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
  37 #include "zram_drv.h"
  38
  39 static DEFINE_IDR(zram_index_idr);
  40 /* idr index must be protected */
  41 static DEFINE_MUTEX(zram_index_mutex);
  42
  43 static int zram_major;
  44 static const char *default_compressor = "lzo";
  45
  46 /* Module params (documentation at end) */
  47 static unsigned int num_devices = 1;
  48 /*
  49  * Pages that compress to sizes equals or greater than this are stored
  50  * uncompressed in memory.
  51  */
  52 static size_t huge_class_size;
  53
  54 static void zram_free_page(struct zram *zram, size_t index);
  55 static int zram_bvec_read(struct zram *zram, struct bio_vec *bvec,
  56                                 u32 index, int offset, struct bio *bio);
  57
  58
  59 static int zram_slot_trylock(struct zram *zram, u32 index)
  60 {
  61         return bit_spin_trylock(ZRAM_LOCK, &zram->table[index].flags);
  62 }
  63
  64 static void zram_slot_lock(struct zram *zram, u32 index)
  65 {
  66         bit_spin_lock(ZRAM_LOCK, &zram->table[index].flags);
  67 }
  68
  69 static void zram_slot_unlock(struct zram *zram, u32 index)
  70 {
  71         bit_spin_unlock(ZRAM_LOCK, &zram->table[index].flags);
  72 }
  73
  74 static inline bool init_done(struct zram *zram)
  75 {
  76         return zram->disksize;
  77 }
  78
  79 static inline struct zram *dev_to_zram(struct device *dev)
  80 {
  81         return (struct zram *)dev_to_disk(dev)->private_data;
  82 }
  83
  84 static unsigned long zram_get_handle(struct zram *zram, u32 index)
  85 {
  86         return zram->table[index].handle;
  87 }
  88
  89 static void zram_set_handle(struct zram *zram, u32 index, unsigned long handle)
  90 {
  91         zram->table[index].handle = handle;
  92 }
  93
  94 /* flag operations require table entry bit_spin_lock() being held */
  95 static bool zram_test_flag(struct zram *zram, u32 index,
  96                         enum zram_pageflags flag)
  97 {
  98         return zram->table[index].flags & BIT(flag);
  99 }
 100
 101 static void zram_set_flag(struct zram *zram, u32 index,
 102                         enum zram_pageflags flag)
 103 {
 104         zram->table[index].flags |= BIT(flag);
 105 }
 106
 107 static void zram_clear_flag(struct zram *zram, u32 index,
 108                         enum zram_pageflags flag)
 109 {
 110         zram->table[index].flags &= ~BIT(flag);
 111 }
 112
 113 static inline void zram_set_element(struct zram *zram, u32 index,
 114                         unsigned long element)
 115 {
 116         zram->table[index].element = element;
 117 }
 118
 119 static unsigned long zram_get_element(struct zram *zram, u32 index)
 120 {
 121         return zram->table[index].element;
 122 }
 123
 124 static size_t zram_get_obj_size(struct zram *zram, u32 index)
 125 {
 126         return zram->table[index].flags & (BIT(ZRAM_FLAG_SHIFT) - 1);
 127 }
 128
 129 static void zram_set_obj_size(struct zram *zram,
 130                                         u32 index, size_t size)
 131 {
 132         unsigned long flags = zram->table[index].flags >> ZRAM_FLAG_SHIFT;
 133
 134         zram->table[index].flags = (flags << ZRAM_FLAG_SHIFT) | size;
 135 }
 136
 137 static inline bool zram_allocated(struct zram *zram, u32 index)
 138 {
 139         return zram_get_obj_size(zram, index) ||
 140                         zram_test_flag(zram, index, ZRAM_SAME) ||
 141                         zram_test_flag(zram, index, ZRAM_WB);
 142 }
 143
 144 #if PAGE_SIZE != 4096
 145 static inline bool is_partial_io(struct bio_vec *bvec)
 146 {
 147         return bvec->bv_len != PAGE_SIZE;
 148 }
 149 #else
 150 static inline bool is_partial_io(struct bio_vec *bvec)
 151 {
 152         return false;
 153 }
 154 #endif
 155
 156 /*
 157  * Check if request is within bounds and aligned on zram logical blocks.
 158  */
 159 static inline bool valid_io_request(struct zram *zram,
 160                 sector_t start, unsigned int size)
 161 {
 162         u64 end, bound;
 163
 164         /* unaligned request */
 165         if (unlikely(start & (ZRAM_SECTOR_PER_LOGICAL_BLOCK - 1)))
 166                 return false;
 167         if (unlikely(size & (ZRAM_LOGICAL_BLOCK_SIZE - 1)))
 168                 return false;
 169
 170         end = start + (size >> SECTOR_SHIFT);
 171         bound = zram->disksize >> SECTOR_SHIFT;
 172         /* out of range range */
 173         if (unlikely(start >= bound || end > bound || start > end))
 174                 return false;
 175
 176         /* I/O request is valid */
 177         return true;
 178 }
 179
 180 static void update_position(u32 *index, int *offset, struct bio_vec *bvec)
 181 {
 182         *index  += (*offset + bvec->bv_len) / PAGE_SIZE;
 183         *offset = (*offset + bvec->bv_len) % PAGE_SIZE;
 184 }
 185
 186 static inline void update_used_max(struct zram *zram,
 187                                         const unsigned long pages)
 188 {
 189         unsigned long old_max, cur_max;
 190
 191         old_max = atomic_long_read(&zram->stats.max_used_pages);
 192
 193         do {
 194                 cur_max = old_max;
 195                 if (pages > cur_max)
 196                         old_max = atomic_long_cmpxchg(
 197                                 &zram->stats.max_used_pages, cur_max, pages);
 198         } while (old_max != cur_max);
 199 }
 200
 201 static inline void zram_fill_page(void *ptr, unsigned long len,
 202                                         unsigned long value)
 203 {
 204         WARN_ON_ONCE(!IS_ALIGNED(len, sizeof(unsigned long)));
 205         memset_l(ptr, value, len / sizeof(unsigned long));
 206 }
 207
 208 static bool page_same_filled(void *ptr, unsigned long *element)
 209 {
 210         unsigned int pos;
 211         unsigned long *page;
 212         unsigned long val;
 213
 214         page = (unsigned long *)ptr;
 215         val = page[0];
 216
 217         for (pos = 1; pos < PAGE_SIZE / sizeof(*page); pos++) {
 218                 if (val != page[pos])
 219                         return false;
 220         }
 221
 222         *element = val;
 223
 224         return true;
 225 }
 226
 227 static ssize_t initstate_show(struct device *dev,
 228                 struct device_attribute *attr, char *buf)
 229 {
 230         u32 val;
 231         struct zram *zram = dev_to_zram(dev);
 232
 233         down_read(&zram->init_lock);
 234         val = init_done(zram);
 235         up_read(&zram->init_lock);
 236
 237         return scnprintf(buf, PAGE_SIZE, "%u\n", val);
 238 }
 239
 240 static ssize_t disksize_show(struct device *dev,
 241                 struct device_attribute *attr, char *buf)
 242 {
 243         struct zram *zram = dev_to_zram(dev);
 244
 245         return scnprintf(buf, PAGE_SIZE, "%llu\n", zram->disksize);
 246 }
 247
 248 static ssize_t mem_limit_store(struct device *dev,
 249                 struct device_attribute *attr, const char *buf, size_t len)
 250 {
 251         u64 limit;
 252         char *tmp;
 253         struct zram *zram = dev_to_zram(dev);
 254
 255         limit = memparse(buf, &tmp);
 256         if (buf == tmp) /* no chars parsed, invalid input */
 257                 return -EINVAL;
 258
 259         down_write(&zram->init_lock);
 260         zram->limit_pages = PAGE_ALIGN(limit) >> PAGE_SHIFT;
 261         up_write(&zram->init_lock);
 262
 263         return len;
 264 }
 265
 266 static ssize_t mem_used_max_store(struct device *dev,
 267                 struct device_attribute *attr, const char *buf, size_t len)
 268 {
 269         int err;
 270         unsigned long val;
 271         struct zram *zram = dev_to_zram(dev);
 272
 273         err = kstrtoul(buf, 10, &val);
 274         if (err || val != 0)
 275                 return -EINVAL;
 276
 277         down_read(&zram->init_lock);
 278         if (init_done(zram)) {
 279                 atomic_long_set(&zram->stats.max_used_pages,
 280                                 zs_get_total_pages(zram->mem_pool));
 281         }
 282         up_read(&zram->init_lock);
 283
 284         return len;
 285 }
 286
 287 static ssize_t idle_store(struct device *dev,
 288                 struct device_attribute *attr, const char *buf, size_t len)
 289 {
 290         struct zram *zram = dev_to_zram(dev);
 291         unsigned long nr_pages = zram->disksize >> PAGE_SHIFT;
 292         int index;
 293         char mode_buf[8];
 294         ssize_t sz;
 295
 296         sz = strscpy(mode_buf, buf, sizeof(mode_buf));
 297         if (sz <= 0)
 298                 return -EINVAL;
 299
 300         /* ignore trailing new line */
 301         if (mode_buf[sz - 1] == '\n')
 302                 mode_buf[sz - 1] = 0x00;
 303
 304         if (strcmp(mode_buf, "all"))
 305                 return -EINVAL;
 306
 307         down_read(&zram->init_lock);
 308         if (!init_done(zram)) {
 309                 up_read(&zram->init_lock);
 310                 return -EINVAL;
 311         }
 312
 313         for (index = 0; index < nr_pages; index++) {
 314                 /*
 315                  * Do not mark ZRAM_UNDER_WB slot as ZRAM_IDLE to close race.
 316                  * See the comment in writeback_store.
 317                  */
 318                 zram_slot_lock(zram, index);
 319                 if (zram_allocated(zram, index) &&
 320                                 !zram_test_flag(zram, index, ZRAM_UNDER_WB))
 321                         zram_set_flag(zram, index, ZRAM_IDLE);
 322                 zram_slot_unlock(zram, index);
 323         }
 324
 325         up_read(&zram->init_lock);
 326
 327         return len;
 328 }
 329
 330 #ifdef CONFIG_ZRAM_WRITEBACK
 331 static ssize_t writeback_limit_enable_store(struct device *dev,
 332                 struct device_attribute *attr, const char *buf, size_t len)
 333 {
 334         struct zram *zram = dev_to_zram(dev);
 335         u64 val;
 336         ssize_t ret = -EINVAL;
 337
 338         if (kstrtoull(buf, 10, &val))
 339                 return ret;
 340
 341         down_read(&zram->init_lock);
 342         spin_lock(&zram->wb_limit_lock);
 343         zram->wb_limit_enable = val;
 344         spin_unlock(&zram->wb_limit_lock);
 345         up_read(&zram->init_lock);
 346         ret = len;
 347
 348         return ret;
 349 }
 350
 351 static ssize_t writeback_limit_enable_show(struct device *dev,
 352                 struct device_attribute *attr, char *buf)
 353 {
 354         bool val;
 355         struct zram *zram = dev_to_zram(dev);
 356
 357         down_read(&zram->init_lock);
 358         spin_lock(&zram->wb_limit_lock);
 359         val = zram->wb_limit_enable;
 360         spin_unlock(&zram->wb_limit_lock);
 361         up_read(&zram->init_lock);
 362
 363         return scnprintf(buf, PAGE_SIZE, "%d\n", val);
 364 }
 365
 366 static ssize_t writeback_limit_store(struct device *dev,
 367                 struct device_attribute *attr, const char *buf, size_t len)
 368 {
 369         struct zram *zram = dev_to_zram(dev);
 370         u64 val;
 371         ssize_t ret = -EINVAL;
 372
 373         if (kstrtoull(buf, 10, &val))
 374                 return ret;
 375
 376         down_read(&zram->init_lock);
 377         spin_lock(&zram->wb_limit_lock);
 378         zram->bd_wb_limit = val;
 379         spin_unlock(&zram->wb_limit_lock);
 380         up_read(&zram->init_lock);
 381         ret = len;
 382
 383         return ret;
 384 }
 385
 386 static ssize_t writeback_limit_show(struct device *dev,
 387                 struct device_attribute *attr, char *buf)
 388 {
 389         u64 val;
 390         struct zram *zram = dev_to_zram(dev);
 391
 392         down_read(&zram->init_lock);
 393         spin_lock(&zram->wb_limit_lock);
 394         val = zram->bd_wb_limit;
 395         spin_unlock(&zram->wb_limit_lock);
 396         up_read(&zram->init_lock);
 397
 398         return scnprintf(buf, PAGE_SIZE, "%llu\n", val);
 399 }
 400
 401 static void reset_bdev(struct zram *zram)
 402 {
 403         struct block_device *bdev;
 404
 405         if (!zram->backing_dev)
 406                 return;
 407
 408         bdev = zram->bdev;
 409         if (zram->old_block_size)
 410                 set_blocksize(bdev, zram->old_block_size);
 411         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 412         /* hope filp_close flush all of IO */
 413         filp_close(zram->backing_dev, NULL);
 414         zram->backing_dev = NULL;
 415         zram->old_block_size = 0;
 416         zram->bdev = NULL;
 417         zram->disk->queue->backing_dev_info->capabilities |=
 418                                 BDI_CAP_SYNCHRONOUS_IO;
 419         kvfree(zram->bitmap);
 420         zram->bitmap = NULL;
 421 }
 422
 423 static ssize_t backing_dev_show(struct device *dev,
 424                 struct device_attribute *attr, char *buf)
 425 {
 426         struct zram *zram = dev_to_zram(dev);
 427         struct file *file = zram->backing_dev;
 428         char *p;
 429         ssize_t ret;
 430
 431         down_read(&zram->init_lock);
 432         if (!zram->backing_dev) {
 433                 memcpy(buf, "none\n", 5);
 434                 up_read(&zram->init_lock);
 435                 return 5;
 436         }
 437
 438         p = file_path(file, buf, PAGE_SIZE - 1);
 439         if (IS_ERR(p)) {
 440                 ret = PTR_ERR(p);
 441                 goto out;
 442         }
 443
 444         ret = strlen(p);
 445         memmove(buf, p, ret);
 446         buf[ret++] = '\n';
 447 out:
 448         up_read(&zram->init_lock);
 449         return ret;
 450 }
 451
 452 static ssize_t backing_dev_store(struct device *dev,
 453                 struct device_attribute *attr, const char *buf, size_t len)
 454 {
 455         char *file_name;
 456         size_t sz;
 457         struct file *backing_dev = NULL;
 458         struct inode *inode;
 459         struct address_space *mapping;
 460         unsigned int bitmap_sz, old_block_size = 0;
 461         unsigned long nr_pages, *bitmap = NULL;
 462         struct block_device *bdev = NULL;
 463         int err;
 464         struct zram *zram = dev_to_zram(dev);
 465
 466         file_name = kmalloc(PATH_MAX, GFP_KERNEL);
 467         if (!file_name)
 468                 return -ENOMEM;
 469
 470         down_write(&zram->init_lock);
 471         if (init_done(zram)) {
 472                 pr_info("Can't setup backing device for initialized device\n");
 473                 err = -EBUSY;
 474                 goto out;
 475         }
 476
 477         strlcpy(file_name, buf, PATH_MAX);
 478         /* ignore trailing newline */
 479         sz = strlen(file_name);
 480         if (sz > 0 && file_name[sz - 1] == '\n')
 481                 file_name[sz - 1] = 0x00;
 482
 483         backing_dev = filp_open(file_name, O_RDWR|O_LARGEFILE, 0);
 484         if (IS_ERR(backing_dev)) {
 485                 err = PTR_ERR(backing_dev);
 486                 backing_dev = NULL;
 487                 goto out;
 488         }
 489
 490         mapping = backing_dev->f_mapping;
 491         inode = mapping->host;
 492
 493         /* Support only block device in this moment */
 494         if (!S_ISBLK(inode->i_mode)) {
 495                 err = -ENOTBLK;
 496                 goto out;
 497         }
 498
 499         bdev = bdgrab(I_BDEV(inode));
 500         err = blkdev_get(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL, zram);
 501         if (err < 0) {
 502                 bdev = NULL;
 503                 goto out;
 504         }
 505
 506         nr_pages = i_size_read(inode) >> PAGE_SHIFT;
 507         bitmap_sz = BITS_TO_LONGS(nr_pages) * sizeof(long);
 508         bitmap = kvzalloc(bitmap_sz, GFP_KERNEL);
 509         if (!bitmap) {
 510                 err = -ENOMEM;
 511                 goto out;
 512         }
 513
 514         old_block_size = block_size(bdev);
 515         err = set_blocksize(bdev, PAGE_SIZE);
 516         if (err)
 517                 goto out;
 518
 519         reset_bdev(zram);
 520
 521         zram->old_block_size = old_block_size;
 522         zram->bdev = bdev;
 523         zram->backing_dev = backing_dev;
 524         zram->bitmap = bitmap;
 525         zram->nr_pages = nr_pages;
 526         /*
 527          * With writeback feature, zram does asynchronous IO so it's no longer
 528          * synchronous device so let's remove synchronous io flag. Othewise,
 529          * upper layer(e.g., swap) could wait IO completion rather than
 530          * (submit and return), which will cause system sluggish.
 531          * Furthermore, when the IO function returns(e.g., swap_readpage),
 532          * upper layer expects IO was done so it could deallocate the page
 533          * freely but in fact, IO is going on so finally could cause
 534          * use-after-free when the IO is really done.
 535          */
 536         zram->disk->queue->backing_dev_info->capabilities &=
 537                         ~BDI_CAP_SYNCHRONOUS_IO;
 538         up_write(&zram->init_lock);
 539
 540         pr_info("setup backing device %s\n", file_name);
 541         kfree(file_name);
 542
 543         return len;
 544 out:
 545         if (bitmap)
 546                 kvfree(bitmap);
 547
 548         if (bdev)
 549                 blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
 550
 551         if (backing_dev)
 552                 filp_close(backing_dev, NULL);
 553
 554         up_write(&zram->init_lock);
 555
 556         kfree(file_name);
 557
 558         return err;
 559 }
 560
 561 static unsigned long alloc_block_bdev(struct zram *zram)
 562 {
 563         unsigned long blk_idx = 1;
 564 retry:
 565         /* skip 0 bit to confuse zram.handle = 0 */
 566         blk_idx = find_next_zero_bit(zram->bitmap, zram->nr_pages, blk_idx);
 567         if (blk_idx == zram->nr_pages)
 568                 return 0;
 569
 570         if (test_and_set_bit(blk_idx, zram->bitmap))
 571                 goto retry;
 572
 573         atomic64_inc(&zram->stats.bd_count);
 574         return blk_idx;
 575 }
 576
 577 static void free_block_bdev(struct zram *zram, unsigned long blk_idx)
 578 {
 579         int was_set;
 580
 581         was_set = test_and_clear_bit(blk_idx, zram->bitmap);
 582         WARN_ON_ONCE(!was_set);
 583         atomic64_dec(&zram->stats.bd_count);
 584 }
 585
 586 static void zram_page_end_io(struct bio *bio)
 587 {
 588         struct page *page = bio_first_page_all(bio);
 589
 590         page_endio(page, op_is_write(bio_op(bio)),
 591                         blk_status_to_errno(bio->bi_status));
 592         bio_put(bio);
 593 }
 594
 595 /*
 596  * Returns 1 if the submission is successful.
 597  */
 598 static int read_from_bdev_async(struct zram *zram, struct bio_vec *bvec,
 599                         unsigned long entry, struct bio *parent)
 600 {
 601         struct bio *bio;
 602
 603         bio = bio_alloc(GFP_ATOMIC, 1);
 604         if (!bio)
 605                 return -ENOMEM;
 606
 607         bio->bi_iter.bi_sector = entry * (PAGE_SIZE >> 9);
 608         bio_set_dev(bio, zram->bdev);
 609         if (!bio_add_page(bio, bvec->bv_page, bvec->bv_len, bvec->bv_offset)) {
 610                 bio_put(bio);
 611                 return -EIO;
 612         }
 613
 614         if (!parent) {
 615                 bio->bi_opf = REQ_OP_READ;
 616                 bio->bi_end_io = zram_page_end_io;
 617         } else {
 618                 bio->bi_opf = parent->bi_opf;
 619                 bio_chain(bio, parent);
 620         }
 621
 622         submit_bio(bio);
 623         return 1;
 624 }
 625
 626 #define HUGE_WRITEBACK 1
 627 #define IDLE_WRITEBACK 2
 628
 629 static ssize_t writeback_store(struct device *dev,
 630                 struct device_attribute *attr, const char *buf, size_t len)
 631 {
 632         struct zram *zram = dev_to_zram(dev);
 633         unsigned long nr_pages = zram->disksize >> PAGE_SHIFT;
 634         unsigned long index;
 635         struct bio bio;
 636         struct bio_vec bio_vec;
 637         struct page *page;
 638         ssize_t ret, sz;
 639         char mode_buf[8];
 640         int mode = -1;
 641         unsigned long blk_idx = 0;
 642
 643         sz = strscpy(mode_buf, buf, sizeof(mode_buf));
 644         if (sz <= 0)
 645                 return -EINVAL;
 646
 647         /* ignore trailing newline */
 648         if (mode_buf[sz - 1] == '\n')
 649                 mode_buf[sz - 1] = 0x00;
 650
 651         if (!strcmp(mode_buf, "idle"))
 652                 mode = IDLE_WRITEBACK;
 653         else if (!strcmp(mode_buf, "huge"))
 654                 mode = HUGE_WRITEBACK;
 655
 656         if (mode == -1)
 657                 return -EINVAL;
 658
 659         down_read(&zram->init_lock);
 660         if (!init_done(zram)) {
 661                 ret = -EINVAL;
 662                 goto release_init_lock;
 663         }
 664
 665         if (!zram->backing_dev) {
 666                 ret = -ENODEV;
 667                 goto release_init_lock;
 668         }
 669
 670         page = alloc_page(GFP_KERNEL);
 671         if (!page) {
 672                 ret = -ENOMEM;
 673                 goto release_init_lock;
 674         }
 675
 676         for (index = 0; index < nr_pages; index++) {
 677                 struct bio_vec bvec;
 678
 679                 bvec.bv_page = page;
 680                 bvec.bv_len = PAGE_SIZE;
 681                 bvec.bv_offset = 0;
 682
 683                 spin_lock(&zram->wb_limit_lock);
 684                 if (zram->wb_limit_enable && !zram->bd_wb_limit) {
 685                         spin_unlock(&zram->wb_limit_lock);
 686                         ret = -EIO;
 687                         break;
 688                 }
 689                 spin_unlock(&zram->wb_limit_lock);
 690
 691                 if (!blk_idx) {
 692                         blk_idx = alloc_block_bdev(zram);
 693                         if (!blk_idx) {
 694                                 ret = -ENOSPC;
 695                                 break;
 696                         }
 697                 }
 698
 699                 zram_slot_lock(zram, index);
 700                 if (!zram_allocated(zram, index))
 701                         goto next;
 702
 703                 if (zram_test_flag(zram, index, ZRAM_WB) ||
 704                                 zram_test_flag(zram, index, ZRAM_SAME) ||
 705                                 zram_test_flag(zram, index, ZRAM_UNDER_WB))
 706                         goto next;
 707
 708                 if (mode == IDLE_WRITEBACK &&
 709                           !zram_test_flag(zram, index, ZRAM_IDLE))
 710                         goto next;
 711                 if (mode == HUGE_WRITEBACK &&
 712                           !zram_test_flag(zram, index, ZRAM_HUGE))
 713                         goto next;
 714                 /*
 715                  * Clearing ZRAM_UNDER_WB is duty of caller.
 716                  * IOW, zram_free_page never clear it.
 717                  */
 718                 zram_set_flag(zram, index, ZRAM_UNDER_WB);
 719                 /* Need for hugepage writeback racing */
 720                 zram_set_flag(zram, index, ZRAM_IDLE);
 721                 zram_slot_unlock(zram, index);
 722                 if (zram_bvec_read(zram, &bvec, index, 0, NULL)) {
 723                         zram_slot_lock(zram, index);
 724                         zram_clear_flag(zram, index, ZRAM_UNDER_WB);
 725                         zram_clear_flag(zram, index, ZRAM_IDLE);
 726                         zram_slot_unlock(zram, index);
 727                         continue;
 728                 }
 729
 730                 bio_init(&bio, &bio_vec, 1);
 731                 bio_set_dev(&bio, zram->bdev);
 732                 bio.bi_iter.bi_sector = blk_idx * (PAGE_SIZE >> 9);
 733                 bio.bi_opf = REQ_OP_WRITE | REQ_SYNC;
 734
 735                 bio_add_page(&bio, bvec.bv_page, bvec.bv_len,
 736                                 bvec.bv_offset);
 737                 /*
 738                  * XXX: A single page IO would be inefficient for write
 739                  * but it would be not bad as starter.
 740                  */
 741                 ret = submit_bio_wait(&bio);
 742                 if (ret) {
 743                         zram_slot_lock(zram, index);
 744                         zram_clear_flag(zram, index, ZRAM_UNDER_WB);
 745                         zram_clear_flag(zram, index, ZRAM_IDLE);
 746                         zram_slot_unlock(zram, index);
 747                         continue;
 748                 }
 749
 750                 atomic64_inc(&zram->stats.bd_writes);
 751                 /*
 752                  * We released zram_slot_lock so need to check if the slot was
 753                  * changed. If there is freeing for the slot, we can catch it
 754                  * easily by zram_allocated.
 755                  * A subtle case is the slot is freed/reallocated/marked as
 756                  * ZRAM_IDLE again. To close the race, idle_store doesn't
 757                  * mark ZRAM_IDLE once it found the slot was ZRAM_UNDER_WB.
 758                  * Thus, we could close the race by checking ZRAM_IDLE bit.
 759                  */
 760                 zram_slot_lock(zram, index);
 761                 if (!zram_allocated(zram, index) ||
 762                           !zram_test_flag(zram, index, ZRAM_IDLE)) {
 763                         zram_clear_flag(zram, index, ZRAM_UNDER_WB);
 764                         zram_clear_flag(zram, index, ZRAM_IDLE);
 765                         goto next;
 766                 }
 767
 768                 zram_free_page(zram, index);
 769                 zram_clear_flag(zram, index, ZRAM_UNDER_WB);
 770                 zram_set_flag(zram, index, ZRAM_WB);
 771                 zram_set_element(zram, index, blk_idx);
 772                 blk_idx = 0;
 773                 atomic64_inc(&zram->stats.pages_stored);
 774                 spin_lock(&zram->wb_limit_lock);
 775                 if (zram->wb_limit_enable && zram->bd_wb_limit > 0)
 776                         zram->bd_wb_limit -=  1UL << (PAGE_SHIFT - 12);
 777                 spin_unlock(&zram->wb_limit_lock);
 778 next:
 779                 zram_slot_unlock(zram, index);
 780         }
 781
 782         if (blk_idx)
 783                 free_block_bdev(zram, blk_idx);
 784         ret = len;
 785         __free_page(page);
 786 release_init_lock:
 787         up_read(&zram->init_lock);
 788
 789         return ret;
 790 }
 791
 792 struct zram_work {
 793         struct work_struct work;
 794         struct zram *zram;
 795         unsigned long entry;
 796         struct bio *bio;
 797 };
 798
 799 #if PAGE_SIZE != 4096
 800 static void zram_sync_read(struct work_struct *work)
 801 {
 802         struct bio_vec bvec;
 803         struct zram_work *zw = container_of(work, struct zram_work, work);
 804         struct zram *zram = zw->zram;
 805         unsigned long entry = zw->entry;
 806         struct bio *bio = zw->bio;
 807
 808         read_from_bdev_async(zram, &bvec, entry, bio);
 809 }
 810
 811 /*
 812  * Block layer want one ->make_request_fn to be active at a time
 813  * so if we use chained IO with parent IO in same context,
 814  * it's a deadlock. To avoid, it, it uses worker thread context.
 815  */
 816 static int read_from_bdev_sync(struct zram *zram, struct bio_vec *bvec,
 817                                 unsigned long entry, struct bio *bio)
 818 {
 819         struct zram_work work;
 820
 821         work.zram = zram;
 822         work.entry = entry;
 823         work.bio = bio;
 824
 825         INIT_WORK_ONSTACK(&work.work, zram_sync_read);
 826         queue_work(system_unbound_wq, &work.work);
 827         flush_work(&work.work);
 828         destroy_work_on_stack(&work.work);
 829
 830         return 1;
 831 }
 832 #else
 833 static int read_from_bdev_sync(struct zram *zram, struct bio_vec *bvec,
 834                                 unsigned long entry, struct bio *bio)
 835 {
 836         WARN_ON(1);
 837         return -EIO;
 838 }
 839 #endif
 840
 841 static int read_from_bdev(struct zram *zram, struct bio_vec *bvec,
 842                         unsigned long entry, struct bio *parent, bool sync)
 843 {
 844         atomic64_inc(&zram->stats.bd_reads);
 845         if (sync)
 846                 return read_from_bdev_sync(zram, bvec, entry, parent);
 847         else
 848                 return read_from_bdev_async(zram, bvec, entry, parent);
 849 }
 850 #else
 851 static inline void reset_bdev(struct zram *zram) {};
 852 static int read_from_bdev(struct zram *zram, struct bio_vec *bvec,
 853                         unsigned long entry, struct bio *parent, bool sync)
 854 {
 855         return -EIO;
 856 }
 857
 858 static void free_block_bdev(struct zram *zram, unsigned long blk_idx) {};
 859 #endif
 860
 861 #ifdef CONFIG_ZRAM_MEMORY_TRACKING
 862
 863 static struct dentry *zram_debugfs_root;
 864
 865 static void zram_debugfs_create(void)
 866 {
 867         zram_debugfs_root = debugfs_create_dir("zram", NULL);
 868 }
 869
 870 static void zram_debugfs_destroy(void)
 871 {
 872         debugfs_remove_recursive(zram_debugfs_root);
 873 }
 874
 875 static void zram_accessed(struct zram *zram, u32 index)
 876 {
 877         zram_clear_flag(zram, index, ZRAM_IDLE);
 878         zram->table[index].ac_time = ktime_get_boottime();
 879 }
 880
 881 static ssize_t read_block_state(struct file *file, char __user *buf,
 882                                 size_t count, loff_t *ppos)
 883 {
 884         char *kbuf;
 885         ssize_t index, written = 0;
 886         struct zram *zram = file->private_data;
 887         unsigned long nr_pages = zram->disksize >> PAGE_SHIFT;
 888         struct timespec64 ts;
 889
 890         kbuf = kvmalloc(count, GFP_KERNEL);
 891         if (!kbuf)
 892                 return -ENOMEM;
 893
 894         down_read(&zram->init_lock);
 895         if (!init_done(zram)) {
 896                 up_read(&zram->init_lock);
 897                 kvfree(kbuf);
 898                 return -EINVAL;
 899         }
 900
 901         for (index = *ppos; index < nr_pages; index++) {
 902                 int copied;
 903
 904                 zram_slot_lock(zram, index);
 905                 if (!zram_allocated(zram, index))
 906                         goto next;
 907
 908                 ts = ktime_to_timespec64(zram->table[index].ac_time);
 909                 copied = snprintf(kbuf + written, count,
 910                         "%12zd %12lld.%06lu %c%c%c%c\n",
 911                         index, (s64)ts.tv_sec,
 912                         ts.tv_nsec / NSEC_PER_USEC,
 913                         zram_test_flag(zram, index, ZRAM_SAME) ? 's' : '.',
 914                         zram_test_flag(zram, index, ZRAM_WB) ? 'w' : '.',
 915                         zram_test_flag(zram, index, ZRAM_HUGE) ? 'h' : '.',
 916                         zram_test_flag(zram, index, ZRAM_IDLE) ? 'i' : '.');
 917
 918                 if (count < copied) {
 919                         zram_slot_unlock(zram, index);
 920                         break;
 921                 }
 922                 written += copied;
 923                 count -= copied;
 924 next:
 925                 zram_slot_unlock(zram, index);
 926                 *ppos += 1;
 927         }
 928
 929         up_read(&zram->init_lock);
 930         if (copy_to_user(buf, kbuf, written))
 931                 written = -EFAULT;
 932         kvfree(kbuf);
 933
 934         return written;
 935 }
 936
 937 static const struct file_operations proc_zram_block_state_op = {
 938         .open = simple_open,
 939         .read = read_block_state,
 940         .llseek = default_llseek,
 941 };
 942
 943 static void zram_debugfs_register(struct zram *zram)
 944 {
 945         if (!zram_debugfs_root)
 946                 return;
 947
 948         zram->debugfs_dir = debugfs_create_dir(zram->disk->disk_name,
 949                                                 zram_debugfs_root);
 950         debugfs_create_file("block_state", 0400, zram->debugfs_dir,
 951                                 zram, &proc_zram_block_state_op);
 952 }
 953
 954 static void zram_debugfs_unregister(struct zram *zram)
 955 {
 956         debugfs_remove_recursive(zram->debugfs_dir);
 957 }
 958 #else
 959 static void zram_debugfs_create(void) {};
 960 static void zram_debugfs_destroy(void) {};
 961 static void zram_accessed(struct zram *zram, u32 index)
 962 {
 963         zram_clear_flag(zram, index, ZRAM_IDLE);
 964 };
 965 static void zram_debugfs_register(struct zram *zram) {};
 966 static void zram_debugfs_unregister(struct zram *zram) {};
 967 #endif
 968
 969 /*
 970  * We switched to per-cpu streams and this attr is not needed anymore.
 971  * However, we will keep it around for some time, because:
 972  * a) we may revert per-cpu streams in the future
 973  * b) it's visible to user space and we need to follow our 2 years
 974  *    retirement rule; but we already have a number of 'soon to be
 975  *    altered' attrs, so max_comp_streams need to wait for the next
 976  *    layoff cycle.
 977  */
 978 static ssize_t max_comp_streams_show(struct device *dev,
 979                 struct device_attribute *attr, char *buf)
 980 {
 981         return scnprintf(buf, PAGE_SIZE, "%d\n", num_online_cpus());
 982 }
 983
 984 static ssize_t max_comp_streams_store(struct device *dev,
 985                 struct device_attribute *attr, const char *buf, size_t len)
 986 {
 987         return len;
 988 }
 989
 990 static ssize_t comp_algorithm_show(struct device *dev,
 991                 struct device_attribute *attr, char *buf)
 992 {
 993         size_t sz;
 994         struct zram *zram = dev_to_zram(dev);
 995
 996         down_read(&zram->init_lock);
 997         sz = zcomp_available_show(zram->compressor, buf);
 998         up_read(&zram->init_lock);
 999
1000         return sz;
1001 }
1002
1003 static ssize_t comp_algorithm_store(struct device *dev,
1004                 struct device_attribute *attr, const char *buf, size_t len)
1005 {
1006         struct zram *zram = dev_to_zram(dev);
1007         char compressor[ARRAY_SIZE(zram->compressor)];
1008         size_t sz;
1009
1010         strlcpy(compressor, buf, sizeof(compressor));
1011         /* ignore trailing newline */
1012         sz = strlen(compressor);
1013         if (sz > 0 && compressor[sz - 1] == '\n')
1014                 compressor[sz - 1] = 0x00;
1015
1016         if (!zcomp_available_algorithm(compressor))
1017                 return -EINVAL;
1018
1019         down_write(&zram->init_lock);
1020         if (init_done(zram)) {
1021                 up_write(&zram->init_lock);
1022                 pr_info("Can't change algorithm for initialized device\n");
1023                 return -EBUSY;
1024         }
1025
1026         strcpy(zram->compressor, compressor);
1027         up_write(&zram->init_lock);
1028         return len;
1029 }
1030
1031 static ssize_t compact_store(struct device *dev,
1032                 struct device_attribute *attr, const char *buf, size_t len)
1033 {
1034         struct zram *zram = dev_to_zram(dev);
1035
1036         down_read(&zram->init_lock);
1037         if (!init_done(zram)) {
1038                 up_read(&zram->init_lock);
1039                 return -EINVAL;
1040         }
1041
1042         zs_compact(zram->mem_pool);
1043         up_read(&zram->init_lock);
1044
1045         return len;
1046 }
1047
1048 static ssize_t io_stat_show(struct device *dev,
1049                 struct device_attribute *attr, char *buf)
1050 {
1051         struct zram *zram = dev_to_zram(dev);
1052         ssize_t ret;
1053
1054         down_read(&zram->init_lock);
1055         ret = scnprintf(buf, PAGE_SIZE,
1056                         "%8llu %8llu %8llu %8llu\n",
1057                         (u64)atomic64_read(&zram->stats.failed_reads),
1058                         (u64)atomic64_read(&zram->stats.failed_writes),
1059                         (u64)atomic64_read(&zram->stats.invalid_io),
1060                         (u64)atomic64_read(&zram->stats.notify_free));
1061         up_read(&zram->init_lock);
1062
1063         return ret;
1064 }
1065
1066 static ssize_t mm_stat_show(struct device *dev,
1067                 struct device_attribute *attr, char *buf)
1068 {
1069         struct zram *zram = dev_to_zram(dev);
1070         struct zs_pool_stats pool_stats;
1071         u64 orig_size, mem_used = 0;
1072         long max_used;
1073         ssize_t ret;
1074
1075         memset(&pool_stats, 0x00, sizeof(struct zs_pool_stats));
1076
1077         down_read(&zram->init_lock);
1078         if (init_done(zram)) {
1079                 mem_used = zs_get_total_pages(zram->mem_pool);
1080                 zs_pool_stats(zram->mem_pool, &pool_stats);
1081         }
1082
1083         orig_size = atomic64_read(&zram->stats.pages_stored);
1084         max_used = atomic_long_read(&zram->stats.max_used_pages);
1085
1086         ret = scnprintf(buf, PAGE_SIZE,
1087                         "%8llu %8llu %8llu %8lu %8ld %8llu %8lu %8llu\n",
1088                         orig_size << PAGE_SHIFT,
1089                         (u64)atomic64_read(&zram->stats.compr_data_size),
1090                         mem_used << PAGE_SHIFT,
1091                         zram->limit_pages << PAGE_SHIFT,
1092                         max_used << PAGE_SHIFT,
1093                         (u64)atomic64_read(&zram->stats.same_pages),
1094                         pool_stats.pages_compacted,
1095                         (u64)atomic64_read(&zram->stats.huge_pages));
1096         up_read(&zram->init_lock);
1097
1098         return ret;
1099 }
1100
1101 #ifdef CONFIG_ZRAM_WRITEBACK
1102 #define FOUR_K(x) ((x) * (1 << (PAGE_SHIFT - 12)))
1103 static ssize_t bd_stat_show(struct device *dev,
1104                 struct device_attribute *attr, char *buf)
1105 {
1106         struct zram *zram = dev_to_zram(dev);
1107         ssize_t ret;
1108
1109         down_read(&zram->init_lock);
1110         ret = scnprintf(buf, PAGE_SIZE,
1111                 "%8llu %8llu %8llu\n",
1112                         FOUR_K((u64)atomic64_read(&zram->stats.bd_count)),
1113                         FOUR_K((u64)atomic64_read(&zram->stats.bd_reads)),
1114                         FOUR_K((u64)atomic64_read(&zram->stats.bd_writes)));
1115         up_read(&zram->init_lock);
1116
1117         return ret;
1118 }
1119 #endif
1120
1121 static ssize_t debug_stat_show(struct device *dev,
1122                 struct device_attribute *attr, char *buf)
1123 {
1124         int version = 1;
1125         struct zram *zram = dev_to_zram(dev);
1126         ssize_t ret;
1127
1128         down_read(&zram->init_lock);
1129         ret = scnprintf(buf, PAGE_SIZE,
1130                         "version: %d\n%8llu %8llu\n",
1131                         version,
1132                         (u64)atomic64_read(&zram->stats.writestall),
1133                         (u64)atomic64_read(&zram->stats.miss_free));
1134         up_read(&zram->init_lock);
1135
1136         return ret;
1137 }
1138
1139 static DEVICE_ATTR_RO(io_stat);
1140 static DEVICE_ATTR_RO(mm_stat);
1141 #ifdef CONFIG_ZRAM_WRITEBACK
1142 static DEVICE_ATTR_RO(bd_stat);
1143 #endif
1144 static DEVICE_ATTR_RO(debug_stat);
1145
1146 static void zram_meta_free(struct zram *zram, u64 disksize)
1147 {
1148         size_t num_pages = disksize >> PAGE_SHIFT;
1149         size_t index;
1150
1151         /* Free all pages that are still in this zram device */
1152         for (index = 0; index < num_pages; index++)
1153                 zram_free_page(zram, index);
1154
1155         zs_destroy_pool(zram->mem_pool);
1156         vfree(zram->table);
1157 }
1158
1159 static bool zram_meta_alloc(struct zram *zram, u64 disksize)
1160 {
1161         size_t num_pages;
1162
1163         num_pages = disksize >> PAGE_SHIFT;
1164         zram->table = vzalloc(array_size(num_pages, sizeof(*zram->table)));
1165         if (!zram->table)
1166                 return false;
1167
1168         zram->mem_pool = zs_create_pool(zram->disk->disk_name);
1169         if (!zram->mem_pool) {
1170                 vfree(zram->table);
1171                 return false;
1172         }
1173
1174         if (!huge_class_size)
1175                 huge_class_size = zs_huge_class_size(zram->mem_pool);
1176         return true;
1177 }
1178
1179 /*
1180  * To protect concurrent access to the same index entry,
1181  * caller should hold this table index entry's bit_spinlock to
1182  * indicate this index entry is accessing.
1183  */
1184 static void zram_free_page(struct zram *zram, size_t index)
1185 {
1186         unsigned long handle;
1187
1188 #ifdef CONFIG_ZRAM_MEMORY_TRACKING
1189         zram->table[index].ac_time = 0;
1190 #endif
1191         if (zram_test_flag(zram, index, ZRAM_IDLE))
1192                 zram_clear_flag(zram, index, ZRAM_IDLE);
1193
1194         if (zram_test_flag(zram, index, ZRAM_HUGE)) {
1195                 zram_clear_flag(zram, index, ZRAM_HUGE);
1196                 atomic64_dec(&zram->stats.huge_pages);
1197         }
1198
1199         if (zram_test_flag(zram, index, ZRAM_WB)) {
1200                 zram_clear_flag(zram, index, ZRAM_WB);
1201                 free_block_bdev(zram, zram_get_element(zram, index));
1202                 goto out;
1203         }
1204
1205         /*
1206          * No memory is allocated for same element filled pages.
1207          * Simply clear same page flag.
1208          */
1209         if (zram_test_flag(zram, index, ZRAM_SAME)) {
1210                 zram_clear_flag(zram, index, ZRAM_SAME);
1211                 atomic64_dec(&zram->stats.same_pages);
1212                 goto out;
1213         }
1214
1215         handle = zram_get_handle(zram, index);
1216         if (!handle)
1217                 return;
1218
1219         zs_free(zram->mem_pool, handle);
1220
1221         atomic64_sub(zram_get_obj_size(zram, index),
1222                         &zram->stats.compr_data_size);
1223 out:
1224         atomic64_dec(&zram->stats.pages_stored);
1225         zram_set_handle(zram, index, 0);
1226         zram_set_obj_size(zram, index, 0);
1227         WARN_ON_ONCE(zram->table[index].flags &
1228                 ~(1UL << ZRAM_LOCK | 1UL << ZRAM_UNDER_WB));
1229 }
1230
1231 static int __zram_bvec_read(struct zram *zram, struct page *page, u32 index,
1232                                 struct bio *bio, bool partial_io)
1233 {
1234         int ret;
1235         unsigned long handle;
1236         unsigned int size;
1237         void *src, *dst;
1238
1239         zram_slot_lock(zram, index);
1240         if (zram_test_flag(zram, index, ZRAM_WB)) {
1241                 struct bio_vec bvec;
1242
1243                 zram_slot_unlock(zram, index);
1244
1245                 bvec.bv_page = page;
1246                 bvec.bv_len = PAGE_SIZE;
1247                 bvec.bv_offset = 0;
1248                 return read_from_bdev(zram, &bvec,
1249                                 zram_get_element(zram, index),
1250                                 bio, partial_io);
1251         }
1252
1253         handle = zram_get_handle(zram, index);
1254         if (!handle || zram_test_flag(zram, index, ZRAM_SAME)) {
1255                 unsigned long value;
1256                 void *mem;
1257
1258                 value = handle ? zram_get_element(zram, index) : 0;
1259                 mem = kmap_atomic(page);
1260                 zram_fill_page(mem, PAGE_SIZE, value);
1261                 kunmap_atomic(mem);
1262                 zram_slot_unlock(zram, index);
1263                 return 0;
1264         }
1265
1266         size = zram_get_obj_size(zram, index);
1267
1268         src = zs_map_object(zram->mem_pool, handle, ZS_MM_RO);
1269         if (size == PAGE_SIZE) {
1270                 dst = kmap_atomic(page);
1271                 memcpy(dst, src, PAGE_SIZE);
1272                 kunmap_atomic(dst);
1273                 ret = 0;
1274         } else {
1275                 struct zcomp_strm *zstrm = zcomp_stream_get(zram->comp);
1276
1277                 dst = kmap_atomic(page);
1278                 ret = zcomp_decompress(zstrm, src, size, dst);
1279                 kunmap_atomic(dst);
1280                 zcomp_stream_put(zram->comp);
1281         }
1282         zs_unmap_object(zram->mem_pool, handle);
1283         zram_slot_unlock(zram, index);
1284
1285         /* Should NEVER happen. Return bio error if it does. */
1286         if (unlikely(ret))
1287                 pr_err("Decompression failed! err=%d, page=%u\n", ret, index);
1288
1289         return ret;
1290 }
1291
1292 static int zram_bvec_read(struct zram *zram, struct bio_vec *bvec,
1293                                 u32 index, int offset, struct bio *bio)
1294 {
1295         int ret;
1296         struct page *page;
1297
1298         page = bvec->bv_page;
1299         if (is_partial_io(bvec)) {
1300                 /* Use a temporary buffer to decompress the page */
1301                 page = alloc_page(GFP_NOIO|__GFP_HIGHMEM);
1302                 if (!page)
1303                         return -ENOMEM;
1304         }
1305
1306         ret = __zram_bvec_read(zram, page, index, bio, is_partial_io(bvec));
1307         if (unlikely(ret))
1308                 goto out;
1309
1310         if (is_partial_io(bvec)) {
1311                 void *dst = kmap_atomic(bvec->bv_page);
1312                 void *src = kmap_atomic(page);
1313
1314                 memcpy(dst + bvec->bv_offset, src + offset, bvec->bv_len);
1315                 kunmap_atomic(src);
1316                 kunmap_atomic(dst);
1317         }
1318 out:
1319         if (is_partial_io(bvec))
1320                 __free_page(page);
1321
1322         return ret;
1323 }
1324
1325 static int __zram_bvec_write(struct zram *zram, struct bio_vec *bvec,
1326                                 u32 index, struct bio *bio)
1327 {
1328         int ret = 0;
1329         unsigned long alloced_pages;
1330         unsigned long handle = 0;
1331         unsigned int comp_len = 0;
1332         void *src, *dst, *mem;
1333         struct zcomp_strm *zstrm;
1334         struct page *page = bvec->bv_page;
1335         unsigned long element = 0;
1336         enum zram_pageflags flags = 0;
1337
1338         mem = kmap_atomic(page);
1339         if (page_same_filled(mem, &element)) {
1340                 kunmap_atomic(mem);
1341                 /* Free memory associated with this sector now. */
1342                 flags = ZRAM_SAME;
1343                 atomic64_inc(&zram->stats.same_pages);
1344                 goto out;
1345         }
1346         kunmap_atomic(mem);
1347
1348 compress_again:
1349         zstrm = zcomp_stream_get(zram->comp);
1350         src = kmap_atomic(page);
1351         ret = zcomp_compress(zstrm, src, &comp_len);
1352         kunmap_atomic(src);
1353
1354         if (unlikely(ret)) {
1355                 zcomp_stream_put(zram->comp);
1356                 pr_err("Compression failed! err=%d\n", ret);
1357                 zs_free(zram->mem_pool, handle);
1358                 return ret;
1359         }
1360
1361         if (comp_len >= huge_class_size)
1362                 comp_len = PAGE_SIZE;
1363         /*
1364          * handle allocation has 2 paths:
1365          * a) fast path is executed with preemption disabled (for
1366          *  per-cpu streams) and has __GFP_DIRECT_RECLAIM bit clear,
1367          *  since we can't sleep;
1368          * b) slow path enables preemption and attempts to allocate
1369          *  the page with __GFP_DIRECT_RECLAIM bit set. we have to
1370          *  put per-cpu compression stream and, thus, to re-do
1371          *  the compression once handle is allocated.
1372          *
1373          * if we have a 'non-null' handle here then we are coming
1374          * from the slow path and handle has already been allocated.
1375          */
1376         if (!handle)
1377                 handle = zs_malloc(zram->mem_pool, comp_len,
1378                                 __GFP_KSWAPD_RECLAIM |
1379                                 __GFP_NOWARN |
1380                                 __GFP_HIGHMEM |
1381                                 __GFP_MOVABLE);
1382         if (!handle) {
1383                 zcomp_stream_put(zram->comp);
1384                 atomic64_inc(&zram->stats.writestall);
1385                 handle = zs_malloc(zram->mem_pool, comp_len,
1386                                 GFP_NOIO | __GFP_HIGHMEM |
1387                                 __GFP_MOVABLE);
1388                 if (handle)
1389                         goto compress_again;
1390                 return -ENOMEM;
1391         }
1392
1393         alloced_pages = zs_get_total_pages(zram->mem_pool);
1394         update_used_max(zram, alloced_pages);
1395
1396         if (zram->limit_pages && alloced_pages > zram->limit_pages) {
1397                 zcomp_stream_put(zram->comp);
1398                 zs_free(zram->mem_pool, handle);
1399                 return -ENOMEM;
1400         }
1401
1402         dst = zs_map_object(zram->mem_pool, handle, ZS_MM_WO);
1403
1404         src = zstrm->buffer;
1405         if (comp_len == PAGE_SIZE)
1406                 src = kmap_atomic(page);
1407         memcpy(dst, src, comp_len);
1408         if (comp_len == PAGE_SIZE)
1409                 kunmap_atomic(src);
1410
1411         zcomp_stream_put(zram->comp);
1412         zs_unmap_object(zram->mem_pool, handle);
1413         atomic64_add(comp_len, &zram->stats.compr_data_size);
1414 out:
1415         /*
1416          * Free memory associated with this sector
1417          * before overwriting unused sectors.
1418          */
1419         zram_slot_lock(zram, index);
1420         zram_free_page(zram, index);
1421
1422         if (comp_len == PAGE_SIZE) {
1423                 zram_set_flag(zram, index, ZRAM_HUGE);
1424                 atomic64_inc(&zram->stats.huge_pages);
1425         }
1426
1427         if (flags) {
1428                 zram_set_flag(zram, index, flags);
1429                 zram_set_element(zram, index, element);
1430         }  else {
1431                 zram_set_handle(zram, index, handle);
1432                 zram_set_obj_size(zram, index, comp_len);
1433         }
1434         zram_slot_unlock(zram, index);
1435
1436         /* Update stats */
1437         atomic64_inc(&zram->stats.pages_stored);
1438         return ret;
1439 }
1440
1441 static int zram_bvec_write(struct zram *zram, struct bio_vec *bvec,
1442                                 u32 index, int offset, struct bio *bio)
1443 {
1444         int ret;
1445         struct page *page = NULL;
1446         void *src;
1447         struct bio_vec vec;
1448
1449         vec = *bvec;
1450         if (is_partial_io(bvec)) {
1451                 void *dst;
1452                 /*
1453                  * This is a partial IO. We need to read the full page
1454                  * before to write the changes.
1455                  */
1456                 page = alloc_page(GFP_NOIO|__GFP_HIGHMEM);
1457                 if (!page)
1458                         return -ENOMEM;
1459
1460                 ret = __zram_bvec_read(zram, page, index, bio, true);
1461                 if (ret)
1462                         goto out;
1463
1464                 src = kmap_atomic(bvec->bv_page);
1465                 dst = kmap_atomic(page);
1466                 memcpy(dst + offset, src + bvec->bv_offset, bvec->bv_len);
1467                 kunmap_atomic(dst);
1468                 kunmap_atomic(src);
1469
1470                 vec.bv_page = page;
1471                 vec.bv_len = PAGE_SIZE;
1472                 vec.bv_offset = 0;
1473         }
1474
1475         ret = __zram_bvec_write(zram, &vec, index, bio);
1476 out:
1477         if (is_partial_io(bvec))
1478                 __free_page(page);
1479         return ret;
1480 }
1481
1482 /*
1483  * zram_bio_discard - handler on discard request
1484  * @index: physical block index in PAGE_SIZE units
1485  * @offset: byte offset within physical block
1486  */
1487 static void zram_bio_discard(struct zram *zram, u32 index,
1488                              int offset, struct bio *bio)
1489 {
1490         size_t n = bio->bi_iter.bi_size;
1491
1492         /*
1493          * zram manages data in physical block size units. Because logical block
1494          * size isn't identical with physical block size on some arch, we
1495          * could get a discard request pointing to a specific offset within a
1496          * certain physical block.  Although we can handle this request by
1497          * reading that physiclal block and decompressing and partially zeroing
1498          * and re-compressing and then re-storing it, this isn't reasonable
1499          * because our intent with a discard request is to save memory.  So
1500          * skipping this logical block is appropriate here.
1501          */
1502         if (offset) {
1503                 if (n <= (PAGE_SIZE - offset))
1504                         return;
1505
1506                 n -= (PAGE_SIZE - offset);
1507                 index++;
1508         }
1509
1510         while (n >= PAGE_SIZE) {
1511                 zram_slot_lock(zram, index);
1512                 zram_free_page(zram, index);
1513                 zram_slot_unlock(zram, index);
1514                 atomic64_inc(&zram->stats.notify_free);
1515                 index++;
1516                 n -= PAGE_SIZE;
1517         }
1518 }
1519
1520 /*
1521  * Returns errno if it has some problem. Otherwise return 0 or 1.
1522  * Returns 0 if IO request was done synchronously
1523  * Returns 1 if IO request was successfully submitted.
1524  */
1525 static int zram_bvec_rw(struct zram *zram, struct bio_vec *bvec, u32 index,
1526                         int offset, unsigned int op, struct bio *bio)
1527 {
1528         unsigned long start_time = jiffies;
1529         struct request_queue *q = zram->disk->queue;
1530         int ret;
1531
1532         generic_start_io_acct(q, op, bvec->bv_len >> SECTOR_SHIFT,
1533                         &zram->disk->part0);
1534
1535         if (!op_is_write(op)) {
1536                 atomic64_inc(&zram->stats.num_reads);
1537                 ret = zram_bvec_read(zram, bvec, index, offset, bio);
1538                 flush_dcache_page(bvec->bv_page);
1539         } else {
1540                 atomic64_inc(&zram->stats.num_writes);
1541                 ret = zram_bvec_write(zram, bvec, index, offset, bio);
1542         }
1543
1544         generic_end_io_acct(q, op, &zram->disk->part0, start_time);
1545
1546         zram_slot_lock(zram, index);
1547         zram_accessed(zram, index);
1548         zram_slot_unlock(zram, index);
1549
1550         if (unlikely(ret < 0)) {
1551                 if (!op_is_write(op))
1552                         atomic64_inc(&zram->stats.failed_reads);
1553                 else
1554                         atomic64_inc(&zram->stats.failed_writes);
1555         }
1556
1557         return ret;
1558 }
1559
1560 static void __zram_make_request(struct zram *zram, struct bio *bio)
1561 {
1562         int offset;
1563         u32 index;
1564         struct bio_vec bvec;
1565         struct bvec_iter iter;
1566
1567         index = bio->bi_iter.bi_sector >> SECTORS_PER_PAGE_SHIFT;
1568         offset = (bio->bi_iter.bi_sector &
1569                   (SECTORS_PER_PAGE - 1)) << SECTOR_SHIFT;
1570
1571         switch (bio_op(bio)) {
1572         case REQ_OP_DISCARD:
1573         case REQ_OP_WRITE_ZEROES:
1574                 zram_bio_discard(zram, index, offset, bio);
1575                 bio_endio(bio);
1576                 return;
1577         default:
1578                 break;
1579         }
1580
1581         bio_for_each_segment(bvec, bio, iter) {
1582                 struct bio_vec bv = bvec;
1583                 unsigned int unwritten = bvec.bv_len;
1584
1585                 do {
1586                         bv.bv_len = min_t(unsigned int, PAGE_SIZE - offset,
1587                                                         unwritten);
1588                         if (zram_bvec_rw(zram, &bv, index, offset,
1589                                          bio_op(bio), bio) < 0)
1590                                 goto out;
1591
1592                         bv.bv_offset += bv.bv_len;
1593                         unwritten -= bv.bv_len;
1594
1595                         update_position(&index, &offset, &bv);
1596                 } while (unwritten);
1597         }
1598
1599         bio_endio(bio);
1600         return;
1601
1602 out:
1603         bio_io_error(bio);
1604 }
1605
1606 /*
1607  * Handler function for all zram I/O requests.
1608  */
1609 static blk_qc_t zram_make_request(struct request_queue *queue, struct bio *bio)
1610 {
1611         struct zram *zram = queue->queuedata;
1612
1613         if (!valid_io_request(zram, bio->bi_iter.bi_sector,
1614                                         bio->bi_iter.bi_size)) {
1615                 atomic64_inc(&zram->stats.invalid_io);
1616                 goto error;
1617         }
1618
1619         __zram_make_request(zram, bio);
1620         return BLK_QC_T_NONE;
1621
1622 error:
1623         bio_io_error(bio);
1624         return BLK_QC_T_NONE;
1625 }
1626
1627 static void zram_slot_free_notify(struct block_device *bdev,
1628                                 unsigned long index)
1629 {
1630         struct zram *zram;
1631
1632         zram = bdev->bd_disk->private_data;
1633
1634         atomic64_inc(&zram->stats.notify_free);
1635         if (!zram_slot_trylock(zram, index)) {
1636                 atomic64_inc(&zram->stats.miss_free);
1637                 return;
1638         }
1639
1640         zram_free_page(zram, index);
1641         zram_slot_unlock(zram, index);
1642 }
1643
1644 static int zram_rw_page(struct block_device *bdev, sector_t sector,
1645                        struct page *page, unsigned int op)
1646 {
1647         int offset, ret;
1648         u32 index;
1649         struct zram *zram;
1650         struct bio_vec bv;
1651
1652         if (PageTransHuge(page))
1653                 return -ENOTSUPP;
1654         zram = bdev->bd_disk->private_data;
1655
1656         if (!valid_io_request(zram, sector, PAGE_SIZE)) {
1657                 atomic64_inc(&zram->stats.invalid_io);
1658                 ret = -EINVAL;
1659                 goto out;
1660         }
1661
1662         index = sector >> SECTORS_PER_PAGE_SHIFT;
1663         offset = (sector & (SECTORS_PER_PAGE - 1)) << SECTOR_SHIFT;
1664
1665         bv.bv_page = page;
1666         bv.bv_len = PAGE_SIZE;
1667         bv.bv_offset = 0;
1668
1669         ret = zram_bvec_rw(zram, &bv, index, offset, op, NULL);
1670 out:
1671         /*
1672          * If I/O fails, just return error(ie, non-zero) without
1673          * calling page_endio.
1674          * It causes resubmit the I/O with bio request by upper functions
1675          * of rw_page(e.g., swap_readpage, __swap_writepage) and
1676          * bio->bi_end_io does things to handle the error
1677          * (e.g., SetPageError, set_page_dirty and extra works).
1678          */
1679         if (unlikely(ret < 0))
1680                 return ret;
1681
1682         switch (ret) {
1683         case 0:
1684                 page_endio(page, op_is_write(op), 0);
1685                 break;
1686         case 1:
1687                 ret = 0;
1688                 break;
1689         default:
1690                 WARN_ON(1);
1691         }
1692         return ret;
1693 }
1694
1695 static void zram_reset_device(struct zram *zram)
1696 {
1697         struct zcomp *comp;
1698         u64 disksize;
1699
1700         down_write(&zram->init_lock);
1701
1702         zram->limit_pages = 0;
1703
1704         if (!init_done(zram)) {
1705                 up_write(&zram->init_lock);
1706                 return;
1707         }
1708
1709         comp = zram->comp;
1710         disksize = zram->disksize;
1711         zram->disksize = 0;
1712
1713         set_capacity(zram->disk, 0);
1714         part_stat_set_all(&zram->disk->part0, 0);
1715
1716         up_write(&zram->init_lock);
1717         /* I/O operation under all of CPU are done so let's free */
1718         zram_meta_free(zram, disksize);
1719         memset(&zram->stats, 0, sizeof(zram->stats));
1720         zcomp_destroy(comp);
1721         reset_bdev(zram);
1722 }
1723
1724 static ssize_t disksize_store(struct device *dev,
1725                 struct device_attribute *attr, const char *buf, size_t len)
1726 {
1727         u64 disksize;
1728         struct zcomp *comp;
1729         struct zram *zram = dev_to_zram(dev);
1730         int err;
1731
1732         disksize = memparse(buf, NULL);
1733         if (!disksize)
1734                 return -EINVAL;
1735
1736         down_write(&zram->init_lock);
1737         if (init_done(zram)) {
1738                 pr_info("Cannot change disksize for initialized device\n");
1739                 err = -EBUSY;
1740                 goto out_unlock;
1741         }
1742
1743         disksize = PAGE_ALIGN(disksize);
1744         if (!zram_meta_alloc(zram, disksize)) {
1745                 err = -ENOMEM;
1746                 goto out_unlock;
1747         }
1748
1749         comp = zcomp_create(zram->compressor);
1750         if (IS_ERR(comp)) {
1751                 pr_err("Cannot initialise %s compressing backend\n",
1752                                 zram->compressor);
1753                 err = PTR_ERR(comp);
1754                 goto out_free_meta;
1755         }
1756
1757         zram->comp = comp;
1758         zram->disksize = disksize;
1759         set_capacity(zram->disk, zram->disksize >> SECTOR_SHIFT);
1760
1761         revalidate_disk(zram->disk);
1762         up_write(&zram->init_lock);
1763
1764         return len;
1765
1766 out_free_meta:
1767         zram_meta_free(zram, disksize);
1768 out_unlock:
1769         up_write(&zram->init_lock);
1770         return err;
1771 }
1772
1773 static ssize_t reset_store(struct device *dev,
1774                 struct device_attribute *attr, const char *buf, size_t len)
1775 {
1776         int ret;
1777         unsigned short do_reset;
1778         struct zram *zram;
1779         struct block_device *bdev;
1780
1781         ret = kstrtou16(buf, 10, &do_reset);
1782         if (ret)
1783                 return ret;
1784
1785         if (!do_reset)
1786                 return -EINVAL;
1787
1788         zram = dev_to_zram(dev);
1789         bdev = bdget_disk(zram->disk, 0);
1790         if (!bdev)
1791                 return -ENOMEM;
1792
1793         mutex_lock(&bdev->bd_mutex);
1794         /* Do not reset an active device or claimed device */
1795         if (bdev->bd_openers || zram->claim) {
1796                 mutex_unlock(&bdev->bd_mutex);
1797                 bdput(bdev);
1798                 return -EBUSY;
1799         }
1800
1801         /* From now on, anyone can't open /dev/zram[0-9] */
1802         zram->claim = true;
1803         mutex_unlock(&bdev->bd_mutex);
1804
1805         /* Make sure all the pending I/O are finished */
1806         fsync_bdev(bdev);
1807         zram_reset_device(zram);
1808         revalidate_disk(zram->disk);
1809         bdput(bdev);
1810
1811         mutex_lock(&bdev->bd_mutex);
1812         zram->claim = false;
1813         mutex_unlock(&bdev->bd_mutex);
1814
1815         return len;
1816 }
1817
1818 static int zram_open(struct block_device *bdev, fmode_t mode)
1819 {
1820         int ret = 0;
1821         struct zram *zram;
1822
1823         WARN_ON(!mutex_is_locked(&bdev->bd_mutex));
1824
1825         zram = bdev->bd_disk->private_data;
1826         /* zram was claimed to reset so open request fails */
1827         if (zram->claim)
1828                 ret = -EBUSY;
1829
1830         return ret;
1831 }
1832
1833 static const struct block_device_operations zram_devops = {
1834         .open = zram_open,
1835         .swap_slot_free_notify = zram_slot_free_notify,
1836         .rw_page = zram_rw_page,
1837         .owner = THIS_MODULE
1838 };
1839
1840 static DEVICE_ATTR_WO(compact);
1841 static DEVICE_ATTR_RW(disksize);
1842 static DEVICE_ATTR_RO(initstate);
1843 static DEVICE_ATTR_WO(reset);
1844 static DEVICE_ATTR_WO(mem_limit);
1845 static DEVICE_ATTR_WO(mem_used_max);
1846 static DEVICE_ATTR_WO(idle);
1847 static DEVICE_ATTR_RW(max_comp_streams);
1848 static DEVICE_ATTR_RW(comp_algorithm);
1849 #ifdef CONFIG_ZRAM_WRITEBACK
1850 static DEVICE_ATTR_RW(backing_dev);
1851 static DEVICE_ATTR_WO(writeback);
1852 static DEVICE_ATTR_RW(writeback_limit);
1853 static DEVICE_ATTR_RW(writeback_limit_enable);
1854 #endif
1855
1856 static struct attribute *zram_disk_attrs[] = {
1857         &dev_attr_disksize.attr,
1858         &dev_attr_initstate.attr,
1859         &dev_attr_reset.attr,
1860         &dev_attr_compact.attr,
1861         &dev_attr_mem_limit.attr,
1862         &dev_attr_mem_used_max.attr,
1863         &dev_attr_idle.attr,
1864         &dev_attr_max_comp_streams.attr,
1865         &dev_attr_comp_algorithm.attr,
1866 #ifdef CONFIG_ZRAM_WRITEBACK
1867         &dev_attr_backing_dev.attr,
1868         &dev_attr_writeback.attr,
1869         &dev_attr_writeback_limit.attr,
1870         &dev_attr_writeback_limit_enable.attr,
1871 #endif
1872         &dev_attr_io_stat.attr,
1873         &dev_attr_mm_stat.attr,
1874 #ifdef CONFIG_ZRAM_WRITEBACK
1875         &dev_attr_bd_stat.attr,
1876 #endif
1877         &dev_attr_debug_stat.attr,
1878         NULL,
1879 };
1880
1881 static const struct attribute_group zram_disk_attr_group = {
1882         .attrs = zram_disk_attrs,
1883 };
1884
1885 static const struct attribute_group *zram_disk_attr_groups[] = {
1886         &zram_disk_attr_group,
1887         NULL,
1888 };
1889
1890 /*
1891  * Allocate and initialize new zram device. the function returns
1892  * '>= 0' device_id upon success, and negative value otherwise.
1893  */
1894 static int zram_add(void)
1895 {
1896         struct zram *zram;
1897         struct request_queue *queue;
1898         int ret, device_id;
1899
1900         zram = kzalloc(sizeof(struct zram), GFP_KERNEL);
1901         if (!zram)
1902                 return -ENOMEM;
1903
1904         ret = idr_alloc(&zram_index_idr, zram, 0, 0, GFP_KERNEL);
1905         if (ret < 0)
1906                 goto out_free_dev;
1907         device_id = ret;
1908
1909         init_rwsem(&zram->init_lock);
1910 #ifdef CONFIG_ZRAM_WRITEBACK
1911         spin_lock_init(&zram->wb_limit_lock);
1912 #endif
1913         queue = blk_alloc_queue(GFP_KERNEL);
1914         if (!queue) {
1915                 pr_err("Error allocating disk queue for device %d\n",
1916                         device_id);
1917                 ret = -ENOMEM;
1918                 goto out_free_idr;
1919         }
1920
1921         blk_queue_make_request(queue, zram_make_request);
1922
1923         /* gendisk structure */
1924         zram->disk = alloc_disk(1);
1925         if (!zram->disk) {
1926                 pr_err("Error allocating disk structure for device %d\n",
1927                         device_id);
1928                 ret = -ENOMEM;
1929                 goto out_free_queue;
1930         }
1931
1932         zram->disk->major = zram_major;
1933         zram->disk->first_minor = device_id;
1934         zram->disk->fops = &zram_devops;
1935         zram->disk->queue = queue;
1936         zram->disk->queue->queuedata = zram;
1937         zram->disk->private_data = zram;
1938         snprintf(zram->disk->disk_name, 16, "zram%d", device_id);
1939
1940         /* Actual capacity set using syfs (/sys/block/zram<id>/disksize */
1941         set_capacity(zram->disk, 0);
1942         /* zram devices sort of resembles non-rotational disks */
1943         blk_queue_flag_set(QUEUE_FLAG_NONROT, zram->disk->queue);
1944         blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, zram->disk->queue);
1945
1946         /*
1947          * To ensure that we always get PAGE_SIZE aligned
1948          * and n*PAGE_SIZED sized I/O requests.
1949          */
1950         blk_queue_physical_block_size(zram->disk->queue, PAGE_SIZE);
1951         blk_queue_logical_block_size(zram->disk->queue,
1952                                         ZRAM_LOGICAL_BLOCK_SIZE);
1953         blk_queue_io_min(zram->disk->queue, PAGE_SIZE);
1954         blk_queue_io_opt(zram->disk->queue, PAGE_SIZE);
1955         zram->disk->queue->limits.discard_granularity = PAGE_SIZE;
1956         blk_queue_max_discard_sectors(zram->disk->queue, UINT_MAX);
1957         blk_queue_flag_set(QUEUE_FLAG_DISCARD, zram->disk->queue);
1958
1959         /*
1960          * zram_bio_discard() will clear all logical blocks if logical block
1961          * size is identical with physical block size(PAGE_SIZE). But if it is
1962          * different, we will skip discarding some parts of logical blocks in
1963          * the part of the request range which isn't aligned to physical block
1964          * size.  So we can't ensure that all discarded logical blocks are
1965          * zeroed.
1966          */
1967         if (ZRAM_LOGICAL_BLOCK_SIZE == PAGE_SIZE)
1968                 blk_queue_max_write_zeroes_sectors(zram->disk->queue, UINT_MAX);
1969
1970         zram->disk->queue->backing_dev_info->capabilities |=
1971                         (BDI_CAP_STABLE_WRITES | BDI_CAP_SYNCHRONOUS_IO);
1972         device_add_disk(NULL, zram->disk, zram_disk_attr_groups);
1973
1974         strlcpy(zram->compressor, default_compressor, sizeof(zram->compressor));
1975
1976         zram_debugfs_register(zram);
1977         pr_info("Added device: %s\n", zram->disk->disk_name);
1978         return device_id;
1979
1980 out_free_queue:
1981         blk_cleanup_queue(queue);
1982 out_free_idr:
1983         idr_remove(&zram_index_idr, device_id);
1984 out_free_dev:
1985         kfree(zram);
1986         return ret;
1987 }
1988
1989 static int zram_remove(struct zram *zram)
1990 {
1991         struct block_device *bdev;
1992
1993         bdev = bdget_disk(zram->disk, 0);
1994         if (!bdev)
1995                 return -ENOMEM;
1996
1997         mutex_lock(&bdev->bd_mutex);
1998         if (bdev->bd_openers || zram->claim) {
1999                 mutex_unlock(&bdev->bd_mutex);
2000                 bdput(bdev);
2001                 return -EBUSY;
2002         }
2003
2004         zram->claim = true;
2005         mutex_unlock(&bdev->bd_mutex);
2006
2007         zram_debugfs_unregister(zram);
2008
2009         /* Make sure all the pending I/O are finished */
2010         fsync_bdev(bdev);
2011         zram_reset_device(zram);
2012         bdput(bdev);
2013
2014         pr_info("Removed device: %s\n", zram->disk->disk_name);
2015
2016         del_gendisk(zram->disk);
2017         blk_cleanup_queue(zram->disk->queue);
2018         put_disk(zram->disk);
2019         kfree(zram);
2020         return 0;
2021 }
2022
2023 /* zram-control sysfs attributes */
2024
2025 /*
2026  * NOTE: hot_add attribute is not the usual read-only sysfs attribute. In a
2027  * sense that reading from this file does alter the state of your system -- it
2028  * creates a new un-initialized zram device and returns back this device's
2029  * device_id (or an error code if it fails to create a new device).
2030  */
2031 static ssize_t hot_add_show(struct class *class,
2032                         struct class_attribute *attr,
2033                         char *buf)
2034 {
2035         int ret;
2036
2037         mutex_lock(&zram_index_mutex);
2038         ret = zram_add();
2039         mutex_unlock(&zram_index_mutex);
2040
2041         if (ret < 0)
2042                 return ret;
2043         return scnprintf(buf, PAGE_SIZE, "%d\n", ret);
2044 }
2045 static CLASS_ATTR_RO(hot_add);
2046
2047 static ssize_t hot_remove_store(struct class *class,
2048                         struct class_attribute *attr,
2049                         const char *buf,
2050                         size_t count)
2051 {
2052         struct zram *zram;
2053         int ret, dev_id;
2054
2055         /* dev_id is gendisk->first_minor, which is `int' */
2056         ret = kstrtoint(buf, 10, &dev_id);
2057         if (ret)
2058                 return ret;
2059         if (dev_id < 0)
2060                 return -EINVAL;
2061
2062         mutex_lock(&zram_index_mutex);
2063
2064         zram = idr_find(&zram_index_idr, dev_id);
2065         if (zram) {
2066                 ret = zram_remove(zram);
2067                 if (!ret)
2068                         idr_remove(&zram_index_idr, dev_id);
2069         } else {
2070                 ret = -ENODEV;
2071         }
2072
2073         mutex_unlock(&zram_index_mutex);
2074         return ret ? ret : count;
2075 }
2076 static CLASS_ATTR_WO(hot_remove);
2077
2078 static struct attribute *zram_control_class_attrs[] = {
2079         &class_attr_hot_add.attr,
2080         &class_attr_hot_remove.attr,
2081         NULL,
2082 };
2083 ATTRIBUTE_GROUPS(zram_control_class);
2084
2085 static struct class zram_control_class = {
2086         .name           = "zram-control",
2087         .owner          = THIS_MODULE,
2088         .class_groups   = zram_control_class_groups,
2089 };
2090
2091 static int zram_remove_cb(int id, void *ptr, void *data)
2092 {
2093         zram_remove(ptr);
2094         return 0;
2095 }
2096
2097 static void destroy_devices(void)
2098 {
2099         class_unregister(&zram_control_class);
2100         idr_for_each(&zram_index_idr, &zram_remove_cb, NULL);
2101         zram_debugfs_destroy();
2102         idr_destroy(&zram_index_idr);
2103         unregister_blkdev(zram_major, "zram");
2104         cpuhp_remove_multi_state(CPUHP_ZCOMP_PREPARE);
2105 }
2106
2107 static int __init zram_init(void)
2108 {
2109         int ret;
2110
2111         ret = cpuhp_setup_state_multi(CPUHP_ZCOMP_PREPARE, "block/zram:prepare",
2112                                       zcomp_cpu_up_prepare, zcomp_cpu_dead);
2113         if (ret < 0)
2114                 return ret;
2115
2116         ret = class_register(&zram_control_class);
2117         if (ret) {
2118                 pr_err("Unable to register zram-control class\n");
2119                 cpuhp_remove_multi_state(CPUHP_ZCOMP_PREPARE);
2120                 return ret;
2121         }
2122
2123         zram_debugfs_create();
2124         zram_major = register_blkdev(0, "zram");
2125         if (zram_major <= 0) {
2126                 pr_err("Unable to get major number\n");
2127                 class_unregister(&zram_control_class);
2128                 cpuhp_remove_multi_state(CPUHP_ZCOMP_PREPARE);
2129                 return -EBUSY;
2130         }
2131
2132         while (num_devices != 0) {
2133                 mutex_lock(&zram_index_mutex);
2134                 ret = zram_add();
2135                 mutex_unlock(&zram_index_mutex);
2136                 if (ret < 0)
2137                         goto out_error;
2138                 num_devices--;
2139         }
2140
2141         return 0;
2142
2143 out_error:
2144         destroy_devices();
2145         return ret;
2146 }
2147
2148 static void __exit zram_exit(void)
2149 {
2150         destroy_devices();
2151 }
2152
2153 module_init(zram_init);
2154 module_exit(zram_exit);
2155
2156 module_param(num_devices, uint, 0);
2157 MODULE_PARM_DESC(num_devices, "Number of pre-created zram devices");
2158
2159 MODULE_LICENSE("Dual BSD/GPL");
2160 MODULE_AUTHOR("Nitin Gupta <ngupta@vflare.org>");
2161 MODULE_DESCRIPTION("Compressed RAM Block Device");