drivers/block/zram/zram_drv.c

   1 /*
   2  * Compressed RAM block device
   3  *
   4  * Copyright (C) 2008, 2009, 2010  Nitin Gupta
   5  *               2012, 2013 Minchan Kim
   6  *
   7  * This code is released using a dual license strategy: BSD/GPL
   8  * You can choose the licence that better fits your requirements.
   9  *
  10  * Released under the terms of 3-clause BSD License
  11  * Released under the terms of GNU General Public License Version 2.0
  12  *
  13  */
  14
  15 #define KMSG_COMPONENT "zram"
  16 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
  17
  18 #include <linux/module.h>
  19 #include <linux/kernel.h>
  20 #include <linux/bio.h>
  21 #include <linux/bitops.h>
  22 #include <linux/blkdev.h>
  23 #include <linux/buffer_head.h>
  24 #include <linux/device.h>
  25 #include <linux/genhd.h>
  26 #include <linux/highmem.h>
  27 #include <linux/slab.h>
  28 #include <linux/backing-dev.h>
  29 #include <linux/string.h>
  30 #include <linux/vmalloc.h>
  31 #include <linux/err.h>
  32 #include <linux/idr.h>
  33 #include <linux/sysfs.h>
  34 #include <linux/debugfs.h>
  35 #include <linux/cpuhotplug.h>
  36 #include <linux/part_stat.h>
  37
  38 #include "zram_drv.h"
  39
  40 static DEFINE_IDR(zram_index_idr);
  41 /* idr index must be protected */
  42 static DEFINE_MUTEX(zram_index_mutex);
  43
  44 static int zram_major;
  45 static const char *default_compressor = CONFIG_ZRAM_DEF_COMP;
  46
  47 /* Module params (documentation at end) */
  48 static unsigned int num_devices = 1;
  49 /*
  50  * Pages that compress to sizes equals or greater than this are stored
  51  * uncompressed in memory.
  52  */
  53 static size_t huge_class_size;
  54
  55 static const struct block_device_operations zram_devops;
  56 static const struct block_device_operations zram_wb_devops;
  57
  58 static void zram_free_page(struct zram *zram, size_t index);
  59 static int zram_bvec_read(struct zram *zram, struct bio_vec *bvec,
  60                                 u32 index, int offset, struct bio *bio);
  61
  62
  63 static int zram_slot_trylock(struct zram *zram, u32 index)
  64 {
  65         return bit_spin_trylock(ZRAM_LOCK, &zram->table[index].flags);
  66 }
  67
  68 static void zram_slot_lock(struct zram *zram, u32 index)
  69 {
  70         bit_spin_lock(ZRAM_LOCK, &zram->table[index].flags);
  71 }
  72
  73 static void zram_slot_unlock(struct zram *zram, u32 index)
  74 {
  75         bit_spin_unlock(ZRAM_LOCK, &zram->table[index].flags);
  76 }
  77
  78 static inline bool init_done(struct zram *zram)
  79 {
  80         return zram->disksize;
  81 }
  82
  83 static inline struct zram *dev_to_zram(struct device *dev)
  84 {
  85         return (struct zram *)dev_to_disk(dev)->private_data;
  86 }
  87
  88 static unsigned long zram_get_handle(struct zram *zram, u32 index)
  89 {
  90         return zram->table[index].handle;
  91 }
  92
  93 static void zram_set_handle(struct zram *zram, u32 index, unsigned long handle)
  94 {
  95         zram->table[index].handle = handle;
  96 }
  97
  98 /* flag operations require table entry bit_spin_lock() being held */
  99 static bool zram_test_flag(struct zram *zram, u32 index,
 100                         enum zram_pageflags flag)
 101 {
 102         return zram->table[index].flags & BIT(flag);
 103 }
 104
 105 static void zram_set_flag(struct zram *zram, u32 index,
 106                         enum zram_pageflags flag)
 107 {
 108         zram->table[index].flags |= BIT(flag);
 109 }
 110
 111 static void zram_clear_flag(struct zram *zram, u32 index,
 112                         enum zram_pageflags flag)
 113 {
 114         zram->table[index].flags &= ~BIT(flag);
 115 }
 116
 117 static inline void zram_set_element(struct zram *zram, u32 index,
 118                         unsigned long element)
 119 {
 120         zram->table[index].element = element;
 121 }
 122
 123 static unsigned long zram_get_element(struct zram *zram, u32 index)
 124 {
 125         return zram->table[index].element;
 126 }
 127
 128 static size_t zram_get_obj_size(struct zram *zram, u32 index)
 129 {
 130         return zram->table[index].flags & (BIT(ZRAM_FLAG_SHIFT) - 1);
 131 }
 132
 133 static void zram_set_obj_size(struct zram *zram,
 134                                         u32 index, size_t size)
 135 {
 136         unsigned long flags = zram->table[index].flags >> ZRAM_FLAG_SHIFT;
 137
 138         zram->table[index].flags = (flags << ZRAM_FLAG_SHIFT) | size;
 139 }
 140
 141 static inline bool zram_allocated(struct zram *zram, u32 index)
 142 {
 143         return zram_get_obj_size(zram, index) ||
 144                         zram_test_flag(zram, index, ZRAM_SAME) ||
 145                         zram_test_flag(zram, index, ZRAM_WB);
 146 }
 147
 148 #if PAGE_SIZE != 4096
 149 static inline bool is_partial_io(struct bio_vec *bvec)
 150 {
 151         return bvec->bv_len != PAGE_SIZE;
 152 }
 153 #else
 154 static inline bool is_partial_io(struct bio_vec *bvec)
 155 {
 156         return false;
 157 }
 158 #endif
 159
 160 /*
 161  * Check if request is within bounds and aligned on zram logical blocks.
 162  */
 163 static inline bool valid_io_request(struct zram *zram,
 164                 sector_t start, unsigned int size)
 165 {
 166         u64 end, bound;
 167
 168         /* unaligned request */
 169         if (unlikely(start & (ZRAM_SECTOR_PER_LOGICAL_BLOCK - 1)))
 170                 return false;
 171         if (unlikely(size & (ZRAM_LOGICAL_BLOCK_SIZE - 1)))
 172                 return false;
 173
 174         end = start + (size >> SECTOR_SHIFT);
 175         bound = zram->disksize >> SECTOR_SHIFT;
 176         /* out of range range */
 177         if (unlikely(start >= bound || end > bound || start > end))
 178                 return false;
 179
 180         /* I/O request is valid */
 181         return true;
 182 }
 183
 184 static void update_position(u32 *index, int *offset, struct bio_vec *bvec)
 185 {
 186         *index  += (*offset + bvec->bv_len) / PAGE_SIZE;
 187         *offset = (*offset + bvec->bv_len) % PAGE_SIZE;
 188 }
 189
 190 static inline void update_used_max(struct zram *zram,
 191                                         const unsigned long pages)
 192 {
 193         unsigned long old_max, cur_max;
 194
 195         old_max = atomic_long_read(&zram->stats.max_used_pages);
 196
 197         do {
 198                 cur_max = old_max;
 199                 if (pages > cur_max)
 200                         old_max = atomic_long_cmpxchg(
 201                                 &zram->stats.max_used_pages, cur_max, pages);
 202         } while (old_max != cur_max);
 203 }
 204
 205 static inline void zram_fill_page(void *ptr, unsigned long len,
 206                                         unsigned long value)
 207 {
 208         WARN_ON_ONCE(!IS_ALIGNED(len, sizeof(unsigned long)));
 209         memset_l(ptr, value, len / sizeof(unsigned long));
 210 }
 211
 212 static bool page_same_filled(void *ptr, unsigned long *element)
 213 {
 214         unsigned long *page;
 215         unsigned long val;
 216         unsigned int pos, last_pos = PAGE_SIZE / sizeof(*page) - 1;
 217
 218         page = (unsigned long *)ptr;
 219         val = page[0];
 220
 221         if (val != page[last_pos])
 222                 return false;
 223
 224         for (pos = 1; pos < last_pos; pos++) {
 225                 if (val != page[pos])
 226                         return false;
 227         }
 228
 229         *element = val;
 230
 231         return true;
 232 }
 233
 234 static ssize_t initstate_show(struct device *dev,
 235                 struct device_attribute *attr, char *buf)
 236 {
 237         u32 val;
 238         struct zram *zram = dev_to_zram(dev);
 239
 240         down_read(&zram->init_lock);
 241         val = init_done(zram);
 242         up_read(&zram->init_lock);
 243
 244         return scnprintf(buf, PAGE_SIZE, "%u\n", val);
 245 }
 246
 247 static ssize_t disksize_show(struct device *dev,
 248                 struct device_attribute *attr, char *buf)
 249 {
 250         struct zram *zram = dev_to_zram(dev);
 251
 252         return scnprintf(buf, PAGE_SIZE, "%llu\n", zram->disksize);
 253 }
 254
 255 static ssize_t mem_limit_store(struct device *dev,
 256                 struct device_attribute *attr, const char *buf, size_t len)
 257 {
 258         u64 limit;
 259         char *tmp;
 260         struct zram *zram = dev_to_zram(dev);
 261
 262         limit = memparse(buf, &tmp);
 263         if (buf == tmp) /* no chars parsed, invalid input */
 264                 return -EINVAL;
 265
 266         down_write(&zram->init_lock);
 267         zram->limit_pages = PAGE_ALIGN(limit) >> PAGE_SHIFT;
 268         up_write(&zram->init_lock);
 269
 270         return len;
 271 }
 272
 273 static ssize_t mem_used_max_store(struct device *dev,
 274                 struct device_attribute *attr, const char *buf, size_t len)
 275 {
 276         int err;
 277         unsigned long val;
 278         struct zram *zram = dev_to_zram(dev);
 279
 280         err = kstrtoul(buf, 10, &val);
 281         if (err || val != 0)
 282                 return -EINVAL;
 283
 284         down_read(&zram->init_lock);
 285         if (init_done(zram)) {
 286                 atomic_long_set(&zram->stats.max_used_pages,
 287                                 zs_get_total_pages(zram->mem_pool));
 288         }
 289         up_read(&zram->init_lock);
 290
 291         return len;
 292 }
 293
 294 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         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 407         /* hope filp_close flush all of IO */
 408         filp_close(zram->backing_dev, NULL);
 409         zram->backing_dev = NULL;
 410         zram->bdev = NULL;
 411         zram->disk->fops = &zram_devops;
 412         kvfree(zram->bitmap);
 413         zram->bitmap = NULL;
 414 }
 415
 416 static ssize_t backing_dev_show(struct device *dev,
 417                 struct device_attribute *attr, char *buf)
 418 {
 419         struct file *file;
 420         struct zram *zram = dev_to_zram(dev);
 421         char *p;
 422         ssize_t ret;
 423
 424         down_read(&zram->init_lock);
 425         file = zram->backing_dev;
 426         if (!file) {
 427                 memcpy(buf, "none\n", 5);
 428                 up_read(&zram->init_lock);
 429                 return 5;
 430         }
 431
 432         p = file_path(file, buf, PAGE_SIZE - 1);
 433         if (IS_ERR(p)) {
 434                 ret = PTR_ERR(p);
 435                 goto out;
 436         }
 437
 438         ret = strlen(p);
 439         memmove(buf, p, ret);
 440         buf[ret++] = '\n';
 441 out:
 442         up_read(&zram->init_lock);
 443         return ret;
 444 }
 445
 446 static ssize_t backing_dev_store(struct device *dev,
 447                 struct device_attribute *attr, const char *buf, size_t len)
 448 {
 449         char *file_name;
 450         size_t sz;
 451         struct file *backing_dev = NULL;
 452         struct inode *inode;
 453         struct address_space *mapping;
 454         unsigned int bitmap_sz;
 455         unsigned long nr_pages, *bitmap = NULL;
 456         struct block_device *bdev = NULL;
 457         int err;
 458         struct zram *zram = dev_to_zram(dev);
 459
 460         file_name = kmalloc(PATH_MAX, GFP_KERNEL);
 461         if (!file_name)
 462                 return -ENOMEM;
 463
 464         down_write(&zram->init_lock);
 465         if (init_done(zram)) {
 466                 pr_info("Can't setup backing device for initialized device\n");
 467                 err = -EBUSY;
 468                 goto out;
 469         }
 470
 471         strlcpy(file_name, buf, PATH_MAX);
 472         /* ignore trailing newline */
 473         sz = strlen(file_name);
 474         if (sz > 0 && file_name[sz - 1] == '\n')
 475                 file_name[sz - 1] = 0x00;
 476
 477         backing_dev = filp_open(file_name, O_RDWR|O_LARGEFILE, 0);
 478         if (IS_ERR(backing_dev)) {
 479                 err = PTR_ERR(backing_dev);
 480                 backing_dev = NULL;
 481                 goto out;
 482         }
 483
 484         mapping = backing_dev->f_mapping;
 485         inode = mapping->host;
 486
 487         /* Support only block device in this moment */
 488         if (!S_ISBLK(inode->i_mode)) {
 489                 err = -ENOTBLK;
 490                 goto out;
 491         }
 492
 493         bdev = blkdev_get_by_dev(inode->i_rdev,
 494                         FMODE_READ | FMODE_WRITE | FMODE_EXCL, zram);
 495         if (IS_ERR(bdev)) {
 496                 err = PTR_ERR(bdev);
 497                 bdev = NULL;
 498                 goto out;
 499         }
 500
 501         nr_pages = i_size_read(inode) >> PAGE_SHIFT;
 502         bitmap_sz = BITS_TO_LONGS(nr_pages) * sizeof(long);
 503         bitmap = kvzalloc(bitmap_sz, GFP_KERNEL);
 504         if (!bitmap) {
 505                 err = -ENOMEM;
 506                 goto out;
 507         }
 508
 509         reset_bdev(zram);
 510
 511         zram->bdev = bdev;
 512         zram->backing_dev = backing_dev;
 513         zram->bitmap = bitmap;
 514         zram->nr_pages = nr_pages;
 515         /*
 516          * With writeback feature, zram does asynchronous IO so it's no longer
 517          * synchronous device so let's remove synchronous io flag. Othewise,
 518          * upper layer(e.g., swap) could wait IO completion rather than
 519          * (submit and return), which will cause system sluggish.
 520          * Furthermore, when the IO function returns(e.g., swap_readpage),
 521          * upper layer expects IO was done so it could deallocate the page
 522          * freely but in fact, IO is going on so finally could cause
 523          * use-after-free when the IO is really done.
 524          */
 525         zram->disk->fops = &zram_wb_devops;
 526         up_write(&zram->init_lock);
 527
 528         pr_info("setup backing device %s\n", file_name);
 529         kfree(file_name);
 530
 531         return len;
 532 out:
 533         if (bitmap)
 534                 kvfree(bitmap);
 535
 536         if (bdev)
 537                 blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
 538
 539         if (backing_dev)
 540                 filp_close(backing_dev, NULL);
 541
 542         up_write(&zram->init_lock);
 543
 544         kfree(file_name);
 545
 546         return err;
 547 }
 548
 549 static unsigned long alloc_block_bdev(struct zram *zram)
 550 {
 551         unsigned long blk_idx = 1;
 552 retry:
 553         /* skip 0 bit to confuse zram.handle = 0 */
 554         blk_idx = find_next_zero_bit(zram->bitmap, zram->nr_pages, blk_idx);
 555         if (blk_idx == zram->nr_pages)
 556                 return 0;
 557
 558         if (test_and_set_bit(blk_idx, zram->bitmap))
 559                 goto retry;
 560
 561         atomic64_inc(&zram->stats.bd_count);
 562         return blk_idx;
 563 }
 564
 565 static void free_block_bdev(struct zram *zram, unsigned long blk_idx)
 566 {
 567         int was_set;
 568
 569         was_set = test_and_clear_bit(blk_idx, zram->bitmap);
 570         WARN_ON_ONCE(!was_set);
 571         atomic64_dec(&zram->stats.bd_count);
 572 }
 573
 574 static void zram_page_end_io(struct bio *bio)
 575 {
 576         struct page *page = bio_first_page_all(bio);
 577
 578         page_endio(page, op_is_write(bio_op(bio)),
 579                         blk_status_to_errno(bio->bi_status));
 580         bio_put(bio);
 581 }
 582
 583 /*
 584  * Returns 1 if the submission is successful.
 585  */
 586 static int read_from_bdev_async(struct zram *zram, struct bio_vec *bvec,
 587                         unsigned long entry, struct bio *parent)
 588 {
 589         struct bio *bio;
 590
 591         bio = bio_alloc(GFP_ATOMIC, 1);
 592         if (!bio)
 593                 return -ENOMEM;
 594
 595         bio->bi_iter.bi_sector = entry * (PAGE_SIZE >> 9);
 596         bio_set_dev(bio, zram->bdev);
 597         if (!bio_add_page(bio, bvec->bv_page, bvec->bv_len, bvec->bv_offset)) {
 598                 bio_put(bio);
 599                 return -EIO;
 600         }
 601
 602         if (!parent) {
 603                 bio->bi_opf = REQ_OP_READ;
 604                 bio->bi_end_io = zram_page_end_io;
 605         } else {
 606                 bio->bi_opf = parent->bi_opf;
 607                 bio_chain(bio, parent);
 608         }
 609
 610         submit_bio(bio);
 611         return 1;
 612 }
 613
 614 #define PAGE_WB_SIG "page_index="
 615
 616 #define PAGE_WRITEBACK 0
 617 #define HUGE_WRITEBACK 1
 618 #define IDLE_WRITEBACK 2
 619
 620
 621 static ssize_t writeback_store(struct device *dev,
 622                 struct device_attribute *attr, const char *buf, size_t len)
 623 {
 624         struct zram *zram = dev_to_zram(dev);
 625         unsigned long nr_pages = zram->disksize >> PAGE_SHIFT;
 626         unsigned long index = 0;
 627         struct bio bio;
 628         struct bio_vec bio_vec;
 629         struct page *page;
 630         ssize_t ret = len;
 631         int mode;
 632         unsigned long blk_idx = 0;
 633
 634         if (sysfs_streq(buf, "idle"))
 635                 mode = IDLE_WRITEBACK;
 636         else if (sysfs_streq(buf, "huge"))
 637                 mode = HUGE_WRITEBACK;
 638         else {
 639                 if (strncmp(buf, PAGE_WB_SIG, sizeof(PAGE_WB_SIG) - 1))
 640                         return -EINVAL;
 641
 642                 ret = kstrtol(buf + sizeof(PAGE_WB_SIG) - 1, 10, &index);
 643                 if (ret || index >= nr_pages)
 644                         return -EINVAL;
 645
 646                 nr_pages = 1;
 647                 mode = PAGE_WRITEBACK;
 648         }
 649
 650         down_read(&zram->init_lock);
 651         if (!init_done(zram)) {
 652                 ret = -EINVAL;
 653                 goto release_init_lock;
 654         }
 655
 656         if (!zram->backing_dev) {
 657                 ret = -ENODEV;
 658                 goto release_init_lock;
 659         }
 660
 661         page = alloc_page(GFP_KERNEL);
 662         if (!page) {
 663                 ret = -ENOMEM;
 664                 goto release_init_lock;
 665         }
 666
 667         while (nr_pages--) {
 668                 struct bio_vec bvec;
 669
 670                 bvec.bv_page = page;
 671                 bvec.bv_len = PAGE_SIZE;
 672                 bvec.bv_offset = 0;
 673
 674                 spin_lock(&zram->wb_limit_lock);
 675                 if (zram->wb_limit_enable && !zram->bd_wb_limit) {
 676                         spin_unlock(&zram->wb_limit_lock);
 677                         ret = -EIO;
 678                         break;
 679                 }
 680                 spin_unlock(&zram->wb_limit_lock);
 681
 682                 if (!blk_idx) {
 683                         blk_idx = alloc_block_bdev(zram);
 684                         if (!blk_idx) {
 685                                 ret = -ENOSPC;
 686                                 break;
 687                         }
 688                 }
 689
 690                 zram_slot_lock(zram, index);
 691                 if (!zram_allocated(zram, index))
 692                         goto next;
 693
 694                 if (zram_test_flag(zram, index, ZRAM_WB) ||
 695                                 zram_test_flag(zram, index, ZRAM_SAME) ||
 696                                 zram_test_flag(zram, index, ZRAM_UNDER_WB))
 697                         goto next;
 698
 699                 if (mode == IDLE_WRITEBACK &&
 700                           !zram_test_flag(zram, index, ZRAM_IDLE))
 701                         goto next;
 702                 if (mode == HUGE_WRITEBACK &&
 703                           !zram_test_flag(zram, index, ZRAM_HUGE))
 704                         goto next;
 705                 /*
 706                  * Clearing ZRAM_UNDER_WB is duty of caller.
 707                  * IOW, zram_free_page never clear it.
 708                  */
 709                 zram_set_flag(zram, index, ZRAM_UNDER_WB);
 710                 /* Need for hugepage writeback racing */
 711                 zram_set_flag(zram, index, ZRAM_IDLE);
 712                 zram_slot_unlock(zram, index);
 713                 if (zram_bvec_read(zram, &bvec, index, 0, NULL)) {
 714                         zram_slot_lock(zram, index);
 715                         zram_clear_flag(zram, index, ZRAM_UNDER_WB);
 716                         zram_clear_flag(zram, index, ZRAM_IDLE);
 717                         zram_slot_unlock(zram, index);
 718                         continue;
 719                 }
 720
 721                 bio_init(&bio, &bio_vec, 1);
 722                 bio_set_dev(&bio, zram->bdev);
 723                 bio.bi_iter.bi_sector = blk_idx * (PAGE_SIZE >> 9);
 724                 bio.bi_opf = REQ_OP_WRITE | REQ_SYNC;
 725
 726                 bio_add_page(&bio, bvec.bv_page, bvec.bv_len,
 727                                 bvec.bv_offset);
 728                 /*
 729                  * XXX: A single page IO would be inefficient for write
 730                  * but it would be not bad as starter.
 731                  */
 732                 ret = submit_bio_wait(&bio);
 733                 if (ret) {
 734                         zram_slot_lock(zram, index);
 735                         zram_clear_flag(zram, index, ZRAM_UNDER_WB);
 736                         zram_clear_flag(zram, index, ZRAM_IDLE);
 737                         zram_slot_unlock(zram, index);
 738                         continue;
 739                 }
 740
 741                 atomic64_inc(&zram->stats.bd_writes);
 742                 /*
 743                  * We released zram_slot_lock so need to check if the slot was
 744                  * changed. If there is freeing for the slot, we can catch it
 745                  * easily by zram_allocated.
 746                  * A subtle case is the slot is freed/reallocated/marked as
 747                  * ZRAM_IDLE again. To close the race, idle_store doesn't
 748                  * mark ZRAM_IDLE once it found the slot was ZRAM_UNDER_WB.
 749                  * Thus, we could close the race by checking ZRAM_IDLE bit.
 750                  */
 751                 zram_slot_lock(zram, index);
 752                 if (!zram_allocated(zram, index) ||
 753                           !zram_test_flag(zram, index, ZRAM_IDLE)) {
 754                         zram_clear_flag(zram, index, ZRAM_UNDER_WB);
 755                         zram_clear_flag(zram, index, ZRAM_IDLE);
 756                         goto next;
 757                 }
 758
 759                 zram_free_page(zram, index);
 760                 zram_clear_flag(zram, index, ZRAM_UNDER_WB);
 761                 zram_set_flag(zram, index, ZRAM_WB);
 762                 zram_set_element(zram, index, blk_idx);
 763                 blk_idx = 0;
 764                 atomic64_inc(&zram->stats.pages_stored);
 765                 spin_lock(&zram->wb_limit_lock);
 766                 if (zram->wb_limit_enable && zram->bd_wb_limit > 0)
 767                         zram->bd_wb_limit -=  1UL << (PAGE_SHIFT - 12);
 768                 spin_unlock(&zram->wb_limit_lock);
 769 next:
 770                 zram_slot_unlock(zram, index);
 771         }
 772
 773         if (blk_idx)
 774                 free_block_bdev(zram, blk_idx);
 775         __free_page(page);
 776 release_init_lock:
 777         up_read(&zram->init_lock);
 778
 779         return ret;
 780 }
 781
 782 struct zram_work {
 783         struct work_struct work;
 784         struct zram *zram;
 785         unsigned long entry;
 786         struct bio *bio;
 787         struct bio_vec bvec;
 788 };
 789
 790 #if PAGE_SIZE != 4096
 791 static void zram_sync_read(struct work_struct *work)
 792 {
 793         struct zram_work *zw = container_of(work, struct zram_work, work);
 794         struct zram *zram = zw->zram;
 795         unsigned long entry = zw->entry;
 796         struct bio *bio = zw->bio;
 797
 798         read_from_bdev_async(zram, &zw->bvec, entry, bio);
 799 }
 800
 801 /*
 802  * Block layer want one ->submit_bio to be active at a time, so if we use
 803  * chained IO with parent IO in same context, it's a deadlock. To avoid that,
 804  * use a worker thread context.
 805  */
 806 static int read_from_bdev_sync(struct zram *zram, struct bio_vec *bvec,
 807                                 unsigned long entry, struct bio *bio)
 808 {
 809         struct zram_work work;
 810
 811         work.bvec = *bvec;
 812         work.zram = zram;
 813         work.entry = entry;
 814         work.bio = bio;
 815
 816         INIT_WORK_ONSTACK(&work.work, zram_sync_read);
 817         queue_work(system_unbound_wq, &work.work);
 818         flush_work(&work.work);
 819         destroy_work_on_stack(&work.work);
 820
 821         return 1;
 822 }
 823 #else
 824 static int read_from_bdev_sync(struct zram *zram, struct bio_vec *bvec,
 825                                 unsigned long entry, struct bio *bio)
 826 {
 827         WARN_ON(1);
 828         return -EIO;
 829 }
 830 #endif
 831
 832 static int read_from_bdev(struct zram *zram, struct bio_vec *bvec,
 833                         unsigned long entry, struct bio *parent, bool sync)
 834 {
 835         atomic64_inc(&zram->stats.bd_reads);
 836         if (sync)
 837                 return read_from_bdev_sync(zram, bvec, entry, parent);
 838         else
 839                 return read_from_bdev_async(zram, bvec, entry, parent);
 840 }
 841 #else
 842 static inline void reset_bdev(struct zram *zram) {};
 843 static int read_from_bdev(struct zram *zram, struct bio_vec *bvec,
 844                         unsigned long entry, struct bio *parent, bool sync)
 845 {
 846         return -EIO;
 847 }
 848
 849 static void free_block_bdev(struct zram *zram, unsigned long blk_idx) {};
 850 #endif
 851
 852 #ifdef CONFIG_ZRAM_MEMORY_TRACKING
 853
 854 static struct dentry *zram_debugfs_root;
 855
 856 static void zram_debugfs_create(void)
 857 {
 858         zram_debugfs_root = debugfs_create_dir("zram", NULL);
 859 }
 860
 861 static void zram_debugfs_destroy(void)
 862 {
 863         debugfs_remove_recursive(zram_debugfs_root);
 864 }
 865
 866 static void zram_accessed(struct zram *zram, u32 index)
 867 {
 868         zram_clear_flag(zram, index, ZRAM_IDLE);
 869         zram->table[index].ac_time = ktime_get_boottime();
 870 }
 871
 872 static ssize_t read_block_state(struct file *file, char __user *buf,
 873                                 size_t count, loff_t *ppos)
 874 {
 875         char *kbuf;
 876         ssize_t index, written = 0;
 877         struct zram *zram = file->private_data;
 878         unsigned long nr_pages = zram->disksize >> PAGE_SHIFT;
 879         struct timespec64 ts;
 880
 881         kbuf = kvmalloc(count, GFP_KERNEL);
 882         if (!kbuf)
 883                 return -ENOMEM;
 884
 885         down_read(&zram->init_lock);
 886         if (!init_done(zram)) {
 887                 up_read(&zram->init_lock);
 888                 kvfree(kbuf);
 889                 return -EINVAL;
 890         }
 891
 892         for (index = *ppos; index < nr_pages; index++) {
 893                 int copied;
 894
 895                 zram_slot_lock(zram, index);
 896                 if (!zram_allocated(zram, index))
 897                         goto next;
 898
 899                 ts = ktime_to_timespec64(zram->table[index].ac_time);
 900                 copied = snprintf(kbuf + written, count,
 901                         "%12zd %12lld.%06lu %c%c%c%c\n",
 902                         index, (s64)ts.tv_sec,
 903                         ts.tv_nsec / NSEC_PER_USEC,
 904                         zram_test_flag(zram, index, ZRAM_SAME) ? 's' : '.',
 905                         zram_test_flag(zram, index, ZRAM_WB) ? 'w' : '.',
 906                         zram_test_flag(zram, index, ZRAM_HUGE) ? 'h' : '.',
 907                         zram_test_flag(zram, index, ZRAM_IDLE) ? 'i' : '.');
 908
 909                 if (count < copied) {
 910                         zram_slot_unlock(zram, index);
 911                         break;
 912                 }
 913                 written += copied;
 914                 count -= copied;
 915 next:
 916                 zram_slot_unlock(zram, index);
 917                 *ppos += 1;
 918         }
 919
 920         up_read(&zram->init_lock);
 921         if (copy_to_user(buf, kbuf, written))
 922                 written = -EFAULT;
 923         kvfree(kbuf);
 924
 925         return written;
 926 }
 927
 928 static const struct file_operations proc_zram_block_state_op = {
 929         .open = simple_open,
 930         .read = read_block_state,
 931         .llseek = default_llseek,
 932 };
 933
 934 static void zram_debugfs_register(struct zram *zram)
 935 {
 936         if (!zram_debugfs_root)
 937                 return;
 938
 939         zram->debugfs_dir = debugfs_create_dir(zram->disk->disk_name,
 940                                                 zram_debugfs_root);
 941         debugfs_create_file("block_state", 0400, zram->debugfs_dir,
 942                                 zram, &proc_zram_block_state_op);
 943 }
 944
 945 static void zram_debugfs_unregister(struct zram *zram)
 946 {
 947         debugfs_remove_recursive(zram->debugfs_dir);
 948 }
 949 #else
 950 static void zram_debugfs_create(void) {};
 951 static void zram_debugfs_destroy(void) {};
 952 static void zram_accessed(struct zram *zram, u32 index)
 953 {
 954         zram_clear_flag(zram, index, ZRAM_IDLE);
 955 };
 956 static void zram_debugfs_register(struct zram *zram) {};
 957 static void zram_debugfs_unregister(struct zram *zram) {};
 958 #endif
 959
 960 /*
 961  * We switched to per-cpu streams and this attr is not needed anymore.
 962  * However, we will keep it around for some time, because:
 963  * a) we may revert per-cpu streams in the future
 964  * b) it's visible to user space and we need to follow our 2 years
 965  *    retirement rule; but we already have a number of 'soon to be
 966  *    altered' attrs, so max_comp_streams need to wait for the next
 967  *    layoff cycle.
 968  */
 969 static ssize_t max_comp_streams_show(struct device *dev,
 970                 struct device_attribute *attr, char *buf)
 971 {
 972         return scnprintf(buf, PAGE_SIZE, "%d\n", num_online_cpus());
 973 }
 974
 975 static ssize_t max_comp_streams_store(struct device *dev,
 976                 struct device_attribute *attr, const char *buf, size_t len)
 977 {
 978         return len;
 979 }
 980
 981 static ssize_t comp_algorithm_show(struct device *dev,
 982                 struct device_attribute *attr, char *buf)
 983 {
 984         size_t sz;
 985         struct zram *zram = dev_to_zram(dev);
 986
 987         down_read(&zram->init_lock);
 988         sz = zcomp_available_show(zram->compressor, buf);
 989         up_read(&zram->init_lock);
 990
 991         return sz;
 992 }
 993
 994 static ssize_t comp_algorithm_store(struct device *dev,
 995                 struct device_attribute *attr, const char *buf, size_t len)
 996 {
 997         struct zram *zram = dev_to_zram(dev);
 998         char compressor[ARRAY_SIZE(zram->compressor)];
 999         size_t sz;
1000
1001         strlcpy(compressor, buf, sizeof(compressor));
1002         /* ignore trailing newline */
1003         sz = strlen(compressor);
1004         if (sz > 0 && compressor[sz - 1] == '\n')
1005                 compressor[sz - 1] = 0x00;
1006
1007         if (!zcomp_available_algorithm(compressor))
1008                 return -EINVAL;
1009
1010         down_write(&zram->init_lock);
1011         if (init_done(zram)) {
1012                 up_write(&zram->init_lock);
1013                 pr_info("Can't change algorithm for initialized device\n");
1014                 return -EBUSY;
1015         }
1016
1017         strcpy(zram->compressor, compressor);
1018         up_write(&zram->init_lock);
1019         return len;
1020 }
1021
1022 static ssize_t compact_store(struct device *dev,
1023                 struct device_attribute *attr, const char *buf, size_t len)
1024 {
1025         struct zram *zram = dev_to_zram(dev);
1026
1027         down_read(&zram->init_lock);
1028         if (!init_done(zram)) {
1029                 up_read(&zram->init_lock);
1030                 return -EINVAL;
1031         }
1032
1033         zs_compact(zram->mem_pool);
1034         up_read(&zram->init_lock);
1035
1036         return len;
1037 }
1038
1039 static ssize_t io_stat_show(struct device *dev,
1040                 struct device_attribute *attr, char *buf)
1041 {
1042         struct zram *zram = dev_to_zram(dev);
1043         ssize_t ret;
1044
1045         down_read(&zram->init_lock);
1046         ret = scnprintf(buf, PAGE_SIZE,
1047                         "%8llu %8llu %8llu %8llu\n",
1048                         (u64)atomic64_read(&zram->stats.failed_reads),
1049                         (u64)atomic64_read(&zram->stats.failed_writes),
1050                         (u64)atomic64_read(&zram->stats.invalid_io),
1051                         (u64)atomic64_read(&zram->stats.notify_free));
1052         up_read(&zram->init_lock);
1053
1054         return ret;
1055 }
1056
1057 static ssize_t mm_stat_show(struct device *dev,
1058                 struct device_attribute *attr, char *buf)
1059 {
1060         struct zram *zram = dev_to_zram(dev);
1061         struct zs_pool_stats pool_stats;
1062         u64 orig_size, mem_used = 0;
1063         long max_used;
1064         ssize_t ret;
1065
1066         memset(&pool_stats, 0x00, sizeof(struct zs_pool_stats));
1067
1068         down_read(&zram->init_lock);
1069         if (init_done(zram)) {
1070                 mem_used = zs_get_total_pages(zram->mem_pool);
1071                 zs_pool_stats(zram->mem_pool, &pool_stats);
1072         }
1073
1074         orig_size = atomic64_read(&zram->stats.pages_stored);
1075         max_used = atomic_long_read(&zram->stats.max_used_pages);
1076
1077         ret = scnprintf(buf, PAGE_SIZE,
1078                         "%8llu %8llu %8llu %8lu %8ld %8llu %8lu %8llu %8llu\n",
1079                         orig_size << PAGE_SHIFT,
1080                         (u64)atomic64_read(&zram->stats.compr_data_size),
1081                         mem_used << PAGE_SHIFT,
1082                         zram->limit_pages << PAGE_SHIFT,
1083                         max_used << PAGE_SHIFT,
1084                         (u64)atomic64_read(&zram->stats.same_pages),
1085                         pool_stats.pages_compacted,
1086                         (u64)atomic64_read(&zram->stats.huge_pages),
1087                         (u64)atomic64_read(&zram->stats.huge_pages_since));
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                 atomic64_inc(&zram->stats.huge_pages_since);
1420         }
1421
1422         if (flags) {
1423                 zram_set_flag(zram, index, flags);
1424                 zram_set_element(zram, index, element);
1425         }  else {
1426                 zram_set_handle(zram, index, handle);
1427                 zram_set_obj_size(zram, index, comp_len);
1428         }
1429         zram_slot_unlock(zram, index);
1430
1431         /* Update stats */
1432         atomic64_inc(&zram->stats.pages_stored);
1433         return ret;
1434 }
1435
1436 static int zram_bvec_write(struct zram *zram, struct bio_vec *bvec,
1437                                 u32 index, int offset, struct bio *bio)
1438 {
1439         int ret;
1440         struct page *page = NULL;
1441         void *src;
1442         struct bio_vec vec;
1443
1444         vec = *bvec;
1445         if (is_partial_io(bvec)) {
1446                 void *dst;
1447                 /*
1448                  * This is a partial IO. We need to read the full page
1449                  * before to write the changes.
1450                  */
1451                 page = alloc_page(GFP_NOIO|__GFP_HIGHMEM);
1452                 if (!page)
1453                         return -ENOMEM;
1454
1455                 ret = __zram_bvec_read(zram, page, index, bio, true);
1456                 if (ret)
1457                         goto out;
1458
1459                 src = kmap_atomic(bvec->bv_page);
1460                 dst = kmap_atomic(page);
1461                 memcpy(dst + offset, src + bvec->bv_offset, bvec->bv_len);
1462                 kunmap_atomic(dst);
1463                 kunmap_atomic(src);
1464
1465                 vec.bv_page = page;
1466                 vec.bv_len = PAGE_SIZE;
1467                 vec.bv_offset = 0;
1468         }
1469
1470         ret = __zram_bvec_write(zram, &vec, index, bio);
1471 out:
1472         if (is_partial_io(bvec))
1473                 __free_page(page);
1474         return ret;
1475 }
1476
1477 /*
1478  * zram_bio_discard - handler on discard request
1479  * @index: physical block index in PAGE_SIZE units
1480  * @offset: byte offset within physical block
1481  */
1482 static void zram_bio_discard(struct zram *zram, u32 index,
1483                              int offset, struct bio *bio)
1484 {
1485         size_t n = bio->bi_iter.bi_size;
1486
1487         /*
1488          * zram manages data in physical block size units. Because logical block
1489          * size isn't identical with physical block size on some arch, we
1490          * could get a discard request pointing to a specific offset within a
1491          * certain physical block.  Although we can handle this request by
1492          * reading that physiclal block and decompressing and partially zeroing
1493          * and re-compressing and then re-storing it, this isn't reasonable
1494          * because our intent with a discard request is to save memory.  So
1495          * skipping this logical block is appropriate here.
1496          */
1497         if (offset) {
1498                 if (n <= (PAGE_SIZE - offset))
1499                         return;
1500
1501                 n -= (PAGE_SIZE - offset);
1502                 index++;
1503         }
1504
1505         while (n >= PAGE_SIZE) {
1506                 zram_slot_lock(zram, index);
1507                 zram_free_page(zram, index);
1508                 zram_slot_unlock(zram, index);
1509                 atomic64_inc(&zram->stats.notify_free);
1510                 index++;
1511                 n -= PAGE_SIZE;
1512         }
1513 }
1514
1515 /*
1516  * Returns errno if it has some problem. Otherwise return 0 or 1.
1517  * Returns 0 if IO request was done synchronously
1518  * Returns 1 if IO request was successfully submitted.
1519  */
1520 static int zram_bvec_rw(struct zram *zram, struct bio_vec *bvec, u32 index,
1521                         int offset, unsigned int op, struct bio *bio)
1522 {
1523         int ret;
1524
1525         if (!op_is_write(op)) {
1526                 atomic64_inc(&zram->stats.num_reads);
1527                 ret = zram_bvec_read(zram, bvec, index, offset, bio);
1528                 flush_dcache_page(bvec->bv_page);
1529         } else {
1530                 atomic64_inc(&zram->stats.num_writes);
1531                 ret = zram_bvec_write(zram, bvec, index, offset, bio);
1532         }
1533
1534         zram_slot_lock(zram, index);
1535         zram_accessed(zram, index);
1536         zram_slot_unlock(zram, index);
1537
1538         if (unlikely(ret < 0)) {
1539                 if (!op_is_write(op))
1540                         atomic64_inc(&zram->stats.failed_reads);
1541                 else
1542                         atomic64_inc(&zram->stats.failed_writes);
1543         }
1544
1545         return ret;
1546 }
1547
1548 static void __zram_make_request(struct zram *zram, struct bio *bio)
1549 {
1550         int offset;
1551         u32 index;
1552         struct bio_vec bvec;
1553         struct bvec_iter iter;
1554         unsigned long start_time;
1555
1556         index = bio->bi_iter.bi_sector >> SECTORS_PER_PAGE_SHIFT;
1557         offset = (bio->bi_iter.bi_sector &
1558                   (SECTORS_PER_PAGE - 1)) << SECTOR_SHIFT;
1559
1560         switch (bio_op(bio)) {
1561         case REQ_OP_DISCARD:
1562         case REQ_OP_WRITE_ZEROES:
1563                 zram_bio_discard(zram, index, offset, bio);
1564                 bio_endio(bio);
1565                 return;
1566         default:
1567                 break;
1568         }
1569
1570         start_time = bio_start_io_acct(bio);
1571         bio_for_each_segment(bvec, bio, iter) {
1572                 struct bio_vec bv = bvec;
1573                 unsigned int unwritten = bvec.bv_len;
1574
1575                 do {
1576                         bv.bv_len = min_t(unsigned int, PAGE_SIZE - offset,
1577                                                         unwritten);
1578                         if (zram_bvec_rw(zram, &bv, index, offset,
1579                                          bio_op(bio), bio) < 0) {
1580                                 bio->bi_status = BLK_STS_IOERR;
1581                                 break;
1582                         }
1583
1584                         bv.bv_offset += bv.bv_len;
1585                         unwritten -= bv.bv_len;
1586
1587                         update_position(&index, &offset, &bv);
1588                 } while (unwritten);
1589         }
1590         bio_end_io_acct(bio, start_time);
1591         bio_endio(bio);
1592 }
1593
1594 /*
1595  * Handler function for all zram I/O requests.
1596  */
1597 static blk_qc_t zram_submit_bio(struct bio *bio)
1598 {
1599         struct zram *zram = bio->bi_disk->private_data;
1600
1601         if (!valid_io_request(zram, bio->bi_iter.bi_sector,
1602                                         bio->bi_iter.bi_size)) {
1603                 atomic64_inc(&zram->stats.invalid_io);
1604                 goto error;
1605         }
1606
1607         __zram_make_request(zram, bio);
1608         return BLK_QC_T_NONE;
1609
1610 error:
1611         bio_io_error(bio);
1612         return BLK_QC_T_NONE;
1613 }
1614
1615 static void zram_slot_free_notify(struct block_device *bdev,
1616                                 unsigned long index)
1617 {
1618         struct zram *zram;
1619
1620         zram = bdev->bd_disk->private_data;
1621
1622         atomic64_inc(&zram->stats.notify_free);
1623         if (!zram_slot_trylock(zram, index)) {
1624                 atomic64_inc(&zram->stats.miss_free);
1625                 return;
1626         }
1627
1628         zram_free_page(zram, index);
1629         zram_slot_unlock(zram, index);
1630 }
1631
1632 static int zram_rw_page(struct block_device *bdev, sector_t sector,
1633                        struct page *page, unsigned int op)
1634 {
1635         int offset, ret;
1636         u32 index;
1637         struct zram *zram;
1638         struct bio_vec bv;
1639         unsigned long start_time;
1640
1641         if (PageTransHuge(page))
1642                 return -ENOTSUPP;
1643         zram = bdev->bd_disk->private_data;
1644
1645         if (!valid_io_request(zram, sector, PAGE_SIZE)) {
1646                 atomic64_inc(&zram->stats.invalid_io);
1647                 ret = -EINVAL;
1648                 goto out;
1649         }
1650
1651         index = sector >> SECTORS_PER_PAGE_SHIFT;
1652         offset = (sector & (SECTORS_PER_PAGE - 1)) << SECTOR_SHIFT;
1653
1654         bv.bv_page = page;
1655         bv.bv_len = PAGE_SIZE;
1656         bv.bv_offset = 0;
1657
1658         start_time = disk_start_io_acct(bdev->bd_disk, SECTORS_PER_PAGE, op);
1659         ret = zram_bvec_rw(zram, &bv, index, offset, op, NULL);
1660         disk_end_io_acct(bdev->bd_disk, op, start_time);
1661 out:
1662         /*
1663          * If I/O fails, just return error(ie, non-zero) without
1664          * calling page_endio.
1665          * It causes resubmit the I/O with bio request by upper functions
1666          * of rw_page(e.g., swap_readpage, __swap_writepage) and
1667          * bio->bi_end_io does things to handle the error
1668          * (e.g., SetPageError, set_page_dirty and extra works).
1669          */
1670         if (unlikely(ret < 0))
1671                 return ret;
1672
1673         switch (ret) {
1674         case 0:
1675                 page_endio(page, op_is_write(op), 0);
1676                 break;
1677         case 1:
1678                 ret = 0;
1679                 break;
1680         default:
1681                 WARN_ON(1);
1682         }
1683         return ret;
1684 }
1685
1686 static void zram_reset_device(struct zram *zram)
1687 {
1688         struct zcomp *comp;
1689         u64 disksize;
1690
1691         down_write(&zram->init_lock);
1692
1693         zram->limit_pages = 0;
1694
1695         if (!init_done(zram)) {
1696                 up_write(&zram->init_lock);
1697                 return;
1698         }
1699
1700         comp = zram->comp;
1701         disksize = zram->disksize;
1702         zram->disksize = 0;
1703
1704         set_capacity_and_notify(zram->disk, 0);
1705         part_stat_set_all(zram->disk->part0, 0);
1706
1707         up_write(&zram->init_lock);
1708         /* I/O operation under all of CPU are done so let's free */
1709         zram_meta_free(zram, disksize);
1710         memset(&zram->stats, 0, sizeof(zram->stats));
1711         zcomp_destroy(comp);
1712         reset_bdev(zram);
1713 }
1714
1715 static ssize_t disksize_store(struct device *dev,
1716                 struct device_attribute *attr, const char *buf, size_t len)
1717 {
1718         u64 disksize;
1719         struct zcomp *comp;
1720         struct zram *zram = dev_to_zram(dev);
1721         int err;
1722
1723         disksize = memparse(buf, NULL);
1724         if (!disksize)
1725                 return -EINVAL;
1726
1727         down_write(&zram->init_lock);
1728         if (init_done(zram)) {
1729                 pr_info("Cannot change disksize for initialized device\n");
1730                 err = -EBUSY;
1731                 goto out_unlock;
1732         }
1733
1734         disksize = PAGE_ALIGN(disksize);
1735         if (!zram_meta_alloc(zram, disksize)) {
1736                 err = -ENOMEM;
1737                 goto out_unlock;
1738         }
1739
1740         comp = zcomp_create(zram->compressor);
1741         if (IS_ERR(comp)) {
1742                 pr_err("Cannot initialise %s compressing backend\n",
1743                                 zram->compressor);
1744                 err = PTR_ERR(comp);
1745                 goto out_free_meta;
1746         }
1747
1748         zram->comp = comp;
1749         zram->disksize = disksize;
1750         set_capacity_and_notify(zram->disk, zram->disksize >> SECTOR_SHIFT);
1751         up_write(&zram->init_lock);
1752
1753         return len;
1754
1755 out_free_meta:
1756         zram_meta_free(zram, disksize);
1757 out_unlock:
1758         up_write(&zram->init_lock);
1759         return err;
1760 }
1761
1762 static ssize_t reset_store(struct device *dev,
1763                 struct device_attribute *attr, const char *buf, size_t len)
1764 {
1765         int ret;
1766         unsigned short do_reset;
1767         struct zram *zram;
1768         struct block_device *bdev;
1769
1770         ret = kstrtou16(buf, 10, &do_reset);
1771         if (ret)
1772                 return ret;
1773
1774         if (!do_reset)
1775                 return -EINVAL;
1776
1777         zram = dev_to_zram(dev);
1778         bdev = zram->disk->part0;
1779
1780         mutex_lock(&bdev->bd_mutex);
1781         /* Do not reset an active device or claimed device */
1782         if (bdev->bd_openers || zram->claim) {
1783                 mutex_unlock(&bdev->bd_mutex);
1784                 return -EBUSY;
1785         }
1786
1787         /* From now on, anyone can't open /dev/zram[0-9] */
1788         zram->claim = true;
1789         mutex_unlock(&bdev->bd_mutex);
1790
1791         /* Make sure all the pending I/O are finished */
1792         fsync_bdev(bdev);
1793         zram_reset_device(zram);
1794
1795         mutex_lock(&bdev->bd_mutex);
1796         zram->claim = false;
1797         mutex_unlock(&bdev->bd_mutex);
1798
1799         return len;
1800 }
1801
1802 static int zram_open(struct block_device *bdev, fmode_t mode)
1803 {
1804         int ret = 0;
1805         struct zram *zram;
1806
1807         WARN_ON(!mutex_is_locked(&bdev->bd_mutex));
1808
1809         zram = bdev->bd_disk->private_data;
1810         /* zram was claimed to reset so open request fails */
1811         if (zram->claim)
1812                 ret = -EBUSY;
1813
1814         return ret;
1815 }
1816
1817 static const struct block_device_operations zram_devops = {
1818         .open = zram_open,
1819         .submit_bio = zram_submit_bio,
1820         .swap_slot_free_notify = zram_slot_free_notify,
1821         .rw_page = zram_rw_page,
1822         .owner = THIS_MODULE
1823 };
1824
1825 static const struct block_device_operations zram_wb_devops = {
1826         .open = zram_open,
1827         .submit_bio = zram_submit_bio,
1828         .swap_slot_free_notify = zram_slot_free_notify,
1829         .owner = THIS_MODULE
1830 };
1831
1832 static DEVICE_ATTR_WO(compact);
1833 static DEVICE_ATTR_RW(disksize);
1834 static DEVICE_ATTR_RO(initstate);
1835 static DEVICE_ATTR_WO(reset);
1836 static DEVICE_ATTR_WO(mem_limit);
1837 static DEVICE_ATTR_WO(mem_used_max);
1838 static DEVICE_ATTR_WO(idle);
1839 static DEVICE_ATTR_RW(max_comp_streams);
1840 static DEVICE_ATTR_RW(comp_algorithm);
1841 #ifdef CONFIG_ZRAM_WRITEBACK
1842 static DEVICE_ATTR_RW(backing_dev);
1843 static DEVICE_ATTR_WO(writeback);
1844 static DEVICE_ATTR_RW(writeback_limit);
1845 static DEVICE_ATTR_RW(writeback_limit_enable);
1846 #endif
1847
1848 static struct attribute *zram_disk_attrs[] = {
1849         &dev_attr_disksize.attr,
1850         &dev_attr_initstate.attr,
1851         &dev_attr_reset.attr,
1852         &dev_attr_compact.attr,
1853         &dev_attr_mem_limit.attr,
1854         &dev_attr_mem_used_max.attr,
1855         &dev_attr_idle.attr,
1856         &dev_attr_max_comp_streams.attr,
1857         &dev_attr_comp_algorithm.attr,
1858 #ifdef CONFIG_ZRAM_WRITEBACK
1859         &dev_attr_backing_dev.attr,
1860         &dev_attr_writeback.attr,
1861         &dev_attr_writeback_limit.attr,
1862         &dev_attr_writeback_limit_enable.attr,
1863 #endif
1864         &dev_attr_io_stat.attr,
1865         &dev_attr_mm_stat.attr,
1866 #ifdef CONFIG_ZRAM_WRITEBACK
1867         &dev_attr_bd_stat.attr,
1868 #endif
1869         &dev_attr_debug_stat.attr,
1870         NULL,
1871 };
1872
1873 static const struct attribute_group zram_disk_attr_group = {
1874         .attrs = zram_disk_attrs,
1875 };
1876
1877 static const struct attribute_group *zram_disk_attr_groups[] = {
1878         &zram_disk_attr_group,
1879         NULL,
1880 };
1881
1882 /*
1883  * Allocate and initialize new zram device. the function returns
1884  * '>= 0' device_id upon success, and negative value otherwise.
1885  */
1886 static int zram_add(void)
1887 {
1888         struct zram *zram;
1889         struct request_queue *queue;
1890         int ret, device_id;
1891
1892         zram = kzalloc(sizeof(struct zram), GFP_KERNEL);
1893         if (!zram)
1894                 return -ENOMEM;
1895
1896         ret = idr_alloc(&zram_index_idr, zram, 0, 0, GFP_KERNEL);
1897         if (ret < 0)
1898                 goto out_free_dev;
1899         device_id = ret;
1900
1901         init_rwsem(&zram->init_lock);
1902 #ifdef CONFIG_ZRAM_WRITEBACK
1903         spin_lock_init(&zram->wb_limit_lock);
1904 #endif
1905         queue = blk_alloc_queue(NUMA_NO_NODE);
1906         if (!queue) {
1907                 pr_err("Error allocating disk queue for device %d\n",
1908                         device_id);
1909                 ret = -ENOMEM;
1910                 goto out_free_idr;
1911         }
1912
1913         /* gendisk structure */
1914         zram->disk = alloc_disk(1);
1915         if (!zram->disk) {
1916                 pr_err("Error allocating disk structure for device %d\n",
1917                         device_id);
1918                 ret = -ENOMEM;
1919                 goto out_free_queue;
1920         }
1921
1922         zram->disk->major = zram_major;
1923         zram->disk->first_minor = device_id;
1924         zram->disk->fops = &zram_devops;
1925         zram->disk->queue = queue;
1926         zram->disk->private_data = zram;
1927         snprintf(zram->disk->disk_name, 16, "zram%d", device_id);
1928
1929         /* Actual capacity set using syfs (/sys/block/zram<id>/disksize */
1930         set_capacity(zram->disk, 0);
1931         /* zram devices sort of resembles non-rotational disks */
1932         blk_queue_flag_set(QUEUE_FLAG_NONROT, zram->disk->queue);
1933         blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, zram->disk->queue);
1934
1935         /*
1936          * To ensure that we always get PAGE_SIZE aligned
1937          * and n*PAGE_SIZED sized I/O requests.
1938          */
1939         blk_queue_physical_block_size(zram->disk->queue, PAGE_SIZE);
1940         blk_queue_logical_block_size(zram->disk->queue,
1941                                         ZRAM_LOGICAL_BLOCK_SIZE);
1942         blk_queue_io_min(zram->disk->queue, PAGE_SIZE);
1943         blk_queue_io_opt(zram->disk->queue, PAGE_SIZE);
1944         zram->disk->queue->limits.discard_granularity = PAGE_SIZE;
1945         blk_queue_max_discard_sectors(zram->disk->queue, UINT_MAX);
1946         blk_queue_flag_set(QUEUE_FLAG_DISCARD, zram->disk->queue);
1947
1948         /*
1949          * zram_bio_discard() will clear all logical blocks if logical block
1950          * size is identical with physical block size(PAGE_SIZE). But if it is
1951          * different, we will skip discarding some parts of logical blocks in
1952          * the part of the request range which isn't aligned to physical block
1953          * size.  So we can't ensure that all discarded logical blocks are
1954          * zeroed.
1955          */
1956         if (ZRAM_LOGICAL_BLOCK_SIZE == PAGE_SIZE)
1957                 blk_queue_max_write_zeroes_sectors(zram->disk->queue, UINT_MAX);
1958
1959         blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, zram->disk->queue);
1960         device_add_disk(NULL, zram->disk, zram_disk_attr_groups);
1961
1962         strlcpy(zram->compressor, default_compressor, sizeof(zram->compressor));
1963
1964         zram_debugfs_register(zram);
1965         pr_info("Added device: %s\n", zram->disk->disk_name);
1966         return device_id;
1967
1968 out_free_queue:
1969         blk_cleanup_queue(queue);
1970 out_free_idr:
1971         idr_remove(&zram_index_idr, device_id);
1972 out_free_dev:
1973         kfree(zram);
1974         return ret;
1975 }
1976
1977 static int zram_remove(struct zram *zram)
1978 {
1979         struct block_device *bdev = zram->disk->part0;
1980
1981         mutex_lock(&bdev->bd_mutex);
1982         if (bdev->bd_openers || zram->claim) {
1983                 mutex_unlock(&bdev->bd_mutex);
1984                 return -EBUSY;
1985         }
1986
1987         zram->claim = true;
1988         mutex_unlock(&bdev->bd_mutex);
1989
1990         zram_debugfs_unregister(zram);
1991
1992         /* Make sure all the pending I/O are finished */
1993         fsync_bdev(bdev);
1994         zram_reset_device(zram);
1995
1996         pr_info("Removed device: %s\n", zram->disk->disk_name);
1997
1998         del_gendisk(zram->disk);
1999         blk_cleanup_queue(zram->disk->queue);
2000         put_disk(zram->disk);
2001         kfree(zram);
2002         return 0;
2003 }
2004
2005 /* zram-control sysfs attributes */
2006
2007 /*
2008  * NOTE: hot_add attribute is not the usual read-only sysfs attribute. In a
2009  * sense that reading from this file does alter the state of your system -- it
2010  * creates a new un-initialized zram device and returns back this device's
2011  * device_id (or an error code if it fails to create a new device).
2012  */
2013 static ssize_t hot_add_show(struct class *class,
2014                         struct class_attribute *attr,
2015                         char *buf)
2016 {
2017         int ret;
2018
2019         mutex_lock(&zram_index_mutex);
2020         ret = zram_add();
2021         mutex_unlock(&zram_index_mutex);
2022
2023         if (ret < 0)
2024                 return ret;
2025         return scnprintf(buf, PAGE_SIZE, "%d\n", ret);
2026 }
2027 static struct class_attribute class_attr_hot_add =
2028         __ATTR(hot_add, 0400, hot_add_show, NULL);
2029
2030 static ssize_t hot_remove_store(struct class *class,
2031                         struct class_attribute *attr,
2032                         const char *buf,
2033                         size_t count)
2034 {
2035         struct zram *zram;
2036         int ret, dev_id;
2037
2038         /* dev_id is gendisk->first_minor, which is `int' */
2039         ret = kstrtoint(buf, 10, &dev_id);
2040         if (ret)
2041                 return ret;
2042         if (dev_id < 0)
2043                 return -EINVAL;
2044
2045         mutex_lock(&zram_index_mutex);
2046
2047         zram = idr_find(&zram_index_idr, dev_id);
2048         if (zram) {
2049                 ret = zram_remove(zram);
2050                 if (!ret)
2051                         idr_remove(&zram_index_idr, dev_id);
2052         } else {
2053                 ret = -ENODEV;
2054         }
2055
2056         mutex_unlock(&zram_index_mutex);
2057         return ret ? ret : count;
2058 }
2059 static CLASS_ATTR_WO(hot_remove);
2060
2061 static struct attribute *zram_control_class_attrs[] = {
2062         &class_attr_hot_add.attr,
2063         &class_attr_hot_remove.attr,
2064         NULL,
2065 };
2066 ATTRIBUTE_GROUPS(zram_control_class);
2067
2068 static struct class zram_control_class = {
2069         .name           = "zram-control",
2070         .owner          = THIS_MODULE,
2071         .class_groups   = zram_control_class_groups,
2072 };
2073
2074 static int zram_remove_cb(int id, void *ptr, void *data)
2075 {
2076         zram_remove(ptr);
2077         return 0;
2078 }
2079
2080 static void destroy_devices(void)
2081 {
2082         class_unregister(&zram_control_class);
2083         idr_for_each(&zram_index_idr, &zram_remove_cb, NULL);
2084         zram_debugfs_destroy();
2085         idr_destroy(&zram_index_idr);
2086         unregister_blkdev(zram_major, "zram");
2087         cpuhp_remove_multi_state(CPUHP_ZCOMP_PREPARE);
2088 }
2089
2090 static int __init zram_init(void)
2091 {
2092         int ret;
2093
2094         ret = cpuhp_setup_state_multi(CPUHP_ZCOMP_PREPARE, "block/zram:prepare",
2095                                       zcomp_cpu_up_prepare, zcomp_cpu_dead);
2096         if (ret < 0)
2097                 return ret;
2098
2099         ret = class_register(&zram_control_class);
2100         if (ret) {
2101                 pr_err("Unable to register zram-control class\n");
2102                 cpuhp_remove_multi_state(CPUHP_ZCOMP_PREPARE);
2103                 return ret;
2104         }
2105
2106         zram_debugfs_create();
2107         zram_major = register_blkdev(0, "zram");
2108         if (zram_major <= 0) {
2109                 pr_err("Unable to get major number\n");
2110                 class_unregister(&zram_control_class);
2111                 cpuhp_remove_multi_state(CPUHP_ZCOMP_PREPARE);
2112                 return -EBUSY;
2113         }
2114
2115         while (num_devices != 0) {
2116                 mutex_lock(&zram_index_mutex);
2117                 ret = zram_add();
2118                 mutex_unlock(&zram_index_mutex);
2119                 if (ret < 0)
2120                         goto out_error;
2121                 num_devices--;
2122         }
2123
2124         return 0;
2125
2126 out_error:
2127         destroy_devices();
2128         return ret;
2129 }
2130
2131 static void __exit zram_exit(void)
2132 {
2133         destroy_devices();
2134 }
2135
2136 module_init(zram_init);
2137 module_exit(zram_exit);
2138
2139 module_param(num_devices, uint, 0);
2140 MODULE_PARM_DESC(num_devices, "Number of pre-created zram devices");
2141
2142 MODULE_LICENSE("Dual BSD/GPL");
2143 MODULE_AUTHOR("Nitin Gupta <ngupta@vflare.org>");
2144 MODULE_DESCRIPTION("Compressed RAM Block Device");