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, err;
 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                 err = submit_bio_wait(&bio);
 729                 if (err) {
 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                         /*
 735                          * Return last IO error unless every IO were
 736                          * not suceeded.
 737                          */
 738                         ret = err;
 739                         continue;
 740                 }
 741
 742                 atomic64_inc(&zram->stats.bd_writes);
 743                 /*
 744                  * We released zram_slot_lock so need to check if the slot was
 745                  * changed. If there is freeing for the slot, we can catch it
 746                  * easily by zram_allocated.
 747                  * A subtle case is the slot is freed/reallocated/marked as
 748                  * ZRAM_IDLE again. To close the race, idle_store doesn't
 749                  * mark ZRAM_IDLE once it found the slot was ZRAM_UNDER_WB.
 750                  * Thus, we could close the race by checking ZRAM_IDLE bit.
 751                  */
 752                 zram_slot_lock(zram, index);
 753                 if (!zram_allocated(zram, index) ||
 754                           !zram_test_flag(zram, index, ZRAM_IDLE)) {
 755                         zram_clear_flag(zram, index, ZRAM_UNDER_WB);
 756                         zram_clear_flag(zram, index, ZRAM_IDLE);
 757                         goto next;
 758                 }
 759
 760                 zram_free_page(zram, index);
 761                 zram_clear_flag(zram, index, ZRAM_UNDER_WB);
 762                 zram_set_flag(zram, index, ZRAM_WB);
 763                 zram_set_element(zram, index, blk_idx);
 764                 blk_idx = 0;
 765                 atomic64_inc(&zram->stats.pages_stored);
 766                 spin_lock(&zram->wb_limit_lock);
 767                 if (zram->wb_limit_enable && zram->bd_wb_limit > 0)
 768                         zram->bd_wb_limit -=  1UL << (PAGE_SHIFT - 12);
 769                 spin_unlock(&zram->wb_limit_lock);
 770 next:
 771                 zram_slot_unlock(zram, index);
 772         }
 773
 774         if (blk_idx)
 775                 free_block_bdev(zram, blk_idx);
 776         __free_page(page);
 777 release_init_lock:
 778         up_read(&zram->init_lock);
 779
 780         return ret;
 781 }
 782
 783 struct zram_work {
 784         struct work_struct work;
 785         struct zram *zram;
 786         unsigned long entry;
 787         struct bio *bio;
 788         struct bio_vec bvec;
 789 };
 790
 791 #if PAGE_SIZE != 4096
 792 static void zram_sync_read(struct work_struct *work)
 793 {
 794         struct zram_work *zw = container_of(work, struct zram_work, work);
 795         struct zram *zram = zw->zram;
 796         unsigned long entry = zw->entry;
 797         struct bio *bio = zw->bio;
 798
 799         read_from_bdev_async(zram, &zw->bvec, entry, bio);
 800 }
 801
 802 /*
 803  * Block layer want one ->submit_bio to be active at a time, so if we use
 804  * chained IO with parent IO in same context, it's a deadlock. To avoid that,
 805  * use a worker thread context.
 806  */
 807 static int read_from_bdev_sync(struct zram *zram, struct bio_vec *bvec,
 808                                 unsigned long entry, struct bio *bio)
 809 {
 810         struct zram_work work;
 811
 812         work.bvec = *bvec;
 813         work.zram = zram;
 814         work.entry = entry;
 815         work.bio = bio;
 816
 817         INIT_WORK_ONSTACK(&work.work, zram_sync_read);
 818         queue_work(system_unbound_wq, &work.work);
 819         flush_work(&work.work);
 820         destroy_work_on_stack(&work.work);
 821
 822         return 1;
 823 }
 824 #else
 825 static int read_from_bdev_sync(struct zram *zram, struct bio_vec *bvec,
 826                                 unsigned long entry, struct bio *bio)
 827 {
 828         WARN_ON(1);
 829         return -EIO;
 830 }
 831 #endif
 832
 833 static int read_from_bdev(struct zram *zram, struct bio_vec *bvec,
 834                         unsigned long entry, struct bio *parent, bool sync)
 835 {
 836         atomic64_inc(&zram->stats.bd_reads);
 837         if (sync)
 838                 return read_from_bdev_sync(zram, bvec, entry, parent);
 839         else
 840                 return read_from_bdev_async(zram, bvec, entry, parent);
 841 }
 842 #else
 843 static inline void reset_bdev(struct zram *zram) {};
 844 static int read_from_bdev(struct zram *zram, struct bio_vec *bvec,
 845                         unsigned long entry, struct bio *parent, bool sync)
 846 {
 847         return -EIO;
 848 }
 849
 850 static void free_block_bdev(struct zram *zram, unsigned long blk_idx) {};
 851 #endif
 852
 853 #ifdef CONFIG_ZRAM_MEMORY_TRACKING
 854
 855 static struct dentry *zram_debugfs_root;
 856
 857 static void zram_debugfs_create(void)
 858 {
 859         zram_debugfs_root = debugfs_create_dir("zram", NULL);
 860 }
 861
 862 static void zram_debugfs_destroy(void)
 863 {
 864         debugfs_remove_recursive(zram_debugfs_root);
 865 }
 866
 867 static void zram_accessed(struct zram *zram, u32 index)
 868 {
 869         zram_clear_flag(zram, index, ZRAM_IDLE);
 870         zram->table[index].ac_time = ktime_get_boottime();
 871 }
 872
 873 static ssize_t read_block_state(struct file *file, char __user *buf,
 874                                 size_t count, loff_t *ppos)
 875 {
 876         char *kbuf;
 877         ssize_t index, written = 0;
 878         struct zram *zram = file->private_data;
 879         unsigned long nr_pages = zram->disksize >> PAGE_SHIFT;
 880         struct timespec64 ts;
 881
 882         kbuf = kvmalloc(count, GFP_KERNEL);
 883         if (!kbuf)
 884                 return -ENOMEM;
 885
 886         down_read(&zram->init_lock);
 887         if (!init_done(zram)) {
 888                 up_read(&zram->init_lock);
 889                 kvfree(kbuf);
 890                 return -EINVAL;
 891         }
 892
 893         for (index = *ppos; index < nr_pages; index++) {
 894                 int copied;
 895
 896                 zram_slot_lock(zram, index);
 897                 if (!zram_allocated(zram, index))
 898                         goto next;
 899
 900                 ts = ktime_to_timespec64(zram->table[index].ac_time);
 901                 copied = snprintf(kbuf + written, count,
 902                         "%12zd %12lld.%06lu %c%c%c%c\n",
 903                         index, (s64)ts.tv_sec,
 904                         ts.tv_nsec / NSEC_PER_USEC,
 905                         zram_test_flag(zram, index, ZRAM_SAME) ? 's' : '.',
 906                         zram_test_flag(zram, index, ZRAM_WB) ? 'w' : '.',
 907                         zram_test_flag(zram, index, ZRAM_HUGE) ? 'h' : '.',
 908                         zram_test_flag(zram, index, ZRAM_IDLE) ? 'i' : '.');
 909
 910                 if (count <= copied) {
 911                         zram_slot_unlock(zram, index);
 912                         break;
 913                 }
 914                 written += copied;
 915                 count -= copied;
 916 next:
 917                 zram_slot_unlock(zram, index);
 918                 *ppos += 1;
 919         }
 920
 921         up_read(&zram->init_lock);
 922         if (copy_to_user(buf, kbuf, written))
 923                 written = -EFAULT;
 924         kvfree(kbuf);
 925
 926         return written;
 927 }
 928
 929 static const struct file_operations proc_zram_block_state_op = {
 930         .open = simple_open,
 931         .read = read_block_state,
 932         .llseek = default_llseek,
 933 };
 934
 935 static void zram_debugfs_register(struct zram *zram)
 936 {
 937         if (!zram_debugfs_root)
 938                 return;
 939
 940         zram->debugfs_dir = debugfs_create_dir(zram->disk->disk_name,
 941                                                 zram_debugfs_root);
 942         debugfs_create_file("block_state", 0400, zram->debugfs_dir,
 943                                 zram, &proc_zram_block_state_op);
 944 }
 945
 946 static void zram_debugfs_unregister(struct zram *zram)
 947 {
 948         debugfs_remove_recursive(zram->debugfs_dir);
 949 }
 950 #else
 951 static void zram_debugfs_create(void) {};
 952 static void zram_debugfs_destroy(void) {};
 953 static void zram_accessed(struct zram *zram, u32 index)
 954 {
 955         zram_clear_flag(zram, index, ZRAM_IDLE);
 956 };
 957 static void zram_debugfs_register(struct zram *zram) {};
 958 static void zram_debugfs_unregister(struct zram *zram) {};
 959 #endif
 960
 961 /*
 962  * We switched to per-cpu streams and this attr is not needed anymore.
 963  * However, we will keep it around for some time, because:
 964  * a) we may revert per-cpu streams in the future
 965  * b) it's visible to user space and we need to follow our 2 years
 966  *    retirement rule; but we already have a number of 'soon to be
 967  *    altered' attrs, so max_comp_streams need to wait for the next
 968  *    layoff cycle.
 969  */
 970 static ssize_t max_comp_streams_show(struct device *dev,
 971                 struct device_attribute *attr, char *buf)
 972 {
 973         return scnprintf(buf, PAGE_SIZE, "%d\n", num_online_cpus());
 974 }
 975
 976 static ssize_t max_comp_streams_store(struct device *dev,
 977                 struct device_attribute *attr, const char *buf, size_t len)
 978 {
 979         return len;
 980 }
 981
 982 static ssize_t comp_algorithm_show(struct device *dev,
 983                 struct device_attribute *attr, char *buf)
 984 {
 985         size_t sz;
 986         struct zram *zram = dev_to_zram(dev);
 987
 988         down_read(&zram->init_lock);
 989         sz = zcomp_available_show(zram->compressor, buf);
 990         up_read(&zram->init_lock);
 991
 992         return sz;
 993 }
 994
 995 static ssize_t comp_algorithm_store(struct device *dev,
 996                 struct device_attribute *attr, const char *buf, size_t len)
 997 {
 998         struct zram *zram = dev_to_zram(dev);
 999         char compressor[ARRAY_SIZE(zram->compressor)];
1000         size_t sz;
1001
1002         strlcpy(compressor, buf, sizeof(compressor));
1003         /* ignore trailing newline */
1004         sz = strlen(compressor);
1005         if (sz > 0 && compressor[sz - 1] == '\n')
1006                 compressor[sz - 1] = 0x00;
1007
1008         if (!zcomp_available_algorithm(compressor))
1009                 return -EINVAL;
1010
1011         down_write(&zram->init_lock);
1012         if (init_done(zram)) {
1013                 up_write(&zram->init_lock);
1014                 pr_info("Can't change algorithm for initialized device\n");
1015                 return -EBUSY;
1016         }
1017
1018         strcpy(zram->compressor, compressor);
1019         up_write(&zram->init_lock);
1020         return len;
1021 }
1022
1023 static ssize_t compact_store(struct device *dev,
1024                 struct device_attribute *attr, const char *buf, size_t len)
1025 {
1026         struct zram *zram = dev_to_zram(dev);
1027
1028         down_read(&zram->init_lock);
1029         if (!init_done(zram)) {
1030                 up_read(&zram->init_lock);
1031                 return -EINVAL;
1032         }
1033
1034         zs_compact(zram->mem_pool);
1035         up_read(&zram->init_lock);
1036
1037         return len;
1038 }
1039
1040 static ssize_t io_stat_show(struct device *dev,
1041                 struct device_attribute *attr, char *buf)
1042 {
1043         struct zram *zram = dev_to_zram(dev);
1044         ssize_t ret;
1045
1046         down_read(&zram->init_lock);
1047         ret = scnprintf(buf, PAGE_SIZE,
1048                         "%8llu %8llu %8llu %8llu\n",
1049                         (u64)atomic64_read(&zram->stats.failed_reads),
1050                         (u64)atomic64_read(&zram->stats.failed_writes),
1051                         (u64)atomic64_read(&zram->stats.invalid_io),
1052                         (u64)atomic64_read(&zram->stats.notify_free));
1053         up_read(&zram->init_lock);
1054
1055         return ret;
1056 }
1057
1058 static ssize_t mm_stat_show(struct device *dev,
1059                 struct device_attribute *attr, char *buf)
1060 {
1061         struct zram *zram = dev_to_zram(dev);
1062         struct zs_pool_stats pool_stats;
1063         u64 orig_size, mem_used = 0;
1064         long max_used;
1065         ssize_t ret;
1066
1067         memset(&pool_stats, 0x00, sizeof(struct zs_pool_stats));
1068
1069         down_read(&zram->init_lock);
1070         if (init_done(zram)) {
1071                 mem_used = zs_get_total_pages(zram->mem_pool);
1072                 zs_pool_stats(zram->mem_pool, &pool_stats);
1073         }
1074
1075         orig_size = atomic64_read(&zram->stats.pages_stored);
1076         max_used = atomic_long_read(&zram->stats.max_used_pages);
1077
1078         ret = scnprintf(buf, PAGE_SIZE,
1079                         "%8llu %8llu %8llu %8lu %8ld %8llu %8lu %8llu\n",
1080                         orig_size << PAGE_SHIFT,
1081                         (u64)atomic64_read(&zram->stats.compr_data_size),
1082                         mem_used << PAGE_SHIFT,
1083                         zram->limit_pages << PAGE_SHIFT,
1084                         max_used << PAGE_SHIFT,
1085                         (u64)atomic64_read(&zram->stats.same_pages),
1086                         atomic_long_read(&pool_stats.pages_compacted),
1087                         (u64)atomic64_read(&zram->stats.huge_pages));
1088         up_read(&zram->init_lock);
1089
1090         return ret;
1091 }
1092
1093 #ifdef CONFIG_ZRAM_WRITEBACK
1094 #define FOUR_K(x) ((x) * (1 << (PAGE_SHIFT - 12)))
1095 static ssize_t bd_stat_show(struct device *dev,
1096                 struct device_attribute *attr, char *buf)
1097 {
1098         struct zram *zram = dev_to_zram(dev);
1099         ssize_t ret;
1100
1101         down_read(&zram->init_lock);
1102         ret = scnprintf(buf, PAGE_SIZE,
1103                 "%8llu %8llu %8llu\n",
1104                         FOUR_K((u64)atomic64_read(&zram->stats.bd_count)),
1105                         FOUR_K((u64)atomic64_read(&zram->stats.bd_reads)),
1106                         FOUR_K((u64)atomic64_read(&zram->stats.bd_writes)));
1107         up_read(&zram->init_lock);
1108
1109         return ret;
1110 }
1111 #endif
1112
1113 static ssize_t debug_stat_show(struct device *dev,
1114                 struct device_attribute *attr, char *buf)
1115 {
1116         int version = 1;
1117         struct zram *zram = dev_to_zram(dev);
1118         ssize_t ret;
1119
1120         down_read(&zram->init_lock);
1121         ret = scnprintf(buf, PAGE_SIZE,
1122                         "version: %d\n%8llu %8llu\n",
1123                         version,
1124                         (u64)atomic64_read(&zram->stats.writestall),
1125                         (u64)atomic64_read(&zram->stats.miss_free));
1126         up_read(&zram->init_lock);
1127
1128         return ret;
1129 }
1130
1131 static DEVICE_ATTR_RO(io_stat);
1132 static DEVICE_ATTR_RO(mm_stat);
1133 #ifdef CONFIG_ZRAM_WRITEBACK
1134 static DEVICE_ATTR_RO(bd_stat);
1135 #endif
1136 static DEVICE_ATTR_RO(debug_stat);
1137
1138 static void zram_meta_free(struct zram *zram, u64 disksize)
1139 {
1140         size_t num_pages = disksize >> PAGE_SHIFT;
1141         size_t index;
1142
1143         /* Free all pages that are still in this zram device */
1144         for (index = 0; index < num_pages; index++)
1145                 zram_free_page(zram, index);
1146
1147         zs_destroy_pool(zram->mem_pool);
1148         vfree(zram->table);
1149 }
1150
1151 static bool zram_meta_alloc(struct zram *zram, u64 disksize)
1152 {
1153         size_t num_pages;
1154
1155         num_pages = disksize >> PAGE_SHIFT;
1156         zram->table = vzalloc(array_size(num_pages, sizeof(*zram->table)));
1157         if (!zram->table)
1158                 return false;
1159
1160         zram->mem_pool = zs_create_pool(zram->disk->disk_name);
1161         if (!zram->mem_pool) {
1162                 vfree(zram->table);
1163                 return false;
1164         }
1165
1166         if (!huge_class_size)
1167                 huge_class_size = zs_huge_class_size(zram->mem_pool);
1168         return true;
1169 }
1170
1171 /*
1172  * To protect concurrent access to the same index entry,
1173  * caller should hold this table index entry's bit_spinlock to
1174  * indicate this index entry is accessing.
1175  */
1176 static void zram_free_page(struct zram *zram, size_t index)
1177 {
1178         unsigned long handle;
1179
1180 #ifdef CONFIG_ZRAM_MEMORY_TRACKING
1181         zram->table[index].ac_time = 0;
1182 #endif
1183         if (zram_test_flag(zram, index, ZRAM_IDLE))
1184                 zram_clear_flag(zram, index, ZRAM_IDLE);
1185
1186         if (zram_test_flag(zram, index, ZRAM_HUGE)) {
1187                 zram_clear_flag(zram, index, ZRAM_HUGE);
1188                 atomic64_dec(&zram->stats.huge_pages);
1189         }
1190
1191         if (zram_test_flag(zram, index, ZRAM_WB)) {
1192                 zram_clear_flag(zram, index, ZRAM_WB);
1193                 free_block_bdev(zram, zram_get_element(zram, index));
1194                 goto out;
1195         }
1196
1197         /*
1198          * No memory is allocated for same element filled pages.
1199          * Simply clear same page flag.
1200          */
1201         if (zram_test_flag(zram, index, ZRAM_SAME)) {
1202                 zram_clear_flag(zram, index, ZRAM_SAME);
1203                 atomic64_dec(&zram->stats.same_pages);
1204                 goto out;
1205         }
1206
1207         handle = zram_get_handle(zram, index);
1208         if (!handle)
1209                 return;
1210
1211         zs_free(zram->mem_pool, handle);
1212
1213         atomic64_sub(zram_get_obj_size(zram, index),
1214                         &zram->stats.compr_data_size);
1215 out:
1216         atomic64_dec(&zram->stats.pages_stored);
1217         zram_set_handle(zram, index, 0);
1218         zram_set_obj_size(zram, index, 0);
1219         WARN_ON_ONCE(zram->table[index].flags &
1220                 ~(1UL << ZRAM_LOCK | 1UL << ZRAM_UNDER_WB));
1221 }
1222
1223 static int __zram_bvec_read(struct zram *zram, struct page *page, u32 index,
1224                                 struct bio *bio, bool partial_io)
1225 {
1226         struct zcomp_strm *zstrm;
1227         unsigned long handle;
1228         unsigned int size;
1229         void *src, *dst;
1230         int ret;
1231
1232         zram_slot_lock(zram, index);
1233         if (zram_test_flag(zram, index, ZRAM_WB)) {
1234                 struct bio_vec bvec;
1235
1236                 zram_slot_unlock(zram, index);
1237
1238                 bvec.bv_page = page;
1239                 bvec.bv_len = PAGE_SIZE;
1240                 bvec.bv_offset = 0;
1241                 return read_from_bdev(zram, &bvec,
1242                                 zram_get_element(zram, index),
1243                                 bio, partial_io);
1244         }
1245
1246         handle = zram_get_handle(zram, index);
1247         if (!handle || zram_test_flag(zram, index, ZRAM_SAME)) {
1248                 unsigned long value;
1249                 void *mem;
1250
1251                 value = handle ? zram_get_element(zram, index) : 0;
1252                 mem = kmap_atomic(page);
1253                 zram_fill_page(mem, PAGE_SIZE, value);
1254                 kunmap_atomic(mem);
1255                 zram_slot_unlock(zram, index);
1256                 return 0;
1257         }
1258
1259         size = zram_get_obj_size(zram, index);
1260
1261         if (size != PAGE_SIZE)
1262                 zstrm = zcomp_stream_get(zram->comp);
1263
1264         src = zs_map_object(zram->mem_pool, handle, ZS_MM_RO);
1265         if (size == PAGE_SIZE) {
1266                 dst = kmap_atomic(page);
1267                 memcpy(dst, src, PAGE_SIZE);
1268                 kunmap_atomic(dst);
1269                 ret = 0;
1270         } else {
1271                 dst = kmap_atomic(page);
1272                 ret = zcomp_decompress(zstrm, src, size, dst);
1273                 kunmap_atomic(dst);
1274                 zcomp_stream_put(zram->comp);
1275         }
1276         zs_unmap_object(zram->mem_pool, handle);
1277         zram_slot_unlock(zram, index);
1278
1279         /* Should NEVER happen. Return bio error if it does. */
1280         if (WARN_ON(ret))
1281                 pr_err("Decompression failed! err=%d, page=%u\n", ret, index);
1282
1283         return ret;
1284 }
1285
1286 static int zram_bvec_read(struct zram *zram, struct bio_vec *bvec,
1287                                 u32 index, int offset, struct bio *bio)
1288 {
1289         int ret;
1290         struct page *page;
1291
1292         page = bvec->bv_page;
1293         if (is_partial_io(bvec)) {
1294                 /* Use a temporary buffer to decompress the page */
1295                 page = alloc_page(GFP_NOIO|__GFP_HIGHMEM);
1296                 if (!page)
1297                         return -ENOMEM;
1298         }
1299
1300         ret = __zram_bvec_read(zram, page, index, bio, is_partial_io(bvec));
1301         if (unlikely(ret))
1302                 goto out;
1303
1304         if (is_partial_io(bvec)) {
1305                 void *dst = kmap_atomic(bvec->bv_page);
1306                 void *src = kmap_atomic(page);
1307
1308                 memcpy(dst + bvec->bv_offset, src + offset, bvec->bv_len);
1309                 kunmap_atomic(src);
1310                 kunmap_atomic(dst);
1311         }
1312 out:
1313         if (is_partial_io(bvec))
1314                 __free_page(page);
1315
1316         return ret;
1317 }
1318
1319 static int __zram_bvec_write(struct zram *zram, struct bio_vec *bvec,
1320                                 u32 index, struct bio *bio)
1321 {
1322         int ret = 0;
1323         unsigned long alloced_pages;
1324         unsigned long handle = 0;
1325         unsigned int comp_len = 0;
1326         void *src, *dst, *mem;
1327         struct zcomp_strm *zstrm;
1328         struct page *page = bvec->bv_page;
1329         unsigned long element = 0;
1330         enum zram_pageflags flags = 0;
1331
1332         mem = kmap_atomic(page);
1333         if (page_same_filled(mem, &element)) {
1334                 kunmap_atomic(mem);
1335                 /* Free memory associated with this sector now. */
1336                 flags = ZRAM_SAME;
1337                 atomic64_inc(&zram->stats.same_pages);
1338                 goto out;
1339         }
1340         kunmap_atomic(mem);
1341
1342 compress_again:
1343         zstrm = zcomp_stream_get(zram->comp);
1344         src = kmap_atomic(page);
1345         ret = zcomp_compress(zstrm, src, &comp_len);
1346         kunmap_atomic(src);
1347
1348         if (unlikely(ret)) {
1349                 zcomp_stream_put(zram->comp);
1350                 pr_err("Compression failed! err=%d\n", ret);
1351                 zs_free(zram->mem_pool, handle);
1352                 return ret;
1353         }
1354
1355         if (comp_len >= huge_class_size)
1356                 comp_len = PAGE_SIZE;
1357         /*
1358          * handle allocation has 2 paths:
1359          * a) fast path is executed with preemption disabled (for
1360          *  per-cpu streams) and has __GFP_DIRECT_RECLAIM bit clear,
1361          *  since we can't sleep;
1362          * b) slow path enables preemption and attempts to allocate
1363          *  the page with __GFP_DIRECT_RECLAIM bit set. we have to
1364          *  put per-cpu compression stream and, thus, to re-do
1365          *  the compression once handle is allocated.
1366          *
1367          * if we have a 'non-null' handle here then we are coming
1368          * from the slow path and handle has already been allocated.
1369          */
1370         if (!handle)
1371                 handle = zs_malloc(zram->mem_pool, comp_len,
1372                                 __GFP_KSWAPD_RECLAIM |
1373                                 __GFP_NOWARN |
1374                                 __GFP_HIGHMEM |
1375                                 __GFP_MOVABLE);
1376         if (!handle) {
1377                 zcomp_stream_put(zram->comp);
1378                 atomic64_inc(&zram->stats.writestall);
1379                 handle = zs_malloc(zram->mem_pool, comp_len,
1380                                 GFP_NOIO | __GFP_HIGHMEM |
1381                                 __GFP_MOVABLE);
1382                 if (handle)
1383                         goto compress_again;
1384                 return -ENOMEM;
1385         }
1386
1387         alloced_pages = zs_get_total_pages(zram->mem_pool);
1388         update_used_max(zram, alloced_pages);
1389
1390         if (zram->limit_pages && alloced_pages > zram->limit_pages) {
1391                 zcomp_stream_put(zram->comp);
1392                 zs_free(zram->mem_pool, handle);
1393                 return -ENOMEM;
1394         }
1395
1396         dst = zs_map_object(zram->mem_pool, handle, ZS_MM_WO);
1397
1398         src = zstrm->buffer;
1399         if (comp_len == PAGE_SIZE)
1400                 src = kmap_atomic(page);
1401         memcpy(dst, src, comp_len);
1402         if (comp_len == PAGE_SIZE)
1403                 kunmap_atomic(src);
1404
1405         zcomp_stream_put(zram->comp);
1406         zs_unmap_object(zram->mem_pool, handle);
1407         atomic64_add(comp_len, &zram->stats.compr_data_size);
1408 out:
1409         /*
1410          * Free memory associated with this sector
1411          * before overwriting unused sectors.
1412          */
1413         zram_slot_lock(zram, index);
1414         zram_free_page(zram, index);
1415
1416         if (comp_len == PAGE_SIZE) {
1417                 zram_set_flag(zram, index, ZRAM_HUGE);
1418                 atomic64_inc(&zram->stats.huge_pages);
1419         }
1420
1421         if (flags) {
1422                 zram_set_flag(zram, index, flags);
1423                 zram_set_element(zram, index, element);
1424         }  else {
1425                 zram_set_handle(zram, index, handle);
1426                 zram_set_obj_size(zram, index, comp_len);
1427         }
1428         zram_slot_unlock(zram, index);
1429
1430         /* Update stats */
1431         atomic64_inc(&zram->stats.pages_stored);
1432         return ret;
1433 }
1434
1435 static int zram_bvec_write(struct zram *zram, struct bio_vec *bvec,
1436                                 u32 index, int offset, struct bio *bio)
1437 {
1438         int ret;
1439         struct page *page = NULL;
1440         void *src;
1441         struct bio_vec vec;
1442
1443         vec = *bvec;
1444         if (is_partial_io(bvec)) {
1445                 void *dst;
1446                 /*
1447                  * This is a partial IO. We need to read the full page
1448                  * before to write the changes.
1449                  */
1450                 page = alloc_page(GFP_NOIO|__GFP_HIGHMEM);
1451                 if (!page)
1452                         return -ENOMEM;
1453
1454                 ret = __zram_bvec_read(zram, page, index, bio, true);
1455                 if (ret)
1456                         goto out;
1457
1458                 src = kmap_atomic(bvec->bv_page);
1459                 dst = kmap_atomic(page);
1460                 memcpy(dst + offset, src + bvec->bv_offset, bvec->bv_len);
1461                 kunmap_atomic(dst);
1462                 kunmap_atomic(src);
1463
1464                 vec.bv_page = page;
1465                 vec.bv_len = PAGE_SIZE;
1466                 vec.bv_offset = 0;
1467         }
1468
1469         ret = __zram_bvec_write(zram, &vec, index, bio);
1470 out:
1471         if (is_partial_io(bvec))
1472                 __free_page(page);
1473         return ret;
1474 }
1475
1476 /*
1477  * zram_bio_discard - handler on discard request
1478  * @index: physical block index in PAGE_SIZE units
1479  * @offset: byte offset within physical block
1480  */
1481 static void zram_bio_discard(struct zram *zram, u32 index,
1482                              int offset, struct bio *bio)
1483 {
1484         size_t n = bio->bi_iter.bi_size;
1485
1486         /*
1487          * zram manages data in physical block size units. Because logical block
1488          * size isn't identical with physical block size on some arch, we
1489          * could get a discard request pointing to a specific offset within a
1490          * certain physical block.  Although we can handle this request by
1491          * reading that physiclal block and decompressing and partially zeroing
1492          * and re-compressing and then re-storing it, this isn't reasonable
1493          * because our intent with a discard request is to save memory.  So
1494          * skipping this logical block is appropriate here.
1495          */
1496         if (offset) {
1497                 if (n <= (PAGE_SIZE - offset))
1498                         return;
1499
1500                 n -= (PAGE_SIZE - offset);
1501                 index++;
1502         }
1503
1504         while (n >= PAGE_SIZE) {
1505                 zram_slot_lock(zram, index);
1506                 zram_free_page(zram, index);
1507                 zram_slot_unlock(zram, index);
1508                 atomic64_inc(&zram->stats.notify_free);
1509                 index++;
1510                 n -= PAGE_SIZE;
1511         }
1512 }
1513
1514 /*
1515  * Returns errno if it has some problem. Otherwise return 0 or 1.
1516  * Returns 0 if IO request was done synchronously
1517  * Returns 1 if IO request was successfully submitted.
1518  */
1519 static int zram_bvec_rw(struct zram *zram, struct bio_vec *bvec, u32 index,
1520                         int offset, unsigned int op, struct bio *bio)
1521 {
1522         int ret;
1523
1524         if (!op_is_write(op)) {
1525                 atomic64_inc(&zram->stats.num_reads);
1526                 ret = zram_bvec_read(zram, bvec, index, offset, bio);
1527                 flush_dcache_page(bvec->bv_page);
1528         } else {
1529                 atomic64_inc(&zram->stats.num_writes);
1530                 ret = zram_bvec_write(zram, bvec, index, offset, bio);
1531         }
1532
1533         zram_slot_lock(zram, index);
1534         zram_accessed(zram, index);
1535         zram_slot_unlock(zram, index);
1536
1537         if (unlikely(ret < 0)) {
1538                 if (!op_is_write(op))
1539                         atomic64_inc(&zram->stats.failed_reads);
1540                 else
1541                         atomic64_inc(&zram->stats.failed_writes);
1542         }
1543
1544         return ret;
1545 }
1546
1547 static void __zram_make_request(struct zram *zram, struct bio *bio)
1548 {
1549         int offset;
1550         u32 index;
1551         struct bio_vec bvec;
1552         struct bvec_iter iter;
1553         unsigned long start_time;
1554
1555         index = bio->bi_iter.bi_sector >> SECTORS_PER_PAGE_SHIFT;
1556         offset = (bio->bi_iter.bi_sector &
1557                   (SECTORS_PER_PAGE - 1)) << SECTOR_SHIFT;
1558
1559         switch (bio_op(bio)) {
1560         case REQ_OP_DISCARD:
1561         case REQ_OP_WRITE_ZEROES:
1562                 zram_bio_discard(zram, index, offset, bio);
1563                 bio_endio(bio);
1564                 return;
1565         default:
1566                 break;
1567         }
1568
1569         start_time = bio_start_io_acct(bio);
1570         bio_for_each_segment(bvec, bio, iter) {
1571                 struct bio_vec bv = bvec;
1572                 unsigned int unwritten = bvec.bv_len;
1573
1574                 do {
1575                         bv.bv_len = min_t(unsigned int, PAGE_SIZE - offset,
1576                                                         unwritten);
1577                         if (zram_bvec_rw(zram, &bv, index, offset,
1578                                          bio_op(bio), bio) < 0) {
1579                                 bio->bi_status = BLK_STS_IOERR;
1580                                 break;
1581                         }
1582
1583                         bv.bv_offset += bv.bv_len;
1584                         unwritten -= bv.bv_len;
1585
1586                         update_position(&index, &offset, &bv);
1587                 } while (unwritten);
1588         }
1589         bio_end_io_acct(bio, start_time);
1590         bio_endio(bio);
1591 }
1592
1593 /*
1594  * Handler function for all zram I/O requests.
1595  */
1596 static blk_qc_t zram_submit_bio(struct bio *bio)
1597 {
1598         struct zram *zram = bio->bi_disk->private_data;
1599
1600         if (!valid_io_request(zram, bio->bi_iter.bi_sector,
1601                                         bio->bi_iter.bi_size)) {
1602                 atomic64_inc(&zram->stats.invalid_io);
1603                 goto error;
1604         }
1605
1606         __zram_make_request(zram, bio);
1607         return BLK_QC_T_NONE;
1608
1609 error:
1610         bio_io_error(bio);
1611         return BLK_QC_T_NONE;
1612 }
1613
1614 static void zram_slot_free_notify(struct block_device *bdev,
1615                                 unsigned long index)
1616 {
1617         struct zram *zram;
1618
1619         zram = bdev->bd_disk->private_data;
1620
1621         atomic64_inc(&zram->stats.notify_free);
1622         if (!zram_slot_trylock(zram, index)) {
1623                 atomic64_inc(&zram->stats.miss_free);
1624                 return;
1625         }
1626
1627         zram_free_page(zram, index);
1628         zram_slot_unlock(zram, index);
1629 }
1630
1631 static int zram_rw_page(struct block_device *bdev, sector_t sector,
1632                        struct page *page, unsigned int op)
1633 {
1634         int offset, ret;
1635         u32 index;
1636         struct zram *zram;
1637         struct bio_vec bv;
1638         unsigned long start_time;
1639
1640         if (PageTransHuge(page))
1641                 return -ENOTSUPP;
1642         zram = bdev->bd_disk->private_data;
1643
1644         if (!valid_io_request(zram, sector, PAGE_SIZE)) {
1645                 atomic64_inc(&zram->stats.invalid_io);
1646                 ret = -EINVAL;
1647                 goto out;
1648         }
1649
1650         index = sector >> SECTORS_PER_PAGE_SHIFT;
1651         offset = (sector & (SECTORS_PER_PAGE - 1)) << SECTOR_SHIFT;
1652
1653         bv.bv_page = page;
1654         bv.bv_len = PAGE_SIZE;
1655         bv.bv_offset = 0;
1656
1657         start_time = disk_start_io_acct(bdev->bd_disk, SECTORS_PER_PAGE, op);
1658         ret = zram_bvec_rw(zram, &bv, index, offset, op, NULL);
1659         disk_end_io_acct(bdev->bd_disk, op, start_time);
1660 out:
1661         /*
1662          * If I/O fails, just return error(ie, non-zero) without
1663          * calling page_endio.
1664          * It causes resubmit the I/O with bio request by upper functions
1665          * of rw_page(e.g., swap_readpage, __swap_writepage) and
1666          * bio->bi_end_io does things to handle the error
1667          * (e.g., SetPageError, set_page_dirty and extra works).
1668          */
1669         if (unlikely(ret < 0))
1670                 return ret;
1671
1672         switch (ret) {
1673         case 0:
1674                 page_endio(page, op_is_write(op), 0);
1675                 break;
1676         case 1:
1677                 ret = 0;
1678                 break;
1679         default:
1680                 WARN_ON(1);
1681         }
1682         return ret;
1683 }
1684
1685 static void zram_reset_device(struct zram *zram)
1686 {
1687         struct zcomp *comp;
1688         u64 disksize;
1689
1690         down_write(&zram->init_lock);
1691
1692         zram->limit_pages = 0;
1693
1694         if (!init_done(zram)) {
1695                 up_write(&zram->init_lock);
1696                 return;
1697         }
1698
1699         comp = zram->comp;
1700         disksize = zram->disksize;
1701         zram->disksize = 0;
1702
1703         set_capacity(zram->disk, 0);
1704         part_stat_set_all(&zram->disk->part0, 0);
1705
1706         up_write(&zram->init_lock);
1707         /* I/O operation under all of CPU are done so let's free */
1708         zram_meta_free(zram, disksize);
1709         memset(&zram->stats, 0, sizeof(zram->stats));
1710         zcomp_destroy(comp);
1711         reset_bdev(zram);
1712 }
1713
1714 static ssize_t disksize_store(struct device *dev,
1715                 struct device_attribute *attr, const char *buf, size_t len)
1716 {
1717         u64 disksize;
1718         struct zcomp *comp;
1719         struct zram *zram = dev_to_zram(dev);
1720         int err;
1721
1722         disksize = memparse(buf, NULL);
1723         if (!disksize)
1724                 return -EINVAL;
1725
1726         down_write(&zram->init_lock);
1727         if (init_done(zram)) {
1728                 pr_info("Cannot change disksize for initialized device\n");
1729                 err = -EBUSY;
1730                 goto out_unlock;
1731         }
1732
1733         disksize = PAGE_ALIGN(disksize);
1734         if (!zram_meta_alloc(zram, disksize)) {
1735                 err = -ENOMEM;
1736                 goto out_unlock;
1737         }
1738
1739         comp = zcomp_create(zram->compressor);
1740         if (IS_ERR(comp)) {
1741                 pr_err("Cannot initialise %s compressing backend\n",
1742                                 zram->compressor);
1743                 err = PTR_ERR(comp);
1744                 goto out_free_meta;
1745         }
1746
1747         zram->comp = comp;
1748         zram->disksize = disksize;
1749         set_capacity(zram->disk, zram->disksize >> SECTOR_SHIFT);
1750
1751         revalidate_disk_size(zram->disk, true);
1752         up_write(&zram->init_lock);
1753
1754         return len;
1755
1756 out_free_meta:
1757         zram_meta_free(zram, disksize);
1758 out_unlock:
1759         up_write(&zram->init_lock);
1760         return err;
1761 }
1762
1763 static ssize_t reset_store(struct device *dev,
1764                 struct device_attribute *attr, const char *buf, size_t len)
1765 {
1766         int ret;
1767         unsigned short do_reset;
1768         struct zram *zram;
1769         struct block_device *bdev;
1770
1771         ret = kstrtou16(buf, 10, &do_reset);
1772         if (ret)
1773                 return ret;
1774
1775         if (!do_reset)
1776                 return -EINVAL;
1777
1778         zram = dev_to_zram(dev);
1779         bdev = bdget_disk(zram->disk, 0);
1780         if (!bdev)
1781                 return -ENOMEM;
1782
1783         mutex_lock(&bdev->bd_mutex);
1784         /* Do not reset an active device or claimed device */
1785         if (bdev->bd_openers || zram->claim) {
1786                 mutex_unlock(&bdev->bd_mutex);
1787                 bdput(bdev);
1788                 return -EBUSY;
1789         }
1790
1791         /* From now on, anyone can't open /dev/zram[0-9] */
1792         zram->claim = true;
1793         mutex_unlock(&bdev->bd_mutex);
1794
1795         /* Make sure all the pending I/O are finished */
1796         fsync_bdev(bdev);
1797         zram_reset_device(zram);
1798         revalidate_disk_size(zram->disk, true);
1799         bdput(bdev);
1800
1801         mutex_lock(&bdev->bd_mutex);
1802         zram->claim = false;
1803         mutex_unlock(&bdev->bd_mutex);
1804
1805         return len;
1806 }
1807
1808 static int zram_open(struct block_device *bdev, fmode_t mode)
1809 {
1810         int ret = 0;
1811         struct zram *zram;
1812
1813         WARN_ON(!mutex_is_locked(&bdev->bd_mutex));
1814
1815         zram = bdev->bd_disk->private_data;
1816         /* zram was claimed to reset so open request fails */
1817         if (zram->claim)
1818                 ret = -EBUSY;
1819
1820         return ret;
1821 }
1822
1823 static const struct block_device_operations zram_devops = {
1824         .open = zram_open,
1825         .submit_bio = zram_submit_bio,
1826         .swap_slot_free_notify = zram_slot_free_notify,
1827         .rw_page = zram_rw_page,
1828         .owner = THIS_MODULE
1829 };
1830
1831 static const struct block_device_operations zram_wb_devops = {
1832         .open = zram_open,
1833         .submit_bio = zram_submit_bio,
1834         .swap_slot_free_notify = zram_slot_free_notify,
1835         .owner = THIS_MODULE
1836 };
1837
1838 static DEVICE_ATTR_WO(compact);
1839 static DEVICE_ATTR_RW(disksize);
1840 static DEVICE_ATTR_RO(initstate);
1841 static DEVICE_ATTR_WO(reset);
1842 static DEVICE_ATTR_WO(mem_limit);
1843 static DEVICE_ATTR_WO(mem_used_max);
1844 static DEVICE_ATTR_WO(idle);
1845 static DEVICE_ATTR_RW(max_comp_streams);
1846 static DEVICE_ATTR_RW(comp_algorithm);
1847 #ifdef CONFIG_ZRAM_WRITEBACK
1848 static DEVICE_ATTR_RW(backing_dev);
1849 static DEVICE_ATTR_WO(writeback);
1850 static DEVICE_ATTR_RW(writeback_limit);
1851 static DEVICE_ATTR_RW(writeback_limit_enable);
1852 #endif
1853
1854 static struct attribute *zram_disk_attrs[] = {
1855         &dev_attr_disksize.attr,
1856         &dev_attr_initstate.attr,
1857         &dev_attr_reset.attr,
1858         &dev_attr_compact.attr,
1859         &dev_attr_mem_limit.attr,
1860         &dev_attr_mem_used_max.attr,
1861         &dev_attr_idle.attr,
1862         &dev_attr_max_comp_streams.attr,
1863         &dev_attr_comp_algorithm.attr,
1864 #ifdef CONFIG_ZRAM_WRITEBACK
1865         &dev_attr_backing_dev.attr,
1866         &dev_attr_writeback.attr,
1867         &dev_attr_writeback_limit.attr,
1868         &dev_attr_writeback_limit_enable.attr,
1869 #endif
1870         &dev_attr_io_stat.attr,
1871         &dev_attr_mm_stat.attr,
1872 #ifdef CONFIG_ZRAM_WRITEBACK
1873         &dev_attr_bd_stat.attr,
1874 #endif
1875         &dev_attr_debug_stat.attr,
1876         NULL,
1877 };
1878
1879 static const struct attribute_group zram_disk_attr_group = {
1880         .attrs = zram_disk_attrs,
1881 };
1882
1883 static const struct attribute_group *zram_disk_attr_groups[] = {
1884         &zram_disk_attr_group,
1885         NULL,
1886 };
1887
1888 /*
1889  * Allocate and initialize new zram device. the function returns
1890  * '>= 0' device_id upon success, and negative value otherwise.
1891  */
1892 static int zram_add(void)
1893 {
1894         struct zram *zram;
1895         struct request_queue *queue;
1896         int ret, device_id;
1897
1898         zram = kzalloc(sizeof(struct zram), GFP_KERNEL);
1899         if (!zram)
1900                 return -ENOMEM;
1901
1902         ret = idr_alloc(&zram_index_idr, zram, 0, 0, GFP_KERNEL);
1903         if (ret < 0)
1904                 goto out_free_dev;
1905         device_id = ret;
1906
1907         init_rwsem(&zram->init_lock);
1908 #ifdef CONFIG_ZRAM_WRITEBACK
1909         spin_lock_init(&zram->wb_limit_lock);
1910 #endif
1911         queue = blk_alloc_queue(NUMA_NO_NODE);
1912         if (!queue) {
1913                 pr_err("Error allocating disk queue for device %d\n",
1914                         device_id);
1915                 ret = -ENOMEM;
1916                 goto out_free_idr;
1917         }
1918
1919         /* gendisk structure */
1920         zram->disk = alloc_disk(1);
1921         if (!zram->disk) {
1922                 pr_err("Error allocating disk structure for device %d\n",
1923                         device_id);
1924                 ret = -ENOMEM;
1925                 goto out_free_queue;
1926         }
1927
1928         zram->disk->major = zram_major;
1929         zram->disk->first_minor = device_id;
1930         zram->disk->fops = &zram_devops;
1931         zram->disk->queue = queue;
1932         zram->disk->private_data = zram;
1933         snprintf(zram->disk->disk_name, 16, "zram%d", device_id);
1934
1935         /* Actual capacity set using syfs (/sys/block/zram<id>/disksize */
1936         set_capacity(zram->disk, 0);
1937         /* zram devices sort of resembles non-rotational disks */
1938         blk_queue_flag_set(QUEUE_FLAG_NONROT, zram->disk->queue);
1939         blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, zram->disk->queue);
1940
1941         /*
1942          * To ensure that we always get PAGE_SIZE aligned
1943          * and n*PAGE_SIZED sized I/O requests.
1944          */
1945         blk_queue_physical_block_size(zram->disk->queue, PAGE_SIZE);
1946         blk_queue_logical_block_size(zram->disk->queue,
1947                                         ZRAM_LOGICAL_BLOCK_SIZE);
1948         blk_queue_io_min(zram->disk->queue, PAGE_SIZE);
1949         blk_queue_io_opt(zram->disk->queue, PAGE_SIZE);
1950         zram->disk->queue->limits.discard_granularity = PAGE_SIZE;
1951         blk_queue_max_discard_sectors(zram->disk->queue, UINT_MAX);
1952         blk_queue_flag_set(QUEUE_FLAG_DISCARD, zram->disk->queue);
1953
1954         /*
1955          * zram_bio_discard() will clear all logical blocks if logical block
1956          * size is identical with physical block size(PAGE_SIZE). But if it is
1957          * different, we will skip discarding some parts of logical blocks in
1958          * the part of the request range which isn't aligned to physical block
1959          * size.  So we can't ensure that all discarded logical blocks are
1960          * zeroed.
1961          */
1962         if (ZRAM_LOGICAL_BLOCK_SIZE == PAGE_SIZE)
1963                 blk_queue_max_write_zeroes_sectors(zram->disk->queue, UINT_MAX);
1964
1965         blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, zram->disk->queue);
1966         device_add_disk(NULL, zram->disk, zram_disk_attr_groups);
1967
1968         strlcpy(zram->compressor, default_compressor, sizeof(zram->compressor));
1969
1970         zram_debugfs_register(zram);
1971         pr_info("Added device: %s\n", zram->disk->disk_name);
1972         return device_id;
1973
1974 out_free_queue:
1975         blk_cleanup_queue(queue);
1976 out_free_idr:
1977         idr_remove(&zram_index_idr, device_id);
1978 out_free_dev:
1979         kfree(zram);
1980         return ret;
1981 }
1982
1983 static int zram_remove(struct zram *zram)
1984 {
1985         struct block_device *bdev;
1986
1987         bdev = bdget_disk(zram->disk, 0);
1988         if (!bdev)
1989                 return -ENOMEM;
1990
1991         mutex_lock(&bdev->bd_mutex);
1992         if (bdev->bd_openers || zram->claim) {
1993                 mutex_unlock(&bdev->bd_mutex);
1994                 bdput(bdev);
1995                 return -EBUSY;
1996         }
1997
1998         zram->claim = true;
1999         mutex_unlock(&bdev->bd_mutex);
2000
2001         zram_debugfs_unregister(zram);
2002
2003         /* Make sure all the pending I/O are finished */
2004         fsync_bdev(bdev);
2005         zram_reset_device(zram);
2006         bdput(bdev);
2007
2008         pr_info("Removed device: %s\n", zram->disk->disk_name);
2009
2010         del_gendisk(zram->disk);
2011         blk_cleanup_queue(zram->disk->queue);
2012         put_disk(zram->disk);
2013         kfree(zram);
2014         return 0;
2015 }
2016
2017 /* zram-control sysfs attributes */
2018
2019 /*
2020  * NOTE: hot_add attribute is not the usual read-only sysfs attribute. In a
2021  * sense that reading from this file does alter the state of your system -- it
2022  * creates a new un-initialized zram device and returns back this device's
2023  * device_id (or an error code if it fails to create a new device).
2024  */
2025 static ssize_t hot_add_show(struct class *class,
2026                         struct class_attribute *attr,
2027                         char *buf)
2028 {
2029         int ret;
2030
2031         mutex_lock(&zram_index_mutex);
2032         ret = zram_add();
2033         mutex_unlock(&zram_index_mutex);
2034
2035         if (ret < 0)
2036                 return ret;
2037         return scnprintf(buf, PAGE_SIZE, "%d\n", ret);
2038 }
2039 static struct class_attribute class_attr_hot_add =
2040         __ATTR(hot_add, 0400, hot_add_show, NULL);
2041
2042 static ssize_t hot_remove_store(struct class *class,
2043                         struct class_attribute *attr,
2044                         const char *buf,
2045                         size_t count)
2046 {
2047         struct zram *zram;
2048         int ret, dev_id;
2049
2050         /* dev_id is gendisk->first_minor, which is `int' */
2051         ret = kstrtoint(buf, 10, &dev_id);
2052         if (ret)
2053                 return ret;
2054         if (dev_id < 0)
2055                 return -EINVAL;
2056
2057         mutex_lock(&zram_index_mutex);
2058
2059         zram = idr_find(&zram_index_idr, dev_id);
2060         if (zram) {
2061                 ret = zram_remove(zram);
2062                 if (!ret)
2063                         idr_remove(&zram_index_idr, dev_id);
2064         } else {
2065                 ret = -ENODEV;
2066         }
2067
2068         mutex_unlock(&zram_index_mutex);
2069         return ret ? ret : count;
2070 }
2071 static CLASS_ATTR_WO(hot_remove);
2072
2073 static struct attribute *zram_control_class_attrs[] = {
2074         &class_attr_hot_add.attr,
2075         &class_attr_hot_remove.attr,
2076         NULL,
2077 };
2078 ATTRIBUTE_GROUPS(zram_control_class);
2079
2080 static struct class zram_control_class = {
2081         .name           = "zram-control",
2082         .owner          = THIS_MODULE,
2083         .class_groups   = zram_control_class_groups,
2084 };
2085
2086 static int zram_remove_cb(int id, void *ptr, void *data)
2087 {
2088         zram_remove(ptr);
2089         return 0;
2090 }
2091
2092 static void destroy_devices(void)
2093 {
2094         class_unregister(&zram_control_class);
2095         idr_for_each(&zram_index_idr, &zram_remove_cb, NULL);
2096         zram_debugfs_destroy();
2097         idr_destroy(&zram_index_idr);
2098         unregister_blkdev(zram_major, "zram");
2099         cpuhp_remove_multi_state(CPUHP_ZCOMP_PREPARE);
2100 }
2101
2102 static int __init zram_init(void)
2103 {
2104         int ret;
2105
2106         ret = cpuhp_setup_state_multi(CPUHP_ZCOMP_PREPARE, "block/zram:prepare",
2107                                       zcomp_cpu_up_prepare, zcomp_cpu_dead);
2108         if (ret < 0)
2109                 return ret;
2110
2111         ret = class_register(&zram_control_class);
2112         if (ret) {
2113                 pr_err("Unable to register zram-control class\n");
2114                 cpuhp_remove_multi_state(CPUHP_ZCOMP_PREPARE);
2115                 return ret;
2116         }
2117
2118         zram_debugfs_create();
2119         zram_major = register_blkdev(0, "zram");
2120         if (zram_major <= 0) {
2121                 pr_err("Unable to get major number\n");
2122                 class_unregister(&zram_control_class);
2123                 cpuhp_remove_multi_state(CPUHP_ZCOMP_PREPARE);
2124                 return -EBUSY;
2125         }
2126
2127         while (num_devices != 0) {
2128                 mutex_lock(&zram_index_mutex);
2129                 ret = zram_add();
2130                 mutex_unlock(&zram_index_mutex);
2131                 if (ret < 0)
2132                         goto out_error;
2133                 num_devices--;
2134         }
2135
2136         return 0;
2137
2138 out_error:
2139         destroy_devices();
2140         return ret;
2141 }
2142
2143 static void __exit zram_exit(void)
2144 {
2145         destroy_devices();
2146 }
2147
2148 module_init(zram_init);
2149 module_exit(zram_exit);
2150
2151 module_param(num_devices, uint, 0);
2152 MODULE_PARM_DESC(num_devices, "Number of pre-created zram devices");
2153
2154 MODULE_LICENSE("Dual BSD/GPL");
2155 MODULE_AUTHOR("Nitin Gupta <ngupta@vflare.org>");
2156 MODULE_DESCRIPTION("Compressed RAM Block Device");