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