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