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/cpuhotplug.h>
  35
  36 #include "zram_drv.h"
  37
  38 static DEFINE_IDR(zram_index_idr);
  39 /* idr index must be protected */
  40 static DEFINE_MUTEX(zram_index_mutex);
  41
  42 static int zram_major;
  43 static const char *default_compressor = "lzo";
  44
  45 /* Module params (documentation at end) */
  46 static unsigned int num_devices = 1;
  47
  48 static void zram_free_page(struct zram *zram, size_t index);
  49
  50 static inline bool init_done(struct zram *zram)
  51 {
  52         return zram->disksize;
  53 }
  54
  55 static inline struct zram *dev_to_zram(struct device *dev)
  56 {
  57         return (struct zram *)dev_to_disk(dev)->private_data;
  58 }
  59
  60 static unsigned long zram_get_handle(struct zram *zram, u32 index)
  61 {
  62         return zram->table[index].handle;
  63 }
  64
  65 static void zram_set_handle(struct zram *zram, u32 index, unsigned long handle)
  66 {
  67         zram->table[index].handle = handle;
  68 }
  69
  70 /* flag operations require table entry bit_spin_lock() being held */
  71 static int zram_test_flag(struct zram *zram, u32 index,
  72                         enum zram_pageflags flag)
  73 {
  74         return zram->table[index].value & BIT(flag);
  75 }
  76
  77 static void zram_set_flag(struct zram *zram, u32 index,
  78                         enum zram_pageflags flag)
  79 {
  80         zram->table[index].value |= BIT(flag);
  81 }
  82
  83 static void zram_clear_flag(struct zram *zram, u32 index,
  84                         enum zram_pageflags flag)
  85 {
  86         zram->table[index].value &= ~BIT(flag);
  87 }
  88
  89 static inline void zram_set_element(struct zram *zram, u32 index,
  90                         unsigned long element)
  91 {
  92         zram->table[index].element = element;
  93 }
  94
  95 static unsigned long zram_get_element(struct zram *zram, u32 index)
  96 {
  97         return zram->table[index].element;
  98 }
  99
 100 static size_t zram_get_obj_size(struct zram *zram, u32 index)
 101 {
 102         return zram->table[index].value & (BIT(ZRAM_FLAG_SHIFT) - 1);
 103 }
 104
 105 static void zram_set_obj_size(struct zram *zram,
 106                                         u32 index, size_t size)
 107 {
 108         unsigned long flags = zram->table[index].value >> ZRAM_FLAG_SHIFT;
 109
 110         zram->table[index].value = (flags << ZRAM_FLAG_SHIFT) | size;
 111 }
 112
 113 #if PAGE_SIZE != 4096
 114 static inline bool is_partial_io(struct bio_vec *bvec)
 115 {
 116         return bvec->bv_len != PAGE_SIZE;
 117 }
 118 #else
 119 static inline bool is_partial_io(struct bio_vec *bvec)
 120 {
 121         return false;
 122 }
 123 #endif
 124
 125 static void zram_revalidate_disk(struct zram *zram)
 126 {
 127         revalidate_disk(zram->disk);
 128         /* revalidate_disk reset the BDI_CAP_STABLE_WRITES so set again */
 129         zram->disk->queue->backing_dev_info->capabilities |=
 130                 BDI_CAP_STABLE_WRITES;
 131 }
 132
 133 /*
 134  * Check if request is within bounds and aligned on zram logical blocks.
 135  */
 136 static inline bool valid_io_request(struct zram *zram,
 137                 sector_t start, unsigned int size)
 138 {
 139         u64 end, bound;
 140
 141         /* unaligned request */
 142         if (unlikely(start & (ZRAM_SECTOR_PER_LOGICAL_BLOCK - 1)))
 143                 return false;
 144         if (unlikely(size & (ZRAM_LOGICAL_BLOCK_SIZE - 1)))
 145                 return false;
 146
 147         end = start + (size >> SECTOR_SHIFT);
 148         bound = zram->disksize >> SECTOR_SHIFT;
 149         /* out of range range */
 150         if (unlikely(start >= bound || end > bound || start > end))
 151                 return false;
 152
 153         /* I/O request is valid */
 154         return true;
 155 }
 156
 157 static void update_position(u32 *index, int *offset, struct bio_vec *bvec)
 158 {
 159         *index  += (*offset + bvec->bv_len) / PAGE_SIZE;
 160         *offset = (*offset + bvec->bv_len) % PAGE_SIZE;
 161 }
 162
 163 static inline void update_used_max(struct zram *zram,
 164                                         const unsigned long pages)
 165 {
 166         unsigned long old_max, cur_max;
 167
 168         old_max = atomic_long_read(&zram->stats.max_used_pages);
 169
 170         do {
 171                 cur_max = old_max;
 172                 if (pages > cur_max)
 173                         old_max = atomic_long_cmpxchg(
 174                                 &zram->stats.max_used_pages, cur_max, pages);
 175         } while (old_max != cur_max);
 176 }
 177
 178 static inline void zram_fill_page(void *ptr, unsigned long len,
 179                                         unsigned long value)
 180 {
 181         WARN_ON_ONCE(!IS_ALIGNED(len, sizeof(unsigned long)));
 182         memset_l(ptr, value, len / sizeof(unsigned long));
 183 }
 184
 185 static bool page_same_filled(void *ptr, unsigned long *element)
 186 {
 187         unsigned int pos;
 188         unsigned long *page;
 189         unsigned long val;
 190
 191         page = (unsigned long *)ptr;
 192         val = page[0];
 193
 194         for (pos = 1; pos < PAGE_SIZE / sizeof(*page); pos++) {
 195                 if (val != page[pos])
 196                         return false;
 197         }
 198
 199         *element = val;
 200
 201         return true;
 202 }
 203
 204 static ssize_t initstate_show(struct device *dev,
 205                 struct device_attribute *attr, char *buf)
 206 {
 207         u32 val;
 208         struct zram *zram = dev_to_zram(dev);
 209
 210         down_read(&zram->init_lock);
 211         val = init_done(zram);
 212         up_read(&zram->init_lock);
 213
 214         return scnprintf(buf, PAGE_SIZE, "%u\n", val);
 215 }
 216
 217 static ssize_t disksize_show(struct device *dev,
 218                 struct device_attribute *attr, char *buf)
 219 {
 220         struct zram *zram = dev_to_zram(dev);
 221
 222         return scnprintf(buf, PAGE_SIZE, "%llu\n", zram->disksize);
 223 }
 224
 225 static ssize_t mem_limit_store(struct device *dev,
 226                 struct device_attribute *attr, const char *buf, size_t len)
 227 {
 228         u64 limit;
 229         char *tmp;
 230         struct zram *zram = dev_to_zram(dev);
 231
 232         limit = memparse(buf, &tmp);
 233         if (buf == tmp) /* no chars parsed, invalid input */
 234                 return -EINVAL;
 235
 236         down_write(&zram->init_lock);
 237         zram->limit_pages = PAGE_ALIGN(limit) >> PAGE_SHIFT;
 238         up_write(&zram->init_lock);
 239
 240         return len;
 241 }
 242
 243 static ssize_t mem_used_max_store(struct device *dev,
 244                 struct device_attribute *attr, const char *buf, size_t len)
 245 {
 246         int err;
 247         unsigned long val;
 248         struct zram *zram = dev_to_zram(dev);
 249
 250         err = kstrtoul(buf, 10, &val);
 251         if (err || val != 0)
 252                 return -EINVAL;
 253
 254         down_read(&zram->init_lock);
 255         if (init_done(zram)) {
 256                 atomic_long_set(&zram->stats.max_used_pages,
 257                                 zs_get_total_pages(zram->mem_pool));
 258         }
 259         up_read(&zram->init_lock);
 260
 261         return len;
 262 }
 263
 264 #ifdef CONFIG_ZRAM_WRITEBACK
 265 static bool zram_wb_enabled(struct zram *zram)
 266 {
 267         return zram->backing_dev;
 268 }
 269
 270 static void reset_bdev(struct zram *zram)
 271 {
 272         struct block_device *bdev;
 273
 274         if (!zram_wb_enabled(zram))
 275                 return;
 276
 277         bdev = zram->bdev;
 278         if (zram->old_block_size)
 279                 set_blocksize(bdev, zram->old_block_size);
 280         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 281         /* hope filp_close flush all of IO */
 282         filp_close(zram->backing_dev, NULL);
 283         zram->backing_dev = NULL;
 284         zram->old_block_size = 0;
 285         zram->bdev = NULL;
 286
 287         kvfree(zram->bitmap);
 288         zram->bitmap = NULL;
 289 }
 290
 291 static ssize_t backing_dev_show(struct device *dev,
 292                 struct device_attribute *attr, char *buf)
 293 {
 294         struct zram *zram = dev_to_zram(dev);
 295         struct file *file = zram->backing_dev;
 296         char *p;
 297         ssize_t ret;
 298
 299         down_read(&zram->init_lock);
 300         if (!zram_wb_enabled(zram)) {
 301                 memcpy(buf, "none\n", 5);
 302                 up_read(&zram->init_lock);
 303                 return 5;
 304         }
 305
 306         p = file_path(file, buf, PAGE_SIZE - 1);
 307         if (IS_ERR(p)) {
 308                 ret = PTR_ERR(p);
 309                 goto out;
 310         }
 311
 312         ret = strlen(p);
 313         memmove(buf, p, ret);
 314         buf[ret++] = '\n';
 315 out:
 316         up_read(&zram->init_lock);
 317         return ret;
 318 }
 319
 320 static ssize_t backing_dev_store(struct device *dev,
 321                 struct device_attribute *attr, const char *buf, size_t len)
 322 {
 323         char *file_name;
 324         struct file *backing_dev = NULL;
 325         struct inode *inode;
 326         struct address_space *mapping;
 327         unsigned int bitmap_sz, old_block_size = 0;
 328         unsigned long nr_pages, *bitmap = NULL;
 329         struct block_device *bdev = NULL;
 330         int err;
 331         struct zram *zram = dev_to_zram(dev);
 332
 333         file_name = kmalloc(PATH_MAX, GFP_KERNEL);
 334         if (!file_name)
 335                 return -ENOMEM;
 336
 337         down_write(&zram->init_lock);
 338         if (init_done(zram)) {
 339                 pr_info("Can't setup backing device for initialized device\n");
 340                 err = -EBUSY;
 341                 goto out;
 342         }
 343
 344         strlcpy(file_name, buf, len);
 345
 346         backing_dev = filp_open(file_name, O_RDWR|O_LARGEFILE, 0);
 347         if (IS_ERR(backing_dev)) {
 348                 err = PTR_ERR(backing_dev);
 349                 backing_dev = NULL;
 350                 goto out;
 351         }
 352
 353         mapping = backing_dev->f_mapping;
 354         inode = mapping->host;
 355
 356         /* Support only block device in this moment */
 357         if (!S_ISBLK(inode->i_mode)) {
 358                 err = -ENOTBLK;
 359                 goto out;
 360         }
 361
 362         bdev = bdgrab(I_BDEV(inode));
 363         err = blkdev_get(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL, zram);
 364         if (err < 0)
 365                 goto out;
 366
 367         nr_pages = i_size_read(inode) >> PAGE_SHIFT;
 368         bitmap_sz = BITS_TO_LONGS(nr_pages) * sizeof(long);
 369         bitmap = kvzalloc(bitmap_sz, GFP_KERNEL);
 370         if (!bitmap) {
 371                 err = -ENOMEM;
 372                 goto out;
 373         }
 374
 375         old_block_size = block_size(bdev);
 376         err = set_blocksize(bdev, PAGE_SIZE);
 377         if (err)
 378                 goto out;
 379
 380         reset_bdev(zram);
 381         spin_lock_init(&zram->bitmap_lock);
 382
 383         zram->old_block_size = old_block_size;
 384         zram->bdev = bdev;
 385         zram->backing_dev = backing_dev;
 386         zram->bitmap = bitmap;
 387         zram->nr_pages = nr_pages;
 388         up_write(&zram->init_lock);
 389
 390         pr_info("setup backing device %s\n", file_name);
 391         kfree(file_name);
 392
 393         return len;
 394 out:
 395         if (bitmap)
 396                 kvfree(bitmap);
 397
 398         if (bdev)
 399                 blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
 400
 401         if (backing_dev)
 402                 filp_close(backing_dev, NULL);
 403
 404         up_write(&zram->init_lock);
 405
 406         kfree(file_name);
 407
 408         return err;
 409 }
 410
 411 static unsigned long get_entry_bdev(struct zram *zram)
 412 {
 413         unsigned long entry;
 414
 415         spin_lock(&zram->bitmap_lock);
 416         /* skip 0 bit to confuse zram.handle = 0 */
 417         entry = find_next_zero_bit(zram->bitmap, zram->nr_pages, 1);
 418         if (entry == zram->nr_pages) {
 419                 spin_unlock(&zram->bitmap_lock);
 420                 return 0;
 421         }
 422
 423         set_bit(entry, zram->bitmap);
 424         spin_unlock(&zram->bitmap_lock);
 425
 426         return entry;
 427 }
 428
 429 static void put_entry_bdev(struct zram *zram, unsigned long entry)
 430 {
 431         int was_set;
 432
 433         spin_lock(&zram->bitmap_lock);
 434         was_set = test_and_clear_bit(entry, zram->bitmap);
 435         spin_unlock(&zram->bitmap_lock);
 436         WARN_ON_ONCE(!was_set);
 437 }
 438
 439 void zram_page_end_io(struct bio *bio)
 440 {
 441         struct page *page = bio->bi_io_vec[0].bv_page;
 442
 443         page_endio(page, op_is_write(bio_op(bio)),
 444                         blk_status_to_errno(bio->bi_status));
 445         bio_put(bio);
 446 }
 447
 448 /*
 449  * Returns 1 if the submission is successful.
 450  */
 451 static int read_from_bdev_async(struct zram *zram, struct bio_vec *bvec,
 452                         unsigned long entry, struct bio *parent)
 453 {
 454         struct bio *bio;
 455
 456         bio = bio_alloc(GFP_ATOMIC, 1);
 457         if (!bio)
 458                 return -ENOMEM;
 459
 460         bio->bi_iter.bi_sector = entry * (PAGE_SIZE >> 9);
 461         bio_set_dev(bio, zram->bdev);
 462         if (!bio_add_page(bio, bvec->bv_page, bvec->bv_len, bvec->bv_offset)) {
 463                 bio_put(bio);
 464                 return -EIO;
 465         }
 466
 467         if (!parent) {
 468                 bio->bi_opf = REQ_OP_READ;
 469                 bio->bi_end_io = zram_page_end_io;
 470         } else {
 471                 bio->bi_opf = parent->bi_opf;
 472                 bio_chain(bio, parent);
 473         }
 474
 475         submit_bio(bio);
 476         return 1;
 477 }
 478
 479 struct zram_work {
 480         struct work_struct work;
 481         struct zram *zram;
 482         unsigned long entry;
 483         struct bio *bio;
 484 };
 485
 486 #if PAGE_SIZE != 4096
 487 static void zram_sync_read(struct work_struct *work)
 488 {
 489         struct bio_vec bvec;
 490         struct zram_work *zw = container_of(work, struct zram_work, work);
 491         struct zram *zram = zw->zram;
 492         unsigned long entry = zw->entry;
 493         struct bio *bio = zw->bio;
 494
 495         read_from_bdev_async(zram, &bvec, entry, bio);
 496 }
 497
 498 /*
 499  * Block layer want one ->make_request_fn to be active at a time
 500  * so if we use chained IO with parent IO in same context,
 501  * it's a deadlock. To avoid, it, it uses worker thread context.
 502  */
 503 static int read_from_bdev_sync(struct zram *zram, struct bio_vec *bvec,
 504                                 unsigned long entry, struct bio *bio)
 505 {
 506         struct zram_work work;
 507
 508         work.zram = zram;
 509         work.entry = entry;
 510         work.bio = bio;
 511
 512         INIT_WORK_ONSTACK(&work.work, zram_sync_read);
 513         queue_work(system_unbound_wq, &work.work);
 514         flush_work(&work.work);
 515         destroy_work_on_stack(&work.work);
 516
 517         return 1;
 518 }
 519 #else
 520 static int read_from_bdev_sync(struct zram *zram, struct bio_vec *bvec,
 521                                 unsigned long entry, struct bio *bio)
 522 {
 523         WARN_ON(1);
 524         return -EIO;
 525 }
 526 #endif
 527
 528 static int read_from_bdev(struct zram *zram, struct bio_vec *bvec,
 529                         unsigned long entry, struct bio *parent, bool sync)
 530 {
 531         if (sync)
 532                 return read_from_bdev_sync(zram, bvec, entry, parent);
 533         else
 534                 return read_from_bdev_async(zram, bvec, entry, parent);
 535 }
 536
 537 static int write_to_bdev(struct zram *zram, struct bio_vec *bvec,
 538                                         u32 index, struct bio *parent,
 539                                         unsigned long *pentry)
 540 {
 541         struct bio *bio;
 542         unsigned long entry;
 543
 544         bio = bio_alloc(GFP_ATOMIC, 1);
 545         if (!bio)
 546                 return -ENOMEM;
 547
 548         entry = get_entry_bdev(zram);
 549         if (!entry) {
 550                 bio_put(bio);
 551                 return -ENOSPC;
 552         }
 553
 554         bio->bi_iter.bi_sector = entry * (PAGE_SIZE >> 9);
 555         bio_set_dev(bio, zram->bdev);
 556         if (!bio_add_page(bio, bvec->bv_page, bvec->bv_len,
 557                                         bvec->bv_offset)) {
 558                 bio_put(bio);
 559                 put_entry_bdev(zram, entry);
 560                 return -EIO;
 561         }
 562
 563         if (!parent) {
 564                 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC;
 565                 bio->bi_end_io = zram_page_end_io;
 566         } else {
 567                 bio->bi_opf = parent->bi_opf;
 568                 bio_chain(bio, parent);
 569         }
 570
 571         submit_bio(bio);
 572         *pentry = entry;
 573
 574         return 0;
 575 }
 576
 577 static void zram_wb_clear(struct zram *zram, u32 index)
 578 {
 579         unsigned long entry;
 580
 581         zram_clear_flag(zram, index, ZRAM_WB);
 582         entry = zram_get_element(zram, index);
 583         zram_set_element(zram, index, 0);
 584         put_entry_bdev(zram, entry);
 585 }
 586
 587 #else
 588 static bool zram_wb_enabled(struct zram *zram) { return false; }
 589 static inline void reset_bdev(struct zram *zram) {};
 590 static int write_to_bdev(struct zram *zram, struct bio_vec *bvec,
 591                                         u32 index, struct bio *parent,
 592                                         unsigned long *pentry)
 593
 594 {
 595         return -EIO;
 596 }
 597
 598 static int read_from_bdev(struct zram *zram, struct bio_vec *bvec,
 599                         unsigned long entry, struct bio *parent, bool sync)
 600 {
 601         return -EIO;
 602 }
 603 static void zram_wb_clear(struct zram *zram, u32 index) {}
 604 #endif
 605
 606
 607 /*
 608  * We switched to per-cpu streams and this attr is not needed anymore.
 609  * However, we will keep it around for some time, because:
 610  * a) we may revert per-cpu streams in the future
 611  * b) it's visible to user space and we need to follow our 2 years
 612  *    retirement rule; but we already have a number of 'soon to be
 613  *    altered' attrs, so max_comp_streams need to wait for the next
 614  *    layoff cycle.
 615  */
 616 static ssize_t max_comp_streams_show(struct device *dev,
 617                 struct device_attribute *attr, char *buf)
 618 {
 619         return scnprintf(buf, PAGE_SIZE, "%d\n", num_online_cpus());
 620 }
 621
 622 static ssize_t max_comp_streams_store(struct device *dev,
 623                 struct device_attribute *attr, const char *buf, size_t len)
 624 {
 625         return len;
 626 }
 627
 628 static ssize_t comp_algorithm_show(struct device *dev,
 629                 struct device_attribute *attr, char *buf)
 630 {
 631         size_t sz;
 632         struct zram *zram = dev_to_zram(dev);
 633
 634         down_read(&zram->init_lock);
 635         sz = zcomp_available_show(zram->compressor, buf);
 636         up_read(&zram->init_lock);
 637
 638         return sz;
 639 }
 640
 641 static ssize_t comp_algorithm_store(struct device *dev,
 642                 struct device_attribute *attr, const char *buf, size_t len)
 643 {
 644         struct zram *zram = dev_to_zram(dev);
 645         char compressor[ARRAY_SIZE(zram->compressor)];
 646         size_t sz;
 647
 648         strlcpy(compressor, buf, sizeof(compressor));
 649         /* ignore trailing newline */
 650         sz = strlen(compressor);
 651         if (sz > 0 && compressor[sz - 1] == '\n')
 652                 compressor[sz - 1] = 0x00;
 653
 654         if (!zcomp_available_algorithm(compressor))
 655                 return -EINVAL;
 656
 657         down_write(&zram->init_lock);
 658         if (init_done(zram)) {
 659                 up_write(&zram->init_lock);
 660                 pr_info("Can't change algorithm for initialized device\n");
 661                 return -EBUSY;
 662         }
 663
 664         strcpy(zram->compressor, compressor);
 665         up_write(&zram->init_lock);
 666         return len;
 667 }
 668
 669 static ssize_t compact_store(struct device *dev,
 670                 struct device_attribute *attr, const char *buf, size_t len)
 671 {
 672         struct zram *zram = dev_to_zram(dev);
 673
 674         down_read(&zram->init_lock);
 675         if (!init_done(zram)) {
 676                 up_read(&zram->init_lock);
 677                 return -EINVAL;
 678         }
 679
 680         zs_compact(zram->mem_pool);
 681         up_read(&zram->init_lock);
 682
 683         return len;
 684 }
 685
 686 static ssize_t io_stat_show(struct device *dev,
 687                 struct device_attribute *attr, char *buf)
 688 {
 689         struct zram *zram = dev_to_zram(dev);
 690         ssize_t ret;
 691
 692         down_read(&zram->init_lock);
 693         ret = scnprintf(buf, PAGE_SIZE,
 694                         "%8llu %8llu %8llu %8llu\n",
 695                         (u64)atomic64_read(&zram->stats.failed_reads),
 696                         (u64)atomic64_read(&zram->stats.failed_writes),
 697                         (u64)atomic64_read(&zram->stats.invalid_io),
 698                         (u64)atomic64_read(&zram->stats.notify_free));
 699         up_read(&zram->init_lock);
 700
 701         return ret;
 702 }
 703
 704 static ssize_t mm_stat_show(struct device *dev,
 705                 struct device_attribute *attr, char *buf)
 706 {
 707         struct zram *zram = dev_to_zram(dev);
 708         struct zs_pool_stats pool_stats;
 709         u64 orig_size, mem_used = 0;
 710         long max_used;
 711         ssize_t ret;
 712
 713         memset(&pool_stats, 0x00, sizeof(struct zs_pool_stats));
 714
 715         down_read(&zram->init_lock);
 716         if (init_done(zram)) {
 717                 mem_used = zs_get_total_pages(zram->mem_pool);
 718                 zs_pool_stats(zram->mem_pool, &pool_stats);
 719         }
 720
 721         orig_size = atomic64_read(&zram->stats.pages_stored);
 722         max_used = atomic_long_read(&zram->stats.max_used_pages);
 723
 724         ret = scnprintf(buf, PAGE_SIZE,
 725                         "%8llu %8llu %8llu %8lu %8ld %8llu %8lu\n",
 726                         orig_size << PAGE_SHIFT,
 727                         (u64)atomic64_read(&zram->stats.compr_data_size),
 728                         mem_used << PAGE_SHIFT,
 729                         zram->limit_pages << PAGE_SHIFT,
 730                         max_used << PAGE_SHIFT,
 731                         (u64)atomic64_read(&zram->stats.same_pages),
 732                         pool_stats.pages_compacted);
 733         up_read(&zram->init_lock);
 734
 735         return ret;
 736 }
 737
 738 static ssize_t debug_stat_show(struct device *dev,
 739                 struct device_attribute *attr, char *buf)
 740 {
 741         int version = 1;
 742         struct zram *zram = dev_to_zram(dev);
 743         ssize_t ret;
 744
 745         down_read(&zram->init_lock);
 746         ret = scnprintf(buf, PAGE_SIZE,
 747                         "version: %d\n%8llu\n",
 748                         version,
 749                         (u64)atomic64_read(&zram->stats.writestall));
 750         up_read(&zram->init_lock);
 751
 752         return ret;
 753 }
 754
 755 static DEVICE_ATTR_RO(io_stat);
 756 static DEVICE_ATTR_RO(mm_stat);
 757 static DEVICE_ATTR_RO(debug_stat);
 758
 759 static void zram_slot_lock(struct zram *zram, u32 index)
 760 {
 761         bit_spin_lock(ZRAM_ACCESS, &zram->table[index].value);
 762 }
 763
 764 static void zram_slot_unlock(struct zram *zram, u32 index)
 765 {
 766         bit_spin_unlock(ZRAM_ACCESS, &zram->table[index].value);
 767 }
 768
 769 static void zram_meta_free(struct zram *zram, u64 disksize)
 770 {
 771         size_t num_pages = disksize >> PAGE_SHIFT;
 772         size_t index;
 773
 774         /* Free all pages that are still in this zram device */
 775         for (index = 0; index < num_pages; index++)
 776                 zram_free_page(zram, index);
 777
 778         zs_destroy_pool(zram->mem_pool);
 779         vfree(zram->table);
 780 }
 781
 782 static bool zram_meta_alloc(struct zram *zram, u64 disksize)
 783 {
 784         size_t num_pages;
 785
 786         num_pages = disksize >> PAGE_SHIFT;
 787         zram->table = vzalloc(num_pages * sizeof(*zram->table));
 788         if (!zram->table)
 789                 return false;
 790
 791         zram->mem_pool = zs_create_pool(zram->disk->disk_name);
 792         if (!zram->mem_pool) {
 793                 vfree(zram->table);
 794                 return false;
 795         }
 796
 797         return true;
 798 }
 799
 800 /*
 801  * To protect concurrent access to the same index entry,
 802  * caller should hold this table index entry's bit_spinlock to
 803  * indicate this index entry is accessing.
 804  */
 805 static void zram_free_page(struct zram *zram, size_t index)
 806 {
 807         unsigned long handle;
 808
 809         if (zram_wb_enabled(zram) && zram_test_flag(zram, index, ZRAM_WB)) {
 810                 zram_wb_clear(zram, index);
 811                 atomic64_dec(&zram->stats.pages_stored);
 812                 return;
 813         }
 814
 815         /*
 816          * No memory is allocated for same element filled pages.
 817          * Simply clear same page flag.
 818          */
 819         if (zram_test_flag(zram, index, ZRAM_SAME)) {
 820                 zram_clear_flag(zram, index, ZRAM_SAME);
 821                 zram_set_element(zram, index, 0);
 822                 atomic64_dec(&zram->stats.same_pages);
 823                 atomic64_dec(&zram->stats.pages_stored);
 824                 return;
 825         }
 826
 827         handle = zram_get_handle(zram, index);
 828         if (!handle)
 829                 return;
 830
 831         zs_free(zram->mem_pool, handle);
 832
 833         atomic64_sub(zram_get_obj_size(zram, index),
 834                         &zram->stats.compr_data_size);
 835         atomic64_dec(&zram->stats.pages_stored);
 836
 837         zram_set_handle(zram, index, 0);
 838         zram_set_obj_size(zram, index, 0);
 839 }
 840
 841 static int __zram_bvec_read(struct zram *zram, struct page *page, u32 index,
 842                                 struct bio *bio, bool partial_io)
 843 {
 844         int ret;
 845         unsigned long handle;
 846         unsigned int size;
 847         void *src, *dst;
 848
 849         if (zram_wb_enabled(zram)) {
 850                 zram_slot_lock(zram, index);
 851                 if (zram_test_flag(zram, index, ZRAM_WB)) {
 852                         struct bio_vec bvec;
 853
 854                         zram_slot_unlock(zram, index);
 855
 856                         bvec.bv_page = page;
 857                         bvec.bv_len = PAGE_SIZE;
 858                         bvec.bv_offset = 0;
 859                         return read_from_bdev(zram, &bvec,
 860                                         zram_get_element(zram, index),
 861                                         bio, partial_io);
 862                 }
 863                 zram_slot_unlock(zram, index);
 864         }
 865
 866         zram_slot_lock(zram, index);
 867         handle = zram_get_handle(zram, index);
 868         if (!handle || zram_test_flag(zram, index, ZRAM_SAME)) {
 869                 unsigned long value;
 870                 void *mem;
 871
 872                 value = handle ? zram_get_element(zram, index) : 0;
 873                 mem = kmap_atomic(page);
 874                 zram_fill_page(mem, PAGE_SIZE, value);
 875                 kunmap_atomic(mem);
 876                 zram_slot_unlock(zram, index);
 877                 return 0;
 878         }
 879
 880         size = zram_get_obj_size(zram, index);
 881
 882         src = zs_map_object(zram->mem_pool, handle, ZS_MM_RO);
 883         if (size == PAGE_SIZE) {
 884                 dst = kmap_atomic(page);
 885                 memcpy(dst, src, PAGE_SIZE);
 886                 kunmap_atomic(dst);
 887                 ret = 0;
 888         } else {
 889                 struct zcomp_strm *zstrm = zcomp_stream_get(zram->comp);
 890
 891                 dst = kmap_atomic(page);
 892                 ret = zcomp_decompress(zstrm, src, size, dst);
 893                 kunmap_atomic(dst);
 894                 zcomp_stream_put(zram->comp);
 895         }
 896         zs_unmap_object(zram->mem_pool, handle);
 897         zram_slot_unlock(zram, index);
 898
 899         /* Should NEVER happen. Return bio error if it does. */
 900         if (unlikely(ret))
 901                 pr_err("Decompression failed! err=%d, page=%u\n", ret, index);
 902
 903         return ret;
 904 }
 905
 906 static int zram_bvec_read(struct zram *zram, struct bio_vec *bvec,
 907                                 u32 index, int offset, struct bio *bio)
 908 {
 909         int ret;
 910         struct page *page;
 911
 912         page = bvec->bv_page;
 913         if (is_partial_io(bvec)) {
 914                 /* Use a temporary buffer to decompress the page */
 915                 page = alloc_page(GFP_NOIO|__GFP_HIGHMEM);
 916                 if (!page)
 917                         return -ENOMEM;
 918         }
 919
 920         ret = __zram_bvec_read(zram, page, index, bio, is_partial_io(bvec));
 921         if (unlikely(ret))
 922                 goto out;
 923
 924         if (is_partial_io(bvec)) {
 925                 void *dst = kmap_atomic(bvec->bv_page);
 926                 void *src = kmap_atomic(page);
 927
 928                 memcpy(dst + bvec->bv_offset, src + offset, bvec->bv_len);
 929                 kunmap_atomic(src);
 930                 kunmap_atomic(dst);
 931         }
 932 out:
 933         if (is_partial_io(bvec))
 934                 __free_page(page);
 935
 936         return ret;
 937 }
 938
 939 static int __zram_bvec_write(struct zram *zram, struct bio_vec *bvec,
 940                                 u32 index, struct bio *bio)
 941 {
 942         int ret = 0;
 943         unsigned long alloced_pages;
 944         unsigned long handle = 0;
 945         unsigned int comp_len = 0;
 946         void *src, *dst, *mem;
 947         struct zcomp_strm *zstrm;
 948         struct page *page = bvec->bv_page;
 949         unsigned long element = 0;
 950         enum zram_pageflags flags = 0;
 951         bool allow_wb = true;
 952
 953         mem = kmap_atomic(page);
 954         if (page_same_filled(mem, &element)) {
 955                 kunmap_atomic(mem);
 956                 /* Free memory associated with this sector now. */
 957                 flags = ZRAM_SAME;
 958                 atomic64_inc(&zram->stats.same_pages);
 959                 goto out;
 960         }
 961         kunmap_atomic(mem);
 962
 963 compress_again:
 964         zstrm = zcomp_stream_get(zram->comp);
 965         src = kmap_atomic(page);
 966         ret = zcomp_compress(zstrm, src, &comp_len);
 967         kunmap_atomic(src);
 968
 969         if (unlikely(ret)) {
 970                 zcomp_stream_put(zram->comp);
 971                 pr_err("Compression failed! err=%d\n", ret);
 972                 zs_free(zram->mem_pool, handle);
 973                 return ret;
 974         }
 975
 976         if (unlikely(comp_len > max_zpage_size)) {
 977                 if (zram_wb_enabled(zram) && allow_wb) {
 978                         zcomp_stream_put(zram->comp);
 979                         ret = write_to_bdev(zram, bvec, index, bio, &element);
 980                         if (!ret) {
 981                                 flags = ZRAM_WB;
 982                                 ret = 1;
 983                                 goto out;
 984                         }
 985                         allow_wb = false;
 986                         goto compress_again;
 987                 }
 988                 comp_len = PAGE_SIZE;
 989         }
 990
 991         /*
 992          * handle allocation has 2 paths:
 993          * a) fast path is executed with preemption disabled (for
 994          *  per-cpu streams) and has __GFP_DIRECT_RECLAIM bit clear,
 995          *  since we can't sleep;
 996          * b) slow path enables preemption and attempts to allocate
 997          *  the page with __GFP_DIRECT_RECLAIM bit set. we have to
 998          *  put per-cpu compression stream and, thus, to re-do
 999          *  the compression once handle is allocated.
1000          *
1001          * if we have a 'non-null' handle here then we are coming
1002          * from the slow path and handle has already been allocated.
1003          */
1004         if (!handle)
1005                 handle = zs_malloc(zram->mem_pool, comp_len,
1006                                 __GFP_KSWAPD_RECLAIM |
1007                                 __GFP_NOWARN |
1008                                 __GFP_HIGHMEM |
1009                                 __GFP_MOVABLE);
1010         if (!handle) {
1011                 zcomp_stream_put(zram->comp);
1012                 atomic64_inc(&zram->stats.writestall);
1013                 handle = zs_malloc(zram->mem_pool, comp_len,
1014                                 GFP_NOIO | __GFP_HIGHMEM |
1015                                 __GFP_MOVABLE);
1016                 if (handle)
1017                         goto compress_again;
1018                 return -ENOMEM;
1019         }
1020
1021         alloced_pages = zs_get_total_pages(zram->mem_pool);
1022         update_used_max(zram, alloced_pages);
1023
1024         if (zram->limit_pages && alloced_pages > zram->limit_pages) {
1025                 zcomp_stream_put(zram->comp);
1026                 zs_free(zram->mem_pool, handle);
1027                 return -ENOMEM;
1028         }
1029
1030         dst = zs_map_object(zram->mem_pool, handle, ZS_MM_WO);
1031
1032         src = zstrm->buffer;
1033         if (comp_len == PAGE_SIZE)
1034                 src = kmap_atomic(page);
1035         memcpy(dst, src, comp_len);
1036         if (comp_len == PAGE_SIZE)
1037                 kunmap_atomic(src);
1038
1039         zcomp_stream_put(zram->comp);
1040         zs_unmap_object(zram->mem_pool, handle);
1041         atomic64_add(comp_len, &zram->stats.compr_data_size);
1042 out:
1043         /*
1044          * Free memory associated with this sector
1045          * before overwriting unused sectors.
1046          */
1047         zram_slot_lock(zram, index);
1048         zram_free_page(zram, index);
1049
1050         if (flags) {
1051                 zram_set_flag(zram, index, flags);
1052                 zram_set_element(zram, index, element);
1053         }  else {
1054                 zram_set_handle(zram, index, handle);
1055                 zram_set_obj_size(zram, index, comp_len);
1056         }
1057         zram_slot_unlock(zram, index);
1058
1059         /* Update stats */
1060         atomic64_inc(&zram->stats.pages_stored);
1061         return ret;
1062 }
1063
1064 static int zram_bvec_write(struct zram *zram, struct bio_vec *bvec,
1065                                 u32 index, int offset, struct bio *bio)
1066 {
1067         int ret;
1068         struct page *page = NULL;
1069         void *src;
1070         struct bio_vec vec;
1071
1072         vec = *bvec;
1073         if (is_partial_io(bvec)) {
1074                 void *dst;
1075                 /*
1076                  * This is a partial IO. We need to read the full page
1077                  * before to write the changes.
1078                  */
1079                 page = alloc_page(GFP_NOIO|__GFP_HIGHMEM);
1080                 if (!page)
1081                         return -ENOMEM;
1082
1083                 ret = __zram_bvec_read(zram, page, index, bio, true);
1084                 if (ret)
1085                         goto out;
1086
1087                 src = kmap_atomic(bvec->bv_page);
1088                 dst = kmap_atomic(page);
1089                 memcpy(dst + offset, src + bvec->bv_offset, bvec->bv_len);
1090                 kunmap_atomic(dst);
1091                 kunmap_atomic(src);
1092
1093                 vec.bv_page = page;
1094                 vec.bv_len = PAGE_SIZE;
1095                 vec.bv_offset = 0;
1096         }
1097
1098         ret = __zram_bvec_write(zram, &vec, index, bio);
1099 out:
1100         if (is_partial_io(bvec))
1101                 __free_page(page);
1102         return ret;
1103 }
1104
1105 /*
1106  * zram_bio_discard - handler on discard request
1107  * @index: physical block index in PAGE_SIZE units
1108  * @offset: byte offset within physical block
1109  */
1110 static void zram_bio_discard(struct zram *zram, u32 index,
1111                              int offset, struct bio *bio)
1112 {
1113         size_t n = bio->bi_iter.bi_size;
1114
1115         /*
1116          * zram manages data in physical block size units. Because logical block
1117          * size isn't identical with physical block size on some arch, we
1118          * could get a discard request pointing to a specific offset within a
1119          * certain physical block.  Although we can handle this request by
1120          * reading that physiclal block and decompressing and partially zeroing
1121          * and re-compressing and then re-storing it, this isn't reasonable
1122          * because our intent with a discard request is to save memory.  So
1123          * skipping this logical block is appropriate here.
1124          */
1125         if (offset) {
1126                 if (n <= (PAGE_SIZE - offset))
1127                         return;
1128
1129                 n -= (PAGE_SIZE - offset);
1130                 index++;
1131         }
1132
1133         while (n >= PAGE_SIZE) {
1134                 zram_slot_lock(zram, index);
1135                 zram_free_page(zram, index);
1136                 zram_slot_unlock(zram, index);
1137                 atomic64_inc(&zram->stats.notify_free);
1138                 index++;
1139                 n -= PAGE_SIZE;
1140         }
1141 }
1142
1143 /*
1144  * Returns errno if it has some problem. Otherwise return 0 or 1.
1145  * Returns 0 if IO request was done synchronously
1146  * Returns 1 if IO request was successfully submitted.
1147  */
1148 static int zram_bvec_rw(struct zram *zram, struct bio_vec *bvec, u32 index,
1149                         int offset, bool is_write, struct bio *bio)
1150 {
1151         unsigned long start_time = jiffies;
1152         int rw_acct = is_write ? REQ_OP_WRITE : REQ_OP_READ;
1153         struct request_queue *q = zram->disk->queue;
1154         int ret;
1155
1156         generic_start_io_acct(q, rw_acct, bvec->bv_len >> SECTOR_SHIFT,
1157                         &zram->disk->part0);
1158
1159         if (!is_write) {
1160                 atomic64_inc(&zram->stats.num_reads);
1161                 ret = zram_bvec_read(zram, bvec, index, offset, bio);
1162                 flush_dcache_page(bvec->bv_page);
1163         } else {
1164                 atomic64_inc(&zram->stats.num_writes);
1165                 ret = zram_bvec_write(zram, bvec, index, offset, bio);
1166         }
1167
1168         generic_end_io_acct(q, rw_acct, &zram->disk->part0, start_time);
1169
1170         if (unlikely(ret < 0)) {
1171                 if (!is_write)
1172                         atomic64_inc(&zram->stats.failed_reads);
1173                 else
1174                         atomic64_inc(&zram->stats.failed_writes);
1175         }
1176
1177         return ret;
1178 }
1179
1180 static void __zram_make_request(struct zram *zram, struct bio *bio)
1181 {
1182         int offset;
1183         u32 index;
1184         struct bio_vec bvec;
1185         struct bvec_iter iter;
1186
1187         index = bio->bi_iter.bi_sector >> SECTORS_PER_PAGE_SHIFT;
1188         offset = (bio->bi_iter.bi_sector &
1189                   (SECTORS_PER_PAGE - 1)) << SECTOR_SHIFT;
1190
1191         switch (bio_op(bio)) {
1192         case REQ_OP_DISCARD:
1193         case REQ_OP_WRITE_ZEROES:
1194                 zram_bio_discard(zram, index, offset, bio);
1195                 bio_endio(bio);
1196                 return;
1197         default:
1198                 break;
1199         }
1200
1201         bio_for_each_segment(bvec, bio, iter) {
1202                 struct bio_vec bv = bvec;
1203                 unsigned int unwritten = bvec.bv_len;
1204
1205                 do {
1206                         bv.bv_len = min_t(unsigned int, PAGE_SIZE - offset,
1207                                                         unwritten);
1208                         if (zram_bvec_rw(zram, &bv, index, offset,
1209                                         op_is_write(bio_op(bio)), bio) < 0)
1210                                 goto out;
1211
1212                         bv.bv_offset += bv.bv_len;
1213                         unwritten -= bv.bv_len;
1214
1215                         update_position(&index, &offset, &bv);
1216                 } while (unwritten);
1217         }
1218
1219         bio_endio(bio);
1220         return;
1221
1222 out:
1223         bio_io_error(bio);
1224 }
1225
1226 /*
1227  * Handler function for all zram I/O requests.
1228  */
1229 static blk_qc_t zram_make_request(struct request_queue *queue, struct bio *bio)
1230 {
1231         struct zram *zram = queue->queuedata;
1232
1233         if (!valid_io_request(zram, bio->bi_iter.bi_sector,
1234                                         bio->bi_iter.bi_size)) {
1235                 atomic64_inc(&zram->stats.invalid_io);
1236                 goto error;
1237         }
1238
1239         __zram_make_request(zram, bio);
1240         return BLK_QC_T_NONE;
1241
1242 error:
1243         bio_io_error(bio);
1244         return BLK_QC_T_NONE;
1245 }
1246
1247 static void zram_slot_free_notify(struct block_device *bdev,
1248                                 unsigned long index)
1249 {
1250         struct zram *zram;
1251
1252         zram = bdev->bd_disk->private_data;
1253
1254         zram_slot_lock(zram, index);
1255         zram_free_page(zram, index);
1256         zram_slot_unlock(zram, index);
1257         atomic64_inc(&zram->stats.notify_free);
1258 }
1259
1260 static int zram_rw_page(struct block_device *bdev, sector_t sector,
1261                        struct page *page, bool is_write)
1262 {
1263         int offset, ret;
1264         u32 index;
1265         struct zram *zram;
1266         struct bio_vec bv;
1267
1268         if (PageTransHuge(page))
1269                 return -ENOTSUPP;
1270         zram = bdev->bd_disk->private_data;
1271
1272         if (!valid_io_request(zram, sector, PAGE_SIZE)) {
1273                 atomic64_inc(&zram->stats.invalid_io);
1274                 ret = -EINVAL;
1275                 goto out;
1276         }
1277
1278         index = sector >> SECTORS_PER_PAGE_SHIFT;
1279         offset = (sector & (SECTORS_PER_PAGE - 1)) << SECTOR_SHIFT;
1280
1281         bv.bv_page = page;
1282         bv.bv_len = PAGE_SIZE;
1283         bv.bv_offset = 0;
1284
1285         ret = zram_bvec_rw(zram, &bv, index, offset, is_write, NULL);
1286 out:
1287         /*
1288          * If I/O fails, just return error(ie, non-zero) without
1289          * calling page_endio.
1290          * It causes resubmit the I/O with bio request by upper functions
1291          * of rw_page(e.g., swap_readpage, __swap_writepage) and
1292          * bio->bi_end_io does things to handle the error
1293          * (e.g., SetPageError, set_page_dirty and extra works).
1294          */
1295         if (unlikely(ret < 0))
1296                 return ret;
1297
1298         switch (ret) {
1299         case 0:
1300                 page_endio(page, is_write, 0);
1301                 break;
1302         case 1:
1303                 ret = 0;
1304                 break;
1305         default:
1306                 WARN_ON(1);
1307         }
1308         return ret;
1309 }
1310
1311 static void zram_reset_device(struct zram *zram)
1312 {
1313         struct zcomp *comp;
1314         u64 disksize;
1315
1316         down_write(&zram->init_lock);
1317
1318         zram->limit_pages = 0;
1319
1320         if (!init_done(zram)) {
1321                 up_write(&zram->init_lock);
1322                 return;
1323         }
1324
1325         comp = zram->comp;
1326         disksize = zram->disksize;
1327         zram->disksize = 0;
1328
1329         set_capacity(zram->disk, 0);
1330         part_stat_set_all(&zram->disk->part0, 0);
1331
1332         up_write(&zram->init_lock);
1333         /* I/O operation under all of CPU are done so let's free */
1334         zram_meta_free(zram, disksize);
1335         memset(&zram->stats, 0, sizeof(zram->stats));
1336         zcomp_destroy(comp);
1337         reset_bdev(zram);
1338 }
1339
1340 static ssize_t disksize_store(struct device *dev,
1341                 struct device_attribute *attr, const char *buf, size_t len)
1342 {
1343         u64 disksize;
1344         struct zcomp *comp;
1345         struct zram *zram = dev_to_zram(dev);
1346         int err;
1347
1348         disksize = memparse(buf, NULL);
1349         if (!disksize)
1350                 return -EINVAL;
1351
1352         down_write(&zram->init_lock);
1353         if (init_done(zram)) {
1354                 pr_info("Cannot change disksize for initialized device\n");
1355                 err = -EBUSY;
1356                 goto out_unlock;
1357         }
1358
1359         disksize = PAGE_ALIGN(disksize);
1360         if (!zram_meta_alloc(zram, disksize)) {
1361                 err = -ENOMEM;
1362                 goto out_unlock;
1363         }
1364
1365         comp = zcomp_create(zram->compressor);
1366         if (IS_ERR(comp)) {
1367                 pr_err("Cannot initialise %s compressing backend\n",
1368                                 zram->compressor);
1369                 err = PTR_ERR(comp);
1370                 goto out_free_meta;
1371         }
1372
1373         zram->comp = comp;
1374         zram->disksize = disksize;
1375         set_capacity(zram->disk, zram->disksize >> SECTOR_SHIFT);
1376         zram_revalidate_disk(zram);
1377         up_write(&zram->init_lock);
1378
1379         return len;
1380
1381 out_free_meta:
1382         zram_meta_free(zram, disksize);
1383 out_unlock:
1384         up_write(&zram->init_lock);
1385         return err;
1386 }
1387
1388 static ssize_t reset_store(struct device *dev,
1389                 struct device_attribute *attr, const char *buf, size_t len)
1390 {
1391         int ret;
1392         unsigned short do_reset;
1393         struct zram *zram;
1394         struct block_device *bdev;
1395
1396         ret = kstrtou16(buf, 10, &do_reset);
1397         if (ret)
1398                 return ret;
1399
1400         if (!do_reset)
1401                 return -EINVAL;
1402
1403         zram = dev_to_zram(dev);
1404         bdev = bdget_disk(zram->disk, 0);
1405         if (!bdev)
1406                 return -ENOMEM;
1407
1408         mutex_lock(&bdev->bd_mutex);
1409         /* Do not reset an active device or claimed device */
1410         if (bdev->bd_openers || zram->claim) {
1411                 mutex_unlock(&bdev->bd_mutex);
1412                 bdput(bdev);
1413                 return -EBUSY;
1414         }
1415
1416         /* From now on, anyone can't open /dev/zram[0-9] */
1417         zram->claim = true;
1418         mutex_unlock(&bdev->bd_mutex);
1419
1420         /* Make sure all the pending I/O are finished */
1421         fsync_bdev(bdev);
1422         zram_reset_device(zram);
1423         zram_revalidate_disk(zram);
1424         bdput(bdev);
1425
1426         mutex_lock(&bdev->bd_mutex);
1427         zram->claim = false;
1428         mutex_unlock(&bdev->bd_mutex);
1429
1430         return len;
1431 }
1432
1433 static int zram_open(struct block_device *bdev, fmode_t mode)
1434 {
1435         int ret = 0;
1436         struct zram *zram;
1437
1438         WARN_ON(!mutex_is_locked(&bdev->bd_mutex));
1439
1440         zram = bdev->bd_disk->private_data;
1441         /* zram was claimed to reset so open request fails */
1442         if (zram->claim)
1443                 ret = -EBUSY;
1444
1445         return ret;
1446 }
1447
1448 static const struct block_device_operations zram_devops = {
1449         .open = zram_open,
1450         .swap_slot_free_notify = zram_slot_free_notify,
1451         .rw_page = zram_rw_page,
1452         .owner = THIS_MODULE
1453 };
1454
1455 static DEVICE_ATTR_WO(compact);
1456 static DEVICE_ATTR_RW(disksize);
1457 static DEVICE_ATTR_RO(initstate);
1458 static DEVICE_ATTR_WO(reset);
1459 static DEVICE_ATTR_WO(mem_limit);
1460 static DEVICE_ATTR_WO(mem_used_max);
1461 static DEVICE_ATTR_RW(max_comp_streams);
1462 static DEVICE_ATTR_RW(comp_algorithm);
1463 #ifdef CONFIG_ZRAM_WRITEBACK
1464 static DEVICE_ATTR_RW(backing_dev);
1465 #endif
1466
1467 static struct attribute *zram_disk_attrs[] = {
1468         &dev_attr_disksize.attr,
1469         &dev_attr_initstate.attr,
1470         &dev_attr_reset.attr,
1471         &dev_attr_compact.attr,
1472         &dev_attr_mem_limit.attr,
1473         &dev_attr_mem_used_max.attr,
1474         &dev_attr_max_comp_streams.attr,
1475         &dev_attr_comp_algorithm.attr,
1476 #ifdef CONFIG_ZRAM_WRITEBACK
1477         &dev_attr_backing_dev.attr,
1478 #endif
1479         &dev_attr_io_stat.attr,
1480         &dev_attr_mm_stat.attr,
1481         &dev_attr_debug_stat.attr,
1482         NULL,
1483 };
1484
1485 static const struct attribute_group zram_disk_attr_group = {
1486         .attrs = zram_disk_attrs,
1487 };
1488
1489 /*
1490  * Allocate and initialize new zram device. the function returns
1491  * '>= 0' device_id upon success, and negative value otherwise.
1492  */
1493 static int zram_add(void)
1494 {
1495         struct zram *zram;
1496         struct request_queue *queue;
1497         int ret, device_id;
1498
1499         zram = kzalloc(sizeof(struct zram), GFP_KERNEL);
1500         if (!zram)
1501                 return -ENOMEM;
1502
1503         ret = idr_alloc(&zram_index_idr, zram, 0, 0, GFP_KERNEL);
1504         if (ret < 0)
1505                 goto out_free_dev;
1506         device_id = ret;
1507
1508         init_rwsem(&zram->init_lock);
1509
1510         queue = blk_alloc_queue(GFP_KERNEL);
1511         if (!queue) {
1512                 pr_err("Error allocating disk queue for device %d\n",
1513                         device_id);
1514                 ret = -ENOMEM;
1515                 goto out_free_idr;
1516         }
1517
1518         blk_queue_make_request(queue, zram_make_request);
1519
1520         /* gendisk structure */
1521         zram->disk = alloc_disk(1);
1522         if (!zram->disk) {
1523                 pr_err("Error allocating disk structure for device %d\n",
1524                         device_id);
1525                 ret = -ENOMEM;
1526                 goto out_free_queue;
1527         }
1528
1529         zram->disk->major = zram_major;
1530         zram->disk->first_minor = device_id;
1531         zram->disk->fops = &zram_devops;
1532         zram->disk->queue = queue;
1533         zram->disk->queue->queuedata = zram;
1534         zram->disk->private_data = zram;
1535         snprintf(zram->disk->disk_name, 16, "zram%d", device_id);
1536
1537         /* Actual capacity set using syfs (/sys/block/zram<id>/disksize */
1538         set_capacity(zram->disk, 0);
1539         /* zram devices sort of resembles non-rotational disks */
1540         queue_flag_set_unlocked(QUEUE_FLAG_NONROT, zram->disk->queue);
1541         queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, zram->disk->queue);
1542         /*
1543          * To ensure that we always get PAGE_SIZE aligned
1544          * and n*PAGE_SIZED sized I/O requests.
1545          */
1546         blk_queue_physical_block_size(zram->disk->queue, PAGE_SIZE);
1547         blk_queue_logical_block_size(zram->disk->queue,
1548                                         ZRAM_LOGICAL_BLOCK_SIZE);
1549         blk_queue_io_min(zram->disk->queue, PAGE_SIZE);
1550         blk_queue_io_opt(zram->disk->queue, PAGE_SIZE);
1551         zram->disk->queue->limits.discard_granularity = PAGE_SIZE;
1552         blk_queue_max_discard_sectors(zram->disk->queue, UINT_MAX);
1553         queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, zram->disk->queue);
1554
1555         /*
1556          * zram_bio_discard() will clear all logical blocks if logical block
1557          * size is identical with physical block size(PAGE_SIZE). But if it is
1558          * different, we will skip discarding some parts of logical blocks in
1559          * the part of the request range which isn't aligned to physical block
1560          * size.  So we can't ensure that all discarded logical blocks are
1561          * zeroed.
1562          */
1563         if (ZRAM_LOGICAL_BLOCK_SIZE == PAGE_SIZE)
1564                 blk_queue_max_write_zeroes_sectors(zram->disk->queue, UINT_MAX);
1565
1566         add_disk(zram->disk);
1567
1568         ret = sysfs_create_group(&disk_to_dev(zram->disk)->kobj,
1569                                 &zram_disk_attr_group);
1570         if (ret < 0) {
1571                 pr_err("Error creating sysfs group for device %d\n",
1572                                 device_id);
1573                 goto out_free_disk;
1574         }
1575         strlcpy(zram->compressor, default_compressor, sizeof(zram->compressor));
1576
1577         pr_info("Added device: %s\n", zram->disk->disk_name);
1578         return device_id;
1579
1580 out_free_disk:
1581         del_gendisk(zram->disk);
1582         put_disk(zram->disk);
1583 out_free_queue:
1584         blk_cleanup_queue(queue);
1585 out_free_idr:
1586         idr_remove(&zram_index_idr, device_id);
1587 out_free_dev:
1588         kfree(zram);
1589         return ret;
1590 }
1591
1592 static int zram_remove(struct zram *zram)
1593 {
1594         struct block_device *bdev;
1595
1596         bdev = bdget_disk(zram->disk, 0);
1597         if (!bdev)
1598                 return -ENOMEM;
1599
1600         mutex_lock(&bdev->bd_mutex);
1601         if (bdev->bd_openers || zram->claim) {
1602                 mutex_unlock(&bdev->bd_mutex);
1603                 bdput(bdev);
1604                 return -EBUSY;
1605         }
1606
1607         zram->claim = true;
1608         mutex_unlock(&bdev->bd_mutex);
1609
1610         /*
1611          * Remove sysfs first, so no one will perform a disksize
1612          * store while we destroy the devices. This also helps during
1613          * hot_remove -- zram_reset_device() is the last holder of
1614          * ->init_lock, no later/concurrent disksize_store() or any
1615          * other sysfs handlers are possible.
1616          */
1617         sysfs_remove_group(&disk_to_dev(zram->disk)->kobj,
1618                         &zram_disk_attr_group);
1619
1620         /* Make sure all the pending I/O are finished */
1621         fsync_bdev(bdev);
1622         zram_reset_device(zram);
1623         bdput(bdev);
1624
1625         pr_info("Removed device: %s\n", zram->disk->disk_name);
1626
1627         blk_cleanup_queue(zram->disk->queue);
1628         del_gendisk(zram->disk);
1629         put_disk(zram->disk);
1630         kfree(zram);
1631         return 0;
1632 }
1633
1634 /* zram-control sysfs attributes */
1635
1636 /*
1637  * NOTE: hot_add attribute is not the usual read-only sysfs attribute. In a
1638  * sense that reading from this file does alter the state of your system -- it
1639  * creates a new un-initialized zram device and returns back this device's
1640  * device_id (or an error code if it fails to create a new device).
1641  */
1642 static ssize_t hot_add_show(struct class *class,
1643                         struct class_attribute *attr,
1644                         char *buf)
1645 {
1646         int ret;
1647
1648         mutex_lock(&zram_index_mutex);
1649         ret = zram_add();
1650         mutex_unlock(&zram_index_mutex);
1651
1652         if (ret < 0)
1653                 return ret;
1654         return scnprintf(buf, PAGE_SIZE, "%d\n", ret);
1655 }
1656 static CLASS_ATTR_RO(hot_add);
1657
1658 static ssize_t hot_remove_store(struct class *class,
1659                         struct class_attribute *attr,
1660                         const char *buf,
1661                         size_t count)
1662 {
1663         struct zram *zram;
1664         int ret, dev_id;
1665
1666         /* dev_id is gendisk->first_minor, which is `int' */
1667         ret = kstrtoint(buf, 10, &dev_id);
1668         if (ret)
1669                 return ret;
1670         if (dev_id < 0)
1671                 return -EINVAL;
1672
1673         mutex_lock(&zram_index_mutex);
1674
1675         zram = idr_find(&zram_index_idr, dev_id);
1676         if (zram) {
1677                 ret = zram_remove(zram);
1678                 if (!ret)
1679                         idr_remove(&zram_index_idr, dev_id);
1680         } else {
1681                 ret = -ENODEV;
1682         }
1683
1684         mutex_unlock(&zram_index_mutex);
1685         return ret ? ret : count;
1686 }
1687 static CLASS_ATTR_WO(hot_remove);
1688
1689 static struct attribute *zram_control_class_attrs[] = {
1690         &class_attr_hot_add.attr,
1691         &class_attr_hot_remove.attr,
1692         NULL,
1693 };
1694 ATTRIBUTE_GROUPS(zram_control_class);
1695
1696 static struct class zram_control_class = {
1697         .name           = "zram-control",
1698         .owner          = THIS_MODULE,
1699         .class_groups   = zram_control_class_groups,
1700 };
1701
1702 static int zram_remove_cb(int id, void *ptr, void *data)
1703 {
1704         zram_remove(ptr);
1705         return 0;
1706 }
1707
1708 static void destroy_devices(void)
1709 {
1710         class_unregister(&zram_control_class);
1711         idr_for_each(&zram_index_idr, &zram_remove_cb, NULL);
1712         idr_destroy(&zram_index_idr);
1713         unregister_blkdev(zram_major, "zram");
1714         cpuhp_remove_multi_state(CPUHP_ZCOMP_PREPARE);
1715 }
1716
1717 static int __init zram_init(void)
1718 {
1719         int ret;
1720
1721         ret = cpuhp_setup_state_multi(CPUHP_ZCOMP_PREPARE, "block/zram:prepare",
1722                                       zcomp_cpu_up_prepare, zcomp_cpu_dead);
1723         if (ret < 0)
1724                 return ret;
1725
1726         ret = class_register(&zram_control_class);
1727         if (ret) {
1728                 pr_err("Unable to register zram-control class\n");
1729                 cpuhp_remove_multi_state(CPUHP_ZCOMP_PREPARE);
1730                 return ret;
1731         }
1732
1733         zram_major = register_blkdev(0, "zram");
1734         if (zram_major <= 0) {
1735                 pr_err("Unable to get major number\n");
1736                 class_unregister(&zram_control_class);
1737                 cpuhp_remove_multi_state(CPUHP_ZCOMP_PREPARE);
1738                 return -EBUSY;
1739         }
1740
1741         while (num_devices != 0) {
1742                 mutex_lock(&zram_index_mutex);
1743                 ret = zram_add();
1744                 mutex_unlock(&zram_index_mutex);
1745                 if (ret < 0)
1746                         goto out_error;
1747                 num_devices--;
1748         }
1749
1750         return 0;
1751
1752 out_error:
1753         destroy_devices();
1754         return ret;
1755 }
1756
1757 static void __exit zram_exit(void)
1758 {
1759         destroy_devices();
1760 }
1761
1762 module_init(zram_init);
1763 module_exit(zram_exit);
1764
1765 module_param(num_devices, uint, 0);
1766 MODULE_PARM_DESC(num_devices, "Number of pre-created zram devices");
1767
1768 MODULE_LICENSE("Dual BSD/GPL");
1769 MODULE_AUTHOR("Nitin Gupta <ngupta@vflare.org>");
1770 MODULE_DESCRIPTION("Compressed RAM Block Device");