drivers/block/zram/zram_drv.c

   1 /*
   2  * Compressed RAM block device
   3  *
   4  * Copyright (C) 2008, 2009, 2010  Nitin Gupta
   5  *               2012, 2013 Minchan Kim
   6  *
   7  * This code is released using a dual license strategy: BSD/GPL
   8  * You can choose the licence that better fits your requirements.
   9  *
  10  * Released under the terms of 3-clause BSD License
  11  * Released under the terms of GNU General Public License Version 2.0
  12  *
  13  */
  14
  15 #define KMSG_COMPONENT "zram"
  16 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
  17
  18 #include <linux/module.h>
  19 #include <linux/kernel.h>
  20 #include <linux/bio.h>
  21 #include <linux/bitops.h>
  22 #include <linux/blkdev.h>
  23 #include <linux/buffer_head.h>
  24 #include <linux/device.h>
  25 #include <linux/genhd.h>
  26 #include <linux/highmem.h>
  27 #include <linux/slab.h>
  28 #include <linux/backing-dev.h>
  29 #include <linux/string.h>
  30 #include <linux/vmalloc.h>
  31 #include <linux/err.h>
  32 #include <linux/idr.h>
  33 #include <linux/sysfs.h>
  34 #include <linux/debugfs.h>
  35 #include <linux/cpuhotplug.h>
  36 #include <linux/part_stat.h>
  37
  38 #include "zram_drv.h"
  39
  40 static DEFINE_IDR(zram_index_idr);
  41 /* idr index must be protected */
  42 static DEFINE_MUTEX(zram_index_mutex);
  43
  44 static int zram_major;
  45 static const char *default_compressor = CONFIG_ZRAM_DEF_COMP;
  46
  47 /* Module params (documentation at end) */
  48 static unsigned int num_devices = 1;
  49 /*
  50  * Pages that compress to sizes equals or greater than this are stored
  51  * uncompressed in memory.
  52  */
  53 static size_t huge_class_size;
  54
  55 static const struct block_device_operations zram_devops;
  56 static const struct block_device_operations zram_wb_devops;
  57
  58 static void zram_free_page(struct zram *zram, size_t index);
  59 static int zram_bvec_read(struct zram *zram, struct bio_vec *bvec,
  60                                 u32 index, int offset, struct bio *bio);
  61
  62
  63 static int zram_slot_trylock(struct zram *zram, u32 index)
  64 {
  65         return bit_spin_trylock(ZRAM_LOCK, &zram->table[index].flags);
  66 }
  67
  68 static void zram_slot_lock(struct zram *zram, u32 index)
  69 {
  70         bit_spin_lock(ZRAM_LOCK, &zram->table[index].flags);
  71 }
  72
  73 static void zram_slot_unlock(struct zram *zram, u32 index)
  74 {
  75         bit_spin_unlock(ZRAM_LOCK, &zram->table[index].flags);
  76 }
  77
  78 static inline bool init_done(struct zram *zram)
  79 {
  80         return zram->disksize;
  81 }
  82
  83 static inline struct zram *dev_to_zram(struct device *dev)
  84 {
  85         return (struct zram *)dev_to_disk(dev)->private_data;
  86 }
  87
  88 static unsigned long zram_get_handle(struct zram *zram, u32 index)
  89 {
  90         return zram->table[index].handle;
  91 }
  92
  93 static void zram_set_handle(struct zram *zram, u32 index, unsigned long handle)
  94 {
  95         zram->table[index].handle = handle;
  96 }
  97
  98 /* flag operations require table entry bit_spin_lock() being held */
  99 static bool zram_test_flag(struct zram *zram, u32 index,
 100                         enum zram_pageflags flag)
 101 {
 102         return zram->table[index].flags & BIT(flag);
 103 }
 104
 105 static void zram_set_flag(struct zram *zram, u32 index,
 106                         enum zram_pageflags flag)
 107 {
 108         zram->table[index].flags |= BIT(flag);
 109 }
 110
 111 static void zram_clear_flag(struct zram *zram, u32 index,
 112                         enum zram_pageflags flag)
 113 {
 114         zram->table[index].flags &= ~BIT(flag);
 115 }
 116
 117 static inline void zram_set_element(struct zram *zram, u32 index,
 118                         unsigned long element)
 119 {
 120         zram->table[index].element = element;
 121 }
 122
 123 static unsigned long zram_get_element(struct zram *zram, u32 index)
 124 {
 125         return zram->table[index].element;
 126 }
 127
 128 static size_t zram_get_obj_size(struct zram *zram, u32 index)
 129 {
 130         return zram->table[index].flags & (BIT(ZRAM_FLAG_SHIFT) - 1);
 131 }
 132
 133 static void zram_set_obj_size(struct zram *zram,
 134                                         u32 index, size_t size)
 135 {
 136         unsigned long flags = zram->table[index].flags >> ZRAM_FLAG_SHIFT;
 137
 138         zram->table[index].flags = (flags << ZRAM_FLAG_SHIFT) | size;
 139 }
 140
 141 static inline bool zram_allocated(struct zram *zram, u32 index)
 142 {
 143         return zram_get_obj_size(zram, index) ||
 144                         zram_test_flag(zram, index, ZRAM_SAME) ||
 145                         zram_test_flag(zram, index, ZRAM_WB);
 146 }
 147
 148 #if PAGE_SIZE != 4096
 149 static inline bool is_partial_io(struct bio_vec *bvec)
 150 {
 151         return bvec->bv_len != PAGE_SIZE;
 152 }
 153 #else
 154 static inline bool is_partial_io(struct bio_vec *bvec)
 155 {
 156         return false;
 157 }
 158 #endif
 159
 160 /*
 161  * Check if request is within bounds and aligned on zram logical blocks.
 162  */
 163 static inline bool valid_io_request(struct zram *zram,
 164                 sector_t start, unsigned int size)
 165 {
 166         u64 end, bound;
 167
 168         /* unaligned request */
 169         if (unlikely(start & (ZRAM_SECTOR_PER_LOGICAL_BLOCK - 1)))
 170                 return false;
 171         if (unlikely(size & (ZRAM_LOGICAL_BLOCK_SIZE - 1)))
 172                 return false;
 173
 174         end = start + (size >> SECTOR_SHIFT);
 175         bound = zram->disksize >> SECTOR_SHIFT;
 176         /* out of range range */
 177         if (unlikely(start >= bound || end > bound || start > end))
 178                 return false;
 179
 180         /* I/O request is valid */
 181         return true;
 182 }
 183
 184 static void update_position(u32 *index, int *offset, struct bio_vec *bvec)
 185 {
 186         *index  += (*offset + bvec->bv_len) / PAGE_SIZE;
 187         *offset = (*offset + bvec->bv_len) % PAGE_SIZE;
 188 }
 189
 190 static inline void update_used_max(struct zram *zram,
 191                                         const unsigned long pages)
 192 {
 193         unsigned long old_max, cur_max;
 194
 195         old_max = atomic_long_read(&zram->stats.max_used_pages);
 196
 197         do {
 198                 cur_max = old_max;
 199                 if (pages > cur_max)
 200                         old_max = atomic_long_cmpxchg(
 201                                 &zram->stats.max_used_pages, cur_max, pages);
 202         } while (old_max != cur_max);
 203 }
 204
 205 static inline void zram_fill_page(void *ptr, unsigned long len,
 206                                         unsigned long value)
 207 {
 208         WARN_ON_ONCE(!IS_ALIGNED(len, sizeof(unsigned long)));
 209         memset_l(ptr, value, len / sizeof(unsigned long));
 210 }
 211
 212 static bool page_same_filled(void *ptr, unsigned long *element)
 213 {
 214         unsigned long *page;
 215         unsigned long val;
 216         unsigned int pos, last_pos = PAGE_SIZE / sizeof(*page) - 1;
 217
 218         page = (unsigned long *)ptr;
 219         val = page[0];
 220
 221         if (val != page[last_pos])
 222                 return false;
 223
 224         for (pos = 1; pos < last_pos; pos++) {
 225                 if (val != page[pos])
 226                         return false;
 227         }
 228
 229         *element = val;
 230
 231         return true;
 232 }
 233
 234 static ssize_t initstate_show(struct device *dev,
 235                 struct device_attribute *attr, char *buf)
 236 {
 237         u32 val;
 238         struct zram *zram = dev_to_zram(dev);
 239
 240         down_read(&zram->init_lock);
 241         val = init_done(zram);
 242         up_read(&zram->init_lock);
 243
 244         return scnprintf(buf, PAGE_SIZE, "%u\n", val);
 245 }
 246
 247 static ssize_t disksize_show(struct device *dev,
 248                 struct device_attribute *attr, char *buf)
 249 {
 250         struct zram *zram = dev_to_zram(dev);
 251
 252         return scnprintf(buf, PAGE_SIZE, "%llu\n", zram->disksize);
 253 }
 254
 255 static ssize_t mem_limit_store(struct device *dev,
 256                 struct device_attribute *attr, const char *buf, size_t len)
 257 {
 258         u64 limit;
 259         char *tmp;
 260         struct zram *zram = dev_to_zram(dev);
 261
 262         limit = memparse(buf, &tmp);
 263         if (buf == tmp) /* no chars parsed, invalid input */
 264                 return -EINVAL;
 265
 266         down_write(&zram->init_lock);
 267         zram->limit_pages = PAGE_ALIGN(limit) >> PAGE_SHIFT;
 268         up_write(&zram->init_lock);
 269
 270         return len;
 271 }
 272
 273 static ssize_t mem_used_max_store(struct device *dev,
 274                 struct device_attribute *attr, const char *buf, size_t len)
 275 {
 276         int err;
 277         unsigned long val;
 278         struct zram *zram = dev_to_zram(dev);
 279
 280         err = kstrtoul(buf, 10, &val);
 281         if (err || val != 0)
 282                 return -EINVAL;
 283
 284         down_read(&zram->init_lock);
 285         if (init_done(zram)) {
 286                 atomic_long_set(&zram->stats.max_used_pages,
 287                                 zs_get_total_pages(zram->mem_pool));
 288         }
 289         up_read(&zram->init_lock);
 290
 291         return len;
 292 }
 293
 294 /*
 295  * Mark all pages which are older than or equal to cutoff as IDLE.
 296  * Callers should hold the zram init lock in read mode
 297  */
 298 static void mark_idle(struct zram *zram, ktime_t cutoff)
 299 {
 300         int is_idle = 1;
 301         unsigned long nr_pages = zram->disksize >> PAGE_SHIFT;
 302         int index;
 303
 304         for (index = 0; index < nr_pages; index++) {
 305                 /*
 306                  * Do not mark ZRAM_UNDER_WB slot as ZRAM_IDLE to close race.
 307                  * See the comment in writeback_store.
 308                  */
 309                 zram_slot_lock(zram, index);
 310                 if (zram_allocated(zram, index) &&
 311                                 !zram_test_flag(zram, index, ZRAM_UNDER_WB)) {
 312 #ifdef CONFIG_ZRAM_MEMORY_TRACKING
 313                         is_idle = !cutoff || ktime_after(cutoff, zram->table[index].ac_time);
 314 #endif
 315                         if (is_idle)
 316                                 zram_set_flag(zram, index, ZRAM_IDLE);
 317                 }
 318                 zram_slot_unlock(zram, index);
 319         }
 320 }
 321
 322 static ssize_t idle_store(struct device *dev,
 323                 struct device_attribute *attr, const char *buf, size_t len)
 324 {
 325         struct zram *zram = dev_to_zram(dev);
 326         ktime_t cutoff_time = 0;
 327         ssize_t rv = -EINVAL;
 328
 329         if (!sysfs_streq(buf, "all")) {
 330                 /*
 331                  * If it did not parse as 'all' try to treat it as an integer when
 332                  * we have memory tracking enabled.
 333                  */
 334                 u64 age_sec;
 335
 336                 if (IS_ENABLED(CONFIG_ZRAM_MEMORY_TRACKING) && !kstrtoull(buf, 0, &age_sec))
 337                         cutoff_time = ktime_sub(ktime_get_boottime(),
 338                                         ns_to_ktime(age_sec * NSEC_PER_SEC));
 339                 else
 340                         goto out;
 341         }
 342
 343         down_read(&zram->init_lock);
 344         if (!init_done(zram))
 345                 goto out_unlock;
 346
 347         /* A cutoff_time of 0 marks everything as idle, this is the "all" behavior */
 348         mark_idle(zram, cutoff_time);
 349         rv = len;
 350
 351 out_unlock:
 352         up_read(&zram->init_lock);
 353 out:
 354         return rv;
 355 }
 356
 357 #ifdef CONFIG_ZRAM_WRITEBACK
 358 static ssize_t writeback_limit_enable_store(struct device *dev,
 359                 struct device_attribute *attr, const char *buf, size_t len)
 360 {
 361         struct zram *zram = dev_to_zram(dev);
 362         u64 val;
 363         ssize_t ret = -EINVAL;
 364
 365         if (kstrtoull(buf, 10, &val))
 366                 return ret;
 367
 368         down_read(&zram->init_lock);
 369         spin_lock(&zram->wb_limit_lock);
 370         zram->wb_limit_enable = val;
 371         spin_unlock(&zram->wb_limit_lock);
 372         up_read(&zram->init_lock);
 373         ret = len;
 374
 375         return ret;
 376 }
 377
 378 static ssize_t writeback_limit_enable_show(struct device *dev,
 379                 struct device_attribute *attr, char *buf)
 380 {
 381         bool val;
 382         struct zram *zram = dev_to_zram(dev);
 383
 384         down_read(&zram->init_lock);
 385         spin_lock(&zram->wb_limit_lock);
 386         val = zram->wb_limit_enable;
 387         spin_unlock(&zram->wb_limit_lock);
 388         up_read(&zram->init_lock);
 389
 390         return scnprintf(buf, PAGE_SIZE, "%d\n", val);
 391 }
 392
 393 static ssize_t writeback_limit_store(struct device *dev,
 394                 struct device_attribute *attr, const char *buf, size_t len)
 395 {
 396         struct zram *zram = dev_to_zram(dev);
 397         u64 val;
 398         ssize_t ret = -EINVAL;
 399
 400         if (kstrtoull(buf, 10, &val))
 401                 return ret;
 402
 403         down_read(&zram->init_lock);
 404         spin_lock(&zram->wb_limit_lock);
 405         zram->bd_wb_limit = val;
 406         spin_unlock(&zram->wb_limit_lock);
 407         up_read(&zram->init_lock);
 408         ret = len;
 409
 410         return ret;
 411 }
 412
 413 static ssize_t writeback_limit_show(struct device *dev,
 414                 struct device_attribute *attr, char *buf)
 415 {
 416         u64 val;
 417         struct zram *zram = dev_to_zram(dev);
 418
 419         down_read(&zram->init_lock);
 420         spin_lock(&zram->wb_limit_lock);
 421         val = zram->bd_wb_limit;
 422         spin_unlock(&zram->wb_limit_lock);
 423         up_read(&zram->init_lock);
 424
 425         return scnprintf(buf, PAGE_SIZE, "%llu\n", val);
 426 }
 427
 428 static void reset_bdev(struct zram *zram)
 429 {
 430         struct block_device *bdev;
 431
 432         if (!zram->backing_dev)
 433                 return;
 434
 435         bdev = zram->bdev;
 436         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 437         /* hope filp_close flush all of IO */
 438         filp_close(zram->backing_dev, NULL);
 439         zram->backing_dev = NULL;
 440         zram->bdev = NULL;
 441         zram->disk->fops = &zram_devops;
 442         kvfree(zram->bitmap);
 443         zram->bitmap = NULL;
 444 }
 445
 446 static ssize_t backing_dev_show(struct device *dev,
 447                 struct device_attribute *attr, char *buf)
 448 {
 449         struct file *file;
 450         struct zram *zram = dev_to_zram(dev);
 451         char *p;
 452         ssize_t ret;
 453
 454         down_read(&zram->init_lock);
 455         file = zram->backing_dev;
 456         if (!file) {
 457                 memcpy(buf, "none\n", 5);
 458                 up_read(&zram->init_lock);
 459                 return 5;
 460         }
 461
 462         p = file_path(file, buf, PAGE_SIZE - 1);
 463         if (IS_ERR(p)) {
 464                 ret = PTR_ERR(p);
 465                 goto out;
 466         }
 467
 468         ret = strlen(p);
 469         memmove(buf, p, ret);
 470         buf[ret++] = '\n';
 471 out:
 472         up_read(&zram->init_lock);
 473         return ret;
 474 }
 475
 476 static ssize_t backing_dev_store(struct device *dev,
 477                 struct device_attribute *attr, const char *buf, size_t len)
 478 {
 479         char *file_name;
 480         size_t sz;
 481         struct file *backing_dev = NULL;
 482         struct inode *inode;
 483         struct address_space *mapping;
 484         unsigned int bitmap_sz;
 485         unsigned long nr_pages, *bitmap = NULL;
 486         struct block_device *bdev = NULL;
 487         int err;
 488         struct zram *zram = dev_to_zram(dev);
 489
 490         file_name = kmalloc(PATH_MAX, GFP_KERNEL);
 491         if (!file_name)
 492                 return -ENOMEM;
 493
 494         down_write(&zram->init_lock);
 495         if (init_done(zram)) {
 496                 pr_info("Can't setup backing device for initialized device\n");
 497                 err = -EBUSY;
 498                 goto out;
 499         }
 500
 501         strlcpy(file_name, buf, PATH_MAX);
 502         /* ignore trailing newline */
 503         sz = strlen(file_name);
 504         if (sz > 0 && file_name[sz - 1] == '\n')
 505                 file_name[sz - 1] = 0x00;
 506
 507         backing_dev = filp_open(file_name, O_RDWR|O_LARGEFILE, 0);
 508         if (IS_ERR(backing_dev)) {
 509                 err = PTR_ERR(backing_dev);
 510                 backing_dev = NULL;
 511                 goto out;
 512         }
 513
 514         mapping = backing_dev->f_mapping;
 515         inode = mapping->host;
 516
 517         /* Support only block device in this moment */
 518         if (!S_ISBLK(inode->i_mode)) {
 519                 err = -ENOTBLK;
 520                 goto out;
 521         }
 522
 523         bdev = blkdev_get_by_dev(inode->i_rdev,
 524                         FMODE_READ | FMODE_WRITE | FMODE_EXCL, zram);
 525         if (IS_ERR(bdev)) {
 526                 err = PTR_ERR(bdev);
 527                 bdev = NULL;
 528                 goto out;
 529         }
 530
 531         nr_pages = i_size_read(inode) >> PAGE_SHIFT;
 532         bitmap_sz = BITS_TO_LONGS(nr_pages) * sizeof(long);
 533         bitmap = kvzalloc(bitmap_sz, GFP_KERNEL);
 534         if (!bitmap) {
 535                 err = -ENOMEM;
 536                 goto out;
 537         }
 538
 539         reset_bdev(zram);
 540
 541         zram->bdev = bdev;
 542         zram->backing_dev = backing_dev;
 543         zram->bitmap = bitmap;
 544         zram->nr_pages = nr_pages;
 545         /*
 546          * With writeback feature, zram does asynchronous IO so it's no longer
 547          * synchronous device so let's remove synchronous io flag. Othewise,
 548          * upper layer(e.g., swap) could wait IO completion rather than
 549          * (submit and return), which will cause system sluggish.
 550          * Furthermore, when the IO function returns(e.g., swap_readpage),
 551          * upper layer expects IO was done so it could deallocate the page
 552          * freely but in fact, IO is going on so finally could cause
 553          * use-after-free when the IO is really done.
 554          */
 555         zram->disk->fops = &zram_wb_devops;
 556         up_write(&zram->init_lock);
 557
 558         pr_info("setup backing device %s\n", file_name);
 559         kfree(file_name);
 560
 561         return len;
 562 out:
 563         kvfree(bitmap);
 564
 565         if (bdev)
 566                 blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
 567
 568         if (backing_dev)
 569                 filp_close(backing_dev, NULL);
 570
 571         up_write(&zram->init_lock);
 572
 573         kfree(file_name);
 574
 575         return err;
 576 }
 577
 578 static unsigned long alloc_block_bdev(struct zram *zram)
 579 {
 580         unsigned long blk_idx = 1;
 581 retry:
 582         /* skip 0 bit to confuse zram.handle = 0 */
 583         blk_idx = find_next_zero_bit(zram->bitmap, zram->nr_pages, blk_idx);
 584         if (blk_idx == zram->nr_pages)
 585                 return 0;
 586
 587         if (test_and_set_bit(blk_idx, zram->bitmap))
 588                 goto retry;
 589
 590         atomic64_inc(&zram->stats.bd_count);
 591         return blk_idx;
 592 }
 593
 594 static void free_block_bdev(struct zram *zram, unsigned long blk_idx)
 595 {
 596         int was_set;
 597
 598         was_set = test_and_clear_bit(blk_idx, zram->bitmap);
 599         WARN_ON_ONCE(!was_set);
 600         atomic64_dec(&zram->stats.bd_count);
 601 }
 602
 603 static void zram_page_end_io(struct bio *bio)
 604 {
 605         struct page *page = bio_first_page_all(bio);
 606
 607         page_endio(page, op_is_write(bio_op(bio)),
 608                         blk_status_to_errno(bio->bi_status));
 609         bio_put(bio);
 610 }
 611
 612 /*
 613  * Returns 1 if the submission is successful.
 614  */
 615 static int read_from_bdev_async(struct zram *zram, struct bio_vec *bvec,
 616                         unsigned long entry, struct bio *parent)
 617 {
 618         struct bio *bio;
 619
 620         bio = bio_alloc(GFP_NOIO, 1);
 621         if (!bio)
 622                 return -ENOMEM;
 623
 624         bio->bi_iter.bi_sector = entry * (PAGE_SIZE >> 9);
 625         bio_set_dev(bio, zram->bdev);
 626         if (!bio_add_page(bio, bvec->bv_page, bvec->bv_len, bvec->bv_offset)) {
 627                 bio_put(bio);
 628                 return -EIO;
 629         }
 630
 631         if (!parent) {
 632                 bio->bi_opf = REQ_OP_READ;
 633                 bio->bi_end_io = zram_page_end_io;
 634         } else {
 635                 bio->bi_opf = parent->bi_opf;
 636                 bio_chain(bio, parent);
 637         }
 638
 639         submit_bio(bio);
 640         return 1;
 641 }
 642
 643 #define PAGE_WB_SIG "page_index="
 644
 645 #define PAGE_WRITEBACK 0
 646 #define HUGE_WRITEBACK 1
 647 #define IDLE_WRITEBACK 2
 648
 649
 650 static ssize_t writeback_store(struct device *dev,
 651                 struct device_attribute *attr, const char *buf, size_t len)
 652 {
 653         struct zram *zram = dev_to_zram(dev);
 654         unsigned long nr_pages = zram->disksize >> PAGE_SHIFT;
 655         unsigned long index = 0;
 656         struct bio bio;
 657         struct bio_vec bio_vec;
 658         struct page *page;
 659         ssize_t ret = len;
 660         int mode, err;
 661         unsigned long blk_idx = 0;
 662
 663         if (sysfs_streq(buf, "idle"))
 664                 mode = IDLE_WRITEBACK;
 665         else if (sysfs_streq(buf, "huge"))
 666                 mode = 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, &bio_vec, 1);
 751                 bio_set_dev(&bio, zram->bdev);
 752                 bio.bi_iter.bi_sector = blk_idx * (PAGE_SIZE >> 9);
 753                 bio.bi_opf = REQ_OP_WRITE | REQ_SYNC;
 754
 755                 bio_add_page(&bio, bvec.bv_page, bvec.bv_len,
 756                                 bvec.bv_offset);
 757                 /*
 758                  * XXX: A single page IO would be inefficient for write
 759                  * but it would be not bad as starter.
 760                  */
 761                 err = submit_bio_wait(&bio);
 762                 if (err) {
 763                         zram_slot_lock(zram, index);
 764                         zram_clear_flag(zram, index, ZRAM_UNDER_WB);
 765                         zram_clear_flag(zram, index, ZRAM_IDLE);
 766                         zram_slot_unlock(zram, index);
 767                         /*
 768                          * Return last IO error unless every IO were
 769                          * not suceeded.
 770                          */
 771                         ret = err;
 772                         continue;
 773                 }
 774
 775                 atomic64_inc(&zram->stats.bd_writes);
 776                 /*
 777                  * We released zram_slot_lock so need to check if the slot was
 778                  * changed. If there is freeing for the slot, we can catch it
 779                  * easily by zram_allocated.
 780                  * A subtle case is the slot is freed/reallocated/marked as
 781                  * ZRAM_IDLE again. To close the race, idle_store doesn't
 782                  * mark ZRAM_IDLE once it found the slot was ZRAM_UNDER_WB.
 783                  * Thus, we could close the race by checking ZRAM_IDLE bit.
 784                  */
 785                 zram_slot_lock(zram, index);
 786                 if (!zram_allocated(zram, index) ||
 787                           !zram_test_flag(zram, index, ZRAM_IDLE)) {
 788                         zram_clear_flag(zram, index, ZRAM_UNDER_WB);
 789                         zram_clear_flag(zram, index, ZRAM_IDLE);
 790                         goto next;
 791                 }
 792
 793                 zram_free_page(zram, index);
 794                 zram_clear_flag(zram, index, ZRAM_UNDER_WB);
 795                 zram_set_flag(zram, index, ZRAM_WB);
 796                 zram_set_element(zram, index, blk_idx);
 797                 blk_idx = 0;
 798                 atomic64_inc(&zram->stats.pages_stored);
 799                 spin_lock(&zram->wb_limit_lock);
 800                 if (zram->wb_limit_enable && zram->bd_wb_limit > 0)
 801                         zram->bd_wb_limit -=  1UL << (PAGE_SHIFT - 12);
 802                 spin_unlock(&zram->wb_limit_lock);
 803 next:
 804                 zram_slot_unlock(zram, index);
 805         }
 806
 807         if (blk_idx)
 808                 free_block_bdev(zram, blk_idx);
 809         __free_page(page);
 810 release_init_lock:
 811         up_read(&zram->init_lock);
 812
 813         return ret;
 814 }
 815
 816 struct zram_work {
 817         struct work_struct work;
 818         struct zram *zram;
 819         unsigned long entry;
 820         struct bio *bio;
 821         struct bio_vec bvec;
 822 };
 823
 824 #if PAGE_SIZE != 4096
 825 static void zram_sync_read(struct work_struct *work)
 826 {
 827         struct zram_work *zw = container_of(work, struct zram_work, work);
 828         struct zram *zram = zw->zram;
 829         unsigned long entry = zw->entry;
 830         struct bio *bio = zw->bio;
 831
 832         read_from_bdev_async(zram, &zw->bvec, entry, bio);
 833 }
 834
 835 /*
 836  * Block layer want one ->submit_bio to be active at a time, so if we use
 837  * chained IO with parent IO in same context, it's a deadlock. To avoid that,
 838  * use a worker thread context.
 839  */
 840 static int read_from_bdev_sync(struct zram *zram, struct bio_vec *bvec,
 841                                 unsigned long entry, struct bio *bio)
 842 {
 843         struct zram_work work;
 844
 845         work.bvec = *bvec;
 846         work.zram = zram;
 847         work.entry = entry;
 848         work.bio = bio;
 849
 850         INIT_WORK_ONSTACK(&work.work, zram_sync_read);
 851         queue_work(system_unbound_wq, &work.work);
 852         flush_work(&work.work);
 853         destroy_work_on_stack(&work.work);
 854
 855         return 1;
 856 }
 857 #else
 858 static int read_from_bdev_sync(struct zram *zram, struct bio_vec *bvec,
 859                                 unsigned long entry, struct bio *bio)
 860 {
 861         WARN_ON(1);
 862         return -EIO;
 863 }
 864 #endif
 865
 866 static int read_from_bdev(struct zram *zram, struct bio_vec *bvec,
 867                         unsigned long entry, struct bio *parent, bool sync)
 868 {
 869         atomic64_inc(&zram->stats.bd_reads);
 870         if (sync)
 871                 return read_from_bdev_sync(zram, bvec, entry, parent);
 872         else
 873                 return read_from_bdev_async(zram, bvec, entry, parent);
 874 }
 875 #else
 876 static inline void reset_bdev(struct zram *zram) {};
 877 static int read_from_bdev(struct zram *zram, struct bio_vec *bvec,
 878                         unsigned long entry, struct bio *parent, bool sync)
 879 {
 880         return -EIO;
 881 }
 882
 883 static void free_block_bdev(struct zram *zram, unsigned long blk_idx) {};
 884 #endif
 885
 886 #ifdef CONFIG_ZRAM_MEMORY_TRACKING
 887
 888 static struct dentry *zram_debugfs_root;
 889
 890 static void zram_debugfs_create(void)
 891 {
 892         zram_debugfs_root = debugfs_create_dir("zram", NULL);
 893 }
 894
 895 static void zram_debugfs_destroy(void)
 896 {
 897         debugfs_remove_recursive(zram_debugfs_root);
 898 }
 899
 900 static void zram_accessed(struct zram *zram, u32 index)
 901 {
 902         zram_clear_flag(zram, index, ZRAM_IDLE);
 903         zram->table[index].ac_time = ktime_get_boottime();
 904 }
 905
 906 static ssize_t read_block_state(struct file *file, char __user *buf,
 907                                 size_t count, loff_t *ppos)
 908 {
 909         char *kbuf;
 910         ssize_t index, written = 0;
 911         struct zram *zram = file->private_data;
 912         unsigned long nr_pages = zram->disksize >> PAGE_SHIFT;
 913         struct timespec64 ts;
 914
 915         kbuf = kvmalloc(count, GFP_KERNEL);
 916         if (!kbuf)
 917                 return -ENOMEM;
 918
 919         down_read(&zram->init_lock);
 920         if (!init_done(zram)) {
 921                 up_read(&zram->init_lock);
 922                 kvfree(kbuf);
 923                 return -EINVAL;
 924         }
 925
 926         for (index = *ppos; index < nr_pages; index++) {
 927                 int copied;
 928
 929                 zram_slot_lock(zram, index);
 930                 if (!zram_allocated(zram, index))
 931                         goto next;
 932
 933                 ts = ktime_to_timespec64(zram->table[index].ac_time);
 934                 copied = snprintf(kbuf + written, count,
 935                         "%12zd %12lld.%06lu %c%c%c%c\n",
 936                         index, (s64)ts.tv_sec,
 937                         ts.tv_nsec / NSEC_PER_USEC,
 938                         zram_test_flag(zram, index, ZRAM_SAME) ? 's' : '.',
 939                         zram_test_flag(zram, index, ZRAM_WB) ? 'w' : '.',
 940                         zram_test_flag(zram, index, ZRAM_HUGE) ? 'h' : '.',
 941                         zram_test_flag(zram, index, ZRAM_IDLE) ? 'i' : '.');
 942
 943                 if (count <= copied) {
 944                         zram_slot_unlock(zram, index);
 945                         break;
 946                 }
 947                 written += copied;
 948                 count -= copied;
 949 next:
 950                 zram_slot_unlock(zram, index);
 951                 *ppos += 1;
 952         }
 953
 954         up_read(&zram->init_lock);
 955         if (copy_to_user(buf, kbuf, written))
 956                 written = -EFAULT;
 957         kvfree(kbuf);
 958
 959         return written;
 960 }
 961
 962 static const struct file_operations proc_zram_block_state_op = {
 963         .open = simple_open,
 964         .read = read_block_state,
 965         .llseek = default_llseek,
 966 };
 967
 968 static void zram_debugfs_register(struct zram *zram)
 969 {
 970         if (!zram_debugfs_root)
 971                 return;
 972
 973         zram->debugfs_dir = debugfs_create_dir(zram->disk->disk_name,
 974                                                 zram_debugfs_root);
 975         debugfs_create_file("block_state", 0400, zram->debugfs_dir,
 976                                 zram, &proc_zram_block_state_op);
 977 }
 978
 979 static void zram_debugfs_unregister(struct zram *zram)
 980 {
 981         debugfs_remove_recursive(zram->debugfs_dir);
 982 }
 983 #else
 984 static void zram_debugfs_create(void) {};
 985 static void zram_debugfs_destroy(void) {};
 986 static void zram_accessed(struct zram *zram, u32 index)
 987 {
 988         zram_clear_flag(zram, index, ZRAM_IDLE);
 989 };
 990 static void zram_debugfs_register(struct zram *zram) {};
 991 static void zram_debugfs_unregister(struct zram *zram) {};
 992 #endif
 993
 994 /*
 995  * We switched to per-cpu streams and this attr is not needed anymore.
 996  * However, we will keep it around for some time, because:
 997  * a) we may revert per-cpu streams in the future
 998  * b) it's visible to user space and we need to follow our 2 years
 999  *    retirement rule; but we already have a number of 'soon to be
1000  *    altered' attrs, so max_comp_streams need to wait for the next
1001  *    layoff cycle.
1002  */
1003 static ssize_t max_comp_streams_show(struct device *dev,
1004                 struct device_attribute *attr, char *buf)
1005 {
1006         return scnprintf(buf, PAGE_SIZE, "%d\n", num_online_cpus());
1007 }
1008
1009 static ssize_t max_comp_streams_store(struct device *dev,
1010                 struct device_attribute *attr, const char *buf, size_t len)
1011 {
1012         return len;
1013 }
1014
1015 static ssize_t comp_algorithm_show(struct device *dev,
1016                 struct device_attribute *attr, char *buf)
1017 {
1018         size_t sz;
1019         struct zram *zram = dev_to_zram(dev);
1020
1021         down_read(&zram->init_lock);
1022         sz = zcomp_available_show(zram->compressor, buf);
1023         up_read(&zram->init_lock);
1024
1025         return sz;
1026 }
1027
1028 static ssize_t comp_algorithm_store(struct device *dev,
1029                 struct device_attribute *attr, const char *buf, size_t len)
1030 {
1031         struct zram *zram = dev_to_zram(dev);
1032         char compressor[ARRAY_SIZE(zram->compressor)];
1033         size_t sz;
1034
1035         strlcpy(compressor, buf, sizeof(compressor));
1036         /* ignore trailing newline */
1037         sz = strlen(compressor);
1038         if (sz > 0 && compressor[sz - 1] == '\n')
1039                 compressor[sz - 1] = 0x00;
1040
1041         if (!zcomp_available_algorithm(compressor))
1042                 return -EINVAL;
1043
1044         down_write(&zram->init_lock);
1045         if (init_done(zram)) {
1046                 up_write(&zram->init_lock);
1047                 pr_info("Can't change algorithm for initialized device\n");
1048                 return -EBUSY;
1049         }
1050
1051         strcpy(zram->compressor, compressor);
1052         up_write(&zram->init_lock);
1053         return len;
1054 }
1055
1056 static ssize_t compact_store(struct device *dev,
1057                 struct device_attribute *attr, const char *buf, size_t len)
1058 {
1059         struct zram *zram = dev_to_zram(dev);
1060
1061         down_read(&zram->init_lock);
1062         if (!init_done(zram)) {
1063                 up_read(&zram->init_lock);
1064                 return -EINVAL;
1065         }
1066
1067         zs_compact(zram->mem_pool);
1068         up_read(&zram->init_lock);
1069
1070         return len;
1071 }
1072
1073 static ssize_t io_stat_show(struct device *dev,
1074                 struct device_attribute *attr, char *buf)
1075 {
1076         struct zram *zram = dev_to_zram(dev);
1077         ssize_t ret;
1078
1079         down_read(&zram->init_lock);
1080         ret = scnprintf(buf, PAGE_SIZE,
1081                         "%8llu %8llu %8llu %8llu\n",
1082                         (u64)atomic64_read(&zram->stats.failed_reads),
1083                         (u64)atomic64_read(&zram->stats.failed_writes),
1084                         (u64)atomic64_read(&zram->stats.invalid_io),
1085                         (u64)atomic64_read(&zram->stats.notify_free));
1086         up_read(&zram->init_lock);
1087
1088         return ret;
1089 }
1090
1091 static ssize_t mm_stat_show(struct device *dev,
1092                 struct device_attribute *attr, char *buf)
1093 {
1094         struct zram *zram = dev_to_zram(dev);
1095         struct zs_pool_stats pool_stats;
1096         u64 orig_size, mem_used = 0;
1097         long max_used;
1098         ssize_t ret;
1099
1100         memset(&pool_stats, 0x00, sizeof(struct zs_pool_stats));
1101
1102         down_read(&zram->init_lock);
1103         if (init_done(zram)) {
1104                 mem_used = zs_get_total_pages(zram->mem_pool);
1105                 zs_pool_stats(zram->mem_pool, &pool_stats);
1106         }
1107
1108         orig_size = atomic64_read(&zram->stats.pages_stored);
1109         max_used = atomic_long_read(&zram->stats.max_used_pages);
1110
1111         ret = scnprintf(buf, PAGE_SIZE,
1112                         "%8llu %8llu %8llu %8lu %8ld %8llu %8lu %8llu %8llu\n",
1113                         orig_size << PAGE_SHIFT,
1114                         (u64)atomic64_read(&zram->stats.compr_data_size),
1115                         mem_used << PAGE_SHIFT,
1116                         zram->limit_pages << PAGE_SHIFT,
1117                         max_used << PAGE_SHIFT,
1118                         (u64)atomic64_read(&zram->stats.same_pages),
1119                         atomic_long_read(&pool_stats.pages_compacted),
1120                         (u64)atomic64_read(&zram->stats.huge_pages),
1121                         (u64)atomic64_read(&zram->stats.huge_pages_since));
1122         up_read(&zram->init_lock);
1123
1124         return ret;
1125 }
1126
1127 #ifdef CONFIG_ZRAM_WRITEBACK
1128 #define FOUR_K(x) ((x) * (1 << (PAGE_SHIFT - 12)))
1129 static ssize_t bd_stat_show(struct device *dev,
1130                 struct device_attribute *attr, char *buf)
1131 {
1132         struct zram *zram = dev_to_zram(dev);
1133         ssize_t ret;
1134
1135         down_read(&zram->init_lock);
1136         ret = scnprintf(buf, PAGE_SIZE,
1137                 "%8llu %8llu %8llu\n",
1138                         FOUR_K((u64)atomic64_read(&zram->stats.bd_count)),
1139                         FOUR_K((u64)atomic64_read(&zram->stats.bd_reads)),
1140                         FOUR_K((u64)atomic64_read(&zram->stats.bd_writes)));
1141         up_read(&zram->init_lock);
1142
1143         return ret;
1144 }
1145 #endif
1146
1147 static ssize_t debug_stat_show(struct device *dev,
1148                 struct device_attribute *attr, char *buf)
1149 {
1150         int version = 1;
1151         struct zram *zram = dev_to_zram(dev);
1152         ssize_t ret;
1153
1154         down_read(&zram->init_lock);
1155         ret = scnprintf(buf, PAGE_SIZE,
1156                         "version: %d\n%8llu %8llu\n",
1157                         version,
1158                         (u64)atomic64_read(&zram->stats.writestall),
1159                         (u64)atomic64_read(&zram->stats.miss_free));
1160         up_read(&zram->init_lock);
1161
1162         return ret;
1163 }
1164
1165 static DEVICE_ATTR_RO(io_stat);
1166 static DEVICE_ATTR_RO(mm_stat);
1167 #ifdef CONFIG_ZRAM_WRITEBACK
1168 static DEVICE_ATTR_RO(bd_stat);
1169 #endif
1170 static DEVICE_ATTR_RO(debug_stat);
1171
1172 static void zram_meta_free(struct zram *zram, u64 disksize)
1173 {
1174         size_t num_pages = disksize >> PAGE_SHIFT;
1175         size_t index;
1176
1177         /* Free all pages that are still in this zram device */
1178         for (index = 0; index < num_pages; index++)
1179                 zram_free_page(zram, index);
1180
1181         zs_destroy_pool(zram->mem_pool);
1182         vfree(zram->table);
1183 }
1184
1185 static bool zram_meta_alloc(struct zram *zram, u64 disksize)
1186 {
1187         size_t num_pages;
1188
1189         num_pages = disksize >> PAGE_SHIFT;
1190         zram->table = vzalloc(array_size(num_pages, sizeof(*zram->table)));
1191         if (!zram->table)
1192                 return false;
1193
1194         zram->mem_pool = zs_create_pool(zram->disk->disk_name);
1195         if (!zram->mem_pool) {
1196                 vfree(zram->table);
1197                 return false;
1198         }
1199
1200         if (!huge_class_size)
1201                 huge_class_size = zs_huge_class_size(zram->mem_pool);
1202         return true;
1203 }
1204
1205 /*
1206  * To protect concurrent access to the same index entry,
1207  * caller should hold this table index entry's bit_spinlock to
1208  * indicate this index entry is accessing.
1209  */
1210 static void zram_free_page(struct zram *zram, size_t index)
1211 {
1212         unsigned long handle;
1213
1214 #ifdef CONFIG_ZRAM_MEMORY_TRACKING
1215         zram->table[index].ac_time = 0;
1216 #endif
1217         if (zram_test_flag(zram, index, ZRAM_IDLE))
1218                 zram_clear_flag(zram, index, ZRAM_IDLE);
1219
1220         if (zram_test_flag(zram, index, ZRAM_HUGE)) {
1221                 zram_clear_flag(zram, index, ZRAM_HUGE);
1222                 atomic64_dec(&zram->stats.huge_pages);
1223         }
1224
1225         if (zram_test_flag(zram, index, ZRAM_WB)) {
1226                 zram_clear_flag(zram, index, ZRAM_WB);
1227                 free_block_bdev(zram, zram_get_element(zram, index));
1228                 goto out;
1229         }
1230
1231         /*
1232          * No memory is allocated for same element filled pages.
1233          * Simply clear same page flag.
1234          */
1235         if (zram_test_flag(zram, index, ZRAM_SAME)) {
1236                 zram_clear_flag(zram, index, ZRAM_SAME);
1237                 atomic64_dec(&zram->stats.same_pages);
1238                 goto out;
1239         }
1240
1241         handle = zram_get_handle(zram, index);
1242         if (!handle)
1243                 return;
1244
1245         zs_free(zram->mem_pool, handle);
1246
1247         atomic64_sub(zram_get_obj_size(zram, index),
1248                         &zram->stats.compr_data_size);
1249 out:
1250         atomic64_dec(&zram->stats.pages_stored);
1251         zram_set_handle(zram, index, 0);
1252         zram_set_obj_size(zram, index, 0);
1253         WARN_ON_ONCE(zram->table[index].flags &
1254                 ~(1UL << ZRAM_LOCK | 1UL << ZRAM_UNDER_WB));
1255 }
1256
1257 static int __zram_bvec_read(struct zram *zram, struct page *page, u32 index,
1258                                 struct bio *bio, bool partial_io)
1259 {
1260         struct zcomp_strm *zstrm;
1261         unsigned long handle;
1262         unsigned int size;
1263         void *src, *dst;
1264         int ret;
1265
1266         zram_slot_lock(zram, index);
1267         if (zram_test_flag(zram, index, ZRAM_WB)) {
1268                 struct bio_vec bvec;
1269
1270                 zram_slot_unlock(zram, index);
1271
1272                 bvec.bv_page = page;
1273                 bvec.bv_len = PAGE_SIZE;
1274                 bvec.bv_offset = 0;
1275                 return read_from_bdev(zram, &bvec,
1276                                 zram_get_element(zram, index),
1277                                 bio, partial_io);
1278         }
1279
1280         handle = zram_get_handle(zram, index);
1281         if (!handle || zram_test_flag(zram, index, ZRAM_SAME)) {
1282                 unsigned long value;
1283                 void *mem;
1284
1285                 value = handle ? zram_get_element(zram, index) : 0;
1286                 mem = kmap_atomic(page);
1287                 zram_fill_page(mem, PAGE_SIZE, value);
1288                 kunmap_atomic(mem);
1289                 zram_slot_unlock(zram, index);
1290                 return 0;
1291         }
1292
1293         size = zram_get_obj_size(zram, index);
1294
1295         if (size != PAGE_SIZE)
1296                 zstrm = zcomp_stream_get(zram->comp);
1297
1298         src = zs_map_object(zram->mem_pool, handle, ZS_MM_RO);
1299         if (size == PAGE_SIZE) {
1300                 dst = kmap_atomic(page);
1301                 memcpy(dst, src, PAGE_SIZE);
1302                 kunmap_atomic(dst);
1303                 ret = 0;
1304         } else {
1305                 dst = kmap_atomic(page);
1306                 ret = zcomp_decompress(zstrm, src, size, dst);
1307                 kunmap_atomic(dst);
1308                 zcomp_stream_put(zram->comp);
1309         }
1310         zs_unmap_object(zram->mem_pool, handle);
1311         zram_slot_unlock(zram, index);
1312
1313         /* Should NEVER happen. Return bio error if it does. */
1314         if (WARN_ON(ret))
1315                 pr_err("Decompression failed! err=%d, page=%u\n", ret, index);
1316
1317         return ret;
1318 }
1319
1320 static int zram_bvec_read(struct zram *zram, struct bio_vec *bvec,
1321                                 u32 index, int offset, struct bio *bio)
1322 {
1323         int ret;
1324         struct page *page;
1325
1326         page = bvec->bv_page;
1327         if (is_partial_io(bvec)) {
1328                 /* Use a temporary buffer to decompress the page */
1329                 page = alloc_page(GFP_NOIO|__GFP_HIGHMEM);
1330                 if (!page)
1331                         return -ENOMEM;
1332         }
1333
1334         ret = __zram_bvec_read(zram, page, index, bio, is_partial_io(bvec));
1335         if (unlikely(ret))
1336                 goto out;
1337
1338         if (is_partial_io(bvec)) {
1339                 void *dst = kmap_atomic(bvec->bv_page);
1340                 void *src = kmap_atomic(page);
1341
1342                 memcpy(dst + bvec->bv_offset, src + offset, bvec->bv_len);
1343                 kunmap_atomic(src);
1344                 kunmap_atomic(dst);
1345         }
1346 out:
1347         if (is_partial_io(bvec))
1348                 __free_page(page);
1349
1350         return ret;
1351 }
1352
1353 static int __zram_bvec_write(struct zram *zram, struct bio_vec *bvec,
1354                                 u32 index, struct bio *bio)
1355 {
1356         int ret = 0;
1357         unsigned long alloced_pages;
1358         unsigned long handle = 0;
1359         unsigned int comp_len = 0;
1360         void *src, *dst, *mem;
1361         struct zcomp_strm *zstrm;
1362         struct page *page = bvec->bv_page;
1363         unsigned long element = 0;
1364         enum zram_pageflags flags = 0;
1365
1366         mem = kmap_atomic(page);
1367         if (page_same_filled(mem, &element)) {
1368                 kunmap_atomic(mem);
1369                 /* Free memory associated with this sector now. */
1370                 flags = ZRAM_SAME;
1371                 atomic64_inc(&zram->stats.same_pages);
1372                 goto out;
1373         }
1374         kunmap_atomic(mem);
1375
1376 compress_again:
1377         zstrm = zcomp_stream_get(zram->comp);
1378         src = kmap_atomic(page);
1379         ret = zcomp_compress(zstrm, src, &comp_len);
1380         kunmap_atomic(src);
1381
1382         if (unlikely(ret)) {
1383                 zcomp_stream_put(zram->comp);
1384                 pr_err("Compression failed! err=%d\n", ret);
1385                 zs_free(zram->mem_pool, handle);
1386                 return ret;
1387         }
1388
1389         if (comp_len >= huge_class_size)
1390                 comp_len = PAGE_SIZE;
1391         /*
1392          * handle allocation has 2 paths:
1393          * a) fast path is executed with preemption disabled (for
1394          *  per-cpu streams) and has __GFP_DIRECT_RECLAIM bit clear,
1395          *  since we can't sleep;
1396          * b) slow path enables preemption and attempts to allocate
1397          *  the page with __GFP_DIRECT_RECLAIM bit set. we have to
1398          *  put per-cpu compression stream and, thus, to re-do
1399          *  the compression once handle is allocated.
1400          *
1401          * if we have a 'non-null' handle here then we are coming
1402          * from the slow path and handle has already been allocated.
1403          */
1404         if (!handle)
1405                 handle = zs_malloc(zram->mem_pool, comp_len,
1406                                 __GFP_KSWAPD_RECLAIM |
1407                                 __GFP_NOWARN |
1408                                 __GFP_HIGHMEM |
1409                                 __GFP_MOVABLE);
1410         if (!handle) {
1411                 zcomp_stream_put(zram->comp);
1412                 atomic64_inc(&zram->stats.writestall);
1413                 handle = zs_malloc(zram->mem_pool, comp_len,
1414                                 GFP_NOIO | __GFP_HIGHMEM |
1415                                 __GFP_MOVABLE);
1416                 if (handle)
1417                         goto compress_again;
1418                 return -ENOMEM;
1419         }
1420
1421         alloced_pages = zs_get_total_pages(zram->mem_pool);
1422         update_used_max(zram, alloced_pages);
1423
1424         if (zram->limit_pages && alloced_pages > zram->limit_pages) {
1425                 zcomp_stream_put(zram->comp);
1426                 zs_free(zram->mem_pool, handle);
1427                 return -ENOMEM;
1428         }
1429
1430         dst = zs_map_object(zram->mem_pool, handle, ZS_MM_WO);
1431
1432         src = zstrm->buffer;
1433         if (comp_len == PAGE_SIZE)
1434                 src = kmap_atomic(page);
1435         memcpy(dst, src, comp_len);
1436         if (comp_len == PAGE_SIZE)
1437                 kunmap_atomic(src);
1438
1439         zcomp_stream_put(zram->comp);
1440         zs_unmap_object(zram->mem_pool, handle);
1441         atomic64_add(comp_len, &zram->stats.compr_data_size);
1442 out:
1443         /*
1444          * Free memory associated with this sector
1445          * before overwriting unused sectors.
1446          */
1447         zram_slot_lock(zram, index);
1448         zram_free_page(zram, index);
1449
1450         if (comp_len == PAGE_SIZE) {
1451                 zram_set_flag(zram, index, ZRAM_HUGE);
1452                 atomic64_inc(&zram->stats.huge_pages);
1453                 atomic64_inc(&zram->stats.huge_pages_since);
1454         }
1455
1456         if (flags) {
1457                 zram_set_flag(zram, index, flags);
1458                 zram_set_element(zram, index, element);
1459         }  else {
1460                 zram_set_handle(zram, index, handle);
1461                 zram_set_obj_size(zram, index, comp_len);
1462         }
1463         zram_slot_unlock(zram, index);
1464
1465         /* Update stats */
1466         atomic64_inc(&zram->stats.pages_stored);
1467         return ret;
1468 }
1469
1470 static int zram_bvec_write(struct zram *zram, struct bio_vec *bvec,
1471                                 u32 index, int offset, struct bio *bio)
1472 {
1473         int ret;
1474         struct page *page = NULL;
1475         void *src;
1476         struct bio_vec vec;
1477
1478         vec = *bvec;
1479         if (is_partial_io(bvec)) {
1480                 void *dst;
1481                 /*
1482                  * This is a partial IO. We need to read the full page
1483                  * before to write the changes.
1484                  */
1485                 page = alloc_page(GFP_NOIO|__GFP_HIGHMEM);
1486                 if (!page)
1487                         return -ENOMEM;
1488
1489                 ret = __zram_bvec_read(zram, page, index, bio, true);
1490                 if (ret)
1491                         goto out;
1492
1493                 src = kmap_atomic(bvec->bv_page);
1494                 dst = kmap_atomic(page);
1495                 memcpy(dst + offset, src + bvec->bv_offset, bvec->bv_len);
1496                 kunmap_atomic(dst);
1497                 kunmap_atomic(src);
1498
1499                 vec.bv_page = page;
1500                 vec.bv_len = PAGE_SIZE;
1501                 vec.bv_offset = 0;
1502         }
1503
1504         ret = __zram_bvec_write(zram, &vec, index, bio);
1505 out:
1506         if (is_partial_io(bvec))
1507                 __free_page(page);
1508         return ret;
1509 }
1510
1511 /*
1512  * zram_bio_discard - handler on discard request
1513  * @index: physical block index in PAGE_SIZE units
1514  * @offset: byte offset within physical block
1515  */
1516 static void zram_bio_discard(struct zram *zram, u32 index,
1517                              int offset, struct bio *bio)
1518 {
1519         size_t n = bio->bi_iter.bi_size;
1520
1521         /*
1522          * zram manages data in physical block size units. Because logical block
1523          * size isn't identical with physical block size on some arch, we
1524          * could get a discard request pointing to a specific offset within a
1525          * certain physical block.  Although we can handle this request by
1526          * reading that physiclal block and decompressing and partially zeroing
1527          * and re-compressing and then re-storing it, this isn't reasonable
1528          * because our intent with a discard request is to save memory.  So
1529          * skipping this logical block is appropriate here.
1530          */
1531         if (offset) {
1532                 if (n <= (PAGE_SIZE - offset))
1533                         return;
1534
1535                 n -= (PAGE_SIZE - offset);
1536                 index++;
1537         }
1538
1539         while (n >= PAGE_SIZE) {
1540                 zram_slot_lock(zram, index);
1541                 zram_free_page(zram, index);
1542                 zram_slot_unlock(zram, index);
1543                 atomic64_inc(&zram->stats.notify_free);
1544                 index++;
1545                 n -= PAGE_SIZE;
1546         }
1547 }
1548
1549 /*
1550  * Returns errno if it has some problem. Otherwise return 0 or 1.
1551  * Returns 0 if IO request was done synchronously
1552  * Returns 1 if IO request was successfully submitted.
1553  */
1554 static int zram_bvec_rw(struct zram *zram, struct bio_vec *bvec, u32 index,
1555                         int offset, unsigned int op, struct bio *bio)
1556 {
1557         int ret;
1558
1559         if (!op_is_write(op)) {
1560                 atomic64_inc(&zram->stats.num_reads);
1561                 ret = zram_bvec_read(zram, bvec, index, offset, bio);
1562                 flush_dcache_page(bvec->bv_page);
1563         } else {
1564                 atomic64_inc(&zram->stats.num_writes);
1565                 ret = zram_bvec_write(zram, bvec, index, offset, bio);
1566         }
1567
1568         zram_slot_lock(zram, index);
1569         zram_accessed(zram, index);
1570         zram_slot_unlock(zram, index);
1571
1572         if (unlikely(ret < 0)) {
1573                 if (!op_is_write(op))
1574                         atomic64_inc(&zram->stats.failed_reads);
1575                 else
1576                         atomic64_inc(&zram->stats.failed_writes);
1577         }
1578
1579         return ret;
1580 }
1581
1582 static void __zram_make_request(struct zram *zram, struct bio *bio)
1583 {
1584         int offset;
1585         u32 index;
1586         struct bio_vec bvec;
1587         struct bvec_iter iter;
1588         unsigned long start_time;
1589
1590         index = bio->bi_iter.bi_sector >> SECTORS_PER_PAGE_SHIFT;
1591         offset = (bio->bi_iter.bi_sector &
1592                   (SECTORS_PER_PAGE - 1)) << SECTOR_SHIFT;
1593
1594         switch (bio_op(bio)) {
1595         case REQ_OP_DISCARD:
1596         case REQ_OP_WRITE_ZEROES:
1597                 zram_bio_discard(zram, index, offset, bio);
1598                 bio_endio(bio);
1599                 return;
1600         default:
1601                 break;
1602         }
1603
1604         start_time = bio_start_io_acct(bio);
1605         bio_for_each_segment(bvec, bio, iter) {
1606                 struct bio_vec bv = bvec;
1607                 unsigned int unwritten = bvec.bv_len;
1608
1609                 do {
1610                         bv.bv_len = min_t(unsigned int, PAGE_SIZE - offset,
1611                                                         unwritten);
1612                         if (zram_bvec_rw(zram, &bv, index, offset,
1613                                          bio_op(bio), bio) < 0) {
1614                                 bio->bi_status = BLK_STS_IOERR;
1615                                 break;
1616                         }
1617
1618                         bv.bv_offset += bv.bv_len;
1619                         unwritten -= bv.bv_len;
1620
1621                         update_position(&index, &offset, &bv);
1622                 } while (unwritten);
1623         }
1624         bio_end_io_acct(bio, start_time);
1625         bio_endio(bio);
1626 }
1627
1628 /*
1629  * Handler function for all zram I/O requests.
1630  */
1631 static void zram_submit_bio(struct bio *bio)
1632 {
1633         struct zram *zram = bio->bi_bdev->bd_disk->private_data;
1634
1635         if (!valid_io_request(zram, bio->bi_iter.bi_sector,
1636                                         bio->bi_iter.bi_size)) {
1637                 atomic64_inc(&zram->stats.invalid_io);
1638                 bio_io_error(bio);
1639                 return;
1640         }
1641
1642         __zram_make_request(zram, bio);
1643 }
1644
1645 static void zram_slot_free_notify(struct block_device *bdev,
1646                                 unsigned long index)
1647 {
1648         struct zram *zram;
1649
1650         zram = bdev->bd_disk->private_data;
1651
1652         atomic64_inc(&zram->stats.notify_free);
1653         if (!zram_slot_trylock(zram, index)) {
1654                 atomic64_inc(&zram->stats.miss_free);
1655                 return;
1656         }
1657
1658         zram_free_page(zram, index);
1659         zram_slot_unlock(zram, index);
1660 }
1661
1662 static int zram_rw_page(struct block_device *bdev, sector_t sector,
1663                        struct page *page, unsigned int op)
1664 {
1665         int offset, ret;
1666         u32 index;
1667         struct zram *zram;
1668         struct bio_vec bv;
1669         unsigned long start_time;
1670
1671         if (PageTransHuge(page))
1672                 return -ENOTSUPP;
1673         zram = bdev->bd_disk->private_data;
1674
1675         if (!valid_io_request(zram, sector, PAGE_SIZE)) {
1676                 atomic64_inc(&zram->stats.invalid_io);
1677                 ret = -EINVAL;
1678                 goto out;
1679         }
1680
1681         index = sector >> SECTORS_PER_PAGE_SHIFT;
1682         offset = (sector & (SECTORS_PER_PAGE - 1)) << SECTOR_SHIFT;
1683
1684         bv.bv_page = page;
1685         bv.bv_len = PAGE_SIZE;
1686         bv.bv_offset = 0;
1687
1688         start_time = disk_start_io_acct(bdev->bd_disk, SECTORS_PER_PAGE, op);
1689         ret = zram_bvec_rw(zram, &bv, index, offset, op, NULL);
1690         disk_end_io_acct(bdev->bd_disk, op, start_time);
1691 out:
1692         /*
1693          * If I/O fails, just return error(ie, non-zero) without
1694          * calling page_endio.
1695          * It causes resubmit the I/O with bio request by upper functions
1696          * of rw_page(e.g., swap_readpage, __swap_writepage) and
1697          * bio->bi_end_io does things to handle the error
1698          * (e.g., SetPageError, set_page_dirty and extra works).
1699          */
1700         if (unlikely(ret < 0))
1701                 return ret;
1702
1703         switch (ret) {
1704         case 0:
1705                 page_endio(page, op_is_write(op), 0);
1706                 break;
1707         case 1:
1708                 ret = 0;
1709                 break;
1710         default:
1711                 WARN_ON(1);
1712         }
1713         return ret;
1714 }
1715
1716 static void zram_reset_device(struct zram *zram)
1717 {
1718         struct zcomp *comp;
1719         u64 disksize;
1720
1721         down_write(&zram->init_lock);
1722
1723         zram->limit_pages = 0;
1724
1725         if (!init_done(zram)) {
1726                 up_write(&zram->init_lock);
1727                 return;
1728         }
1729
1730         comp = zram->comp;
1731         disksize = zram->disksize;
1732         zram->disksize = 0;
1733
1734         set_capacity_and_notify(zram->disk, 0);
1735         part_stat_set_all(zram->disk->part0, 0);
1736
1737         /* I/O operation under all of CPU are done so let's free */
1738         zram_meta_free(zram, disksize);
1739         memset(&zram->stats, 0, sizeof(zram->stats));
1740         zcomp_destroy(comp);
1741         reset_bdev(zram);
1742
1743         up_write(&zram->init_lock);
1744 }
1745
1746 static ssize_t disksize_store(struct device *dev,
1747                 struct device_attribute *attr, const char *buf, size_t len)
1748 {
1749         u64 disksize;
1750         struct zcomp *comp;
1751         struct zram *zram = dev_to_zram(dev);
1752         int err;
1753
1754         disksize = memparse(buf, NULL);
1755         if (!disksize)
1756                 return -EINVAL;
1757
1758         down_write(&zram->init_lock);
1759         if (init_done(zram)) {
1760                 pr_info("Cannot change disksize for initialized device\n");
1761                 err = -EBUSY;
1762                 goto out_unlock;
1763         }
1764
1765         disksize = PAGE_ALIGN(disksize);
1766         if (!zram_meta_alloc(zram, disksize)) {
1767                 err = -ENOMEM;
1768                 goto out_unlock;
1769         }
1770
1771         comp = zcomp_create(zram->compressor);
1772         if (IS_ERR(comp)) {
1773                 pr_err("Cannot initialise %s compressing backend\n",
1774                                 zram->compressor);
1775                 err = PTR_ERR(comp);
1776                 goto out_free_meta;
1777         }
1778
1779         zram->comp = comp;
1780         zram->disksize = disksize;
1781         set_capacity_and_notify(zram->disk, zram->disksize >> SECTOR_SHIFT);
1782         up_write(&zram->init_lock);
1783
1784         return len;
1785
1786 out_free_meta:
1787         zram_meta_free(zram, disksize);
1788 out_unlock:
1789         up_write(&zram->init_lock);
1790         return err;
1791 }
1792
1793 static ssize_t reset_store(struct device *dev,
1794                 struct device_attribute *attr, const char *buf, size_t len)
1795 {
1796         int ret;
1797         unsigned short do_reset;
1798         struct zram *zram;
1799         struct block_device *bdev;
1800
1801         ret = kstrtou16(buf, 10, &do_reset);
1802         if (ret)
1803                 return ret;
1804
1805         if (!do_reset)
1806                 return -EINVAL;
1807
1808         zram = dev_to_zram(dev);
1809         bdev = zram->disk->part0;
1810
1811         mutex_lock(&bdev->bd_disk->open_mutex);
1812         /* Do not reset an active device or claimed device */
1813         if (bdev->bd_openers || zram->claim) {
1814                 mutex_unlock(&bdev->bd_disk->open_mutex);
1815                 return -EBUSY;
1816         }
1817
1818         /* From now on, anyone can't open /dev/zram[0-9] */
1819         zram->claim = true;
1820         mutex_unlock(&bdev->bd_disk->open_mutex);
1821
1822         /* Make sure all the pending I/O are finished */
1823         sync_blockdev(bdev);
1824         zram_reset_device(zram);
1825
1826         mutex_lock(&bdev->bd_disk->open_mutex);
1827         zram->claim = false;
1828         mutex_unlock(&bdev->bd_disk->open_mutex);
1829
1830         return len;
1831 }
1832
1833 static int zram_open(struct block_device *bdev, fmode_t mode)
1834 {
1835         int ret = 0;
1836         struct zram *zram;
1837
1838         WARN_ON(!mutex_is_locked(&bdev->bd_disk->open_mutex));
1839
1840         zram = bdev->bd_disk->private_data;
1841         /* zram was claimed to reset so open request fails */
1842         if (zram->claim)
1843                 ret = -EBUSY;
1844
1845         return ret;
1846 }
1847
1848 static const struct block_device_operations zram_devops = {
1849         .open = zram_open,
1850         .submit_bio = zram_submit_bio,
1851         .swap_slot_free_notify = zram_slot_free_notify,
1852         .rw_page = zram_rw_page,
1853         .owner = THIS_MODULE
1854 };
1855
1856 static const struct block_device_operations zram_wb_devops = {
1857         .open = zram_open,
1858         .submit_bio = zram_submit_bio,
1859         .swap_slot_free_notify = zram_slot_free_notify,
1860         .owner = THIS_MODULE
1861 };
1862
1863 static DEVICE_ATTR_WO(compact);
1864 static DEVICE_ATTR_RW(disksize);
1865 static DEVICE_ATTR_RO(initstate);
1866 static DEVICE_ATTR_WO(reset);
1867 static DEVICE_ATTR_WO(mem_limit);
1868 static DEVICE_ATTR_WO(mem_used_max);
1869 static DEVICE_ATTR_WO(idle);
1870 static DEVICE_ATTR_RW(max_comp_streams);
1871 static DEVICE_ATTR_RW(comp_algorithm);
1872 #ifdef CONFIG_ZRAM_WRITEBACK
1873 static DEVICE_ATTR_RW(backing_dev);
1874 static DEVICE_ATTR_WO(writeback);
1875 static DEVICE_ATTR_RW(writeback_limit);
1876 static DEVICE_ATTR_RW(writeback_limit_enable);
1877 #endif
1878
1879 static struct attribute *zram_disk_attrs[] = {
1880         &dev_attr_disksize.attr,
1881         &dev_attr_initstate.attr,
1882         &dev_attr_reset.attr,
1883         &dev_attr_compact.attr,
1884         &dev_attr_mem_limit.attr,
1885         &dev_attr_mem_used_max.attr,
1886         &dev_attr_idle.attr,
1887         &dev_attr_max_comp_streams.attr,
1888         &dev_attr_comp_algorithm.attr,
1889 #ifdef CONFIG_ZRAM_WRITEBACK
1890         &dev_attr_backing_dev.attr,
1891         &dev_attr_writeback.attr,
1892         &dev_attr_writeback_limit.attr,
1893         &dev_attr_writeback_limit_enable.attr,
1894 #endif
1895         &dev_attr_io_stat.attr,
1896         &dev_attr_mm_stat.attr,
1897 #ifdef CONFIG_ZRAM_WRITEBACK
1898         &dev_attr_bd_stat.attr,
1899 #endif
1900         &dev_attr_debug_stat.attr,
1901         NULL,
1902 };
1903
1904 static const struct attribute_group zram_disk_attr_group = {
1905         .attrs = zram_disk_attrs,
1906 };
1907
1908 static const struct attribute_group *zram_disk_attr_groups[] = {
1909         &zram_disk_attr_group,
1910         NULL,
1911 };
1912
1913 /*
1914  * Allocate and initialize new zram device. the function returns
1915  * '>= 0' device_id upon success, and negative value otherwise.
1916  */
1917 static int zram_add(void)
1918 {
1919         struct zram *zram;
1920         int ret, device_id;
1921
1922         zram = kzalloc(sizeof(struct zram), GFP_KERNEL);
1923         if (!zram)
1924                 return -ENOMEM;
1925
1926         ret = idr_alloc(&zram_index_idr, zram, 0, 0, GFP_KERNEL);
1927         if (ret < 0)
1928                 goto out_free_dev;
1929         device_id = ret;
1930
1931         init_rwsem(&zram->init_lock);
1932 #ifdef CONFIG_ZRAM_WRITEBACK
1933         spin_lock_init(&zram->wb_limit_lock);
1934 #endif
1935
1936         /* gendisk structure */
1937         zram->disk = blk_alloc_disk(NUMA_NO_NODE);
1938         if (!zram->disk) {
1939                 pr_err("Error allocating disk structure for device %d\n",
1940                         device_id);
1941                 ret = -ENOMEM;
1942                 goto out_free_idr;
1943         }
1944
1945         zram->disk->major = zram_major;
1946         zram->disk->first_minor = device_id;
1947         zram->disk->minors = 1;
1948         zram->disk->fops = &zram_devops;
1949         zram->disk->private_data = zram;
1950         snprintf(zram->disk->disk_name, 16, "zram%d", device_id);
1951
1952         /* Actual capacity set using syfs (/sys/block/zram<id>/disksize */
1953         set_capacity(zram->disk, 0);
1954         /* zram devices sort of resembles non-rotational disks */
1955         blk_queue_flag_set(QUEUE_FLAG_NONROT, zram->disk->queue);
1956         blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, zram->disk->queue);
1957
1958         /*
1959          * To ensure that we always get PAGE_SIZE aligned
1960          * and n*PAGE_SIZED sized I/O requests.
1961          */
1962         blk_queue_physical_block_size(zram->disk->queue, PAGE_SIZE);
1963         blk_queue_logical_block_size(zram->disk->queue,
1964                                         ZRAM_LOGICAL_BLOCK_SIZE);
1965         blk_queue_io_min(zram->disk->queue, PAGE_SIZE);
1966         blk_queue_io_opt(zram->disk->queue, PAGE_SIZE);
1967         zram->disk->queue->limits.discard_granularity = PAGE_SIZE;
1968         blk_queue_max_discard_sectors(zram->disk->queue, UINT_MAX);
1969         blk_queue_flag_set(QUEUE_FLAG_DISCARD, zram->disk->queue);
1970
1971         /*
1972          * zram_bio_discard() will clear all logical blocks if logical block
1973          * size is identical with physical block size(PAGE_SIZE). But if it is
1974          * different, we will skip discarding some parts of logical blocks in
1975          * the part of the request range which isn't aligned to physical block
1976          * size.  So we can't ensure that all discarded logical blocks are
1977          * zeroed.
1978          */
1979         if (ZRAM_LOGICAL_BLOCK_SIZE == PAGE_SIZE)
1980                 blk_queue_max_write_zeroes_sectors(zram->disk->queue, UINT_MAX);
1981
1982         blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, zram->disk->queue);
1983         ret = device_add_disk(NULL, zram->disk, zram_disk_attr_groups);
1984         if (ret)
1985                 goto out_cleanup_disk;
1986
1987         strlcpy(zram->compressor, default_compressor, sizeof(zram->compressor));
1988
1989         zram_debugfs_register(zram);
1990         pr_info("Added device: %s\n", zram->disk->disk_name);
1991         return device_id;
1992
1993 out_cleanup_disk:
1994         blk_cleanup_disk(zram->disk);
1995 out_free_idr:
1996         idr_remove(&zram_index_idr, device_id);
1997 out_free_dev:
1998         kfree(zram);
1999         return ret;
2000 }
2001
2002 static int zram_remove(struct zram *zram)
2003 {
2004         struct block_device *bdev = zram->disk->part0;
2005         bool claimed;
2006
2007         mutex_lock(&bdev->bd_disk->open_mutex);
2008         if (bdev->bd_openers) {
2009                 mutex_unlock(&bdev->bd_disk->open_mutex);
2010                 return -EBUSY;
2011         }
2012
2013         claimed = zram->claim;
2014         if (!claimed)
2015                 zram->claim = true;
2016         mutex_unlock(&bdev->bd_disk->open_mutex);
2017
2018         zram_debugfs_unregister(zram);
2019
2020         if (claimed) {
2021                 /*
2022                  * If we were claimed by reset_store(), del_gendisk() will
2023                  * wait until reset_store() is done, so nothing need to do.
2024                  */
2025                 ;
2026         } else {
2027                 /* Make sure all the pending I/O are finished */
2028                 sync_blockdev(bdev);
2029                 zram_reset_device(zram);
2030         }
2031
2032         pr_info("Removed device: %s\n", zram->disk->disk_name);
2033
2034         del_gendisk(zram->disk);
2035
2036         /* del_gendisk drains pending reset_store */
2037         WARN_ON_ONCE(claimed && zram->claim);
2038
2039         /*
2040          * disksize_store() may be called in between zram_reset_device()
2041          * and del_gendisk(), so run the last reset to avoid leaking
2042          * anything allocated with disksize_store()
2043          */
2044         zram_reset_device(zram);
2045
2046         blk_cleanup_disk(zram->disk);
2047         kfree(zram);
2048         return 0;
2049 }
2050
2051 /* zram-control sysfs attributes */
2052
2053 /*
2054  * NOTE: hot_add attribute is not the usual read-only sysfs attribute. In a
2055  * sense that reading from this file does alter the state of your system -- it
2056  * creates a new un-initialized zram device and returns back this device's
2057  * device_id (or an error code if it fails to create a new device).
2058  */
2059 static ssize_t hot_add_show(struct class *class,
2060                         struct class_attribute *attr,
2061                         char *buf)
2062 {
2063         int ret;
2064
2065         mutex_lock(&zram_index_mutex);
2066         ret = zram_add();
2067         mutex_unlock(&zram_index_mutex);
2068
2069         if (ret < 0)
2070                 return ret;
2071         return scnprintf(buf, PAGE_SIZE, "%d\n", ret);
2072 }
2073 static struct class_attribute class_attr_hot_add =
2074         __ATTR(hot_add, 0400, hot_add_show, NULL);
2075
2076 static ssize_t hot_remove_store(struct class *class,
2077                         struct class_attribute *attr,
2078                         const char *buf,
2079                         size_t count)
2080 {
2081         struct zram *zram;
2082         int ret, dev_id;
2083
2084         /* dev_id is gendisk->first_minor, which is `int' */
2085         ret = kstrtoint(buf, 10, &dev_id);
2086         if (ret)
2087                 return ret;
2088         if (dev_id < 0)
2089                 return -EINVAL;
2090
2091         mutex_lock(&zram_index_mutex);
2092
2093         zram = idr_find(&zram_index_idr, dev_id);
2094         if (zram) {
2095                 ret = zram_remove(zram);
2096                 if (!ret)
2097                         idr_remove(&zram_index_idr, dev_id);
2098         } else {
2099                 ret = -ENODEV;
2100         }
2101
2102         mutex_unlock(&zram_index_mutex);
2103         return ret ? ret : count;
2104 }
2105 static CLASS_ATTR_WO(hot_remove);
2106
2107 static struct attribute *zram_control_class_attrs[] = {
2108         &class_attr_hot_add.attr,
2109         &class_attr_hot_remove.attr,
2110         NULL,
2111 };
2112 ATTRIBUTE_GROUPS(zram_control_class);
2113
2114 static struct class zram_control_class = {
2115         .name           = "zram-control",
2116         .owner          = THIS_MODULE,
2117         .class_groups   = zram_control_class_groups,
2118 };
2119
2120 static int zram_remove_cb(int id, void *ptr, void *data)
2121 {
2122         WARN_ON_ONCE(zram_remove(ptr));
2123         return 0;
2124 }
2125
2126 static void destroy_devices(void)
2127 {
2128         class_unregister(&zram_control_class);
2129         idr_for_each(&zram_index_idr, &zram_remove_cb, NULL);
2130         zram_debugfs_destroy();
2131         idr_destroy(&zram_index_idr);
2132         unregister_blkdev(zram_major, "zram");
2133         cpuhp_remove_multi_state(CPUHP_ZCOMP_PREPARE);
2134 }
2135
2136 static int __init zram_init(void)
2137 {
2138         int ret;
2139
2140         ret = cpuhp_setup_state_multi(CPUHP_ZCOMP_PREPARE, "block/zram:prepare",
2141                                       zcomp_cpu_up_prepare, zcomp_cpu_dead);
2142         if (ret < 0)
2143                 return ret;
2144
2145         ret = class_register(&zram_control_class);
2146         if (ret) {
2147                 pr_err("Unable to register zram-control class\n");
2148                 cpuhp_remove_multi_state(CPUHP_ZCOMP_PREPARE);
2149                 return ret;
2150         }
2151
2152         zram_debugfs_create();
2153         zram_major = register_blkdev(0, "zram");
2154         if (zram_major <= 0) {
2155                 pr_err("Unable to get major number\n");
2156                 class_unregister(&zram_control_class);
2157                 cpuhp_remove_multi_state(CPUHP_ZCOMP_PREPARE);
2158                 return -EBUSY;
2159         }
2160
2161         while (num_devices != 0) {
2162                 mutex_lock(&zram_index_mutex);
2163                 ret = zram_add();
2164                 mutex_unlock(&zram_index_mutex);
2165                 if (ret < 0)
2166                         goto out_error;
2167                 num_devices--;
2168         }
2169
2170         return 0;
2171
2172 out_error:
2173         destroy_devices();
2174         return ret;
2175 }
2176
2177 static void __exit zram_exit(void)
2178 {
2179         destroy_devices();
2180 }
2181
2182 module_init(zram_init);
2183 module_exit(zram_exit);
2184
2185 module_param(num_devices, uint, 0);
2186 MODULE_PARM_DESC(num_devices, "Number of pre-created zram devices");
2187
2188 MODULE_LICENSE("Dual BSD/GPL");
2189 MODULE_AUTHOR("Nitin Gupta <ngupta@vflare.org>");
2190 MODULE_DESCRIPTION("Compressed RAM Block Device");