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 file *file;
 295         struct zram *zram = dev_to_zram(dev);
 296         char *p;
 297         ssize_t ret;
 298
 299         down_read(&zram->init_lock);
 300         file = zram->backing_dev;
 301         if (!file) {
 302                 memcpy(buf, "none\n", 5);
 303                 up_read(&zram->init_lock);
 304                 return 5;
 305         }
 306
 307         p = file_path(file, buf, PAGE_SIZE - 1);
 308         if (IS_ERR(p)) {
 309                 ret = PTR_ERR(p);
 310                 goto out;
 311         }
 312
 313         ret = strlen(p);
 314         memmove(buf, p, ret);
 315         buf[ret++] = '\n';
 316 out:
 317         up_read(&zram->init_lock);
 318         return ret;
 319 }
 320
 321 static ssize_t backing_dev_store(struct device *dev,
 322                 struct device_attribute *attr, const char *buf, size_t len)
 323 {
 324         char *file_name;
 325         size_t sz;
 326         struct file *backing_dev = NULL;
 327         struct inode *inode;
 328         struct address_space *mapping;
 329         unsigned int bitmap_sz, old_block_size = 0;
 330         unsigned long nr_pages, *bitmap = NULL;
 331         struct block_device *bdev = NULL;
 332         int err;
 333         struct zram *zram = dev_to_zram(dev);
 334
 335         file_name = kmalloc(PATH_MAX, GFP_KERNEL);
 336         if (!file_name)
 337                 return -ENOMEM;
 338
 339         down_write(&zram->init_lock);
 340         if (init_done(zram)) {
 341                 pr_info("Can't setup backing device for initialized device\n");
 342                 err = -EBUSY;
 343                 goto out;
 344         }
 345
 346         strlcpy(file_name, buf, PATH_MAX);
 347         /* ignore trailing newline */
 348         sz = strlen(file_name);
 349         if (sz > 0 && file_name[sz - 1] == '\n')
 350                 file_name[sz - 1] = 0x00;
 351
 352         backing_dev = filp_open(file_name, O_RDWR|O_LARGEFILE, 0);
 353         if (IS_ERR(backing_dev)) {
 354                 err = PTR_ERR(backing_dev);
 355                 backing_dev = NULL;
 356                 goto out;
 357         }
 358
 359         mapping = backing_dev->f_mapping;
 360         inode = mapping->host;
 361
 362         /* Support only block device in this moment */
 363         if (!S_ISBLK(inode->i_mode)) {
 364                 err = -ENOTBLK;
 365                 goto out;
 366         }
 367
 368         bdev = bdgrab(I_BDEV(inode));
 369         err = blkdev_get(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL, zram);
 370         if (err < 0) {
 371                 bdev = NULL;
 372                 goto out;
 373         }
 374
 375         nr_pages = i_size_read(inode) >> PAGE_SHIFT;
 376         bitmap_sz = BITS_TO_LONGS(nr_pages) * sizeof(long);
 377         bitmap = kvzalloc(bitmap_sz, GFP_KERNEL);
 378         if (!bitmap) {
 379                 err = -ENOMEM;
 380                 goto out;
 381         }
 382
 383         old_block_size = block_size(bdev);
 384         err = set_blocksize(bdev, PAGE_SIZE);
 385         if (err)
 386                 goto out;
 387
 388         reset_bdev(zram);
 389         spin_lock_init(&zram->bitmap_lock);
 390
 391         zram->old_block_size = old_block_size;
 392         zram->bdev = bdev;
 393         zram->backing_dev = backing_dev;
 394         zram->bitmap = bitmap;
 395         zram->nr_pages = nr_pages;
 396         up_write(&zram->init_lock);
 397
 398         pr_info("setup backing device %s\n", file_name);
 399         kfree(file_name);
 400
 401         return len;
 402 out:
 403         if (bitmap)
 404                 kvfree(bitmap);
 405
 406         if (bdev)
 407                 blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
 408
 409         if (backing_dev)
 410                 filp_close(backing_dev, NULL);
 411
 412         up_write(&zram->init_lock);
 413
 414         kfree(file_name);
 415
 416         return err;
 417 }
 418
 419 static unsigned long get_entry_bdev(struct zram *zram)
 420 {
 421         unsigned long entry;
 422
 423         spin_lock(&zram->bitmap_lock);
 424         /* skip 0 bit to confuse zram.handle = 0 */
 425         entry = find_next_zero_bit(zram->bitmap, zram->nr_pages, 1);
 426         if (entry == zram->nr_pages) {
 427                 spin_unlock(&zram->bitmap_lock);
 428                 return 0;
 429         }
 430
 431         set_bit(entry, zram->bitmap);
 432         spin_unlock(&zram->bitmap_lock);
 433
 434         return entry;
 435 }
 436
 437 static void put_entry_bdev(struct zram *zram, unsigned long entry)
 438 {
 439         int was_set;
 440
 441         spin_lock(&zram->bitmap_lock);
 442         was_set = test_and_clear_bit(entry, zram->bitmap);
 443         spin_unlock(&zram->bitmap_lock);
 444         WARN_ON_ONCE(!was_set);
 445 }
 446
 447 void zram_page_end_io(struct bio *bio)
 448 {
 449         struct page *page = bio->bi_io_vec[0].bv_page;
 450
 451         page_endio(page, op_is_write(bio_op(bio)),
 452                         blk_status_to_errno(bio->bi_status));
 453         bio_put(bio);
 454 }
 455
 456 /*
 457  * Returns 1 if the submission is successful.
 458  */
 459 static int read_from_bdev_async(struct zram *zram, struct bio_vec *bvec,
 460                         unsigned long entry, struct bio *parent)
 461 {
 462         struct bio *bio;
 463
 464         bio = bio_alloc(GFP_ATOMIC, 1);
 465         if (!bio)
 466                 return -ENOMEM;
 467
 468         bio->bi_iter.bi_sector = entry * (PAGE_SIZE >> 9);
 469         bio_set_dev(bio, zram->bdev);
 470         if (!bio_add_page(bio, bvec->bv_page, bvec->bv_len, bvec->bv_offset)) {
 471                 bio_put(bio);
 472                 return -EIO;
 473         }
 474
 475         if (!parent) {
 476                 bio->bi_opf = REQ_OP_READ;
 477                 bio->bi_end_io = zram_page_end_io;
 478         } else {
 479                 bio->bi_opf = parent->bi_opf;
 480                 bio_chain(bio, parent);
 481         }
 482
 483         submit_bio(bio);
 484         return 1;
 485 }
 486
 487 struct zram_work {
 488         struct work_struct work;
 489         struct zram *zram;
 490         unsigned long entry;
 491         struct bio *bio;
 492         struct bio_vec bvec;
 493 };
 494
 495 #if PAGE_SIZE != 4096
 496 static void zram_sync_read(struct work_struct *work)
 497 {
 498         struct zram_work *zw = container_of(work, struct zram_work, work);
 499         struct zram *zram = zw->zram;
 500         unsigned long entry = zw->entry;
 501         struct bio *bio = zw->bio;
 502
 503         read_from_bdev_async(zram, &zw->bvec, entry, bio);
 504 }
 505
 506 /*
 507  * Block layer want one ->make_request_fn to be active at a time
 508  * so if we use chained IO with parent IO in same context,
 509  * it's a deadlock. To avoid, it, it uses worker thread context.
 510  */
 511 static int read_from_bdev_sync(struct zram *zram, struct bio_vec *bvec,
 512                                 unsigned long entry, struct bio *bio)
 513 {
 514         struct zram_work work;
 515
 516         work.bvec = *bvec;
 517         work.zram = zram;
 518         work.entry = entry;
 519         work.bio = bio;
 520
 521         INIT_WORK_ONSTACK(&work.work, zram_sync_read);
 522         queue_work(system_unbound_wq, &work.work);
 523         flush_work(&work.work);
 524         destroy_work_on_stack(&work.work);
 525
 526         return 1;
 527 }
 528 #else
 529 static int read_from_bdev_sync(struct zram *zram, struct bio_vec *bvec,
 530                                 unsigned long entry, struct bio *bio)
 531 {
 532         WARN_ON(1);
 533         return -EIO;
 534 }
 535 #endif
 536
 537 static int read_from_bdev(struct zram *zram, struct bio_vec *bvec,
 538                         unsigned long entry, struct bio *parent, bool sync)
 539 {
 540         if (sync)
 541                 return read_from_bdev_sync(zram, bvec, entry, parent);
 542         else
 543                 return read_from_bdev_async(zram, bvec, entry, parent);
 544 }
 545
 546 static int write_to_bdev(struct zram *zram, struct bio_vec *bvec,
 547                                         u32 index, struct bio *parent,
 548                                         unsigned long *pentry)
 549 {
 550         struct bio *bio;
 551         unsigned long entry;
 552
 553         bio = bio_alloc(GFP_ATOMIC, 1);
 554         if (!bio)
 555                 return -ENOMEM;
 556
 557         entry = get_entry_bdev(zram);
 558         if (!entry) {
 559                 bio_put(bio);
 560                 return -ENOSPC;
 561         }
 562
 563         bio->bi_iter.bi_sector = entry * (PAGE_SIZE >> 9);
 564         bio_set_dev(bio, zram->bdev);
 565         if (!bio_add_page(bio, bvec->bv_page, bvec->bv_len,
 566                                         bvec->bv_offset)) {
 567                 bio_put(bio);
 568                 put_entry_bdev(zram, entry);
 569                 return -EIO;
 570         }
 571
 572         if (!parent) {
 573                 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC;
 574                 bio->bi_end_io = zram_page_end_io;
 575         } else {
 576                 bio->bi_opf = parent->bi_opf;
 577                 bio_chain(bio, parent);
 578         }
 579
 580         submit_bio(bio);
 581         *pentry = entry;
 582
 583         return 0;
 584 }
 585
 586 static void zram_wb_clear(struct zram *zram, u32 index)
 587 {
 588         unsigned long entry;
 589
 590         zram_clear_flag(zram, index, ZRAM_WB);
 591         entry = zram_get_element(zram, index);
 592         zram_set_element(zram, index, 0);
 593         put_entry_bdev(zram, entry);
 594 }
 595
 596 #else
 597 static bool zram_wb_enabled(struct zram *zram) { return false; }
 598 static inline void reset_bdev(struct zram *zram) {};
 599 static int write_to_bdev(struct zram *zram, struct bio_vec *bvec,
 600                                         u32 index, struct bio *parent,
 601                                         unsigned long *pentry)
 602
 603 {
 604         return -EIO;
 605 }
 606
 607 static int read_from_bdev(struct zram *zram, struct bio_vec *bvec,
 608                         unsigned long entry, struct bio *parent, bool sync)
 609 {
 610         return -EIO;
 611 }
 612 static void zram_wb_clear(struct zram *zram, u32 index) {}
 613 #endif
 614
 615
 616 /*
 617  * We switched to per-cpu streams and this attr is not needed anymore.
 618  * However, we will keep it around for some time, because:
 619  * a) we may revert per-cpu streams in the future
 620  * b) it's visible to user space and we need to follow our 2 years
 621  *    retirement rule; but we already have a number of 'soon to be
 622  *    altered' attrs, so max_comp_streams need to wait for the next
 623  *    layoff cycle.
 624  */
 625 static ssize_t max_comp_streams_show(struct device *dev,
 626                 struct device_attribute *attr, char *buf)
 627 {
 628         return scnprintf(buf, PAGE_SIZE, "%d\n", num_online_cpus());
 629 }
 630
 631 static ssize_t max_comp_streams_store(struct device *dev,
 632                 struct device_attribute *attr, const char *buf, size_t len)
 633 {
 634         return len;
 635 }
 636
 637 static ssize_t comp_algorithm_show(struct device *dev,
 638                 struct device_attribute *attr, char *buf)
 639 {
 640         size_t sz;
 641         struct zram *zram = dev_to_zram(dev);
 642
 643         down_read(&zram->init_lock);
 644         sz = zcomp_available_show(zram->compressor, buf);
 645         up_read(&zram->init_lock);
 646
 647         return sz;
 648 }
 649
 650 static ssize_t comp_algorithm_store(struct device *dev,
 651                 struct device_attribute *attr, const char *buf, size_t len)
 652 {
 653         struct zram *zram = dev_to_zram(dev);
 654         char compressor[ARRAY_SIZE(zram->compressor)];
 655         size_t sz;
 656
 657         strlcpy(compressor, buf, sizeof(compressor));
 658         /* ignore trailing newline */
 659         sz = strlen(compressor);
 660         if (sz > 0 && compressor[sz - 1] == '\n')
 661                 compressor[sz - 1] = 0x00;
 662
 663         if (!zcomp_available_algorithm(compressor))
 664                 return -EINVAL;
 665
 666         down_write(&zram->init_lock);
 667         if (init_done(zram)) {
 668                 up_write(&zram->init_lock);
 669                 pr_info("Can't change algorithm for initialized device\n");
 670                 return -EBUSY;
 671         }
 672
 673         strcpy(zram->compressor, compressor);
 674         up_write(&zram->init_lock);
 675         return len;
 676 }
 677
 678 static ssize_t compact_store(struct device *dev,
 679                 struct device_attribute *attr, const char *buf, size_t len)
 680 {
 681         struct zram *zram = dev_to_zram(dev);
 682
 683         down_read(&zram->init_lock);
 684         if (!init_done(zram)) {
 685                 up_read(&zram->init_lock);
 686                 return -EINVAL;
 687         }
 688
 689         zs_compact(zram->mem_pool);
 690         up_read(&zram->init_lock);
 691
 692         return len;
 693 }
 694
 695 static ssize_t io_stat_show(struct device *dev,
 696                 struct device_attribute *attr, char *buf)
 697 {
 698         struct zram *zram = dev_to_zram(dev);
 699         ssize_t ret;
 700
 701         down_read(&zram->init_lock);
 702         ret = scnprintf(buf, PAGE_SIZE,
 703                         "%8llu %8llu %8llu %8llu\n",
 704                         (u64)atomic64_read(&zram->stats.failed_reads),
 705                         (u64)atomic64_read(&zram->stats.failed_writes),
 706                         (u64)atomic64_read(&zram->stats.invalid_io),
 707                         (u64)atomic64_read(&zram->stats.notify_free));
 708         up_read(&zram->init_lock);
 709
 710         return ret;
 711 }
 712
 713 static ssize_t mm_stat_show(struct device *dev,
 714                 struct device_attribute *attr, char *buf)
 715 {
 716         struct zram *zram = dev_to_zram(dev);
 717         struct zs_pool_stats pool_stats;
 718         u64 orig_size, mem_used = 0;
 719         long max_used;
 720         ssize_t ret;
 721
 722         memset(&pool_stats, 0x00, sizeof(struct zs_pool_stats));
 723
 724         down_read(&zram->init_lock);
 725         if (init_done(zram)) {
 726                 mem_used = zs_get_total_pages(zram->mem_pool);
 727                 zs_pool_stats(zram->mem_pool, &pool_stats);
 728         }
 729
 730         orig_size = atomic64_read(&zram->stats.pages_stored);
 731         max_used = atomic_long_read(&zram->stats.max_used_pages);
 732
 733         ret = scnprintf(buf, PAGE_SIZE,
 734                         "%8llu %8llu %8llu %8lu %8ld %8llu %8lu\n",
 735                         orig_size << PAGE_SHIFT,
 736                         (u64)atomic64_read(&zram->stats.compr_data_size),
 737                         mem_used << PAGE_SHIFT,
 738                         zram->limit_pages << PAGE_SHIFT,
 739                         max_used << PAGE_SHIFT,
 740                         (u64)atomic64_read(&zram->stats.same_pages),
 741                         atomic_long_read(&pool_stats.pages_compacted));
 742         up_read(&zram->init_lock);
 743
 744         return ret;
 745 }
 746
 747 static ssize_t debug_stat_show(struct device *dev,
 748                 struct device_attribute *attr, char *buf)
 749 {
 750         int version = 1;
 751         struct zram *zram = dev_to_zram(dev);
 752         ssize_t ret;
 753
 754         down_read(&zram->init_lock);
 755         ret = scnprintf(buf, PAGE_SIZE,
 756                         "version: %d\n%8llu\n",
 757                         version,
 758                         (u64)atomic64_read(&zram->stats.writestall));
 759         up_read(&zram->init_lock);
 760
 761         return ret;
 762 }
 763
 764 static DEVICE_ATTR_RO(io_stat);
 765 static DEVICE_ATTR_RO(mm_stat);
 766 static DEVICE_ATTR_RO(debug_stat);
 767
 768 static void zram_slot_lock(struct zram *zram, u32 index)
 769 {
 770         bit_spin_lock(ZRAM_ACCESS, &zram->table[index].value);
 771 }
 772
 773 static void zram_slot_unlock(struct zram *zram, u32 index)
 774 {
 775         bit_spin_unlock(ZRAM_ACCESS, &zram->table[index].value);
 776 }
 777
 778 static void zram_meta_free(struct zram *zram, u64 disksize)
 779 {
 780         size_t num_pages = disksize >> PAGE_SHIFT;
 781         size_t index;
 782
 783         /* Free all pages that are still in this zram device */
 784         for (index = 0; index < num_pages; index++)
 785                 zram_free_page(zram, index);
 786
 787         zs_destroy_pool(zram->mem_pool);
 788         vfree(zram->table);
 789 }
 790
 791 static bool zram_meta_alloc(struct zram *zram, u64 disksize)
 792 {
 793         size_t num_pages;
 794
 795         num_pages = disksize >> PAGE_SHIFT;
 796         zram->table = vzalloc(num_pages * sizeof(*zram->table));
 797         if (!zram->table)
 798                 return false;
 799
 800         zram->mem_pool = zs_create_pool(zram->disk->disk_name);
 801         if (!zram->mem_pool) {
 802                 vfree(zram->table);
 803                 return false;
 804         }
 805
 806         return true;
 807 }
 808
 809 /*
 810  * To protect concurrent access to the same index entry,
 811  * caller should hold this table index entry's bit_spinlock to
 812  * indicate this index entry is accessing.
 813  */
 814 static void zram_free_page(struct zram *zram, size_t index)
 815 {
 816         unsigned long handle;
 817
 818         if (zram_wb_enabled(zram) && zram_test_flag(zram, index, ZRAM_WB)) {
 819                 zram_wb_clear(zram, index);
 820                 atomic64_dec(&zram->stats.pages_stored);
 821                 return;
 822         }
 823
 824         /*
 825          * No memory is allocated for same element filled pages.
 826          * Simply clear same page flag.
 827          */
 828         if (zram_test_flag(zram, index, ZRAM_SAME)) {
 829                 zram_clear_flag(zram, index, ZRAM_SAME);
 830                 zram_set_element(zram, index, 0);
 831                 atomic64_dec(&zram->stats.same_pages);
 832                 atomic64_dec(&zram->stats.pages_stored);
 833                 return;
 834         }
 835
 836         handle = zram_get_handle(zram, index);
 837         if (!handle)
 838                 return;
 839
 840         zs_free(zram->mem_pool, handle);
 841
 842         atomic64_sub(zram_get_obj_size(zram, index),
 843                         &zram->stats.compr_data_size);
 844         atomic64_dec(&zram->stats.pages_stored);
 845
 846         zram_set_handle(zram, index, 0);
 847         zram_set_obj_size(zram, index, 0);
 848 }
 849
 850 static int __zram_bvec_read(struct zram *zram, struct page *page, u32 index,
 851                                 struct bio *bio, bool partial_io)
 852 {
 853         int ret;
 854         unsigned long handle;
 855         unsigned int size;
 856         void *src, *dst;
 857
 858         if (zram_wb_enabled(zram)) {
 859                 zram_slot_lock(zram, index);
 860                 if (zram_test_flag(zram, index, ZRAM_WB)) {
 861                         struct bio_vec bvec;
 862
 863                         zram_slot_unlock(zram, index);
 864
 865                         bvec.bv_page = page;
 866                         bvec.bv_len = PAGE_SIZE;
 867                         bvec.bv_offset = 0;
 868                         return read_from_bdev(zram, &bvec,
 869                                         zram_get_element(zram, index),
 870                                         bio, partial_io);
 871                 }
 872                 zram_slot_unlock(zram, index);
 873         }
 874
 875         zram_slot_lock(zram, index);
 876         handle = zram_get_handle(zram, index);
 877         if (!handle || zram_test_flag(zram, index, ZRAM_SAME)) {
 878                 unsigned long value;
 879                 void *mem;
 880
 881                 value = handle ? zram_get_element(zram, index) : 0;
 882                 mem = kmap_atomic(page);
 883                 zram_fill_page(mem, PAGE_SIZE, value);
 884                 kunmap_atomic(mem);
 885                 zram_slot_unlock(zram, index);
 886                 return 0;
 887         }
 888
 889         size = zram_get_obj_size(zram, index);
 890
 891         src = zs_map_object(zram->mem_pool, handle, ZS_MM_RO);
 892         if (size == PAGE_SIZE) {
 893                 dst = kmap_atomic(page);
 894                 memcpy(dst, src, PAGE_SIZE);
 895                 kunmap_atomic(dst);
 896                 ret = 0;
 897         } else {
 898                 struct zcomp_strm *zstrm = zcomp_stream_get(zram->comp);
 899
 900                 dst = kmap_atomic(page);
 901                 ret = zcomp_decompress(zstrm, src, size, dst);
 902                 kunmap_atomic(dst);
 903                 zcomp_stream_put(zram->comp);
 904         }
 905         zs_unmap_object(zram->mem_pool, handle);
 906         zram_slot_unlock(zram, index);
 907
 908         /* Should NEVER happen. Return bio error if it does. */
 909         if (unlikely(ret))
 910                 pr_err("Decompression failed! err=%d, page=%u\n", ret, index);
 911
 912         return ret;
 913 }
 914
 915 static int zram_bvec_read(struct zram *zram, struct bio_vec *bvec,
 916                                 u32 index, int offset, struct bio *bio)
 917 {
 918         int ret;
 919         struct page *page;
 920
 921         page = bvec->bv_page;
 922         if (is_partial_io(bvec)) {
 923                 /* Use a temporary buffer to decompress the page */
 924                 page = alloc_page(GFP_NOIO|__GFP_HIGHMEM);
 925                 if (!page)
 926                         return -ENOMEM;
 927         }
 928
 929         ret = __zram_bvec_read(zram, page, index, bio, is_partial_io(bvec));
 930         if (unlikely(ret))
 931                 goto out;
 932
 933         if (is_partial_io(bvec)) {
 934                 void *dst = kmap_atomic(bvec->bv_page);
 935                 void *src = kmap_atomic(page);
 936
 937                 memcpy(dst + bvec->bv_offset, src + offset, bvec->bv_len);
 938                 kunmap_atomic(src);
 939                 kunmap_atomic(dst);
 940         }
 941 out:
 942         if (is_partial_io(bvec))
 943                 __free_page(page);
 944
 945         return ret;
 946 }
 947
 948 static int __zram_bvec_write(struct zram *zram, struct bio_vec *bvec,
 949                                 u32 index, struct bio *bio)
 950 {
 951         int ret = 0;
 952         unsigned long alloced_pages;
 953         unsigned long handle = 0;
 954         unsigned int comp_len = 0;
 955         void *src, *dst, *mem;
 956         struct zcomp_strm *zstrm;
 957         struct page *page = bvec->bv_page;
 958         unsigned long element = 0;
 959         enum zram_pageflags flags = 0;
 960         bool allow_wb = true;
 961
 962         mem = kmap_atomic(page);
 963         if (page_same_filled(mem, &element)) {
 964                 kunmap_atomic(mem);
 965                 /* Free memory associated with this sector now. */
 966                 flags = ZRAM_SAME;
 967                 atomic64_inc(&zram->stats.same_pages);
 968                 goto out;
 969         }
 970         kunmap_atomic(mem);
 971
 972 compress_again:
 973         zstrm = zcomp_stream_get(zram->comp);
 974         src = kmap_atomic(page);
 975         ret = zcomp_compress(zstrm, src, &comp_len);
 976         kunmap_atomic(src);
 977
 978         if (unlikely(ret)) {
 979                 zcomp_stream_put(zram->comp);
 980                 pr_err("Compression failed! err=%d\n", ret);
 981                 zs_free(zram->mem_pool, handle);
 982                 return ret;
 983         }
 984
 985         if (unlikely(comp_len > max_zpage_size)) {
 986                 if (zram_wb_enabled(zram) && allow_wb) {
 987                         zcomp_stream_put(zram->comp);
 988                         ret = write_to_bdev(zram, bvec, index, bio, &element);
 989                         if (!ret) {
 990                                 flags = ZRAM_WB;
 991                                 ret = 1;
 992                                 goto out;
 993                         }
 994                         allow_wb = false;
 995                         goto compress_again;
 996                 }
 997                 comp_len = PAGE_SIZE;
 998         }
 999
1000         /*
1001          * handle allocation has 2 paths:
1002          * a) fast path is executed with preemption disabled (for
1003          *  per-cpu streams) and has __GFP_DIRECT_RECLAIM bit clear,
1004          *  since we can't sleep;
1005          * b) slow path enables preemption and attempts to allocate
1006          *  the page with __GFP_DIRECT_RECLAIM bit set. we have to
1007          *  put per-cpu compression stream and, thus, to re-do
1008          *  the compression once handle is allocated.
1009          *
1010          * if we have a 'non-null' handle here then we are coming
1011          * from the slow path and handle has already been allocated.
1012          */
1013         if (!handle)
1014                 handle = zs_malloc(zram->mem_pool, comp_len,
1015                                 __GFP_KSWAPD_RECLAIM |
1016                                 __GFP_NOWARN |
1017                                 __GFP_HIGHMEM |
1018                                 __GFP_MOVABLE);
1019         if (!handle) {
1020                 zcomp_stream_put(zram->comp);
1021                 atomic64_inc(&zram->stats.writestall);
1022                 handle = zs_malloc(zram->mem_pool, comp_len,
1023                                 GFP_NOIO | __GFP_HIGHMEM |
1024                                 __GFP_MOVABLE);
1025                 if (handle)
1026                         goto compress_again;
1027                 return -ENOMEM;
1028         }
1029
1030         alloced_pages = zs_get_total_pages(zram->mem_pool);
1031         update_used_max(zram, alloced_pages);
1032
1033         if (zram->limit_pages && alloced_pages > zram->limit_pages) {
1034                 zcomp_stream_put(zram->comp);
1035                 zs_free(zram->mem_pool, handle);
1036                 return -ENOMEM;
1037         }
1038
1039         dst = zs_map_object(zram->mem_pool, handle, ZS_MM_WO);
1040
1041         src = zstrm->buffer;
1042         if (comp_len == PAGE_SIZE)
1043                 src = kmap_atomic(page);
1044         memcpy(dst, src, comp_len);
1045         if (comp_len == PAGE_SIZE)
1046                 kunmap_atomic(src);
1047
1048         zcomp_stream_put(zram->comp);
1049         zs_unmap_object(zram->mem_pool, handle);
1050         atomic64_add(comp_len, &zram->stats.compr_data_size);
1051 out:
1052         /*
1053          * Free memory associated with this sector
1054          * before overwriting unused sectors.
1055          */
1056         zram_slot_lock(zram, index);
1057         zram_free_page(zram, index);
1058
1059         if (flags) {
1060                 zram_set_flag(zram, index, flags);
1061                 zram_set_element(zram, index, element);
1062         }  else {
1063                 zram_set_handle(zram, index, handle);
1064                 zram_set_obj_size(zram, index, comp_len);
1065         }
1066         zram_slot_unlock(zram, index);
1067
1068         /* Update stats */
1069         atomic64_inc(&zram->stats.pages_stored);
1070         return ret;
1071 }
1072
1073 static int zram_bvec_write(struct zram *zram, struct bio_vec *bvec,
1074                                 u32 index, int offset, struct bio *bio)
1075 {
1076         int ret;
1077         struct page *page = NULL;
1078         void *src;
1079         struct bio_vec vec;
1080
1081         vec = *bvec;
1082         if (is_partial_io(bvec)) {
1083                 void *dst;
1084                 /*
1085                  * This is a partial IO. We need to read the full page
1086                  * before to write the changes.
1087                  */
1088                 page = alloc_page(GFP_NOIO|__GFP_HIGHMEM);
1089                 if (!page)
1090                         return -ENOMEM;
1091
1092                 ret = __zram_bvec_read(zram, page, index, bio, true);
1093                 if (ret)
1094                         goto out;
1095
1096                 src = kmap_atomic(bvec->bv_page);
1097                 dst = kmap_atomic(page);
1098                 memcpy(dst + offset, src + bvec->bv_offset, bvec->bv_len);
1099                 kunmap_atomic(dst);
1100                 kunmap_atomic(src);
1101
1102                 vec.bv_page = page;
1103                 vec.bv_len = PAGE_SIZE;
1104                 vec.bv_offset = 0;
1105         }
1106
1107         ret = __zram_bvec_write(zram, &vec, index, bio);
1108 out:
1109         if (is_partial_io(bvec))
1110                 __free_page(page);
1111         return ret;
1112 }
1113
1114 /*
1115  * zram_bio_discard - handler on discard request
1116  * @index: physical block index in PAGE_SIZE units
1117  * @offset: byte offset within physical block
1118  */
1119 static void zram_bio_discard(struct zram *zram, u32 index,
1120                              int offset, struct bio *bio)
1121 {
1122         size_t n = bio->bi_iter.bi_size;
1123
1124         /*
1125          * zram manages data in physical block size units. Because logical block
1126          * size isn't identical with physical block size on some arch, we
1127          * could get a discard request pointing to a specific offset within a
1128          * certain physical block.  Although we can handle this request by
1129          * reading that physiclal block and decompressing and partially zeroing
1130          * and re-compressing and then re-storing it, this isn't reasonable
1131          * because our intent with a discard request is to save memory.  So
1132          * skipping this logical block is appropriate here.
1133          */
1134         if (offset) {
1135                 if (n <= (PAGE_SIZE - offset))
1136                         return;
1137
1138                 n -= (PAGE_SIZE - offset);
1139                 index++;
1140         }
1141
1142         while (n >= PAGE_SIZE) {
1143                 zram_slot_lock(zram, index);
1144                 zram_free_page(zram, index);
1145                 zram_slot_unlock(zram, index);
1146                 atomic64_inc(&zram->stats.notify_free);
1147                 index++;
1148                 n -= PAGE_SIZE;
1149         }
1150 }
1151
1152 /*
1153  * Returns errno if it has some problem. Otherwise return 0 or 1.
1154  * Returns 0 if IO request was done synchronously
1155  * Returns 1 if IO request was successfully submitted.
1156  */
1157 static int zram_bvec_rw(struct zram *zram, struct bio_vec *bvec, u32 index,
1158                         int offset, bool is_write, struct bio *bio)
1159 {
1160         unsigned long start_time = jiffies;
1161         int rw_acct = is_write ? REQ_OP_WRITE : REQ_OP_READ;
1162         struct request_queue *q = zram->disk->queue;
1163         int ret;
1164
1165         generic_start_io_acct(q, rw_acct, bvec->bv_len >> SECTOR_SHIFT,
1166                         &zram->disk->part0);
1167
1168         if (!is_write) {
1169                 atomic64_inc(&zram->stats.num_reads);
1170                 ret = zram_bvec_read(zram, bvec, index, offset, bio);
1171                 flush_dcache_page(bvec->bv_page);
1172         } else {
1173                 atomic64_inc(&zram->stats.num_writes);
1174                 ret = zram_bvec_write(zram, bvec, index, offset, bio);
1175         }
1176
1177         generic_end_io_acct(q, rw_acct, &zram->disk->part0, start_time);
1178
1179         if (unlikely(ret < 0)) {
1180                 if (!is_write)
1181                         atomic64_inc(&zram->stats.failed_reads);
1182                 else
1183                         atomic64_inc(&zram->stats.failed_writes);
1184         }
1185
1186         return ret;
1187 }
1188
1189 static void __zram_make_request(struct zram *zram, struct bio *bio)
1190 {
1191         int offset;
1192         u32 index;
1193         struct bio_vec bvec;
1194         struct bvec_iter iter;
1195
1196         index = bio->bi_iter.bi_sector >> SECTORS_PER_PAGE_SHIFT;
1197         offset = (bio->bi_iter.bi_sector &
1198                   (SECTORS_PER_PAGE - 1)) << SECTOR_SHIFT;
1199
1200         switch (bio_op(bio)) {
1201         case REQ_OP_DISCARD:
1202         case REQ_OP_WRITE_ZEROES:
1203                 zram_bio_discard(zram, index, offset, bio);
1204                 bio_endio(bio);
1205                 return;
1206         default:
1207                 break;
1208         }
1209
1210         bio_for_each_segment(bvec, bio, iter) {
1211                 struct bio_vec bv = bvec;
1212                 unsigned int unwritten = bvec.bv_len;
1213
1214                 do {
1215                         bv.bv_len = min_t(unsigned int, PAGE_SIZE - offset,
1216                                                         unwritten);
1217                         if (zram_bvec_rw(zram, &bv, index, offset,
1218                                         op_is_write(bio_op(bio)), bio) < 0)
1219                                 goto out;
1220
1221                         bv.bv_offset += bv.bv_len;
1222                         unwritten -= bv.bv_len;
1223
1224                         update_position(&index, &offset, &bv);
1225                 } while (unwritten);
1226         }
1227
1228         bio_endio(bio);
1229         return;
1230
1231 out:
1232         bio_io_error(bio);
1233 }
1234
1235 /*
1236  * Handler function for all zram I/O requests.
1237  */
1238 static blk_qc_t zram_make_request(struct request_queue *queue, struct bio *bio)
1239 {
1240         struct zram *zram = queue->queuedata;
1241
1242         if (!valid_io_request(zram, bio->bi_iter.bi_sector,
1243                                         bio->bi_iter.bi_size)) {
1244                 atomic64_inc(&zram->stats.invalid_io);
1245                 goto error;
1246         }
1247
1248         __zram_make_request(zram, bio);
1249         return BLK_QC_T_NONE;
1250
1251 error:
1252         bio_io_error(bio);
1253         return BLK_QC_T_NONE;
1254 }
1255
1256 static void zram_slot_free_notify(struct block_device *bdev,
1257                                 unsigned long index)
1258 {
1259         struct zram *zram;
1260
1261         zram = bdev->bd_disk->private_data;
1262
1263         zram_slot_lock(zram, index);
1264         zram_free_page(zram, index);
1265         zram_slot_unlock(zram, index);
1266         atomic64_inc(&zram->stats.notify_free);
1267 }
1268
1269 static int zram_rw_page(struct block_device *bdev, sector_t sector,
1270                        struct page *page, bool is_write)
1271 {
1272         int offset, ret;
1273         u32 index;
1274         struct zram *zram;
1275         struct bio_vec bv;
1276
1277         if (PageTransHuge(page))
1278                 return -ENOTSUPP;
1279         zram = bdev->bd_disk->private_data;
1280
1281         if (!valid_io_request(zram, sector, PAGE_SIZE)) {
1282                 atomic64_inc(&zram->stats.invalid_io);
1283                 ret = -EINVAL;
1284                 goto out;
1285         }
1286
1287         index = sector >> SECTORS_PER_PAGE_SHIFT;
1288         offset = (sector & (SECTORS_PER_PAGE - 1)) << SECTOR_SHIFT;
1289
1290         bv.bv_page = page;
1291         bv.bv_len = PAGE_SIZE;
1292         bv.bv_offset = 0;
1293
1294         ret = zram_bvec_rw(zram, &bv, index, offset, is_write, NULL);
1295 out:
1296         /*
1297          * If I/O fails, just return error(ie, non-zero) without
1298          * calling page_endio.
1299          * It causes resubmit the I/O with bio request by upper functions
1300          * of rw_page(e.g., swap_readpage, __swap_writepage) and
1301          * bio->bi_end_io does things to handle the error
1302          * (e.g., SetPageError, set_page_dirty and extra works).
1303          */
1304         if (unlikely(ret < 0))
1305                 return ret;
1306
1307         switch (ret) {
1308         case 0:
1309                 page_endio(page, is_write, 0);
1310                 break;
1311         case 1:
1312                 ret = 0;
1313                 break;
1314         default:
1315                 WARN_ON(1);
1316         }
1317         return ret;
1318 }
1319
1320 static void zram_reset_device(struct zram *zram)
1321 {
1322         struct zcomp *comp;
1323         u64 disksize;
1324
1325         down_write(&zram->init_lock);
1326
1327         zram->limit_pages = 0;
1328
1329         if (!init_done(zram)) {
1330                 up_write(&zram->init_lock);
1331                 return;
1332         }
1333
1334         comp = zram->comp;
1335         disksize = zram->disksize;
1336         zram->disksize = 0;
1337
1338         set_capacity(zram->disk, 0);
1339         part_stat_set_all(&zram->disk->part0, 0);
1340
1341         up_write(&zram->init_lock);
1342         /* I/O operation under all of CPU are done so let's free */
1343         zram_meta_free(zram, disksize);
1344         memset(&zram->stats, 0, sizeof(zram->stats));
1345         zcomp_destroy(comp);
1346         reset_bdev(zram);
1347 }
1348
1349 static ssize_t disksize_store(struct device *dev,
1350                 struct device_attribute *attr, const char *buf, size_t len)
1351 {
1352         u64 disksize;
1353         struct zcomp *comp;
1354         struct zram *zram = dev_to_zram(dev);
1355         int err;
1356
1357         disksize = memparse(buf, NULL);
1358         if (!disksize)
1359                 return -EINVAL;
1360
1361         down_write(&zram->init_lock);
1362         if (init_done(zram)) {
1363                 pr_info("Cannot change disksize for initialized device\n");
1364                 err = -EBUSY;
1365                 goto out_unlock;
1366         }
1367
1368         disksize = PAGE_ALIGN(disksize);
1369         if (!zram_meta_alloc(zram, disksize)) {
1370                 err = -ENOMEM;
1371                 goto out_unlock;
1372         }
1373
1374         comp = zcomp_create(zram->compressor);
1375         if (IS_ERR(comp)) {
1376                 pr_err("Cannot initialise %s compressing backend\n",
1377                                 zram->compressor);
1378                 err = PTR_ERR(comp);
1379                 goto out_free_meta;
1380         }
1381
1382         zram->comp = comp;
1383         zram->disksize = disksize;
1384         set_capacity(zram->disk, zram->disksize >> SECTOR_SHIFT);
1385         zram_revalidate_disk(zram);
1386         up_write(&zram->init_lock);
1387
1388         return len;
1389
1390 out_free_meta:
1391         zram_meta_free(zram, disksize);
1392 out_unlock:
1393         up_write(&zram->init_lock);
1394         return err;
1395 }
1396
1397 static ssize_t reset_store(struct device *dev,
1398                 struct device_attribute *attr, const char *buf, size_t len)
1399 {
1400         int ret;
1401         unsigned short do_reset;
1402         struct zram *zram;
1403         struct block_device *bdev;
1404
1405         ret = kstrtou16(buf, 10, &do_reset);
1406         if (ret)
1407                 return ret;
1408
1409         if (!do_reset)
1410                 return -EINVAL;
1411
1412         zram = dev_to_zram(dev);
1413         bdev = bdget_disk(zram->disk, 0);
1414         if (!bdev)
1415                 return -ENOMEM;
1416
1417         mutex_lock(&bdev->bd_mutex);
1418         /* Do not reset an active device or claimed device */
1419         if (bdev->bd_openers || zram->claim) {
1420                 mutex_unlock(&bdev->bd_mutex);
1421                 bdput(bdev);
1422                 return -EBUSY;
1423         }
1424
1425         /* From now on, anyone can't open /dev/zram[0-9] */
1426         zram->claim = true;
1427         mutex_unlock(&bdev->bd_mutex);
1428
1429         /* Make sure all the pending I/O are finished */
1430         fsync_bdev(bdev);
1431         zram_reset_device(zram);
1432         zram_revalidate_disk(zram);
1433         bdput(bdev);
1434
1435         mutex_lock(&bdev->bd_mutex);
1436         zram->claim = false;
1437         mutex_unlock(&bdev->bd_mutex);
1438
1439         return len;
1440 }
1441
1442 static int zram_open(struct block_device *bdev, fmode_t mode)
1443 {
1444         int ret = 0;
1445         struct zram *zram;
1446
1447         WARN_ON(!mutex_is_locked(&bdev->bd_mutex));
1448
1449         zram = bdev->bd_disk->private_data;
1450         /* zram was claimed to reset so open request fails */
1451         if (zram->claim)
1452                 ret = -EBUSY;
1453
1454         return ret;
1455 }
1456
1457 static const struct block_device_operations zram_devops = {
1458         .open = zram_open,
1459         .swap_slot_free_notify = zram_slot_free_notify,
1460         .rw_page = zram_rw_page,
1461         .owner = THIS_MODULE
1462 };
1463
1464 static DEVICE_ATTR_WO(compact);
1465 static DEVICE_ATTR_RW(disksize);
1466 static DEVICE_ATTR_RO(initstate);
1467 static DEVICE_ATTR_WO(reset);
1468 static DEVICE_ATTR_WO(mem_limit);
1469 static DEVICE_ATTR_WO(mem_used_max);
1470 static DEVICE_ATTR_RW(max_comp_streams);
1471 static DEVICE_ATTR_RW(comp_algorithm);
1472 #ifdef CONFIG_ZRAM_WRITEBACK
1473 static DEVICE_ATTR_RW(backing_dev);
1474 #endif
1475
1476 static struct attribute *zram_disk_attrs[] = {
1477         &dev_attr_disksize.attr,
1478         &dev_attr_initstate.attr,
1479         &dev_attr_reset.attr,
1480         &dev_attr_compact.attr,
1481         &dev_attr_mem_limit.attr,
1482         &dev_attr_mem_used_max.attr,
1483         &dev_attr_max_comp_streams.attr,
1484         &dev_attr_comp_algorithm.attr,
1485 #ifdef CONFIG_ZRAM_WRITEBACK
1486         &dev_attr_backing_dev.attr,
1487 #endif
1488         &dev_attr_io_stat.attr,
1489         &dev_attr_mm_stat.attr,
1490         &dev_attr_debug_stat.attr,
1491         NULL,
1492 };
1493
1494 static const struct attribute_group zram_disk_attr_group = {
1495         .attrs = zram_disk_attrs,
1496 };
1497
1498 static const struct attribute_group *zram_disk_attr_groups[] = {
1499         &zram_disk_attr_group,
1500         NULL,
1501 };
1502
1503 /*
1504  * Allocate and initialize new zram device. the function returns
1505  * '>= 0' device_id upon success, and negative value otherwise.
1506  */
1507 static int zram_add(void)
1508 {
1509         struct zram *zram;
1510         struct request_queue *queue;
1511         int ret, device_id;
1512
1513         zram = kzalloc(sizeof(struct zram), GFP_KERNEL);
1514         if (!zram)
1515                 return -ENOMEM;
1516
1517         ret = idr_alloc(&zram_index_idr, zram, 0, 0, GFP_KERNEL);
1518         if (ret < 0)
1519                 goto out_free_dev;
1520         device_id = ret;
1521
1522         init_rwsem(&zram->init_lock);
1523
1524         queue = blk_alloc_queue(GFP_KERNEL);
1525         if (!queue) {
1526                 pr_err("Error allocating disk queue for device %d\n",
1527                         device_id);
1528                 ret = -ENOMEM;
1529                 goto out_free_idr;
1530         }
1531
1532         blk_queue_make_request(queue, zram_make_request);
1533
1534         /* gendisk structure */
1535         zram->disk = alloc_disk(1);
1536         if (!zram->disk) {
1537                 pr_err("Error allocating disk structure for device %d\n",
1538                         device_id);
1539                 ret = -ENOMEM;
1540                 goto out_free_queue;
1541         }
1542
1543         zram->disk->major = zram_major;
1544         zram->disk->first_minor = device_id;
1545         zram->disk->fops = &zram_devops;
1546         zram->disk->queue = queue;
1547         zram->disk->queue->queuedata = zram;
1548         zram->disk->private_data = zram;
1549         snprintf(zram->disk->disk_name, 16, "zram%d", device_id);
1550
1551         /* Actual capacity set using syfs (/sys/block/zram<id>/disksize */
1552         set_capacity(zram->disk, 0);
1553         /* zram devices sort of resembles non-rotational disks */
1554         queue_flag_set_unlocked(QUEUE_FLAG_NONROT, zram->disk->queue);
1555         queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, zram->disk->queue);
1556         /*
1557          * To ensure that we always get PAGE_SIZE aligned
1558          * and n*PAGE_SIZED sized I/O requests.
1559          */
1560         blk_queue_physical_block_size(zram->disk->queue, PAGE_SIZE);
1561         blk_queue_logical_block_size(zram->disk->queue,
1562                                         ZRAM_LOGICAL_BLOCK_SIZE);
1563         blk_queue_io_min(zram->disk->queue, PAGE_SIZE);
1564         blk_queue_io_opt(zram->disk->queue, PAGE_SIZE);
1565         zram->disk->queue->limits.discard_granularity = PAGE_SIZE;
1566         blk_queue_max_discard_sectors(zram->disk->queue, UINT_MAX);
1567         queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, zram->disk->queue);
1568
1569         /*
1570          * zram_bio_discard() will clear all logical blocks if logical block
1571          * size is identical with physical block size(PAGE_SIZE). But if it is
1572          * different, we will skip discarding some parts of logical blocks in
1573          * the part of the request range which isn't aligned to physical block
1574          * size.  So we can't ensure that all discarded logical blocks are
1575          * zeroed.
1576          */
1577         if (ZRAM_LOGICAL_BLOCK_SIZE == PAGE_SIZE)
1578                 blk_queue_max_write_zeroes_sectors(zram->disk->queue, UINT_MAX);
1579
1580         disk_to_dev(zram->disk)->groups = zram_disk_attr_groups;
1581         add_disk(zram->disk);
1582
1583         strlcpy(zram->compressor, default_compressor, sizeof(zram->compressor));
1584
1585         pr_info("Added device: %s\n", zram->disk->disk_name);
1586         return device_id;
1587
1588 out_free_queue:
1589         blk_cleanup_queue(queue);
1590 out_free_idr:
1591         idr_remove(&zram_index_idr, device_id);
1592 out_free_dev:
1593         kfree(zram);
1594         return ret;
1595 }
1596
1597 static int zram_remove(struct zram *zram)
1598 {
1599         struct block_device *bdev;
1600
1601         bdev = bdget_disk(zram->disk, 0);
1602         if (!bdev)
1603                 return -ENOMEM;
1604
1605         mutex_lock(&bdev->bd_mutex);
1606         if (bdev->bd_openers || zram->claim) {
1607                 mutex_unlock(&bdev->bd_mutex);
1608                 bdput(bdev);
1609                 return -EBUSY;
1610         }
1611
1612         zram->claim = true;
1613         mutex_unlock(&bdev->bd_mutex);
1614
1615         /* Make sure all the pending I/O are finished */
1616         fsync_bdev(bdev);
1617         zram_reset_device(zram);
1618         bdput(bdev);
1619
1620         pr_info("Removed device: %s\n", zram->disk->disk_name);
1621
1622         blk_cleanup_queue(zram->disk->queue);
1623         del_gendisk(zram->disk);
1624         put_disk(zram->disk);
1625         kfree(zram);
1626         return 0;
1627 }
1628
1629 /* zram-control sysfs attributes */
1630
1631 /*
1632  * NOTE: hot_add attribute is not the usual read-only sysfs attribute. In a
1633  * sense that reading from this file does alter the state of your system -- it
1634  * creates a new un-initialized zram device and returns back this device's
1635  * device_id (or an error code if it fails to create a new device).
1636  */
1637 static ssize_t hot_add_show(struct class *class,
1638                         struct class_attribute *attr,
1639                         char *buf)
1640 {
1641         int ret;
1642
1643         mutex_lock(&zram_index_mutex);
1644         ret = zram_add();
1645         mutex_unlock(&zram_index_mutex);
1646
1647         if (ret < 0)
1648                 return ret;
1649         return scnprintf(buf, PAGE_SIZE, "%d\n", ret);
1650 }
1651 static struct class_attribute class_attr_hot_add =
1652         __ATTR(hot_add, 0400, hot_add_show, NULL);
1653
1654 static ssize_t hot_remove_store(struct class *class,
1655                         struct class_attribute *attr,
1656                         const char *buf,
1657                         size_t count)
1658 {
1659         struct zram *zram;
1660         int ret, dev_id;
1661
1662         /* dev_id is gendisk->first_minor, which is `int' */
1663         ret = kstrtoint(buf, 10, &dev_id);
1664         if (ret)
1665                 return ret;
1666         if (dev_id < 0)
1667                 return -EINVAL;
1668
1669         mutex_lock(&zram_index_mutex);
1670
1671         zram = idr_find(&zram_index_idr, dev_id);
1672         if (zram) {
1673                 ret = zram_remove(zram);
1674                 if (!ret)
1675                         idr_remove(&zram_index_idr, dev_id);
1676         } else {
1677                 ret = -ENODEV;
1678         }
1679
1680         mutex_unlock(&zram_index_mutex);
1681         return ret ? ret : count;
1682 }
1683 static CLASS_ATTR_WO(hot_remove);
1684
1685 static struct attribute *zram_control_class_attrs[] = {
1686         &class_attr_hot_add.attr,
1687         &class_attr_hot_remove.attr,
1688         NULL,
1689 };
1690 ATTRIBUTE_GROUPS(zram_control_class);
1691
1692 static struct class zram_control_class = {
1693         .name           = "zram-control",
1694         .owner          = THIS_MODULE,
1695         .class_groups   = zram_control_class_groups,
1696 };
1697
1698 static int zram_remove_cb(int id, void *ptr, void *data)
1699 {
1700         zram_remove(ptr);
1701         return 0;
1702 }
1703
1704 static void destroy_devices(void)
1705 {
1706         class_unregister(&zram_control_class);
1707         idr_for_each(&zram_index_idr, &zram_remove_cb, NULL);
1708         idr_destroy(&zram_index_idr);
1709         unregister_blkdev(zram_major, "zram");
1710         cpuhp_remove_multi_state(CPUHP_ZCOMP_PREPARE);
1711 }
1712
1713 static int __init zram_init(void)
1714 {
1715         int ret;
1716
1717         ret = cpuhp_setup_state_multi(CPUHP_ZCOMP_PREPARE, "block/zram:prepare",
1718                                       zcomp_cpu_up_prepare, zcomp_cpu_dead);
1719         if (ret < 0)
1720                 return ret;
1721
1722         ret = class_register(&zram_control_class);
1723         if (ret) {
1724                 pr_err("Unable to register zram-control class\n");
1725                 cpuhp_remove_multi_state(CPUHP_ZCOMP_PREPARE);
1726                 return ret;
1727         }
1728
1729         zram_major = register_blkdev(0, "zram");
1730         if (zram_major <= 0) {
1731                 pr_err("Unable to get major number\n");
1732                 class_unregister(&zram_control_class);
1733                 cpuhp_remove_multi_state(CPUHP_ZCOMP_PREPARE);
1734                 return -EBUSY;
1735         }
1736
1737         while (num_devices != 0) {
1738                 mutex_lock(&zram_index_mutex);
1739                 ret = zram_add();
1740                 mutex_unlock(&zram_index_mutex);
1741                 if (ret < 0)
1742                         goto out_error;
1743                 num_devices--;
1744         }
1745
1746         return 0;
1747
1748 out_error:
1749         destroy_devices();
1750         return ret;
1751 }
1752
1753 static void __exit zram_exit(void)
1754 {
1755         destroy_devices();
1756 }
1757
1758 module_init(zram_init);
1759 module_exit(zram_exit);
1760
1761 module_param(num_devices, uint, 0);
1762 MODULE_PARM_DESC(num_devices, "Number of pre-created zram devices");
1763
1764 MODULE_LICENSE("Dual BSD/GPL");
1765 MODULE_AUTHOR("Nitin Gupta <ngupta@vflare.org>");
1766 MODULE_DESCRIPTION("Compressed RAM Block Device");