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