->mem_pool);
1200         return true;
1201 }
1202
1203 /*
1204  * To protect concurrent access to the same index entry,
1205  * caller should hold this table index entry's bit_spinlock to
1206  * indicate this index entry is accessing.
1207  */
1208 static void zram_free_page(struct zram *zram, size_t index)
1209 {
1210         unsigned long handle;
1211
1212 #ifdef CONFIG_ZRAM_MEMORY_TRACKING
1213         zram->table[index].ac_time = 0;
1214 #endif
1215         if (zram_test_flag(zram, index, ZRAM_IDLE))
1216                 zram_clear_flag(zram, index, ZRAM_IDLE);
1217
1218         if (zram_test_flag(zram, index, ZRAM_HUGE)) {
1219                 zram_clear_flag(zram, index, ZRAM_HUGE);
1220                 atomic64_dec(&zram->stats.huge_pages);
1221         }
1222
1223         if (zram_test_flag(zram, index, ZRAM_WB)) {
1224                 zram_clear_flag(zram, index, ZRAM_WB);
1225                 free_block_bdev(zram, zram_get_element(zram, index));
1226                 goto out;
1227         }
1228
1229         /*
1230          * No memory is allocated for same element filled pages.
1231          * Simply clear same page flag.
1232          */
1233         if (zram_test_flag(zram, index, ZRAM_SAME)) {
1234                 zram_clear_flag(zram, index, ZRAM_SAME);
1235                 atomic64_dec(&zram->stats.same_pages);
1236                 goto out;
1237         }
1238
1239         handle = zram_get_handle(zram, index);
1240         if (!handle)
1241                 return;
1242
1243         zs_free(zram->mem_pool, handle);
1244
1245         atomic64_sub(zram_get_obj_size(zram, index),
1246                         &zram->stats.compr_data_size);
1247 out:
1248         atomic64_dec(&zram->stats.pages_stored);
1249         zram_set_handle(zram, index, 0);
1250         zram_set_obj_size(zram, index, 0);
1251         WARN_ON_ONCE(zram->table[index].flags &
1252                 ~(1UL << ZRAM_LOCK | 1UL << ZRAM_UNDER_WB));
1253 }
1254
1255 static int __zram_bvec_read(struct zram *zram, struct page *page, u32 index,
1256                                 struct bio *bio, bool partial_io)
1257 {
1258         struct zcomp_strm *zstrm;
1259         unsigned long handle;
1260         unsigned int size;
1261         void *src, *dst;
1262         int ret;
1263
1264         zram_slot_lock(zram, index);
1265         if (zram_test_flag(zram, index, ZRAM_WB)) {
1266                 struct bio_vec bvec;
1267
1268                 zram_slot_unlock(zram, index);
1269
1270                 bvec.bv_page = page;
1271                 bvec.bv_len = PAGE_SIZE;
1272                 bvec.bv_offset = 0;
1273                 return read_from_bdev(zram, &bvec,
1274                                 zram_get_element(zram, index),
1275                                 bio, partial_io);
1276         }
1277
1278         handle = zram_get_handle(zram, index);
1279         if (!handle || zram_test_flag(zram, index, ZRAM_SAME)) {
1280                 unsigned long value;
1281                 void *mem;
1282
1283                 value = handle ? zram_get_element(zram, index) : 0;
1284                 mem = kmap_atomic(page);
1285                 zram_fill_page(mem, PAGE_SIZE, value);
1286                 kunmap_atomic(mem);
1287                 zram_slot_unlock(zram, index);
1288                 return 0;
1289         }
1290
1291         size = zram_get_obj_size(zram, index);
1292
1293         if (size != PAGE_SIZE)
1294                 zstrm = zcomp_stream_get(zram->comp);
1295
1296         src = zs_map_object(zram->mem_pool, handle, ZS_MM_RO);
1297         if (size == PAGE_SIZE) {
1298                 dst = kmap_atomic(page);
1299                 memcpy(dst, src, PAGE_SIZE);
1300                 kunmap_atomic(dst);
1301                 ret = 0;
1302         } else {
1303                 dst = kmap_atomic(page);
1304                 ret = zcomp_decompress(zstrm, src, size, dst);
1305                 kunmap_atomic(dst);
1306                 zcomp_stream_put(zram->comp);
1307         }
1308         zs_unmap_object(zram->mem_pool, handle);
1309         zram_slot_unlock(zram, index);
1310
1311         /* Should NEVER happen. Return bio error if it does. */
1312         if (WARN_ON(ret))
1313                 pr_err("Decompression failed! err=%d, page=%u\n", ret, index);
1314
1315         return ret;
1316 }
1317
1318 static int zram_bvec_read(struct zram *zram, struct bio_vec *bvec,
1319                                 u32 index, int offset, struct bio *bio)
1320 {
1321         int ret;
1322         struct page *page;
1323
1324         page = bvec->bv_page;
1325         if (is_partial_io(bvec)) {
1326                 /* Use a temporary buffer to decompress the page */
1327                 page = alloc_page(GFP_NOIO|__GFP_HIGHMEM);
1328                 if (!page)
1329                         return -ENOMEM;
1330         }
1331
1332         ret = __zram_bvec_read(zram, page, index, bio, is_partial_io(bvec));
1333         if (unlikely(ret))
1334                 goto out;
1335
1336         if (is_partial_io(bvec)) {
1337                 void *src = kmap_atomic(page);
1338
1339                 memcpy_to_bvec(bvec, src + offset);
1340                 kunmap_atomic(src);
1341         }
1342 out:
1343         if (is_partial_io(bvec))
1344                 __free_page(page);
1345
1346         return ret;
1347 }
1348
1349 static int __zram_bvec_write(struct zram *zram, struct bio_vec *bvec,
1350                                 u32 index, struct bio *bio)
1351 {
1352         int ret = 0;
1353         unsigned long alloced_pages;
1354         unsigned long handle = 0;
1355         unsigned int comp_len = 0;
1356         void *src, *dst, *mem;
1357         struct zcomp_strm *zstrm;
1358         struct page *page = bvec->bv_page;
1359         unsigned long element = 0;
1360         enum zram_pageflags flags = 0;
1361
1362         mem = kmap_atomic(page);
1363         if (page_same_filled(mem, &element)) {
1364                 kunmap_atomic(mem);
1365                 /* Free memory associated with this sector now. */
1366                 flags = ZRAM_SAME;
1367                 atomic64_inc(&zram->stats.same_pages);
1368                 goto out;
1369         }
1370         kunmap_atomic(mem);
1371
1372         zstrm = zcomp_stream_get(zram->comp);
1373         src = kmap_atomic(page);
1374         ret = zcomp_compress(zstrm, src, &comp_len);
1375         kunmap_atomic(src);
1376
1377         if (unlikely(ret)) {
1378                 zcomp_stream_put(zram->comp);
1379                 pr_err("Compression failed! err=%d\n", ret);
1380                 return ret;
1381         }
1382
1383         if (comp_len >= huge_class_size)
1384                 comp_len = PAGE_SIZE;
1385
1386         handle = zs_malloc(zram->mem_pool, comp_len,
1387                         __GFP_KSWAPD_RECLAIM |
1388                         __GFP_NOWARN |
1389                         __GFP_HIGHMEM |
1390                         __GFP_MOVABLE);
1391
1392         if (IS_ERR((void *)handle)) {
1393                 zcomp_stream_put(zram->comp);
1394                 return PTR_ERR((void *)handle);
1395         }
1396
1397         alloced_pages = zs_get_total_pages(zram->mem_pool);
1398         update_used_max(zram, alloced_pages);
1399
1400         if (zram->limit_pages && alloced_pages > zram->limit_pages) {
1401                 zcomp_stream_put(zram->comp);
1402                 zs_free(zram->mem_pool, handle);
1403                 return -ENOMEM;
1404         }
1405
1406         dst = zs_map_object(zram->mem_pool, handle, ZS_MM_WO);
1407
1408         src = zstrm->buffer;
1409         if (comp_len == PAGE_SIZE)
1410                 src = kmap_atomic(page);
1411         memcpy(dst, src, comp_len);
1412         if (comp_len == PAGE_SIZE)
1413                 kunmap_atomic(src);
1414
1415         zcomp_stream_put(zram->comp);
1416         zs_unmap_object(zram->mem_pool, handle);
1417         atomic64_add(comp_len, &zram->stats.compr_data_size);
1418 out:
1419         /*
1420          * Free memory associated with this sector
1421          * before overwriting unused sectors.
1422          */
1423         zram_slot_lock(zram, index);
1424         zram_free_page(zram, index);
1425
1426         if (comp_len == PAGE_SIZE) {
1427                 zram_set_flag(zram, index, ZRAM_HUGE);
1428                 atomic64_inc(&zram->stats.huge_pages);
1429                 atomic64_inc(&zram->stats.huge_pages_since);
1430         }
1431
1432         if (flags) {
1433                 zram_set_flag(zram, index, flags);
1434                 zram_set_element(zram, index, element);
1435         }  else {
1436                 zram_set_handle(zram, index, handle);
1437                 zram_set_obj_size(zram, index, comp_len);
1438         }
1439         zram_slot_unlock(zram, index);
1440
1441         /* Update stats */
1442         atomic64_inc(&zram->stats.pages_stored);
1443         return ret;
1444 }
1445
1446 static int zram_bvec_write(struct zram *zram, struct bio_vec *bvec,
1447                                 u32 index, int offset, struct bio *bio)
1448 {
1449         int ret;
1450         struct page *page = NULL;
1451         struct bio_vec vec;
1452
1453         vec = *bvec;
1454         if (is_partial_io(bvec)) {
1455                 void *dst;
1456                 /*
1457                  * This is a partial IO. We need to read the full page
1458                  * before to write the changes.
1459                  */
1460                 page = alloc_page(GFP_NOIO|__GFP_HIGHMEM);
1461                 if (!page)
1462                         return -ENOMEM;
1463
1464                 ret = __zram_bvec_read(zram, page, index, bio, true);
1465                 if (ret)
1466                         goto out;
1467
1468                 dst = kmap_atomic(page);
1469                 memcpy_from_bvec(dst + offset, bvec);
1470                 kunmap_atomic(dst);
1471
1472                 vec.bv_page = page;
1473                 vec.bv_len = PAGE_SIZE;
1474                 vec.bv_offset = 0;
1475         }
1476
1477         ret = __zram_bvec_write(zram, &vec, index, bio);
1478 out:
1479         if (is_partial_io(bvec))
1480                 __free_page(page);
1481         return ret;
1482 }
1483
1484 /*
1485  * zram_bio_discard - handler on discard request
1486  * @index: physical block index in PAGE_SIZE units
1487  * @offset: byte offset within physical block
1488  */
1489 static void zram_bio_discard(struct zram *zram, u32 index,
1490                              int offset, struct bio *bio)
1491 {
1492         size_t n = bio->bi_iter.bi_size;
1493
1494         /*
1495          * zram manages data in physical block size units. Because logical block
1496          * size isn't identical with physical block size on some arch, we
1497          * could get a discard request pointing to a specific offset within a
1498          * certain physical block.  Although we can handle this request by
1499          * reading that physiclal block and decompressing and partially zeroing
1500          * and re-compressing and then re-storing it, this isn't reasonable
1501          * because our intent with a discard request is to save memory.  So
1502          * skipping this logical block is appropriate here.
1503          */
1504         if (offset) {
1505                 if (n <= (PAGE_SIZE - offset))
1506                         return;
1507
1508                 n -= (PAGE_SIZE - offset);
1509                 index++;
1510         }
1511
1512         while (n >= PAGE_SIZE) {
1513                 zram_slot_lock(zram, index);
1514                 zram_free_page(zram, index);
1515                 zram_slot_unlock(zram, index);
1516                 atomic64_inc(&zram->stats.notify_free);
1517                 index++;
1518                 n -= PAGE_SIZE;
1519         }
1520 }
1521
1522 /*
1523  * Returns errno if it has some problem. Otherwise return 0 or 1.
1524  * Returns 0 if IO request was done synchronously
1525  * Returns 1 if IO request was successfully submitted.
1526  */
1527 static int zram_bvec_rw(struct zram *zram, struct bio_vec *bvec, u32 index,
1528                         int offset, enum req_op op, struct bio *bio)
1529 {
1530         int ret;
1531
1532         if (!op_is_write(op)) {
1533                 atomic64_inc(&zram->stats.num_reads);
1534                 ret = zram_bvec_read(zram, bvec, index, offset, bio);
1535                 flush_dcache_page(bvec->bv_page);
1536         } else {
1537                 atomic64_inc(&zram->stats.num_writes);
1538                 ret = zram_bvec_write(zram, bvec, index, offset, bio);
1539         }
1540
1541         zram_slot_lock(zram, index);
1542         zram_accessed(zram, index);
1543         zram_slot_unlock(zram, index);
1544
1545         if (unlikely(ret < 0)) {
1546                 if (!op_is_write(op))
1547                         atomic64_inc(&zram->stats.failed_reads);
1548                 else
1549                         atomic64_inc(&zram->stats.failed_writes);
1550         }
1551
1552         return ret;
1553 }
1554
1555 static void __zram_make_request(struct zram *zram, struct bio *bio)
1556 {
1557         int offset;
1558         u32 index;
1559         struct bio_vec bvec;
1560         struct bvec_iter iter;
1561         unsigned long start_time;
1562
1563         index = bio->bi_iter.bi_sector >> SECTORS_PER_PAGE_SHIFT;
1564         offset = (bio->bi_iter.bi_sector &
1565                   (SECTORS_PER_PAGE - 1)) << SECTOR_SHIFT;
1566
1567         switch (bio_op(bio)) {
1568         case REQ_OP_DISCARD:
1569         case REQ_OP_WRITE_ZEROES:
1570                 zram_bio_discard(zram, index, offset, bio);
1571                 bio_endio(bio);
1572                 return;
1573         default:
1574                 break;
1575         }
1576
1577         start_time = bio_start_io_acct(bio);
1578         bio_for_each_segment(bvec, bio, iter) {
1579                 struct bio_vec bv = bvec;
1580                 unsigned int unwritten = bvec.bv_len;
1581
1582                 do {
1583                         bv.bv_len = min_t(unsigned int, PAGE_SIZE - offset,
1584                                                         unwritten);
1585                         if (zram_bvec_rw(zram, &bv, index, offset,
1586                                          bio_op(bio), bio) < 0) {
1587                                 bio->bi_status = BLK_STS_IOERR;
1588                                 break;
1589                         }
1590
1591                         bv.bv_offset += bv.bv_len;
1592                         unwritten -= bv.bv_len;
1593
1594                         update_position(&index, &offset, &bv);
1595                 } while (unwritten);
1596         }
1597         bio_end_io_acct(bio, start_time);
1598         bio_endio(bio);
1599 }
1600
1601 /*
1602  * Handler function for all zram I/O requests.
1603  */
1604 static void zram_submit_bio(struct bio *bio)
1605 {
1606         struct zram *zram = bio->bi_bdev->bd_disk->private_data;
1607
1608         if (!valid_io_request(zram, bio->bi_iter.bi_sector,
1609                                         bio->bi_iter.bi_size)) {
1610                 atomic64_inc(&zram->stats.invalid_io);
1611                 bio_io_error(bio);
1612                 return;
1613         }
1614
1615         __zram_make_request(zram, bio);
1616 }
1617
1618 static void zram_slot_free_notify(struct block_device *bdev,
1619                                 unsigned long index)
1620 {
1621         struct zram *zram;
1622
1623         zram = bdev->bd_disk->private_data;
1624
1625         atomic64_inc(&zram->stats.notify_free);
1626         if (!zram_slot_trylock(zram, index)) {
1627                 atomic64_inc(&zram->stats.miss_free);
1628                 return;
1629         }
1630
1631         zram_free_page(zram, index);
1632         zram_slot_unlock(zram, index);
1633 }
1634
1635 static int zram_rw_page(struct block_device *bdev, sector_t sector,
1636                        struct page *page, enum req_op op)
1637 {
1638         int offset, ret;
1639         u32 index;
1640         struct zram *zram;
1641         struct bio_vec bv;
1642         unsigned long start_time;
1643
1644         if (PageTransHuge(page))
1645                 return -ENOTSUPP;
1646         zram = bdev->bd_disk->private_data;
1647
1648         if (!valid_io_request(zram, sector, PAGE_SIZE)) {
1649                 atomic64_inc(&zram->stats.invalid_io);
1650                 ret = -EINVAL;
1651                 goto out;
1652         }
1653
1654         index = sector >> SECTORS_PER_PAGE_SHIFT;
1655         offset = (sector & (SECTORS_PER_PAGE - 1)) << SECTOR_SHIFT;
1656
1657         bv.bv_page = page;
1658         bv.bv_len = PAGE_SIZE;
1659         bv.bv_offset = 0;
1660
1661         start_time = bdev_start_io_acct(bdev->bd_disk->part0,
1662                         SECTORS_PER_PAGE, op, jiffies);
1663         ret = zram_bvec_rw(zram, &bv, index, offset, op, NULL);
1664         bdev_end_io_acct(bdev->bd_disk->part0, op, start_time);
1665 out:
1666         /*
1667          * If I/O fails, just return error(ie, non-zero) without
1668          * calling page_endio.
1669          * It causes resubmit the I/O with bio request by upper functions
1670          * of rw_page(e.g., swap_readpage, __swap_writepage) and
1671          * bio->bi_end_io does things to handle the error
1672          * (e.g., SetPageError, set_page_dirty and extra works).
1673          */
1674         if (unlikely(ret < 0))
1675                 return ret;
1676
1677         switch (ret) {
1678         case 0:
1679                 page_endio(page, op_is_write(op), 0);
1680                 break;
1681         case 1:
1682                 ret = 0;
1683                 break;
1684         default:
1685                 WARN_ON(1);
1686         }
1687         return ret;
1688 }
1689
1690 static void zram_reset_device(struct zram *zram)
1691 {
1692         struct zcomp *comp;
1693         u64 disksize;
1694
1695         down_write(&zram->init_lock);
1696
1697         zram->limit_pages = 0;
1698
1699         if (!init_done(zram)) {
1700                 up_write(&zram->init_lock);
1701                 return;
1702         }
1703
1704         comp = zram->comp;
1705         disksize = zram->disksize;
1706         zram->disksize = 0;
1707
1708         set_capacity_and_notify(zram->disk, 0);
1709         part_stat_set_all(zram->disk->part0, 0);
1710
1711         /* I/O operation under all of CPU are done so let's free */
1712         zram_meta_free(zram, disksize);
1713         memset(&zram->stats, 0, sizeof(zram->stats));
1714         zcomp_destroy(comp);
1715         reset_bdev(zram);
1716
1717         up_write(&zram->init_lock);
1718 }
1719
1720 static ssize_t disksize_store(struct device *dev,
1721                 struct device_attribute *attr, const char *buf, size_t len)
1722 {
1723         u64 disksize;
1724         struct zcomp *comp;
1725         struct zram *zram = dev_to_zram(dev);
1726         int err;
1727
1728         disksize = memparse(buf, NULL);
1729         if (!disksize)
1730                 return -EINVAL;
1731
1732         down_write(&zram->init_lock);
1733         if (init_done(zram)) {
1734                 pr_info("Cannot change disksize for initialized device\n");
1735                 err = -EBUSY;
1736                 goto out_unlock;
1737         }
1738
1739         disksize = PAGE_ALIGN(disksize);
1740         if (!zram_meta_alloc(zram, disksize)) {
1741                 err = -ENOMEM;
1742                 goto out_unlock;
1743         }
1744
1745         comp = zcomp_create(zram->compressor);
1746         if (IS_ERR(comp)) {
1747                 pr_err("Cannot initialise %s compressing backend\n",
1748                                 zram->compressor);
1749                 err = PTR_ERR(comp);
1750                 goto out_free_meta;
1751         }
1752
1753         zram->comp = comp;
1754         zram->disksize = disksize;
1755         set_capacity_and_notify(zram->disk, zram->disksize >> SECTOR_SHIFT);
1756         up_write(&zram->init_lock);
1757
1758         return len;
1759
1760 out_free_meta:
1761         zram_meta_free(zram, disksize);
1762 out_unlock:
1763         up_write(&zram->init_lock);
1764         return err;
1765 }
1766
1767 static ssize_t reset_store(struct device *dev,
1768                 struct device_attribute *attr, const char *buf, size_t len)
1769 {
1770         int ret;
1771         unsigned short do_reset;
1772         struct zram *zram;
1773         struct gendisk *disk;
1774
1775         ret = kstrtou16(buf, 10, &do_reset);
1776         if (ret)
1777                 return ret;
1778
1779         if (!do_reset)
1780                 return -EINVAL;
1781
1782         zram = dev_to_zram(dev);
1783         disk = zram->disk;
1784
1785         mutex_lock(&disk->open_mutex);
1786         /* Do not reset an active device or claimed device */
1787         if (disk_openers(disk) || zram->claim) {
1788                 mutex_unlock(&disk->open_mutex);
1789                 return -EBUSY;
1790         }
1791
1792         /* From now on, anyone can't open /dev/zram[0-9] */
1793         zram->claim = true;
1794         mutex_unlock(&disk->open_mutex);
1795
1796         /* Make sure all the pending I/O are finished */
1797         sync_blockdev(disk->part0);
1798         zram_reset_device(zram);
1799
1800         mutex_lock(&disk->open_mutex);
1801         zram->claim = false;
1802         mutex_unlock(&disk->open_mutex);
1803
1804         return len;
1805 }
1806
1807 static int zram_open(struct block_device *bdev, fmode_t mode)
1808 {
1809         int ret = 0;
1810         struct zram *zram;
1811
1812         WARN_ON(!mutex_is_locked(&bdev->bd_disk->open_mutex));
1813
1814         zram = bdev->bd_disk->private_data;
1815         /* zram was claimed to reset so open request fails */
1816         if (zram->claim)
1817                 ret = -EBUSY;
1818
1819         return ret;
1820 }
1821
1822 static const struct block_device_operations zram_devops = {
1823         .open = zram_open,
1824         .submit_bio = zram_submit_bio,
1825         .swap_slot_free_notify = zram_slot_free_notify,
1826         .rw_page = zram_rw_page,
1827         .owner = THIS_MODULE
1828 };
1829
1830 #ifdef CONFIG_ZRAM_WRITEBACK
1831 static const struct block_device_operations zram_wb_devops = {
1832         .open = zram_open,
1833         .submit_bio = zram_submit_bio,
1834         .swap_slot_free_notify = zram_slot_free_notify,
1835         .owner = THIS_MODULE
1836 };
1837 #endif
1838
1839 static DEVICE_ATTR_WO(compact);
1840 static DEVICE_ATTR_RW(disksize);
1841 static DEVICE_ATTR_RO(initstate);
1842 static DEVICE_ATTR_WO(reset);
1843 static DEVICE_ATTR_WO(mem_limit);
1844 static DEVICE_ATTR_WO(mem_used_max);
1845 static DEVICE_ATTR_WO(idle);
1846 static DEVICE_ATTR_RW(max_comp_streams);
1847 static DEVICE_ATTR_RW(comp_algorithm);
1848 #ifdef CONFIG_ZRAM_WRITEBACK
1849 static DEVICE_ATTR_RW(backing_dev);
1850 static DEVICE_ATTR_WO(writeback);
1851 static DEVICE_ATTR_RW(writeback_limit);
1852 static DEVICE_ATTR_RW(writeback_limit_enable);
1853 #endif
1854
1855 static struct attribute *zram_disk_attrs[] = {
1856         &dev_attr_disksize.attr,
1857         &dev_attr_initstate.attr,
1858         &dev_attr_reset.attr,
1859         &dev_attr_compact.attr,
1860         &dev_attr_mem_limit.attr,
1861         &dev_attr_mem_used_max.attr,
1862         &dev_attr_idle.attr,
1863         &dev_attr_max_comp_streams.attr,
1864         &dev_attr_comp_algorithm.attr,
1865 #ifdef CONFIG_ZRAM_WRITEBACK
1866         &dev_attr_backing_dev.attr,
1867         &dev_attr_writeback.attr,
1868         &dev_attr_writeback_limit.attr,
1869         &dev_attr_writeback_limit_enable.attr,
1870 #endif
1871         &dev_attr_io_stat.attr,
1872         &dev_attr_mm_stat.attr,
1873 #ifdef CONFIG_ZRAM_WRITEBACK
1874         &dev_attr_bd_stat.attr,
1875 #endif
1876         &dev_attr_debug_stat.attr,
1877         NULL,
1878 };
1879
1880 ATTRIBUTE_GROUPS(zram_disk);
1881
1882 /*
1883  * Allocate and initialize new zram device. the function returns
1884  * '>= 0' device_id upon success, and negative value otherwise.
1885  */
1886 static int zram_add(void)
1887 {
1888         struct zram *zram;
1889         int ret, device_id;
1890
1891         zram = kzalloc(sizeof(struct zram), GFP_KERNEL);
1892         if (!zram)
1893                 return -ENOMEM;
1894
1895         ret = idr_alloc(&zram_index_idr, zram, 0, 0, GFP_KERNEL);
1896         if (ret < 0)
1897                 goto out_free_dev;
1898         device_id = ret;
1899
1900         init_rwsem(&zram->init_lock);
1901 #ifdef CONFIG_ZRAM_WRITEBACK
1902         spin_lock_init(&zram->wb_limit_lock);
1903 #endif
1904
1905         /* gendisk structure */
1906         zram->disk = blk_alloc_disk(NUMA_NO_NODE);
1907         if (!zram->disk) {
1908                 pr_err("Error allocating disk structure for device %d\n",
1909                         device_id);
1910                 ret = -ENOMEM;
1911                 goto out_free_idr;
1912         }
1913
1914         zram->disk->major = zram_major;
1915         zram->disk->first_minor = device_id;
1916         zram->disk->minors = 1;
1917         zram->disk->flags |= GENHD_FL_NO_PART;
1918         zram->disk->fops = &zram_devops;
1919         zram->disk->private_data = zram;
1920         snprintf(zram->disk->disk_name, 16, "zram%d", device_id);
1921
1922         /* Actual capacity set using syfs (/sys/block/zram<id>/disksize */
1923         set_capacity(zram->disk, 0);
1924         /* zram devices sort of resembles non-rotational disks */
1925         blk_queue_flag_set(QUEUE_FLAG_NONROT, zram->disk->queue);
1926         blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, zram->disk->queue);
1927
1928         /*
1929          * To ensure that we always get PAGE_SIZE aligned
1930          * and n*PAGE_SIZED sized I/O requests.
1931          */
1932         blk_queue_physical_block_size(zram->disk->queue, PAGE_SIZE);
1933         blk_queue_logical_block_size(zram->disk->queue,
1934                                         ZRAM_LOGICAL_BLOCK_SIZE);
1935         blk_queue_io_min(zram->disk->queue, PAGE_SIZE);
1936         blk_queue_io_opt(zram->disk->queue, PAGE_SIZE);
1937         zram->disk->queue->limits.discard_granularity = PAGE_SIZE;
1938         blk_queue_max_discard_sectors(zram->disk->queue, UINT_MAX);
1939
1940         /*
1941          * zram_bio_discard() will clear all logical blocks if logical block
1942          * size is identical with physical block size(PAGE_SIZE). But if it is
1943          * different, we will skip discarding some parts of logical blocks in
1944          * the part of the request range which isn't aligned to physical block
1945          * size.  So we can't ensure that all discarded logical blocks are
1946          * zeroed.
1947          */
1948         if (ZRAM_LOGICAL_BLOCK_SIZE == PAGE_SIZE)
1949                 blk_queue_max_write_zeroes_sectors(zram->disk->queue, UINT_MAX);
1950
1951         ret = device_add_disk(NULL, zram->disk, zram_disk_groups);
1952         if (ret)
1953                 goto out_cleanup_disk;
1954
1955         strlcpy(zram->compressor, default_compressor, sizeof(zram->compressor));
1956
1957         zram_debugfs_register(zram);
1958         pr_info("Added device: %s\n", zram->disk->disk_name);
1959         return device_id;
1960
1961 out_cleanup_disk:
1962         put_disk(zram->disk);
1963 out_free_idr:
1964         idr_remove(&zram_index_idr, device_id);
1965 out_free_dev:
1966         kfree(zram);
1967         return ret;
1968 }
1969
1970 static int zram_remove(struct zram *zram)
1971 {
1972         bool claimed;
1973
1974         mutex_lock(&zram->disk->open_mutex);
1975         if (disk_openers(zram->disk)) {
1976                 mutex_unlock(&zram->disk->open_mutex);
1977                 return -EBUSY;
1978         }
1979
1980         claimed = zram->claim;
1981         if (!claimed)
1982                 zram->claim = true;
1983         mutex_unlock(&zram->disk->open_mutex);
1984
1985         zram_debugfs_unregister(zram);
1986
1987         if (claimed) {
1988                 /*
1989                  * If we were claimed by reset_store(), del_gendisk() will
1990                  * wait until reset_store() is done, so nothing need to do.
1991                  */
1992                 ;
1993         } else {
1994                 /* Make sure all the pending I/O are finished */
1995                 sync_blockdev(zram->disk->part0);
1996                 zram_reset_device(zram);
1997         }
1998
1999         pr_info("Removed device: %s\n", zram->disk->disk_name);
2000
2001         del_gendisk(zram->disk);
2002
2003         /* del_gendisk drains pending reset_store */
2004         WARN_ON_ONCE(claimed && zram->claim);
2005
2006         /*
2007          * disksize_store() may be called in between zram_reset_device()
2008          * and del_gendisk(), so run the last reset to avoid leaking
2009          * anything allocated with disksize_store()
2010          */
2011         zram_reset_device(zram);
2012
2013         put_disk(zram->disk);
2014         kfree(zram);
2015         return 0;
2016 }
2017
2018 /* zram-control sysfs attributes */
2019
2020 /*
2021  * NOTE: hot_add attribute is not the usual read-only sysfs attribute. In a
2022  * sense that reading from this file does alter the state of your system -- it
2023  * creates a new un-initialized zram device and returns back this device's
2024  * device_id (or an error code if it fails to create a new device).
2025  */
2026 static ssize_t hot_add_show(struct class *class,
2027                         struct class_attribute *attr,
2028                         char *buf)
2029 {
2030         int ret;
2031
2032         mutex_lock(&zram_index_mutex);
2033         ret = zram_add();
2034         mutex_unlock(&zram_index_mutex);
2035
2036         if (ret < 0)
2037                 return ret;
2038         return scnprintf(buf, PAGE_SIZE, "%d\n", ret);
2039 }
2040 static struct class_attribute class_attr_hot_add =
2041         __ATTR(hot_add, 0400, hot_add_show, NULL);
2042
2043 static ssize_t hot_remove_store(struct class *class,
2044                         struct class_attribute *attr,
2045                         const char *buf,
2046                         size_t count)
2047 {
2048         struct zram *zram;
2049         int ret, dev_id;
2050
2051         /* dev_id is gendisk->first_minor, which is `int' */
2052         ret = kstrtoint(buf, 10, &dev_id);
2053         if (ret)
2054                 return ret;
2055         if (dev_id < 0)
2056                 return -EINVAL;
2057
2058         mutex_lock(&zram_index_mutex);
2059
2060         zram = idr_find(&zram_index_idr, dev_id);
2061         if (zram) {
2062                 ret = zram_remove(zram);
2063                 if (!ret)
2064                         idr_remove(&zram_index_idr, dev_id);
2065         } else {
2066                 ret = -ENODEV;
2067         }
2068
2069         mutex_unlock(&zram_index_mutex);
2070         return ret ? ret : count;
2071 }
2072 static CLASS_ATTR_WO(hot_remove);
2073
2074 static struct attribute *zram_control_class_attrs[] = {
2075         &class_attr_hot_add.attr,
2076         &class_attr_hot_remove.attr,
2077         NULL,
2078 };
2079 ATTRIBUTE_GROUPS(zram_control_class);
2080
2081 static struct class zram_control_class = {
2082         .name           = "zram-control",
2083         .owner          = THIS_MODULE,
2084         .class_groups   = zram_control_class_groups,
2085 };
2086
2087 static int zram_remove_cb(int id, void *ptr, void *data)
2088 {
2089         WARN_ON_ONCE(zram_remove(ptr));
2090         return 0;
2091 }
2092
2093 static void destroy_devices(void)
2094 {
2095         class_unregister(&zram_control_class);
2096         idr_for_each(&zram_index_idr, &zram_remove_cb, NULL);
2097         zram_debugfs_destroy();
2098         idr_destroy(&zram_index_idr);
2099         unregister_blkdev(zram_major, "zram");
2100         cpuhp_remove_multi_state(CPUHP_ZCOMP_PREPARE);
2101 }
2102
2103 static int __init zram_init(void)
2104 {
2105         int ret;
2106
2107         ret = cpuhp_setup_state_multi(CPUHP_ZCOMP_PREPARE, "block/zram:prepare",
2108                                       zcomp_cpu_up_prepare, zcomp_cpu_dead);
2109         if (ret < 0)
2110                 return ret;
2111
2112         ret = class_register(&zram_control_class);
2113         if (ret) {
2114                 pr_err("Unable to register zram-control class\n");
2115                 cpuhp_remove_multi_state(CPUHP_ZCOMP_PREPARE);
2116                 return ret;
2117         }
2118
2119         zram_debugfs_create();
2120         zram_major = register_blkdev(0, "zram");
2121         if (zram_major <= 0) {
2122                 pr_err("Unable to get major number\n");
2123                 class_unregister(&zram_control_class);
2124                 cpuhp_remove_multi_state(CPUHP_ZCOMP_PREPARE);
2125                 return -EBUSY;
2126         }
2127
2128         while (num_devices != 0) {
2129                 mutex_lock(&zram_index_mutex);
2130                 ret = zram_add();
2131                 mutex_unlock(&zram_index_mutex);
2132                 if (ret < 0)
2133                         goto out_error;
2134                 num_devices--;
2135         }
2136
2137         return 0;
2138
2139 out_error:
2140         destroy_devices();
2141         return ret;
2142 }
2143
2144 static void __exit zram_exit(void)
2145 {
2146         destroy_devices();
2147 }
2148
2149 module_init(zram_init);
2150 module_exit(zram_exit);
2151
2152 module_param(num_devices, uint, 0);
2153 MODULE_PARM_DESC(num_devices, "Number of pre-created zram devices");
2154
2155 MODULE_LICENSE("Dual BSD/GPL");
2156 MODULE_AUTHOR("Nitin Gupta <ngupta@vflare.org>");
2157 MODULE_DESCRIPTION("Compressed RAM Block Device");