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