drivers/block/zram/zram_drv.c

   1 /*
   2  * Compressed RAM block device
   3  *
   4  * Copyright (C) 2008, 2009, 2010  Nitin Gupta
   5  *               2012, 2013 Minchan Kim
   6  *
   7  * This code is released using a dual license strategy: BSD/GPL
   8  * You can choose the licence that better fits your requirements.
   9  *
  10  * Released under the terms of 3-clause BSD License
  11  * Released under the terms of GNU General Public License Version 2.0
  12  *
  13  */
  14
  15 #define KMSG_COMPONENT "zram"
  16 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
  17
  18 #include <linux/module.h>
  19 #include <linux/kernel.h>
  20 #include <linux/bio.h>
  21 #include <linux/bitops.h>
  22 #include <linux/blkdev.h>
  23 #include <linux/buffer_head.h>
  24 #include <linux/device.h>
  25 #include <linux/genhd.h>
  26 #include <linux/highmem.h>
  27 #include <linux/slab.h>
  28 #include <linux/backing-dev.h>
  29 #include <linux/string.h>
  30 #include <linux/vmalloc.h>
  31 #include <linux/err.h>
  32 #include <linux/idr.h>
  33 #include <linux/sysfs.h>
  34 #include <linux/debugfs.h>
  35 #include <linux/cpuhotplug.h>
  36
  37 #include "zram_drv.h"
  38
  39 static DEFINE_IDR(zram_index_idr);
  40 /* idr index must be protected */
  41 static DEFINE_MUTEX(zram_index_mutex);
  42
  43 static int zram_major;
  44 static const char *default_compressor = "lzo";
  45
  46 /* Module params (documentation at end) */
  47 static unsigned int num_devices = 1;
  48 /*
  49  * Pages that compress to sizes equals or greater than this are stored
  50  * uncompressed in memory.
  51  */
  52 static size_t huge_class_size;
  53
  54 static void zram_free_page(struct zram *zram, size_t index);
  55
  56 static int zram_slot_trylock(struct zram *zram, u32 index)
  57 {
  58         return bit_spin_trylock(ZRAM_LOCK, &zram->table[index].value);
  59 }
  60
  61 static void zram_slot_lock(struct zram *zram, u32 index)
  62 {
  63         bit_spin_lock(ZRAM_LOCK, &zram->table[index].value);
  64 }
  65
  66 static void zram_slot_unlock(struct zram *zram, u32 index)
  67 {
  68         bit_spin_unlock(ZRAM_LOCK, &zram->table[index].value);
  69 }
  70
  71 static inline bool init_done(struct zram *zram)
  72 {
  73         return zram->disksize;
  74 }
  75
  76 static inline bool zram_allocated(struct zram *zram, u32 index)
  77 {
  78
  79         return (zram->table[index].value >> (ZRAM_FLAG_SHIFT + 1)) ||
  80                                         zram->table[index].handle;
  81 }
  82
  83 static inline struct zram *dev_to_zram(struct device *dev)
  84 {
  85         return (struct zram *)dev_to_disk(dev)->private_data;
  86 }
  87
  88 static unsigned long zram_get_handle(struct zram *zram, u32 index)
  89 {
  90         return zram->table[index].handle;
  91 }
  92
  93 static void zram_set_handle(struct zram *zram, u32 index, unsigned long handle)
  94 {
  95         zram->table[index].handle = handle;
  96 }
  97
  98 /* flag operations require table entry bit_spin_lock() being held */
  99 static bool zram_test_flag(struct zram *zram, u32 index,
 100                         enum zram_pageflags flag)
 101 {
 102         return zram->table[index].value & BIT(flag);
 103 }
 104
 105 static void zram_set_flag(struct zram *zram, u32 index,
 106                         enum zram_pageflags flag)
 107 {
 108         zram->table[index].value |= BIT(flag);
 109 }
 110
 111 static void zram_clear_flag(struct zram *zram, u32 index,
 112                         enum zram_pageflags flag)
 113 {
 114         zram->table[index].value &= ~BIT(flag);
 115 }
 116
 117 static inline void zram_set_element(struct zram *zram, u32 index,
 118                         unsigned long element)
 119 {
 120         zram->table[index].element = element;
 121 }
 122
 123 static unsigned long zram_get_element(struct zram *zram, u32 index)
 124 {
 125         return zram->table[index].element;
 126 }
 127
 128 static size_t zram_get_obj_size(struct zram *zram, u32 index)
 129 {
 130         return zram->table[index].value & (BIT(ZRAM_FLAG_SHIFT) - 1);
 131 }
 132
 133 static void zram_set_obj_size(struct zram *zram,
 134                                         u32 index, size_t size)
 135 {
 136         unsigned long flags = zram->table[index].value >> ZRAM_FLAG_SHIFT;
 137
 138         zram->table[index].value = (flags << ZRAM_FLAG_SHIFT) | size;
 139 }
 140
 141 #if PAGE_SIZE != 4096
 142 static inline bool is_partial_io(struct bio_vec *bvec)
 143 {
 144         return bvec->bv_len != PAGE_SIZE;
 145 }
 146 #else
 147 static inline bool is_partial_io(struct bio_vec *bvec)
 148 {
 149         return false;
 150 }
 151 #endif
 152
 153 /*
 154  * Check if request is within bounds and aligned on zram logical blocks.
 155  */
 156 static inline bool valid_io_request(struct zram *zram,
 157                 sector_t start, unsigned int size)
 158 {
 159         u64 end, bound;
 160
 161         /* unaligned request */
 162         if (unlikely(start & (ZRAM_SECTOR_PER_LOGICAL_BLOCK - 1)))
 163                 return false;
 164         if (unlikely(size & (ZRAM_LOGICAL_BLOCK_SIZE - 1)))
 165                 return false;
 166
 167         end = start + (size >> SECTOR_SHIFT);
 168         bound = zram->disksize >> SECTOR_SHIFT;
 169         /* out of range range */
 170         if (unlikely(start >= bound || end > bound || start > end))
 171                 return false;
 172
 173         /* I/O request is valid */
 174         return true;
 175 }
 176
 177 static void update_position(u32 *index, int *offset, struct bio_vec *bvec)
 178 {
 179         *index  += (*offset + bvec->bv_len) / PAGE_SIZE;
 180         *offset = (*offset + bvec->bv_len) % PAGE_SIZE;
 181 }
 182
 183 static inline void update_used_max(struct zram *zram,
 184                                         const unsigned long pages)
 185 {
 186         unsigned long old_max, cur_max;
 187
 188         old_max = atomic_long_read(&zram->stats.max_used_pages);
 189
 190         do {
 191                 cur_max = old_max;
 192                 if (pages > cur_max)
 193                         old_max = atomic_long_cmpxchg(
 194                                 &zram->stats.max_used_pages, cur_max, pages);
 195         } while (old_max != cur_max);
 196 }
 197
 198 static inline void zram_fill_page(void *ptr, unsigned long len,
 199                                         unsigned long value)
 200 {
 201         WARN_ON_ONCE(!IS_ALIGNED(len, sizeof(unsigned long)));
 202         memset_l(ptr, value, len / sizeof(unsigned long));
 203 }
 204
 205 static bool page_same_filled(void *ptr, unsigned long *element)
 206 {
 207         unsigned int pos;
 208         unsigned long *page;
 209         unsigned long val;
 210
 211         page = (unsigned long *)ptr;
 212         val = page[0];
 213
 214         for (pos = 1; pos < PAGE_SIZE / sizeof(*page); pos++) {
 215                 if (val != page[pos])
 216                         return false;
 217         }
 218
 219         *element = val;
 220
 221         return true;
 222 }
 223
 224 static ssize_t initstate_show(struct device *dev,
 225                 struct device_attribute *attr, char *buf)
 226 {
 227         u32 val;
 228         struct zram *zram = dev_to_zram(dev);
 229
 230         down_read(&zram->init_lock);
 231         val = init_done(zram);
 232         up_read(&zram->init_lock);
 233
 234         return scnprintf(buf, PAGE_SIZE, "%u\n", val);
 235 }
 236
 237 static ssize_t disksize_show(struct device *dev,
 238                 struct device_attribute *attr, char *buf)
 239 {
 240         struct zram *zram = dev_to_zram(dev);
 241
 242         return scnprintf(buf, PAGE_SIZE, "%llu\n", zram->disksize);
 243 }
 244
 245 static ssize_t mem_limit_store(struct device *dev,
 246                 struct device_attribute *attr, const char *buf, size_t len)
 247 {
 248         u64 limit;
 249         char *tmp;
 250         struct zram *zram = dev_to_zram(dev);
 251
 252         limit = memparse(buf, &tmp);
 253         if (buf == tmp) /* no chars parsed, invalid input */
 254                 return -EINVAL;
 255
 256         down_write(&zram->init_lock);
 257         zram->limit_pages = PAGE_ALIGN(limit) >> PAGE_SHIFT;
 258         up_write(&zram->init_lock);
 259
 260         return len;
 261 }
 262
 263 static ssize_t mem_used_max_store(struct device *dev,
 264                 struct device_attribute *attr, const char *buf, size_t len)
 265 {
 266         int err;
 267         unsigned long val;
 268         struct zram *zram = dev_to_zram(dev);
 269
 270         err = kstrtoul(buf, 10, &val);
 271         if (err || val != 0)
 272                 return -EINVAL;
 273
 274         down_read(&zram->init_lock);
 275         if (init_done(zram)) {
 276                 atomic_long_set(&zram->stats.max_used_pages,
 277                                 zs_get_total_pages(zram->mem_pool));
 278         }
 279         up_read(&zram->init_lock);
 280
 281         return len;
 282 }
 283
 284 #ifdef CONFIG_ZRAM_WRITEBACK
 285 static bool zram_wb_enabled(struct zram *zram)
 286 {
 287         return zram->backing_dev;
 288 }
 289
 290 static void reset_bdev(struct zram *zram)
 291 {
 292         struct block_device *bdev;
 293
 294         if (!zram_wb_enabled(zram))
 295                 return;
 296
 297         bdev = zram->bdev;
 298         if (zram->old_block_size)
 299                 set_blocksize(bdev, zram->old_block_size);
 300         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 301         /* hope filp_close flush all of IO */
 302         filp_close(zram->backing_dev, NULL);
 303         zram->backing_dev = NULL;
 304         zram->old_block_size = 0;
 305         zram->bdev = NULL;
 306         zram->disk->queue->backing_dev_info->capabilities |=
 307                                 BDI_CAP_SYNCHRONOUS_IO;
 308         kvfree(zram->bitmap);
 309         zram->bitmap = NULL;
 310 }
 311
 312 static ssize_t backing_dev_show(struct device *dev,
 313                 struct device_attribute *attr, char *buf)
 314 {
 315         struct file *file;
 316         struct zram *zram = dev_to_zram(dev);
 317         char *p;
 318         ssize_t ret;
 319
 320         down_read(&zram->init_lock);
 321         file = zram->backing_dev;
 322         if (!file) {
 323                 memcpy(buf, "none\n", 5);
 324                 up_read(&zram->init_lock);
 325                 return 5;
 326         }
 327
 328         p = file_path(file, buf, PAGE_SIZE - 1);
 329         if (IS_ERR(p)) {
 330                 ret = PTR_ERR(p);
 331                 goto out;
 332         }
 333
 334         ret = strlen(p);
 335         memmove(buf, p, ret);
 336         buf[ret++] = '\n';
 337 out:
 338         up_read(&zram->init_lock);
 339         return ret;
 340 }
 341
 342 static ssize_t backing_dev_store(struct device *dev,
 343                 struct device_attribute *attr, const char *buf, size_t len)
 344 {
 345         char *file_name;
 346         size_t sz;
 347         struct file *backing_dev = NULL;
 348         struct inode *inode;
 349         struct address_space *mapping;
 350         unsigned int bitmap_sz, old_block_size = 0;
 351         unsigned long nr_pages, *bitmap = NULL;
 352         struct block_device *bdev = NULL;
 353         int err;
 354         struct zram *zram = dev_to_zram(dev);
 355
 356         file_name = kmalloc(PATH_MAX, GFP_KERNEL);
 357         if (!file_name)
 358                 return -ENOMEM;
 359
 360         down_write(&zram->init_lock);
 361         if (init_done(zram)) {
 362                 pr_info("Can't setup backing device for initialized device\n");
 363                 err = -EBUSY;
 364                 goto out;
 365         }
 366
 367         strlcpy(file_name, buf, PATH_MAX);
 368         /* ignore trailing newline */
 369         sz = strlen(file_name);
 370         if (sz > 0 && file_name[sz - 1] == '\n')
 371                 file_name[sz - 1] = 0x00;
 372
 373         backing_dev = filp_open(file_name, O_RDWR|O_LARGEFILE, 0);
 374         if (IS_ERR(backing_dev)) {
 375                 err = PTR_ERR(backing_dev);
 376                 backing_dev = NULL;
 377                 goto out;
 378         }
 379
 380         mapping = backing_dev->f_mapping;
 381         inode = mapping->host;
 382
 383         /* Support only block device in this moment */
 384         if (!S_ISBLK(inode->i_mode)) {
 385                 err = -ENOTBLK;
 386                 goto out;
 387         }
 388
 389         bdev = bdgrab(I_BDEV(inode));
 390         err = blkdev_get(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL, zram);
 391         if (err < 0) {
 392                 bdev = NULL;
 393                 goto out;
 394         }
 395
 396         nr_pages = i_size_read(inode) >> PAGE_SHIFT;
 397         bitmap_sz = BITS_TO_LONGS(nr_pages) * sizeof(long);
 398         bitmap = kvzalloc(bitmap_sz, GFP_KERNEL);
 399         if (!bitmap) {
 400                 err = -ENOMEM;
 401                 goto out;
 402         }
 403
 404         old_block_size = block_size(bdev);
 405         err = set_blocksize(bdev, PAGE_SIZE);
 406         if (err)
 407                 goto out;
 408
 409         reset_bdev(zram);
 410
 411         zram->old_block_size = old_block_size;
 412         zram->bdev = bdev;
 413         zram->backing_dev = backing_dev;
 414         zram->bitmap = bitmap;
 415         zram->nr_pages = nr_pages;
 416         /*
 417          * With writeback feature, zram does asynchronous IO so it's no longer
 418          * synchronous device so let's remove synchronous io flag. Othewise,
 419          * upper layer(e.g., swap) could wait IO completion rather than
 420          * (submit and return), which will cause system sluggish.
 421          * Furthermore, when the IO function returns(e.g., swap_readpage),
 422          * upper layer expects IO was done so it could deallocate the page
 423          * freely but in fact, IO is going on so finally could cause
 424          * use-after-free when the IO is really done.
 425          */
 426         zram->disk->queue->backing_dev_info->capabilities &=
 427                         ~BDI_CAP_SYNCHRONOUS_IO;
 428         up_write(&zram->init_lock);
 429
 430         pr_info("setup backing device %s\n", file_name);
 431         kfree(file_name);
 432
 433         return len;
 434 out:
 435         if (bitmap)
 436                 kvfree(bitmap);
 437
 438         if (bdev)
 439                 blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
 440
 441         if (backing_dev)
 442                 filp_close(backing_dev, NULL);
 443
 444         up_write(&zram->init_lock);
 445
 446         kfree(file_name);
 447
 448         return err;
 449 }
 450
 451 static unsigned long get_entry_bdev(struct zram *zram)
 452 {
 453         unsigned long blk_idx = 1;
 454 retry:
 455         /* skip 0 bit to confuse zram.handle = 0 */
 456         blk_idx = find_next_zero_bit(zram->bitmap, zram->nr_pages, blk_idx);
 457         if (blk_idx == zram->nr_pages)
 458                 return 0;
 459
 460         if (test_and_set_bit(blk_idx, zram->bitmap))
 461                 goto retry;
 462
 463         return blk_idx;
 464 }
 465
 466 static void put_entry_bdev(struct zram *zram, unsigned long entry)
 467 {
 468         int was_set;
 469
 470         was_set = test_and_clear_bit(entry, zram->bitmap);
 471         WARN_ON_ONCE(!was_set);
 472 }
 473
 474 static void zram_page_end_io(struct bio *bio)
 475 {
 476         struct page *page = bio_first_page_all(bio);
 477
 478         page_endio(page, op_is_write(bio_op(bio)),
 479                         blk_status_to_errno(bio->bi_status));
 480         bio_put(bio);
 481 }
 482
 483 /*
 484  * Returns 1 if the submission is successful.
 485  */
 486 static int read_from_bdev_async(struct zram *zram, struct bio_vec *bvec,
 487                         unsigned long entry, struct bio *parent)
 488 {
 489         struct bio *bio;
 490
 491         bio = bio_alloc(GFP_ATOMIC, 1);
 492         if (!bio)
 493                 return -ENOMEM;
 494
 495         bio->bi_iter.bi_sector = entry * (PAGE_SIZE >> 9);
 496         bio_set_dev(bio, zram->bdev);
 497         if (!bio_add_page(bio, bvec->bv_page, bvec->bv_len, bvec->bv_offset)) {
 498                 bio_put(bio);
 499                 return -EIO;
 500         }
 501
 502         if (!parent) {
 503                 bio->bi_opf = REQ_OP_READ;
 504                 bio->bi_end_io = zram_page_end_io;
 505         } else {
 506                 bio->bi_opf = parent->bi_opf;
 507                 bio_chain(bio, parent);
 508         }
 509
 510         submit_bio(bio);
 511         return 1;
 512 }
 513
 514 struct zram_work {
 515         struct work_struct work;
 516         struct zram *zram;
 517         unsigned long entry;
 518         struct bio *bio;
 519         struct bio_vec bvec;
 520 };
 521
 522 #if PAGE_SIZE != 4096
 523 static void zram_sync_read(struct work_struct *work)
 524 {
 525         struct zram_work *zw = container_of(work, struct zram_work, work);
 526         struct zram *zram = zw->zram;
 527         unsigned long entry = zw->entry;
 528         struct bio *bio = zw->bio;
 529
 530         read_from_bdev_async(zram, &zw->bvec, entry, bio);
 531 }
 532
 533 /*
 534  * Block layer want one ->make_request_fn to be active at a time
 535  * so if we use chained IO with parent IO in same context,
 536  * it's a deadlock. To avoid, it, it uses worker thread context.
 537  */
 538 static int read_from_bdev_sync(struct zram *zram, struct bio_vec *bvec,
 539                                 unsigned long entry, struct bio *bio)
 540 {
 541         struct zram_work work;
 542
 543         work.bvec = *bvec;
 544         work.zram = zram;
 545         work.entry = entry;
 546         work.bio = bio;
 547
 548         INIT_WORK_ONSTACK(&work.work, zram_sync_read);
 549         queue_work(system_unbound_wq, &work.work);
 550         flush_work(&work.work);
 551         destroy_work_on_stack(&work.work);
 552
 553         return 1;
 554 }
 555 #else
 556 static int read_from_bdev_sync(struct zram *zram, struct bio_vec *bvec,
 557                                 unsigned long entry, struct bio *bio)
 558 {
 559         WARN_ON(1);
 560         return -EIO;
 561 }
 562 #endif
 563
 564 static int read_from_bdev(struct zram *zram, struct bio_vec *bvec,
 565                         unsigned long entry, struct bio *parent, bool sync)
 566 {
 567         if (sync)
 568                 return read_from_bdev_sync(zram, bvec, entry, parent);
 569         else
 570                 return read_from_bdev_async(zram, bvec, entry, parent);
 571 }
 572
 573 static int write_to_bdev(struct zram *zram, struct bio_vec *bvec,
 574                                         u32 index, struct bio *parent,
 575                                         unsigned long *pentry)
 576 {
 577         struct bio *bio;
 578         unsigned long entry;
 579
 580         bio = bio_alloc(GFP_ATOMIC, 1);
 581         if (!bio)
 582                 return -ENOMEM;
 583
 584         entry = get_entry_bdev(zram);
 585         if (!entry) {
 586                 bio_put(bio);
 587                 return -ENOSPC;
 588         }
 589
 590         bio->bi_iter.bi_sector = entry * (PAGE_SIZE >> 9);
 591         bio_set_dev(bio, zram->bdev);
 592         if (!bio_add_page(bio, bvec->bv_page, bvec->bv_len,
 593                                         bvec->bv_offset)) {
 594                 bio_put(bio);
 595                 put_entry_bdev(zram, entry);
 596                 return -EIO;
 597         }
 598
 599         if (!parent) {
 600                 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC;
 601                 bio->bi_end_io = zram_page_end_io;
 602         } else {
 603                 bio->bi_opf = parent->bi_opf;
 604                 bio_chain(bio, parent);
 605         }
 606
 607         submit_bio(bio);
 608         *pentry = entry;
 609
 610         return 0;
 611 }
 612
 613 static void zram_wb_clear(struct zram *zram, u32 index)
 614 {
 615         unsigned long entry;
 616
 617         zram_clear_flag(zram, index, ZRAM_WB);
 618         entry = zram_get_element(zram, index);
 619         zram_set_element(zram, index, 0);
 620         put_entry_bdev(zram, entry);
 621 }
 622
 623 #else
 624 static bool zram_wb_enabled(struct zram *zram) { return false; }
 625 static inline void reset_bdev(struct zram *zram) {};
 626 static int write_to_bdev(struct zram *zram, struct bio_vec *bvec,
 627                                         u32 index, struct bio *parent,
 628                                         unsigned long *pentry)
 629
 630 {
 631         return -EIO;
 632 }
 633
 634 static int read_from_bdev(struct zram *zram, struct bio_vec *bvec,
 635                         unsigned long entry, struct bio *parent, bool sync)
 636 {
 637         return -EIO;
 638 }
 639 static void zram_wb_clear(struct zram *zram, u32 index) {}
 640 #endif
 641
 642 #ifdef CONFIG_ZRAM_MEMORY_TRACKING
 643
 644 static struct dentry *zram_debugfs_root;
 645
 646 static void zram_debugfs_create(void)
 647 {
 648         zram_debugfs_root = debugfs_create_dir("zram", NULL);
 649 }
 650
 651 static void zram_debugfs_destroy(void)
 652 {
 653         debugfs_remove_recursive(zram_debugfs_root);
 654 }
 655
 656 static void zram_accessed(struct zram *zram, u32 index)
 657 {
 658         zram->table[index].ac_time = ktime_get_boottime();
 659 }
 660
 661 static void zram_reset_access(struct zram *zram, u32 index)
 662 {
 663         zram->table[index].ac_time = 0;
 664 }
 665
 666 static ssize_t read_block_state(struct file *file, char __user *buf,
 667                                 size_t count, loff_t *ppos)
 668 {
 669         char *kbuf;
 670         ssize_t index, written = 0;
 671         struct zram *zram = file->private_data;
 672         unsigned long nr_pages = zram->disksize >> PAGE_SHIFT;
 673         struct timespec64 ts;
 674
 675         kbuf = kvmalloc(count, GFP_KERNEL);
 676         if (!kbuf)
 677                 return -ENOMEM;
 678
 679         down_read(&zram->init_lock);
 680         if (!init_done(zram)) {
 681                 up_read(&zram->init_lock);
 682                 kvfree(kbuf);
 683                 return -EINVAL;
 684         }
 685
 686         for (index = *ppos; index < nr_pages; index++) {
 687                 int copied;
 688
 689                 zram_slot_lock(zram, index);
 690                 if (!zram_allocated(zram, index))
 691                         goto next;
 692
 693                 ts = ktime_to_timespec64(zram->table[index].ac_time);
 694                 copied = snprintf(kbuf + written, count,
 695                         "%12zd %12lld.%06lu %c%c%c\n",
 696                         index, (s64)ts.tv_sec,
 697                         ts.tv_nsec / NSEC_PER_USEC,
 698                         zram_test_flag(zram, index, ZRAM_SAME) ? 's' : '.',
 699                         zram_test_flag(zram, index, ZRAM_WB) ? 'w' : '.',
 700                         zram_test_flag(zram, index, ZRAM_HUGE) ? 'h' : '.');
 701
 702                 if (count <= copied) {
 703                         zram_slot_unlock(zram, index);
 704                         break;
 705                 }
 706                 written += copied;
 707                 count -= copied;
 708 next:
 709                 zram_slot_unlock(zram, index);
 710                 *ppos += 1;
 711         }
 712
 713         up_read(&zram->init_lock);
 714         if (copy_to_user(buf, kbuf, written))
 715                 written = -EFAULT;
 716         kvfree(kbuf);
 717
 718         return written;
 719 }
 720
 721 static const struct file_operations proc_zram_block_state_op = {
 722         .open = simple_open,
 723         .read = read_block_state,
 724         .llseek = default_llseek,
 725 };
 726
 727 static void zram_debugfs_register(struct zram *zram)
 728 {
 729         if (!zram_debugfs_root)
 730                 return;
 731
 732         zram->debugfs_dir = debugfs_create_dir(zram->disk->disk_name,
 733                                                 zram_debugfs_root);
 734         debugfs_create_file("block_state", 0400, zram->debugfs_dir,
 735                                 zram, &proc_zram_block_state_op);
 736 }
 737
 738 static void zram_debugfs_unregister(struct zram *zram)
 739 {
 740         debugfs_remove_recursive(zram->debugfs_dir);
 741 }
 742 #else
 743 static void zram_debugfs_create(void) {};
 744 static void zram_debugfs_destroy(void) {};
 745 static void zram_accessed(struct zram *zram, u32 index) {};
 746 static void zram_reset_access(struct zram *zram, u32 index) {};
 747 static void zram_debugfs_register(struct zram *zram) {};
 748 static void zram_debugfs_unregister(struct zram *zram) {};
 749 #endif
 750
 751 /*
 752  * We switched to per-cpu streams and this attr is not needed anymore.
 753  * However, we will keep it around for some time, because:
 754  * a) we may revert per-cpu streams in the future
 755  * b) it's visible to user space and we need to follow our 2 years
 756  *    retirement rule; but we already have a number of 'soon to be
 757  *    altered' attrs, so max_comp_streams need to wait for the next
 758  *    layoff cycle.
 759  */
 760 static ssize_t max_comp_streams_show(struct device *dev,
 761                 struct device_attribute *attr, char *buf)
 762 {
 763         return scnprintf(buf, PAGE_SIZE, "%d\n", num_online_cpus());
 764 }
 765
 766 static ssize_t max_comp_streams_store(struct device *dev,
 767                 struct device_attribute *attr, const char *buf, size_t len)
 768 {
 769         return len;
 770 }
 771
 772 static ssize_t comp_algorithm_show(struct device *dev,
 773                 struct device_attribute *attr, char *buf)
 774 {
 775         size_t sz;
 776         struct zram *zram = dev_to_zram(dev);
 777
 778         down_read(&zram->init_lock);
 779         sz = zcomp_available_show(zram->compressor, buf);
 780         up_read(&zram->init_lock);
 781
 782         return sz;
 783 }
 784
 785 static ssize_t comp_algorithm_store(struct device *dev,
 786                 struct device_attribute *attr, const char *buf, size_t len)
 787 {
 788         struct zram *zram = dev_to_zram(dev);
 789         char compressor[ARRAY_SIZE(zram->compressor)];
 790         size_t sz;
 791
 792         strlcpy(compressor, buf, sizeof(compressor));
 793         /* ignore trailing newline */
 794         sz = strlen(compressor);
 795         if (sz > 0 && compressor[sz - 1] == '\n')
 796                 compressor[sz - 1] = 0x00;
 797
 798         if (!zcomp_available_algorithm(compressor))
 799                 return -EINVAL;
 800
 801         down_write(&zram->init_lock);
 802         if (init_done(zram)) {
 803                 up_write(&zram->init_lock);
 804                 pr_info("Can't change algorithm for initialized device\n");
 805                 return -EBUSY;
 806         }
 807
 808         strcpy(zram->compressor, compressor);
 809         up_write(&zram->init_lock);
 810         return len;
 811 }
 812
 813 static ssize_t compact_store(struct device *dev,
 814                 struct device_attribute *attr, const char *buf, size_t len)
 815 {
 816         struct zram *zram = dev_to_zram(dev);
 817
 818         down_read(&zram->init_lock);
 819         if (!init_done(zram)) {
 820                 up_read(&zram->init_lock);
 821                 return -EINVAL;
 822         }
 823
 824         zs_compact(zram->mem_pool);
 825         up_read(&zram->init_lock);
 826
 827         return len;
 828 }
 829
 830 static ssize_t io_stat_show(struct device *dev,
 831                 struct device_attribute *attr, char *buf)
 832 {
 833         struct zram *zram = dev_to_zram(dev);
 834         ssize_t ret;
 835
 836         down_read(&zram->init_lock);
 837         ret = scnprintf(buf, PAGE_SIZE,
 838                         "%8llu %8llu %8llu %8llu\n",
 839                         (u64)atomic64_read(&zram->stats.failed_reads),
 840                         (u64)atomic64_read(&zram->stats.failed_writes),
 841                         (u64)atomic64_read(&zram->stats.invalid_io),
 842                         (u64)atomic64_read(&zram->stats.notify_free));
 843         up_read(&zram->init_lock);
 844
 845         return ret;
 846 }
 847
 848 static ssize_t mm_stat_show(struct device *dev,
 849                 struct device_attribute *attr, char *buf)
 850 {
 851         struct zram *zram = dev_to_zram(dev);
 852         struct zs_pool_stats pool_stats;
 853         u64 orig_size, mem_used = 0;
 854         long max_used;
 855         ssize_t ret;
 856
 857         memset(&pool_stats, 0x00, sizeof(struct zs_pool_stats));
 858
 859         down_read(&zram->init_lock);
 860         if (init_done(zram)) {
 861                 mem_used = zs_get_total_pages(zram->mem_pool);
 862                 zs_pool_stats(zram->mem_pool, &pool_stats);
 863         }
 864
 865         orig_size = atomic64_read(&zram->stats.pages_stored);
 866         max_used = atomic_long_read(&zram->stats.max_used_pages);
 867
 868         ret = scnprintf(buf, PAGE_SIZE,
 869                         "%8llu %8llu %8llu %8lu %8ld %8llu %8lu %8llu\n",
 870                         orig_size << PAGE_SHIFT,
 871                         (u64)atomic64_read(&zram->stats.compr_data_size),
 872                         mem_used << PAGE_SHIFT,
 873                         zram->limit_pages << PAGE_SHIFT,
 874                         max_used << PAGE_SHIFT,
 875                         (u64)atomic64_read(&zram->stats.same_pages),
 876                         atomic_long_read(&pool_stats.pages_compacted),
 877                         (u64)atomic64_read(&zram->stats.huge_pages));
 878         up_read(&zram->init_lock);
 879
 880         return ret;
 881 }
 882
 883 static ssize_t debug_stat_show(struct device *dev,
 884                 struct device_attribute *attr, char *buf)
 885 {
 886         int version = 1;
 887         struct zram *zram = dev_to_zram(dev);
 888         ssize_t ret;
 889
 890         down_read(&zram->init_lock);
 891         ret = scnprintf(buf, PAGE_SIZE,
 892                         "version: %d\n%8llu %8llu\n",
 893                         version,
 894                         (u64)atomic64_read(&zram->stats.writestall),
 895                         (u64)atomic64_read(&zram->stats.miss_free));
 896         up_read(&zram->init_lock);
 897
 898         return ret;
 899 }
 900
 901 static DEVICE_ATTR_RO(io_stat);
 902 static DEVICE_ATTR_RO(mm_stat);
 903 static DEVICE_ATTR_RO(debug_stat);
 904
 905 static void zram_meta_free(struct zram *zram, u64 disksize)
 906 {
 907         size_t num_pages = disksize >> PAGE_SHIFT;
 908         size_t index;
 909
 910         /* Free all pages that are still in this zram device */
 911         for (index = 0; index < num_pages; index++)
 912                 zram_free_page(zram, index);
 913
 914         zs_destroy_pool(zram->mem_pool);
 915         vfree(zram->table);
 916 }
 917
 918 static bool zram_meta_alloc(struct zram *zram, u64 disksize)
 919 {
 920         size_t num_pages;
 921
 922         num_pages = disksize >> PAGE_SHIFT;
 923         zram->table = vzalloc(array_size(num_pages, sizeof(*zram->table)));
 924         if (!zram->table)
 925                 return false;
 926
 927         zram->mem_pool = zs_create_pool(zram->disk->disk_name);
 928         if (!zram->mem_pool) {
 929                 vfree(zram->table);
 930                 return false;
 931         }
 932
 933         if (!huge_class_size)
 934                 huge_class_size = zs_huge_class_size(zram->mem_pool);
 935         return true;
 936 }
 937
 938 /*
 939  * To protect concurrent access to the same index entry,
 940  * caller should hold this table index entry's bit_spinlock to
 941  * indicate this index entry is accessing.
 942  */
 943 static void zram_free_page(struct zram *zram, size_t index)
 944 {
 945         unsigned long handle;
 946
 947         zram_reset_access(zram, index);
 948
 949         if (zram_test_flag(zram, index, ZRAM_HUGE)) {
 950                 zram_clear_flag(zram, index, ZRAM_HUGE);
 951                 atomic64_dec(&zram->stats.huge_pages);
 952         }
 953
 954         if (zram_wb_enabled(zram) && zram_test_flag(zram, index, ZRAM_WB)) {
 955                 zram_wb_clear(zram, index);
 956                 atomic64_dec(&zram->stats.pages_stored);
 957                 return;
 958         }
 959
 960         /*
 961          * No memory is allocated for same element filled pages.
 962          * Simply clear same page flag.
 963          */
 964         if (zram_test_flag(zram, index, ZRAM_SAME)) {
 965                 zram_clear_flag(zram, index, ZRAM_SAME);
 966                 zram_set_element(zram, index, 0);
 967                 atomic64_dec(&zram->stats.same_pages);
 968                 atomic64_dec(&zram->stats.pages_stored);
 969                 return;
 970         }
 971
 972         handle = zram_get_handle(zram, index);
 973         if (!handle)
 974                 return;
 975
 976         zs_free(zram->mem_pool, handle);
 977
 978         atomic64_sub(zram_get_obj_size(zram, index),
 979                         &zram->stats.compr_data_size);
 980         atomic64_dec(&zram->stats.pages_stored);
 981
 982         zram_set_handle(zram, index, 0);
 983         zram_set_obj_size(zram, index, 0);
 984 }
 985
 986 static int __zram_bvec_read(struct zram *zram, struct page *page, u32 index,
 987                                 struct bio *bio, bool partial_io)
 988 {
 989         int ret;
 990         unsigned long handle;
 991         unsigned int size;
 992         void *src, *dst;
 993
 994         if (zram_wb_enabled(zram)) {
 995                 zram_slot_lock(zram, index);
 996                 if (zram_test_flag(zram, index, ZRAM_WB)) {
 997                         struct bio_vec bvec;
 998
 999                         zram_slot_unlock(zram, index);
1000
1001                         bvec.bv_page = page;
1002                         bvec.bv_len = PAGE_SIZE;
1003                         bvec.bv_offset = 0;
1004                         return read_from_bdev(zram, &bvec,
1005                                         zram_get_element(zram, index),
1006                                         bio, partial_io);
1007                 }
1008                 zram_slot_unlock(zram, index);
1009         }
1010
1011         zram_slot_lock(zram, index);
1012         handle = zram_get_handle(zram, index);
1013         if (!handle || zram_test_flag(zram, index, ZRAM_SAME)) {
1014                 unsigned long value;
1015                 void *mem;
1016
1017                 value = handle ? zram_get_element(zram, index) : 0;
1018                 mem = kmap_atomic(page);
1019                 zram_fill_page(mem, PAGE_SIZE, value);
1020                 kunmap_atomic(mem);
1021                 zram_slot_unlock(zram, index);
1022                 return 0;
1023         }
1024
1025         size = zram_get_obj_size(zram, index);
1026
1027         src = zs_map_object(zram->mem_pool, handle, ZS_MM_RO);
1028         if (size == PAGE_SIZE) {
1029                 dst = kmap_atomic(page);
1030                 memcpy(dst, src, PAGE_SIZE);
1031                 kunmap_atomic(dst);
1032                 ret = 0;
1033         } else {
1034                 struct zcomp_strm *zstrm = zcomp_stream_get(zram->comp);
1035
1036                 dst = kmap_atomic(page);
1037                 ret = zcomp_decompress(zstrm, src, size, dst);
1038                 kunmap_atomic(dst);
1039                 zcomp_stream_put(zram->comp);
1040         }
1041         zs_unmap_object(zram->mem_pool, handle);
1042         zram_slot_unlock(zram, index);
1043
1044         /* Should NEVER happen. Return bio error if it does. */
1045         if (unlikely(ret))
1046                 pr_err("Decompression failed! err=%d, page=%u\n", ret, index);
1047
1048         return ret;
1049 }
1050
1051 static int zram_bvec_read(struct zram *zram, struct bio_vec *bvec,
1052                                 u32 index, int offset, struct bio *bio)
1053 {
1054         int ret;
1055         struct page *page;
1056
1057         page = bvec->bv_page;
1058         if (is_partial_io(bvec)) {
1059                 /* Use a temporary buffer to decompress the page */
1060                 page = alloc_page(GFP_NOIO|__GFP_HIGHMEM);
1061                 if (!page)
1062                         return -ENOMEM;
1063         }
1064
1065         ret = __zram_bvec_read(zram, page, index, bio, is_partial_io(bvec));
1066         if (unlikely(ret))
1067                 goto out;
1068
1069         if (is_partial_io(bvec)) {
1070                 void *dst = kmap_atomic(bvec->bv_page);
1071                 void *src = kmap_atomic(page);
1072
1073                 memcpy(dst + bvec->bv_offset, src + offset, bvec->bv_len);
1074                 kunmap_atomic(src);
1075                 kunmap_atomic(dst);
1076         }
1077 out:
1078         if (is_partial_io(bvec))
1079                 __free_page(page);
1080
1081         return ret;
1082 }
1083
1084 static int __zram_bvec_write(struct zram *zram, struct bio_vec *bvec,
1085                                 u32 index, struct bio *bio)
1086 {
1087         int ret = 0;
1088         unsigned long alloced_pages;
1089         unsigned long handle = 0;
1090         unsigned int comp_len = 0;
1091         void *src, *dst, *mem;
1092         struct zcomp_strm *zstrm;
1093         struct page *page = bvec->bv_page;
1094         unsigned long element = 0;
1095         enum zram_pageflags flags = 0;
1096         bool allow_wb = true;
1097
1098         mem = kmap_atomic(page);
1099         if (page_same_filled(mem, &element)) {
1100                 kunmap_atomic(mem);
1101                 /* Free memory associated with this sector now. */
1102                 flags = ZRAM_SAME;
1103                 atomic64_inc(&zram->stats.same_pages);
1104                 goto out;
1105         }
1106         kunmap_atomic(mem);
1107
1108 compress_again:
1109         zstrm = zcomp_stream_get(zram->comp);
1110         src = kmap_atomic(page);
1111         ret = zcomp_compress(zstrm, src, &comp_len);
1112         kunmap_atomic(src);
1113
1114         if (unlikely(ret)) {
1115                 zcomp_stream_put(zram->comp);
1116                 pr_err("Compression failed! err=%d\n", ret);
1117                 zs_free(zram->mem_pool, handle);
1118                 return ret;
1119         }
1120
1121         if (unlikely(comp_len >= huge_class_size)) {
1122                 comp_len = PAGE_SIZE;
1123                 if (zram_wb_enabled(zram) && allow_wb) {
1124                         zcomp_stream_put(zram->comp);
1125                         ret = write_to_bdev(zram, bvec, index, bio, &element);
1126                         if (!ret) {
1127                                 flags = ZRAM_WB;
1128                                 ret = 1;
1129                                 goto out;
1130                         }
1131                         allow_wb = false;
1132                         goto compress_again;
1133                 }
1134         }
1135
1136         /*
1137          * handle allocation has 2 paths:
1138          * a) fast path is executed with preemption disabled (for
1139          *  per-cpu streams) and has __GFP_DIRECT_RECLAIM bit clear,
1140          *  since we can't sleep;
1141          * b) slow path enables preemption and attempts to allocate
1142          *  the page with __GFP_DIRECT_RECLAIM bit set. we have to
1143          *  put per-cpu compression stream and, thus, to re-do
1144          *  the compression once handle is allocated.
1145          *
1146          * if we have a 'non-null' handle here then we are coming
1147          * from the slow path and handle has already been allocated.
1148          */
1149         if (!handle)
1150                 handle = zs_malloc(zram->mem_pool, comp_len,
1151                                 __GFP_KSWAPD_RECLAIM |
1152                                 __GFP_NOWARN |
1153                                 __GFP_HIGHMEM |
1154                                 __GFP_MOVABLE);
1155         if (!handle) {
1156                 zcomp_stream_put(zram->comp);
1157                 atomic64_inc(&zram->stats.writestall);
1158                 handle = zs_malloc(zram->mem_pool, comp_len,
1159                                 GFP_NOIO | __GFP_HIGHMEM |
1160                                 __GFP_MOVABLE);
1161                 if (handle)
1162                         goto compress_again;
1163                 return -ENOMEM;
1164         }
1165
1166         alloced_pages = zs_get_total_pages(zram->mem_pool);
1167         update_used_max(zram, alloced_pages);
1168
1169         if (zram->limit_pages && alloced_pages > zram->limit_pages) {
1170                 zcomp_stream_put(zram->comp);
1171                 zs_free(zram->mem_pool, handle);
1172                 return -ENOMEM;
1173         }
1174
1175         dst = zs_map_object(zram->mem_pool, handle, ZS_MM_WO);
1176
1177         src = zstrm->buffer;
1178         if (comp_len == PAGE_SIZE)
1179                 src = kmap_atomic(page);
1180         memcpy(dst, src, comp_len);
1181         if (comp_len == PAGE_SIZE)
1182                 kunmap_atomic(src);
1183
1184         zcomp_stream_put(zram->comp);
1185         zs_unmap_object(zram->mem_pool, handle);
1186         atomic64_add(comp_len, &zram->stats.compr_data_size);
1187 out:
1188         /*
1189          * Free memory associated with this sector
1190          * before overwriting unused sectors.
1191          */
1192         zram_slot_lock(zram, index);
1193         zram_free_page(zram, index);
1194
1195         if (comp_len == PAGE_SIZE) {
1196                 zram_set_flag(zram, index, ZRAM_HUGE);
1197                 atomic64_inc(&zram->stats.huge_pages);
1198         }
1199
1200         if (flags) {
1201                 zram_set_flag(zram, index, flags);
1202                 zram_set_element(zram, index, element);
1203         }  else {
1204                 zram_set_handle(zram, index, handle);
1205                 zram_set_obj_size(zram, index, comp_len);
1206         }
1207         zram_slot_unlock(zram, index);
1208
1209         /* Update stats */
1210         atomic64_inc(&zram->stats.pages_stored);
1211         return ret;
1212 }
1213
1214 static int zram_bvec_write(struct zram *zram, struct bio_vec *bvec,
1215                                 u32 index, int offset, struct bio *bio)
1216 {
1217         int ret;
1218         struct page *page = NULL;
1219         void *src;
1220         struct bio_vec vec;
1221
1222         vec = *bvec;
1223         if (is_partial_io(bvec)) {
1224                 void *dst;
1225                 /*
1226                  * This is a partial IO. We need to read the full page
1227                  * before to write the changes.
1228                  */
1229                 page = alloc_page(GFP_NOIO|__GFP_HIGHMEM);
1230                 if (!page)
1231                         return -ENOMEM;
1232
1233                 ret = __zram_bvec_read(zram, page, index, bio, true);
1234                 if (ret)
1235                         goto out;
1236
1237                 src = kmap_atomic(bvec->bv_page);
1238                 dst = kmap_atomic(page);
1239                 memcpy(dst + offset, src + bvec->bv_offset, bvec->bv_len);
1240                 kunmap_atomic(dst);
1241                 kunmap_atomic(src);
1242
1243                 vec.bv_page = page;
1244                 vec.bv_len = PAGE_SIZE;
1245                 vec.bv_offset = 0;
1246         }
1247
1248         ret = __zram_bvec_write(zram, &vec, index, bio);
1249 out:
1250         if (is_partial_io(bvec))
1251                 __free_page(page);
1252         return ret;
1253 }
1254
1255 /*
1256  * zram_bio_discard - handler on discard request
1257  * @index: physical block index in PAGE_SIZE units
1258  * @offset: byte offset within physical block
1259  */
1260 static void zram_bio_discard(struct zram *zram, u32 index,
1261                              int offset, struct bio *bio)
1262 {
1263         size_t n = bio->bi_iter.bi_size;
1264
1265         /*
1266          * zram manages data in physical block size units. Because logical block
1267          * size isn't identical with physical block size on some arch, we
1268          * could get a discard request pointing to a specific offset within a
1269          * certain physical block.  Although we can handle this request by
1270          * reading that physiclal block and decompressing and partially zeroing
1271          * and re-compressing and then re-storing it, this isn't reasonable
1272          * because our intent with a discard request is to save memory.  So
1273          * skipping this logical block is appropriate here.
1274          */
1275         if (offset) {
1276                 if (n <= (PAGE_SIZE - offset))
1277                         return;
1278
1279                 n -= (PAGE_SIZE - offset);
1280                 index++;
1281         }
1282
1283         while (n >= PAGE_SIZE) {
1284                 zram_slot_lock(zram, index);
1285                 zram_free_page(zram, index);
1286                 zram_slot_unlock(zram, index);
1287                 atomic64_inc(&zram->stats.notify_free);
1288                 index++;
1289                 n -= PAGE_SIZE;
1290         }
1291 }
1292
1293 /*
1294  * Returns errno if it has some problem. Otherwise return 0 or 1.
1295  * Returns 0 if IO request was done synchronously
1296  * Returns 1 if IO request was successfully submitted.
1297  */
1298 static int zram_bvec_rw(struct zram *zram, struct bio_vec *bvec, u32 index,
1299                         int offset, unsigned int op, struct bio *bio)
1300 {
1301         unsigned long start_time = jiffies;
1302         struct request_queue *q = zram->disk->queue;
1303         int ret;
1304
1305         generic_start_io_acct(q, op, bvec->bv_len >> SECTOR_SHIFT,
1306                         &zram->disk->part0);
1307
1308         if (!op_is_write(op)) {
1309                 atomic64_inc(&zram->stats.num_reads);
1310                 ret = zram_bvec_read(zram, bvec, index, offset, bio);
1311                 flush_dcache_page(bvec->bv_page);
1312         } else {
1313                 atomic64_inc(&zram->stats.num_writes);
1314                 ret = zram_bvec_write(zram, bvec, index, offset, bio);
1315         }
1316
1317         generic_end_io_acct(q, op, &zram->disk->part0, start_time);
1318
1319         zram_slot_lock(zram, index);
1320         zram_accessed(zram, index);
1321         zram_slot_unlock(zram, index);
1322
1323         if (unlikely(ret < 0)) {
1324                 if (!op_is_write(op))
1325                         atomic64_inc(&zram->stats.failed_reads);
1326                 else
1327                         atomic64_inc(&zram->stats.failed_writes);
1328         }
1329
1330         return ret;
1331 }
1332
1333 static void __zram_make_request(struct zram *zram, struct bio *bio)
1334 {
1335         int offset;
1336         u32 index;
1337         struct bio_vec bvec;
1338         struct bvec_iter iter;
1339
1340         index = bio->bi_iter.bi_sector >> SECTORS_PER_PAGE_SHIFT;
1341         offset = (bio->bi_iter.bi_sector &
1342                   (SECTORS_PER_PAGE - 1)) << SECTOR_SHIFT;
1343
1344         switch (bio_op(bio)) {
1345         case REQ_OP_DISCARD:
1346         case REQ_OP_WRITE_ZEROES:
1347                 zram_bio_discard(zram, index, offset, bio);
1348                 bio_endio(bio);
1349                 return;
1350         default:
1351                 break;
1352         }
1353
1354         bio_for_each_segment(bvec, bio, iter) {
1355                 struct bio_vec bv = bvec;
1356                 unsigned int unwritten = bvec.bv_len;
1357
1358                 do {
1359                         bv.bv_len = min_t(unsigned int, PAGE_SIZE - offset,
1360                                                         unwritten);
1361                         if (zram_bvec_rw(zram, &bv, index, offset,
1362                                          bio_op(bio), bio) < 0)
1363                                 goto out;
1364
1365                         bv.bv_offset += bv.bv_len;
1366                         unwritten -= bv.bv_len;
1367
1368                         update_position(&index, &offset, &bv);
1369                 } while (unwritten);
1370         }
1371
1372         bio_endio(bio);
1373         return;
1374
1375 out:
1376         bio_io_error(bio);
1377 }
1378
1379 /*
1380  * Handler function for all zram I/O requests.
1381  */
1382 static blk_qc_t zram_make_request(struct request_queue *queue, struct bio *bio)
1383 {
1384         struct zram *zram = queue->queuedata;
1385
1386         if (!valid_io_request(zram, bio->bi_iter.bi_sector,
1387                                         bio->bi_iter.bi_size)) {
1388                 atomic64_inc(&zram->stats.invalid_io);
1389                 goto error;
1390         }
1391
1392         __zram_make_request(zram, bio);
1393         return BLK_QC_T_NONE;
1394
1395 error:
1396         bio_io_error(bio);
1397         return BLK_QC_T_NONE;
1398 }
1399
1400 static void zram_slot_free_notify(struct block_device *bdev,
1401                                 unsigned long index)
1402 {
1403         struct zram *zram;
1404
1405         zram = bdev->bd_disk->private_data;
1406
1407         atomic64_inc(&zram->stats.notify_free);
1408         if (!zram_slot_trylock(zram, index)) {
1409                 atomic64_inc(&zram->stats.miss_free);
1410                 return;
1411         }
1412
1413         zram_free_page(zram, index);
1414         zram_slot_unlock(zram, index);
1415 }
1416
1417 static int zram_rw_page(struct block_device *bdev, sector_t sector,
1418                        struct page *page, unsigned int op)
1419 {
1420         int offset, ret;
1421         u32 index;
1422         struct zram *zram;
1423         struct bio_vec bv;
1424
1425         if (PageTransHuge(page))
1426                 return -ENOTSUPP;
1427         zram = bdev->bd_disk->private_data;
1428
1429         if (!valid_io_request(zram, sector, PAGE_SIZE)) {
1430                 atomic64_inc(&zram->stats.invalid_io);
1431                 ret = -EINVAL;
1432                 goto out;
1433         }
1434
1435         index = sector >> SECTORS_PER_PAGE_SHIFT;
1436         offset = (sector & (SECTORS_PER_PAGE - 1)) << SECTOR_SHIFT;
1437
1438         bv.bv_page = page;
1439         bv.bv_len = PAGE_SIZE;
1440         bv.bv_offset = 0;
1441
1442         ret = zram_bvec_rw(zram, &bv, index, offset, op, NULL);
1443 out:
1444         /*
1445          * If I/O fails, just return error(ie, non-zero) without
1446          * calling page_endio.
1447          * It causes resubmit the I/O with bio request by upper functions
1448          * of rw_page(e.g., swap_readpage, __swap_writepage) and
1449          * bio->bi_end_io does things to handle the error
1450          * (e.g., SetPageError, set_page_dirty and extra works).
1451          */
1452         if (unlikely(ret < 0))
1453                 return ret;
1454
1455         switch (ret) {
1456         case 0:
1457                 page_endio(page, op_is_write(op), 0);
1458                 break;
1459         case 1:
1460                 ret = 0;
1461                 break;
1462         default:
1463                 WARN_ON(1);
1464         }
1465         return ret;
1466 }
1467
1468 static void zram_reset_device(struct zram *zram)
1469 {
1470         struct zcomp *comp;
1471         u64 disksize;
1472
1473         down_write(&zram->init_lock);
1474
1475         zram->limit_pages = 0;
1476
1477         if (!init_done(zram)) {
1478                 up_write(&zram->init_lock);
1479                 return;
1480         }
1481
1482         comp = zram->comp;
1483         disksize = zram->disksize;
1484         zram->disksize = 0;
1485
1486         set_capacity(zram->disk, 0);
1487         part_stat_set_all(&zram->disk->part0, 0);
1488
1489         up_write(&zram->init_lock);
1490         /* I/O operation under all of CPU are done so let's free */
1491         zram_meta_free(zram, disksize);
1492         memset(&zram->stats, 0, sizeof(zram->stats));
1493         zcomp_destroy(comp);
1494         reset_bdev(zram);
1495 }
1496
1497 static ssize_t disksize_store(struct device *dev,
1498                 struct device_attribute *attr, const char *buf, size_t len)
1499 {
1500         u64 disksize;
1501         struct zcomp *comp;
1502         struct zram *zram = dev_to_zram(dev);
1503         int err;
1504
1505         disksize = memparse(buf, NULL);
1506         if (!disksize)
1507                 return -EINVAL;
1508
1509         down_write(&zram->init_lock);
1510         if (init_done(zram)) {
1511                 pr_info("Cannot change disksize for initialized device\n");
1512                 err = -EBUSY;
1513                 goto out_unlock;
1514         }
1515
1516         disksize = PAGE_ALIGN(disksize);
1517         if (!zram_meta_alloc(zram, disksize)) {
1518                 err = -ENOMEM;
1519                 goto out_unlock;
1520         }
1521
1522         comp = zcomp_create(zram->compressor);
1523         if (IS_ERR(comp)) {
1524                 pr_err("Cannot initialise %s compressing backend\n",
1525                                 zram->compressor);
1526                 err = PTR_ERR(comp);
1527                 goto out_free_meta;
1528         }
1529
1530         zram->comp = comp;
1531         zram->disksize = disksize;
1532         set_capacity(zram->disk, zram->disksize >> SECTOR_SHIFT);
1533
1534         revalidate_disk(zram->disk);
1535         up_write(&zram->init_lock);
1536
1537         return len;
1538
1539 out_free_meta:
1540         zram_meta_free(zram, disksize);
1541 out_unlock:
1542         up_write(&zram->init_lock);
1543         return err;
1544 }
1545
1546 static ssize_t reset_store(struct device *dev,
1547                 struct device_attribute *attr, const char *buf, size_t len)
1548 {
1549         int ret;
1550         unsigned short do_reset;
1551         struct zram *zram;
1552         struct block_device *bdev;
1553
1554         ret = kstrtou16(buf, 10, &do_reset);
1555         if (ret)
1556                 return ret;
1557
1558         if (!do_reset)
1559                 return -EINVAL;
1560
1561         zram = dev_to_zram(dev);
1562         bdev = bdget_disk(zram->disk, 0);
1563         if (!bdev)
1564                 return -ENOMEM;
1565
1566         mutex_lock(&bdev->bd_mutex);
1567         /* Do not reset an active device or claimed device */
1568         if (bdev->bd_openers || zram->claim) {
1569                 mutex_unlock(&bdev->bd_mutex);
1570                 bdput(bdev);
1571                 return -EBUSY;
1572         }
1573
1574         /* From now on, anyone can't open /dev/zram[0-9] */
1575         zram->claim = true;
1576         mutex_unlock(&bdev->bd_mutex);
1577
1578         /* Make sure all the pending I/O are finished */
1579         fsync_bdev(bdev);
1580         zram_reset_device(zram);
1581         revalidate_disk(zram->disk);
1582         bdput(bdev);
1583
1584         mutex_lock(&bdev->bd_mutex);
1585         zram->claim = false;
1586         mutex_unlock(&bdev->bd_mutex);
1587
1588         return len;
1589 }
1590
1591 static int zram_open(struct block_device *bdev, fmode_t mode)
1592 {
1593         int ret = 0;
1594         struct zram *zram;
1595
1596         WARN_ON(!mutex_is_locked(&bdev->bd_mutex));
1597
1598         zram = bdev->bd_disk->private_data;
1599         /* zram was claimed to reset so open request fails */
1600         if (zram->claim)
1601                 ret = -EBUSY;
1602
1603         return ret;
1604 }
1605
1606 static const struct block_device_operations zram_devops = {
1607         .open = zram_open,
1608         .swap_slot_free_notify = zram_slot_free_notify,
1609         .rw_page = zram_rw_page,
1610         .owner = THIS_MODULE
1611 };
1612
1613 static DEVICE_ATTR_WO(compact);
1614 static DEVICE_ATTR_RW(disksize);
1615 static DEVICE_ATTR_RO(initstate);
1616 static DEVICE_ATTR_WO(reset);
1617 static DEVICE_ATTR_WO(mem_limit);
1618 static DEVICE_ATTR_WO(mem_used_max);
1619 static DEVICE_ATTR_RW(max_comp_streams);
1620 static DEVICE_ATTR_RW(comp_algorithm);
1621 #ifdef CONFIG_ZRAM_WRITEBACK
1622 static DEVICE_ATTR_RW(backing_dev);
1623 #endif
1624
1625 static struct attribute *zram_disk_attrs[] = {
1626         &dev_attr_disksize.attr,
1627         &dev_attr_initstate.attr,
1628         &dev_attr_reset.attr,
1629         &dev_attr_compact.attr,
1630         &dev_attr_mem_limit.attr,
1631         &dev_attr_mem_used_max.attr,
1632         &dev_attr_max_comp_streams.attr,
1633         &dev_attr_comp_algorithm.attr,
1634 #ifdef CONFIG_ZRAM_WRITEBACK
1635         &dev_attr_backing_dev.attr,
1636 #endif
1637         &dev_attr_io_stat.attr,
1638         &dev_attr_mm_stat.attr,
1639         &dev_attr_debug_stat.attr,
1640         NULL,
1641 };
1642
1643 static const struct attribute_group zram_disk_attr_group = {
1644         .attrs = zram_disk_attrs,
1645 };
1646
1647 static const struct attribute_group *zram_disk_attr_groups[] = {
1648         &zram_disk_attr_group,
1649         NULL,
1650 };
1651
1652 /*
1653  * Allocate and initialize new zram device. the function returns
1654  * '>= 0' device_id upon success, and negative value otherwise.
1655  */
1656 static int zram_add(void)
1657 {
1658         struct zram *zram;
1659         struct request_queue *queue;
1660         int ret, device_id;
1661
1662         zram = kzalloc(sizeof(struct zram), GFP_KERNEL);
1663         if (!zram)
1664                 return -ENOMEM;
1665
1666         ret = idr_alloc(&zram_index_idr, zram, 0, 0, GFP_KERNEL);
1667         if (ret < 0)
1668                 goto out_free_dev;
1669         device_id = ret;
1670
1671         init_rwsem(&zram->init_lock);
1672
1673         queue = blk_alloc_queue(GFP_KERNEL);
1674         if (!queue) {
1675                 pr_err("Error allocating disk queue for device %d\n",
1676                         device_id);
1677                 ret = -ENOMEM;
1678                 goto out_free_idr;
1679         }
1680
1681         blk_queue_make_request(queue, zram_make_request);
1682
1683         /* gendisk structure */
1684         zram->disk = alloc_disk(1);
1685         if (!zram->disk) {
1686                 pr_err("Error allocating disk structure for device %d\n",
1687                         device_id);
1688                 ret = -ENOMEM;
1689                 goto out_free_queue;
1690         }
1691
1692         zram->disk->major = zram_major;
1693         zram->disk->first_minor = device_id;
1694         zram->disk->fops = &zram_devops;
1695         zram->disk->queue = queue;
1696         zram->disk->queue->queuedata = zram;
1697         zram->disk->private_data = zram;
1698         snprintf(zram->disk->disk_name, 16, "zram%d", device_id);
1699
1700         /* Actual capacity set using syfs (/sys/block/zram<id>/disksize */
1701         set_capacity(zram->disk, 0);
1702         /* zram devices sort of resembles non-rotational disks */
1703         blk_queue_flag_set(QUEUE_FLAG_NONROT, zram->disk->queue);
1704         blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, zram->disk->queue);
1705
1706         /*
1707          * To ensure that we always get PAGE_SIZE aligned
1708          * and n*PAGE_SIZED sized I/O requests.
1709          */
1710         blk_queue_physical_block_size(zram->disk->queue, PAGE_SIZE);
1711         blk_queue_logical_block_size(zram->disk->queue,
1712                                         ZRAM_LOGICAL_BLOCK_SIZE);
1713         blk_queue_io_min(zram->disk->queue, PAGE_SIZE);
1714         blk_queue_io_opt(zram->disk->queue, PAGE_SIZE);
1715         zram->disk->queue->limits.discard_granularity = PAGE_SIZE;
1716         blk_queue_max_discard_sectors(zram->disk->queue, UINT_MAX);
1717         blk_queue_flag_set(QUEUE_FLAG_DISCARD, zram->disk->queue);
1718
1719         /*
1720          * zram_bio_discard() will clear all logical blocks if logical block
1721          * size is identical with physical block size(PAGE_SIZE). But if it is
1722          * different, we will skip discarding some parts of logical blocks in
1723          * the part of the request range which isn't aligned to physical block
1724          * size.  So we can't ensure that all discarded logical blocks are
1725          * zeroed.
1726          */
1727         if (ZRAM_LOGICAL_BLOCK_SIZE == PAGE_SIZE)
1728                 blk_queue_max_write_zeroes_sectors(zram->disk->queue, UINT_MAX);
1729
1730         zram->disk->queue->backing_dev_info->capabilities |=
1731                         (BDI_CAP_STABLE_WRITES | BDI_CAP_SYNCHRONOUS_IO);
1732         device_add_disk(NULL, zram->disk, zram_disk_attr_groups);
1733
1734         strlcpy(zram->compressor, default_compressor, sizeof(zram->compressor));
1735
1736         zram_debugfs_register(zram);
1737         pr_info("Added device: %s\n", zram->disk->disk_name);
1738         return device_id;
1739
1740 out_free_queue:
1741         blk_cleanup_queue(queue);
1742 out_free_idr:
1743         idr_remove(&zram_index_idr, device_id);
1744 out_free_dev:
1745         kfree(zram);
1746         return ret;
1747 }
1748
1749 static int zram_remove(struct zram *zram)
1750 {
1751         struct block_device *bdev;
1752
1753         bdev = bdget_disk(zram->disk, 0);
1754         if (!bdev)
1755                 return -ENOMEM;
1756
1757         mutex_lock(&bdev->bd_mutex);
1758         if (bdev->bd_openers || zram->claim) {
1759                 mutex_unlock(&bdev->bd_mutex);
1760                 bdput(bdev);
1761                 return -EBUSY;
1762         }
1763
1764         zram->claim = true;
1765         mutex_unlock(&bdev->bd_mutex);
1766
1767         zram_debugfs_unregister(zram);
1768
1769         /* Make sure all the pending I/O are finished */
1770         fsync_bdev(bdev);
1771         zram_reset_device(zram);
1772         bdput(bdev);
1773
1774         pr_info("Removed device: %s\n", zram->disk->disk_name);
1775
1776         del_gendisk(zram->disk);
1777         blk_cleanup_queue(zram->disk->queue);
1778         put_disk(zram->disk);
1779         kfree(zram);
1780         return 0;
1781 }
1782
1783 /* zram-control sysfs attributes */
1784
1785 /*
1786  * NOTE: hot_add attribute is not the usual read-only sysfs attribute. In a
1787  * sense that reading from this file does alter the state of your system -- it
1788  * creates a new un-initialized zram device and returns back this device's
1789  * device_id (or an error code if it fails to create a new device).
1790  */
1791 static ssize_t hot_add_show(struct class *class,
1792                         struct class_attribute *attr,
1793                         char *buf)
1794 {
1795         int ret;
1796
1797         mutex_lock(&zram_index_mutex);
1798         ret = zram_add();
1799         mutex_unlock(&zram_index_mutex);
1800
1801         if (ret < 0)
1802                 return ret;
1803         return scnprintf(buf, PAGE_SIZE, "%d\n", ret);
1804 }
1805 static struct class_attribute class_attr_hot_add =
1806         __ATTR(hot_add, 0400, hot_add_show, NULL);
1807
1808 static ssize_t hot_remove_store(struct class *class,
1809                         struct class_attribute *attr,
1810                         const char *buf,
1811                         size_t count)
1812 {
1813         struct zram *zram;
1814         int ret, dev_id;
1815
1816         /* dev_id is gendisk->first_minor, which is `int' */
1817         ret = kstrtoint(buf, 10, &dev_id);
1818         if (ret)
1819                 return ret;
1820         if (dev_id < 0)
1821                 return -EINVAL;
1822
1823         mutex_lock(&zram_index_mutex);
1824
1825         zram = idr_find(&zram_index_idr, dev_id);
1826         if (zram) {
1827                 ret = zram_remove(zram);
1828                 if (!ret)
1829                         idr_remove(&zram_index_idr, dev_id);
1830         } else {
1831                 ret = -ENODEV;
1832         }
1833
1834         mutex_unlock(&zram_index_mutex);
1835         return ret ? ret : count;
1836 }
1837 static CLASS_ATTR_WO(hot_remove);
1838
1839 static struct attribute *zram_control_class_attrs[] = {
1840         &class_attr_hot_add.attr,
1841         &class_attr_hot_remove.attr,
1842         NULL,
1843 };
1844 ATTRIBUTE_GROUPS(zram_control_class);
1845
1846 static struct class zram_control_class = {
1847         .name           = "zram-control",
1848         .owner          = THIS_MODULE,
1849         .class_groups   = zram_control_class_groups,
1850 };
1851
1852 static int zram_remove_cb(int id, void *ptr, void *data)
1853 {
1854         zram_remove(ptr);
1855         return 0;
1856 }
1857
1858 static void destroy_devices(void)
1859 {
1860         class_unregister(&zram_control_class);
1861         idr_for_each(&zram_index_idr, &zram_remove_cb, NULL);
1862         zram_debugfs_destroy();
1863         idr_destroy(&zram_index_idr);
1864         unregister_blkdev(zram_major, "zram");
1865         cpuhp_remove_multi_state(CPUHP_ZCOMP_PREPARE);
1866 }
1867
1868 static int __init zram_init(void)
1869 {
1870         int ret;
1871
1872         ret = cpuhp_setup_state_multi(CPUHP_ZCOMP_PREPARE, "block/zram:prepare",
1873                                       zcomp_cpu_up_prepare, zcomp_cpu_dead);
1874         if (ret < 0)
1875                 return ret;
1876
1877         ret = class_register(&zram_control_class);
1878         if (ret) {
1879                 pr_err("Unable to register zram-control class\n");
1880                 cpuhp_remove_multi_state(CPUHP_ZCOMP_PREPARE);
1881                 return ret;
1882         }
1883
1884         zram_debugfs_create();
1885         zram_major = register_blkdev(0, "zram");
1886         if (zram_major <= 0) {
1887                 pr_err("Unable to get major number\n");
1888                 class_unregister(&zram_control_class);
1889                 cpuhp_remove_multi_state(CPUHP_ZCOMP_PREPARE);
1890                 return -EBUSY;
1891         }
1892
1893         while (num_devices != 0) {
1894                 mutex_lock(&zram_index_mutex);
1895                 ret = zram_add();
1896                 mutex_unlock(&zram_index_mutex);
1897                 if (ret < 0)
1898                         goto out_error;
1899                 num_devices--;
1900         }
1901
1902         return 0;
1903
1904 out_error:
1905         destroy_devices();
1906         return ret;
1907 }
1908
1909 static void __exit zram_exit(void)
1910 {
1911         destroy_devices();
1912 }
1913
1914 module_init(zram_init);
1915 module_exit(zram_exit);
1916
1917 module_param(num_devices, uint, 0);
1918 MODULE_PARM_DESC(num_devices, "Number of pre-created zram devices");
1919
1920 MODULE_LICENSE("Dual BSD/GPL");
1921 MODULE_AUTHOR("Nitin Gupta <ngupta@vflare.org>");
1922 MODULE_DESCRIPTION("Compressed RAM Block Device");