drivers/block/zram/zram_drv.c

   1 /*
   2  * Compressed RAM block device
   3  *
   4  * Copyright (C) 2008, 2009, 2010  Nitin Gupta
   5  *               2012, 2013 Minchan Kim
   6  *
   7  * This code is released using a dual license strategy: BSD/GPL
   8  * You can choose the licence that better fits your requirements.
   9  *
  10  * Released under the terms of 3-clause BSD License
  11  * Released under the terms of GNU General Public License Version 2.0
  12  *
  13  */
  14
  15 #define KMSG_COMPONENT "zram"
  16 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
  17
  18 #include <linux/module.h>
  19 #include <linux/kernel.h>
  20 #include <linux/bio.h>
  21 #include <linux/bitops.h>
  22 #include <linux/blkdev.h>
  23 #include <linux/buffer_head.h>
  24 #include <linux/device.h>
  25 #include <linux/genhd.h>
  26 #include <linux/highmem.h>
  27 #include <linux/slab.h>
  28 #include <linux/backing-dev.h>
  29 #include <linux/string.h>
  30 #include <linux/vmalloc.h>
  31 #include <linux/err.h>
  32 #include <linux/idr.h>
  33 #include <linux/sysfs.h>
  34 #include <linux/debugfs.h>
  35 #include <linux/cpuhotplug.h>
  36
  37 #include "zram_drv.h"
  38
  39 static DEFINE_IDR(zram_index_idr);
  40 /* idr index must be protected */
  41 static DEFINE_MUTEX(zram_index_mutex);
  42
  43 static int zram_major;
  44 static const char *default_compressor = "lzo";
  45
  46 /* Module params (documentation at end) */
  47 static unsigned int num_devices = 1;
  48 /*
  49  * Pages that compress to sizes equals or greater than this are stored
  50  * uncompressed in memory.
  51  */
  52 static size_t huge_class_size;
  53
  54 static void zram_free_page(struct zram *zram, size_t index);
  55
  56 static int zram_slot_trylock(struct zram *zram, u32 index)
  57 {
  58         return bit_spin_trylock(ZRAM_LOCK, &zram->table[index].flags);
  59 }
  60
  61 static void zram_slot_lock(struct zram *zram, u32 index)
  62 {
  63         bit_spin_lock(ZRAM_LOCK, &zram->table[index].flags);
  64 }
  65
  66 static void zram_slot_unlock(struct zram *zram, u32 index)
  67 {
  68         bit_spin_unlock(ZRAM_LOCK, &zram->table[index].flags);
  69 }
  70
  71 static inline bool init_done(struct zram *zram)
  72 {
  73         return zram->disksize;
  74 }
  75
  76 static inline bool zram_allocated(struct zram *zram, u32 index)
  77 {
  78
  79         return (zram->table[index].flags >> (ZRAM_FLAG_SHIFT + 1)) ||
  80                                         zram->table[index].handle;
  81 }
  82
  83 static inline struct zram *dev_to_zram(struct device *dev)
  84 {
  85         return (struct zram *)dev_to_disk(dev)->private_data;
  86 }
  87
  88 static unsigned long zram_get_handle(struct zram *zram, u32 index)
  89 {
  90         return zram->table[index].handle;
  91 }
  92
  93 static void zram_set_handle(struct zram *zram, u32 index, unsigned long handle)
  94 {
  95         zram->table[index].handle = handle;
  96 }
  97
  98 /* flag operations require table entry bit_spin_lock() being held */
  99 static bool zram_test_flag(struct zram *zram, u32 index,
 100                         enum zram_pageflags flag)
 101 {
 102         return zram->table[index].flags & BIT(flag);
 103 }
 104
 105 static void zram_set_flag(struct zram *zram, u32 index,
 106                         enum zram_pageflags flag)
 107 {
 108         zram->table[index].flags |= BIT(flag);
 109 }
 110
 111 static void zram_clear_flag(struct zram *zram, u32 index,
 112                         enum zram_pageflags flag)
 113 {
 114         zram->table[index].flags &= ~BIT(flag);
 115 }
 116
 117 static inline void zram_set_element(struct zram *zram, u32 index,
 118                         unsigned long element)
 119 {
 120         zram->table[index].element = element;
 121 }
 122
 123 static unsigned long zram_get_element(struct zram *zram, u32 index)
 124 {
 125         return zram->table[index].element;
 126 }
 127
 128 static size_t zram_get_obj_size(struct zram *zram, u32 index)
 129 {
 130         return zram->table[index].flags & (BIT(ZRAM_FLAG_SHIFT) - 1);
 131 }
 132
 133 static void zram_set_obj_size(struct zram *zram,
 134                                         u32 index, size_t size)
 135 {
 136         unsigned long flags = zram->table[index].flags >> ZRAM_FLAG_SHIFT;
 137
 138         zram->table[index].flags = (flags << ZRAM_FLAG_SHIFT) | size;
 139 }
 140
 141 #if PAGE_SIZE != 4096
 142 static inline bool is_partial_io(struct bio_vec *bvec)
 143 {
 144         return bvec->bv_len != PAGE_SIZE;
 145 }
 146 #else
 147 static inline bool is_partial_io(struct bio_vec *bvec)
 148 {
 149         return false;
 150 }
 151 #endif
 152
 153 /*
 154  * Check if request is within bounds and aligned on zram logical blocks.
 155  */
 156 static inline bool valid_io_request(struct zram *zram,
 157                 sector_t start, unsigned int size)
 158 {
 159         u64 end, bound;
 160
 161         /* unaligned request */
 162         if (unlikely(start & (ZRAM_SECTOR_PER_LOGICAL_BLOCK - 1)))
 163                 return false;
 164         if (unlikely(size & (ZRAM_LOGICAL_BLOCK_SIZE - 1)))
 165                 return false;
 166
 167         end = start + (size >> SECTOR_SHIFT);
 168         bound = zram->disksize >> SECTOR_SHIFT;
 169         /* out of range range */
 170         if (unlikely(start >= bound || end > bound || start > end))
 171                 return false;
 172
 173         /* I/O request is valid */
 174         return true;
 175 }
 176
 177 static void update_position(u32 *index, int *offset, struct bio_vec *bvec)
 178 {
 179         *index  += (*offset + bvec->bv_len) / PAGE_SIZE;
 180         *offset = (*offset + bvec->bv_len) % PAGE_SIZE;
 181 }
 182
 183 static inline void update_used_max(struct zram *zram,
 184                                         const unsigned long pages)
 185 {
 186         unsigned long old_max, cur_max;
 187
 188         old_max = atomic_long_read(&zram->stats.max_used_pages);
 189
 190         do {
 191                 cur_max = old_max;
 192                 if (pages > cur_max)
 193                         old_max = atomic_long_cmpxchg(
 194                                 &zram->stats.max_used_pages, cur_max, pages);
 195         } while (old_max != cur_max);
 196 }
 197
 198 static inline void zram_fill_page(void *ptr, unsigned long len,
 199                                         unsigned long value)
 200 {
 201         WARN_ON_ONCE(!IS_ALIGNED(len, sizeof(unsigned long)));
 202         memset_l(ptr, value, len / sizeof(unsigned long));
 203 }
 204
 205 static bool page_same_filled(void *ptr, unsigned long *element)
 206 {
 207         unsigned int pos;
 208         unsigned long *page;
 209         unsigned long val;
 210
 211         page = (unsigned long *)ptr;
 212         val = page[0];
 213
 214         for (pos = 1; pos < PAGE_SIZE / sizeof(*page); pos++) {
 215                 if (val != page[pos])
 216                         return false;
 217         }
 218
 219         *element = val;
 220
 221         return true;
 222 }
 223
 224 static ssize_t initstate_show(struct device *dev,
 225                 struct device_attribute *attr, char *buf)
 226 {
 227         u32 val;
 228         struct zram *zram = dev_to_zram(dev);
 229
 230         down_read(&zram->init_lock);
 231         val = init_done(zram);
 232         up_read(&zram->init_lock);
 233
 234         return scnprintf(buf, PAGE_SIZE, "%u\n", val);
 235 }
 236
 237 static ssize_t disksize_show(struct device *dev,
 238                 struct device_attribute *attr, char *buf)
 239 {
 240         struct zram *zram = dev_to_zram(dev);
 241
 242         return scnprintf(buf, PAGE_SIZE, "%llu\n", zram->disksize);
 243 }
 244
 245 static ssize_t mem_limit_store(struct device *dev,
 246                 struct device_attribute *attr, const char *buf, size_t len)
 247 {
 248         u64 limit;
 249         char *tmp;
 250         struct zram *zram = dev_to_zram(dev);
 251
 252         limit = memparse(buf, &tmp);
 253         if (buf == tmp) /* no chars parsed, invalid input */
 254                 return -EINVAL;
 255
 256         down_write(&zram->init_lock);
 257         zram->limit_pages = PAGE_ALIGN(limit) >> PAGE_SHIFT;
 258         up_write(&zram->init_lock);
 259
 260         return len;
 261 }
 262
 263 static ssize_t mem_used_max_store(struct device *dev,
 264                 struct device_attribute *attr, const char *buf, size_t len)
 265 {
 266         int err;
 267         unsigned long val;
 268         struct zram *zram = dev_to_zram(dev);
 269
 270         err = kstrtoul(buf, 10, &val);
 271         if (err || val != 0)
 272                 return -EINVAL;
 273
 274         down_read(&zram->init_lock);
 275         if (init_done(zram)) {
 276                 atomic_long_set(&zram->stats.max_used_pages,
 277                                 zs_get_total_pages(zram->mem_pool));
 278         }
 279         up_read(&zram->init_lock);
 280
 281         return len;
 282 }
 283
 284 static ssize_t idle_store(struct device *dev,
 285                 struct device_attribute *attr, const char *buf, size_t len)
 286 {
 287         struct zram *zram = dev_to_zram(dev);
 288         unsigned long nr_pages = zram->disksize >> PAGE_SHIFT;
 289         int index;
 290         char mode_buf[8];
 291         ssize_t sz;
 292
 293         sz = strscpy(mode_buf, buf, sizeof(mode_buf));
 294         if (sz <= 0)
 295                 return -EINVAL;
 296
 297         /* ignore trailing new line */
 298         if (mode_buf[sz - 1] == '\n')
 299                 mode_buf[sz - 1] = 0x00;
 300
 301         if (strcmp(mode_buf, "all"))
 302                 return -EINVAL;
 303
 304         down_read(&zram->init_lock);
 305         if (!init_done(zram)) {
 306                 up_read(&zram->init_lock);
 307                 return -EINVAL;
 308         }
 309
 310         for (index = 0; index < nr_pages; index++) {
 311                 zram_slot_lock(zram, index);
 312                 if (!zram_allocated(zram, index))
 313                         goto next;
 314
 315                 zram_set_flag(zram, index, ZRAM_IDLE);
 316 next:
 317                 zram_slot_unlock(zram, index);
 318         }
 319
 320         up_read(&zram->init_lock);
 321
 322         return len;
 323 }
 324
 325 #ifdef CONFIG_ZRAM_WRITEBACK
 326 static void reset_bdev(struct zram *zram)
 327 {
 328         struct block_device *bdev;
 329
 330         if (!zram->backing_dev)
 331                 return;
 332
 333         bdev = zram->bdev;
 334         if (zram->old_block_size)
 335                 set_blocksize(bdev, zram->old_block_size);
 336         blkdev_put(bdev, FMODE_READ|FMODE_WRITE|FMODE_EXCL);
 337         /* hope filp_close flush all of IO */
 338         filp_close(zram->backing_dev, NULL);
 339         zram->backing_dev = NULL;
 340         zram->old_block_size = 0;
 341         zram->bdev = NULL;
 342         zram->disk->queue->backing_dev_info->capabilities |=
 343                                 BDI_CAP_SYNCHRONOUS_IO;
 344         kvfree(zram->bitmap);
 345         zram->bitmap = NULL;
 346 }
 347
 348 static ssize_t backing_dev_show(struct device *dev,
 349                 struct device_attribute *attr, char *buf)
 350 {
 351         struct zram *zram = dev_to_zram(dev);
 352         struct file *file = zram->backing_dev;
 353         char *p;
 354         ssize_t ret;
 355
 356         down_read(&zram->init_lock);
 357         if (!zram->backing_dev) {
 358                 memcpy(buf, "none\n", 5);
 359                 up_read(&zram->init_lock);
 360                 return 5;
 361         }
 362
 363         p = file_path(file, buf, PAGE_SIZE - 1);
 364         if (IS_ERR(p)) {
 365                 ret = PTR_ERR(p);
 366                 goto out;
 367         }
 368
 369         ret = strlen(p);
 370         memmove(buf, p, ret);
 371         buf[ret++] = '\n';
 372 out:
 373         up_read(&zram->init_lock);
 374         return ret;
 375 }
 376
 377 static ssize_t backing_dev_store(struct device *dev,
 378                 struct device_attribute *attr, const char *buf, size_t len)
 379 {
 380         char *file_name;
 381         size_t sz;
 382         struct file *backing_dev = NULL;
 383         struct inode *inode;
 384         struct address_space *mapping;
 385         unsigned int bitmap_sz, old_block_size = 0;
 386         unsigned long nr_pages, *bitmap = NULL;
 387         struct block_device *bdev = NULL;
 388         int err;
 389         struct zram *zram = dev_to_zram(dev);
 390
 391         file_name = kmalloc(PATH_MAX, GFP_KERNEL);
 392         if (!file_name)
 393                 return -ENOMEM;
 394
 395         down_write(&zram->init_lock);
 396         if (init_done(zram)) {
 397                 pr_info("Can't setup backing device for initialized device\n");
 398                 err = -EBUSY;
 399                 goto out;
 400         }
 401
 402         strlcpy(file_name, buf, PATH_MAX);
 403         /* ignore trailing newline */
 404         sz = strlen(file_name);
 405         if (sz > 0 && file_name[sz - 1] == '\n')
 406                 file_name[sz - 1] = 0x00;
 407
 408         backing_dev = filp_open(file_name, O_RDWR|O_LARGEFILE, 0);
 409         if (IS_ERR(backing_dev)) {
 410                 err = PTR_ERR(backing_dev);
 411                 backing_dev = NULL;
 412                 goto out;
 413         }
 414
 415         mapping = backing_dev->f_mapping;
 416         inode = mapping->host;
 417
 418         /* Support only block device in this moment */
 419         if (!S_ISBLK(inode->i_mode)) {
 420                 err = -ENOTBLK;
 421                 goto out;
 422         }
 423
 424         bdev = bdgrab(I_BDEV(inode));
 425         err = blkdev_get(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL, zram);
 426         if (err < 0) {
 427                 bdev = NULL;
 428                 goto out;
 429         }
 430
 431         nr_pages = i_size_read(inode) >> PAGE_SHIFT;
 432         bitmap_sz = BITS_TO_LONGS(nr_pages) * sizeof(long);
 433         bitmap = kvzalloc(bitmap_sz, GFP_KERNEL);
 434         if (!bitmap) {
 435                 err = -ENOMEM;
 436                 goto out;
 437         }
 438
 439         old_block_size = block_size(bdev);
 440         err = set_blocksize(bdev, PAGE_SIZE);
 441         if (err)
 442                 goto out;
 443
 444         reset_bdev(zram);
 445
 446         zram->old_block_size = old_block_size;
 447         zram->bdev = bdev;
 448         zram->backing_dev = backing_dev;
 449         zram->bitmap = bitmap;
 450         zram->nr_pages = nr_pages;
 451         /*
 452          * With writeback feature, zram does asynchronous IO so it's no longer
 453          * synchronous device so let's remove synchronous io flag. Othewise,
 454          * upper layer(e.g., swap) could wait IO completion rather than
 455          * (submit and return), which will cause system sluggish.
 456          * Furthermore, when the IO function returns(e.g., swap_readpage),
 457          * upper layer expects IO was done so it could deallocate the page
 458          * freely but in fact, IO is going on so finally could cause
 459          * use-after-free when the IO is really done.
 460          */
 461         zram->disk->queue->backing_dev_info->capabilities &=
 462                         ~BDI_CAP_SYNCHRONOUS_IO;
 463         up_write(&zram->init_lock);
 464
 465         pr_info("setup backing device %s\n", file_name);
 466         kfree(file_name);
 467
 468         return len;
 469 out:
 470         if (bitmap)
 471                 kvfree(bitmap);
 472
 473         if (bdev)
 474                 blkdev_put(bdev, FMODE_READ | FMODE_WRITE | FMODE_EXCL);
 475
 476         if (backing_dev)
 477                 filp_close(backing_dev, NULL);
 478
 479         up_write(&zram->init_lock);
 480
 481         kfree(file_name);
 482
 483         return err;
 484 }
 485
 486 static unsigned long alloc_block_bdev(struct zram *zram)
 487 {
 488         unsigned long blk_idx = 1;
 489 retry:
 490         /* skip 0 bit to confuse zram.handle = 0 */
 491         blk_idx = find_next_zero_bit(zram->bitmap, zram->nr_pages, blk_idx);
 492         if (blk_idx == zram->nr_pages)
 493                 return 0;
 494
 495         if (test_and_set_bit(blk_idx, zram->bitmap))
 496                 goto retry;
 497
 498         return blk_idx;
 499 }
 500
 501 static void free_block_bdev(struct zram *zram, unsigned long blk_idx)
 502 {
 503         int was_set;
 504
 505         was_set = test_and_clear_bit(blk_idx, zram->bitmap);
 506         WARN_ON_ONCE(!was_set);
 507 }
 508
 509 static void zram_page_end_io(struct bio *bio)
 510 {
 511         struct page *page = bio_first_page_all(bio);
 512
 513         page_endio(page, op_is_write(bio_op(bio)),
 514                         blk_status_to_errno(bio->bi_status));
 515         bio_put(bio);
 516 }
 517
 518 /*
 519  * Returns 1 if the submission is successful.
 520  */
 521 static int read_from_bdev_async(struct zram *zram, struct bio_vec *bvec,
 522                         unsigned long entry, struct bio *parent)
 523 {
 524         struct bio *bio;
 525
 526         bio = bio_alloc(GFP_ATOMIC, 1);
 527         if (!bio)
 528                 return -ENOMEM;
 529
 530         bio->bi_iter.bi_sector = entry * (PAGE_SIZE >> 9);
 531         bio_set_dev(bio, zram->bdev);
 532         if (!bio_add_page(bio, bvec->bv_page, bvec->bv_len, bvec->bv_offset)) {
 533                 bio_put(bio);
 534                 return -EIO;
 535         }
 536
 537         if (!parent) {
 538                 bio->bi_opf = REQ_OP_READ;
 539                 bio->bi_end_io = zram_page_end_io;
 540         } else {
 541                 bio->bi_opf = parent->bi_opf;
 542                 bio_chain(bio, parent);
 543         }
 544
 545         submit_bio(bio);
 546         return 1;
 547 }
 548
 549 struct zram_work {
 550         struct work_struct work;
 551         struct zram *zram;
 552         unsigned long entry;
 553         struct bio *bio;
 554 };
 555
 556 #if PAGE_SIZE != 4096
 557 static void zram_sync_read(struct work_struct *work)
 558 {
 559         struct bio_vec bvec;
 560         struct zram_work *zw = container_of(work, struct zram_work, work);
 561         struct zram *zram = zw->zram;
 562         unsigned long entry = zw->entry;
 563         struct bio *bio = zw->bio;
 564
 565         read_from_bdev_async(zram, &bvec, entry, bio);
 566 }
 567
 568 /*
 569  * Block layer want one ->make_request_fn to be active at a time
 570  * so if we use chained IO with parent IO in same context,
 571  * it's a deadlock. To avoid, it, it uses worker thread context.
 572  */
 573 static int read_from_bdev_sync(struct zram *zram, struct bio_vec *bvec,
 574                                 unsigned long entry, struct bio *bio)
 575 {
 576         struct zram_work work;
 577
 578         work.zram = zram;
 579         work.entry = entry;
 580         work.bio = bio;
 581
 582         INIT_WORK_ONSTACK(&work.work, zram_sync_read);
 583         queue_work(system_unbound_wq, &work.work);
 584         flush_work(&work.work);
 585         destroy_work_on_stack(&work.work);
 586
 587         return 1;
 588 }
 589 #else
 590 static int read_from_bdev_sync(struct zram *zram, struct bio_vec *bvec,
 591                                 unsigned long entry, struct bio *bio)
 592 {
 593         WARN_ON(1);
 594         return -EIO;
 595 }
 596 #endif
 597
 598 static int read_from_bdev(struct zram *zram, struct bio_vec *bvec,
 599                         unsigned long entry, struct bio *parent, bool sync)
 600 {
 601         if (sync)
 602                 return read_from_bdev_sync(zram, bvec, entry, parent);
 603         else
 604                 return read_from_bdev_async(zram, bvec, entry, parent);
 605 }
 606
 607 static int write_to_bdev(struct zram *zram, struct bio_vec *bvec,
 608                                         u32 index, struct bio *parent,
 609                                         unsigned long *pentry)
 610 {
 611         struct bio *bio;
 612         unsigned long entry;
 613
 614         bio = bio_alloc(GFP_ATOMIC, 1);
 615         if (!bio)
 616                 return -ENOMEM;
 617
 618         entry = alloc_block_bdev(zram);
 619         if (!entry) {
 620                 bio_put(bio);
 621                 return -ENOSPC;
 622         }
 623
 624         bio->bi_iter.bi_sector = entry * (PAGE_SIZE >> 9);
 625         bio_set_dev(bio, zram->bdev);
 626         if (!bio_add_page(bio, bvec->bv_page, bvec->bv_len,
 627                                         bvec->bv_offset)) {
 628                 bio_put(bio);
 629                 free_block_bdev(zram, entry);
 630                 return -EIO;
 631         }
 632
 633         if (!parent) {
 634                 bio->bi_opf = REQ_OP_WRITE | REQ_SYNC;
 635                 bio->bi_end_io = zram_page_end_io;
 636         } else {
 637                 bio->bi_opf = parent->bi_opf;
 638                 bio_chain(bio, parent);
 639         }
 640
 641         submit_bio(bio);
 642         *pentry = entry;
 643
 644         return 0;
 645 }
 646
 647 #else
 648 static inline void reset_bdev(struct zram *zram) {};
 649 static int write_to_bdev(struct zram *zram, struct bio_vec *bvec,
 650                                         u32 index, struct bio *parent,
 651                                         unsigned long *pentry)
 652
 653 {
 654         return -EIO;
 655 }
 656
 657 static int read_from_bdev(struct zram *zram, struct bio_vec *bvec,
 658                         unsigned long entry, struct bio *parent, bool sync)
 659 {
 660         return -EIO;
 661 }
 662
 663 static void free_block_bdev(struct zram *zram, unsigned long blk_idx) {};
 664 #endif
 665
 666 #ifdef CONFIG_ZRAM_MEMORY_TRACKING
 667
 668 static struct dentry *zram_debugfs_root;
 669
 670 static void zram_debugfs_create(void)
 671 {
 672         zram_debugfs_root = debugfs_create_dir("zram", NULL);
 673 }
 674
 675 static void zram_debugfs_destroy(void)
 676 {
 677         debugfs_remove_recursive(zram_debugfs_root);
 678 }
 679
 680 static void zram_accessed(struct zram *zram, u32 index)
 681 {
 682         zram_clear_flag(zram, index, ZRAM_IDLE);
 683         zram->table[index].ac_time = ktime_get_boottime();
 684 }
 685
 686 static ssize_t read_block_state(struct file *file, char __user *buf,
 687                                 size_t count, loff_t *ppos)
 688 {
 689         char *kbuf;
 690         ssize_t index, written = 0;
 691         struct zram *zram = file->private_data;
 692         unsigned long nr_pages = zram->disksize >> PAGE_SHIFT;
 693         struct timespec64 ts;
 694
 695         kbuf = kvmalloc(count, GFP_KERNEL);
 696         if (!kbuf)
 697                 return -ENOMEM;
 698
 699         down_read(&zram->init_lock);
 700         if (!init_done(zram)) {
 701                 up_read(&zram->init_lock);
 702                 kvfree(kbuf);
 703                 return -EINVAL;
 704         }
 705
 706         for (index = *ppos; index < nr_pages; index++) {
 707                 int copied;
 708
 709                 zram_slot_lock(zram, index);
 710                 if (!zram_allocated(zram, index))
 711                         goto next;
 712
 713                 ts = ktime_to_timespec64(zram->table[index].ac_time);
 714                 copied = snprintf(kbuf + written, count,
 715                         "%12zd %12lld.%06lu %c%c%c%c\n",
 716                         index, (s64)ts.tv_sec,
 717                         ts.tv_nsec / NSEC_PER_USEC,
 718                         zram_test_flag(zram, index, ZRAM_SAME) ? 's' : '.',
 719                         zram_test_flag(zram, index, ZRAM_WB) ? 'w' : '.',
 720                         zram_test_flag(zram, index, ZRAM_HUGE) ? 'h' : '.',
 721                         zram_test_flag(zram, index, ZRAM_IDLE) ? 'i' : '.');
 722
 723                 if (count < copied) {
 724                         zram_slot_unlock(zram, index);
 725                         break;
 726                 }
 727                 written += copied;
 728                 count -= copied;
 729 next:
 730                 zram_slot_unlock(zram, index);
 731                 *ppos += 1;
 732         }
 733
 734         up_read(&zram->init_lock);
 735         if (copy_to_user(buf, kbuf, written))
 736                 written = -EFAULT;
 737         kvfree(kbuf);
 738
 739         return written;
 740 }
 741
 742 static const struct file_operations proc_zram_block_state_op = {
 743         .open = simple_open,
 744         .read = read_block_state,
 745         .llseek = default_llseek,
 746 };
 747
 748 static void zram_debugfs_register(struct zram *zram)
 749 {
 750         if (!zram_debugfs_root)
 751                 return;
 752
 753         zram->debugfs_dir = debugfs_create_dir(zram->disk->disk_name,
 754                                                 zram_debugfs_root);
 755         debugfs_create_file("block_state", 0400, zram->debugfs_dir,
 756                                 zram, &proc_zram_block_state_op);
 757 }
 758
 759 static void zram_debugfs_unregister(struct zram *zram)
 760 {
 761         debugfs_remove_recursive(zram->debugfs_dir);
 762 }
 763 #else
 764 static void zram_debugfs_create(void) {};
 765 static void zram_debugfs_destroy(void) {};
 766 static void zram_accessed(struct zram *zram, u32 index)
 767 {
 768         zram_clear_flag(zram, index, ZRAM_IDLE);
 769 };
 770 static void zram_debugfs_register(struct zram *zram) {};
 771 static void zram_debugfs_unregister(struct zram *zram) {};
 772 #endif
 773
 774 /*
 775  * We switched to per-cpu streams and this attr is not needed anymore.
 776  * However, we will keep it around for some time, because:
 777  * a) we may revert per-cpu streams in the future
 778  * b) it's visible to user space and we need to follow our 2 years
 779  *    retirement rule; but we already have a number of 'soon to be
 780  *    altered' attrs, so max_comp_streams need to wait for the next
 781  *    layoff cycle.
 782  */
 783 static ssize_t max_comp_streams_show(struct device *dev,
 784                 struct device_attribute *attr, char *buf)
 785 {
 786         return scnprintf(buf, PAGE_SIZE, "%d\n", num_online_cpus());
 787 }
 788
 789 static ssize_t max_comp_streams_store(struct device *dev,
 790                 struct device_attribute *attr, const char *buf, size_t len)
 791 {
 792         return len;
 793 }
 794
 795 static ssize_t comp_algorithm_show(struct device *dev,
 796                 struct device_attribute *attr, char *buf)
 797 {
 798         size_t sz;
 799         struct zram *zram = dev_to_zram(dev);
 800
 801         down_read(&zram->init_lock);
 802         sz = zcomp_available_show(zram->compressor, buf);
 803         up_read(&zram->init_lock);
 804
 805         return sz;
 806 }
 807
 808 static ssize_t comp_algorithm_store(struct device *dev,
 809                 struct device_attribute *attr, const char *buf, size_t len)
 810 {
 811         struct zram *zram = dev_to_zram(dev);
 812         char compressor[ARRAY_SIZE(zram->compressor)];
 813         size_t sz;
 814
 815         strlcpy(compressor, buf, sizeof(compressor));
 816         /* ignore trailing newline */
 817         sz = strlen(compressor);
 818         if (sz > 0 && compressor[sz - 1] == '\n')
 819                 compressor[sz - 1] = 0x00;
 820
 821         if (!zcomp_available_algorithm(compressor))
 822                 return -EINVAL;
 823
 824         down_write(&zram->init_lock);
 825         if (init_done(zram)) {
 826                 up_write(&zram->init_lock);
 827                 pr_info("Can't change algorithm for initialized device\n");
 828                 return -EBUSY;
 829         }
 830
 831         strcpy(zram->compressor, compressor);
 832         up_write(&zram->init_lock);
 833         return len;
 834 }
 835
 836 static ssize_t compact_store(struct device *dev,
 837                 struct device_attribute *attr, const char *buf, size_t len)
 838 {
 839         struct zram *zram = dev_to_zram(dev);
 840
 841         down_read(&zram->init_lock);
 842         if (!init_done(zram)) {
 843                 up_read(&zram->init_lock);
 844                 return -EINVAL;
 845         }
 846
 847         zs_compact(zram->mem_pool);
 848         up_read(&zram->init_lock);
 849
 850         return len;
 851 }
 852
 853 static ssize_t io_stat_show(struct device *dev,
 854                 struct device_attribute *attr, char *buf)
 855 {
 856         struct zram *zram = dev_to_zram(dev);
 857         ssize_t ret;
 858
 859         down_read(&zram->init_lock);
 860         ret = scnprintf(buf, PAGE_SIZE,
 861                         "%8llu %8llu %8llu %8llu\n",
 862                         (u64)atomic64_read(&zram->stats.failed_reads),
 863                         (u64)atomic64_read(&zram->stats.failed_writes),
 864                         (u64)atomic64_read(&zram->stats.invalid_io),
 865                         (u64)atomic64_read(&zram->stats.notify_free));
 866         up_read(&zram->init_lock);
 867
 868         return ret;
 869 }
 870
 871 static ssize_t mm_stat_show(struct device *dev,
 872                 struct device_attribute *attr, char *buf)
 873 {
 874         struct zram *zram = dev_to_zram(dev);
 875         struct zs_pool_stats pool_stats;
 876         u64 orig_size, mem_used = 0;
 877         long max_used;
 878         ssize_t ret;
 879
 880         memset(&pool_stats, 0x00, sizeof(struct zs_pool_stats));
 881
 882         down_read(&zram->init_lock);
 883         if (init_done(zram)) {
 884                 mem_used = zs_get_total_pages(zram->mem_pool);
 885                 zs_pool_stats(zram->mem_pool, &pool_stats);
 886         }
 887
 888         orig_size = atomic64_read(&zram->stats.pages_stored);
 889         max_used = atomic_long_read(&zram->stats.max_used_pages);
 890
 891         ret = scnprintf(buf, PAGE_SIZE,
 892                         "%8llu %8llu %8llu %8lu %8ld %8llu %8lu %8llu\n",
 893                         orig_size << PAGE_SHIFT,
 894                         (u64)atomic64_read(&zram->stats.compr_data_size),
 895                         mem_used << PAGE_SHIFT,
 896                         zram->limit_pages << PAGE_SHIFT,
 897                         max_used << PAGE_SHIFT,
 898                         (u64)atomic64_read(&zram->stats.same_pages),
 899                         pool_stats.pages_compacted,
 900                         (u64)atomic64_read(&zram->stats.huge_pages));
 901         up_read(&zram->init_lock);
 902
 903         return ret;
 904 }
 905
 906 static ssize_t debug_stat_show(struct device *dev,
 907                 struct device_attribute *attr, char *buf)
 908 {
 909         int version = 1;
 910         struct zram *zram = dev_to_zram(dev);
 911         ssize_t ret;
 912
 913         down_read(&zram->init_lock);
 914         ret = scnprintf(buf, PAGE_SIZE,
 915                         "version: %d\n%8llu %8llu\n",
 916                         version,
 917                         (u64)atomic64_read(&zram->stats.writestall),
 918                         (u64)atomic64_read(&zram->stats.miss_free));
 919         up_read(&zram->init_lock);
 920
 921         return ret;
 922 }
 923
 924 static DEVICE_ATTR_RO(io_stat);
 925 static DEVICE_ATTR_RO(mm_stat);
 926 static DEVICE_ATTR_RO(debug_stat);
 927
 928 static void zram_meta_free(struct zram *zram, u64 disksize)
 929 {
 930         size_t num_pages = disksize >> PAGE_SHIFT;
 931         size_t index;
 932
 933         /* Free all pages that are still in this zram device */
 934         for (index = 0; index < num_pages; index++)
 935                 zram_free_page(zram, index);
 936
 937         zs_destroy_pool(zram->mem_pool);
 938         vfree(zram->table);
 939 }
 940
 941 static bool zram_meta_alloc(struct zram *zram, u64 disksize)
 942 {
 943         size_t num_pages;
 944
 945         num_pages = disksize >> PAGE_SHIFT;
 946         zram->table = vzalloc(array_size(num_pages, sizeof(*zram->table)));
 947         if (!zram->table)
 948                 return false;
 949
 950         zram->mem_pool = zs_create_pool(zram->disk->disk_name);
 951         if (!zram->mem_pool) {
 952                 vfree(zram->table);
 953                 return false;
 954         }
 955
 956         if (!huge_class_size)
 957                 huge_class_size = zs_huge_class_size(zram->mem_pool);
 958         return true;
 959 }
 960
 961 /*
 962  * To protect concurrent access to the same index entry,
 963  * caller should hold this table index entry's bit_spinlock to
 964  * indicate this index entry is accessing.
 965  */
 966 static void zram_free_page(struct zram *zram, size_t index)
 967 {
 968         unsigned long handle;
 969
 970 #ifdef CONFIG_ZRAM_MEMORY_TRACKING
 971         zram->table[index].ac_time = 0;
 972 #endif
 973         if (zram_test_flag(zram, index, ZRAM_IDLE))
 974                 zram_clear_flag(zram, index, ZRAM_IDLE);
 975
 976         if (zram_test_flag(zram, index, ZRAM_HUGE)) {
 977                 zram_clear_flag(zram, index, ZRAM_HUGE);
 978                 atomic64_dec(&zram->stats.huge_pages);
 979         }
 980
 981         if (zram_test_flag(zram, index, ZRAM_WB)) {
 982                 zram_clear_flag(zram, index, ZRAM_WB);
 983                 free_block_bdev(zram, zram_get_element(zram, index));
 984                 goto out;
 985         }
 986
 987         /*
 988          * No memory is allocated for same element filled pages.
 989          * Simply clear same page flag.
 990          */
 991         if (zram_test_flag(zram, index, ZRAM_SAME)) {
 992                 zram_clear_flag(zram, index, ZRAM_SAME);
 993                 atomic64_dec(&zram->stats.same_pages);
 994                 goto out;
 995         }
 996
 997         handle = zram_get_handle(zram, index);
 998         if (!handle)
 999                 return;
1000
1001         zs_free(zram->mem_pool, handle);
1002
1003         atomic64_sub(zram_get_obj_size(zram, index),
1004                         &zram->stats.compr_data_size);
1005 out:
1006         atomic64_dec(&zram->stats.pages_stored);
1007         zram_set_handle(zram, index, 0);
1008         zram_set_obj_size(zram, index, 0);
1009         WARN_ON_ONCE(zram->table[index].flags & ~(1UL << ZRAM_LOCK));
1010 }
1011
1012 static int __zram_bvec_read(struct zram *zram, struct page *page, u32 index,
1013                                 struct bio *bio, bool partial_io)
1014 {
1015         int ret;
1016         unsigned long handle;
1017         unsigned int size;
1018         void *src, *dst;
1019
1020         zram_slot_lock(zram, index);
1021         if (zram_test_flag(zram, index, ZRAM_WB)) {
1022                 struct bio_vec bvec;
1023
1024                 zram_slot_unlock(zram, index);
1025
1026                 bvec.bv_page = page;
1027                 bvec.bv_len = PAGE_SIZE;
1028                 bvec.bv_offset = 0;
1029                 return read_from_bdev(zram, &bvec,
1030                                 zram_get_element(zram, index),
1031                                 bio, partial_io);
1032         }
1033
1034         handle = zram_get_handle(zram, index);
1035         if (!handle || zram_test_flag(zram, index, ZRAM_SAME)) {
1036                 unsigned long value;
1037                 void *mem;
1038
1039                 value = handle ? zram_get_element(zram, index) : 0;
1040                 mem = kmap_atomic(page);
1041                 zram_fill_page(mem, PAGE_SIZE, value);
1042                 kunmap_atomic(mem);
1043                 zram_slot_unlock(zram, index);
1044                 return 0;
1045         }
1046
1047         size = zram_get_obj_size(zram, index);
1048
1049         src = zs_map_object(zram->mem_pool, handle, ZS_MM_RO);
1050         if (size == PAGE_SIZE) {
1051                 dst = kmap_atomic(page);
1052                 memcpy(dst, src, PAGE_SIZE);
1053                 kunmap_atomic(dst);
1054                 ret = 0;
1055         } else {
1056                 struct zcomp_strm *zstrm = zcomp_stream_get(zram->comp);
1057
1058                 dst = kmap_atomic(page);
1059                 ret = zcomp_decompress(zstrm, src, size, dst);
1060                 kunmap_atomic(dst);
1061                 zcomp_stream_put(zram->comp);
1062         }
1063         zs_unmap_object(zram->mem_pool, handle);
1064         zram_slot_unlock(zram, index);
1065
1066         /* Should NEVER happen. Return bio error if it does. */
1067         if (unlikely(ret))
1068                 pr_err("Decompression failed! err=%d, page=%u\n", ret, index);
1069
1070         return ret;
1071 }
1072
1073 static int zram_bvec_read(struct zram *zram, struct bio_vec *bvec,
1074                                 u32 index, int offset, struct bio *bio)
1075 {
1076         int ret;
1077         struct page *page;
1078
1079         page = bvec->bv_page;
1080         if (is_partial_io(bvec)) {
1081                 /* Use a temporary buffer to decompress the page */
1082                 page = alloc_page(GFP_NOIO|__GFP_HIGHMEM);
1083                 if (!page)
1084                         return -ENOMEM;
1085         }
1086
1087         ret = __zram_bvec_read(zram, page, index, bio, is_partial_io(bvec));
1088         if (unlikely(ret))
1089                 goto out;
1090
1091         if (is_partial_io(bvec)) {
1092                 void *dst = kmap_atomic(bvec->bv_page);
1093                 void *src = kmap_atomic(page);
1094
1095                 memcpy(dst + bvec->bv_offset, src + offset, bvec->bv_len);
1096                 kunmap_atomic(src);
1097                 kunmap_atomic(dst);
1098         }
1099 out:
1100         if (is_partial_io(bvec))
1101                 __free_page(page);
1102
1103         return ret;
1104 }
1105
1106 static int __zram_bvec_write(struct zram *zram, struct bio_vec *bvec,
1107                                 u32 index, struct bio *bio)
1108 {
1109         int ret = 0;
1110         unsigned long alloced_pages;
1111         unsigned long handle = 0;
1112         unsigned int comp_len = 0;
1113         void *src, *dst, *mem;
1114         struct zcomp_strm *zstrm;
1115         struct page *page = bvec->bv_page;
1116         unsigned long element = 0;
1117         enum zram_pageflags flags = 0;
1118         bool allow_wb = true;
1119
1120         mem = kmap_atomic(page);
1121         if (page_same_filled(mem, &element)) {
1122                 kunmap_atomic(mem);
1123                 /* Free memory associated with this sector now. */
1124                 flags = ZRAM_SAME;
1125                 atomic64_inc(&zram->stats.same_pages);
1126                 goto out;
1127         }
1128         kunmap_atomic(mem);
1129
1130 compress_again:
1131         zstrm = zcomp_stream_get(zram->comp);
1132         src = kmap_atomic(page);
1133         ret = zcomp_compress(zstrm, src, &comp_len);
1134         kunmap_atomic(src);
1135
1136         if (unlikely(ret)) {
1137                 zcomp_stream_put(zram->comp);
1138                 pr_err("Compression failed! err=%d\n", ret);
1139                 zs_free(zram->mem_pool, handle);
1140                 return ret;
1141         }
1142
1143         if (unlikely(comp_len >= huge_class_size)) {
1144                 comp_len = PAGE_SIZE;
1145                 if (zram->backing_dev && allow_wb) {
1146                         zcomp_stream_put(zram->comp);
1147                         ret = write_to_bdev(zram, bvec, index, bio, &element);
1148                         if (!ret) {
1149                                 flags = ZRAM_WB;
1150                                 ret = 1;
1151                                 goto out;
1152                         }
1153                         allow_wb = false;
1154                         goto compress_again;
1155                 }
1156         }
1157
1158         /*
1159          * handle allocation has 2 paths:
1160          * a) fast path is executed with preemption disabled (for
1161          *  per-cpu streams) and has __GFP_DIRECT_RECLAIM bit clear,
1162          *  since we can't sleep;
1163          * b) slow path enables preemption and attempts to allocate
1164          *  the page with __GFP_DIRECT_RECLAIM bit set. we have to
1165          *  put per-cpu compression stream and, thus, to re-do
1166          *  the compression once handle is allocated.
1167          *
1168          * if we have a 'non-null' handle here then we are coming
1169          * from the slow path and handle has already been allocated.
1170          */
1171         if (!handle)
1172                 handle = zs_malloc(zram->mem_pool, comp_len,
1173                                 __GFP_KSWAPD_RECLAIM |
1174                                 __GFP_NOWARN |
1175                                 __GFP_HIGHMEM |
1176                                 __GFP_MOVABLE);
1177         if (!handle) {
1178                 zcomp_stream_put(zram->comp);
1179                 atomic64_inc(&zram->stats.writestall);
1180                 handle = zs_malloc(zram->mem_pool, comp_len,
1181                                 GFP_NOIO | __GFP_HIGHMEM |
1182                                 __GFP_MOVABLE);
1183                 if (handle)
1184                         goto compress_again;
1185                 return -ENOMEM;
1186         }
1187
1188         alloced_pages = zs_get_total_pages(zram->mem_pool);
1189         update_used_max(zram, alloced_pages);
1190
1191         if (zram->limit_pages && alloced_pages > zram->limit_pages) {
1192                 zcomp_stream_put(zram->comp);
1193                 zs_free(zram->mem_pool, handle);
1194                 return -ENOMEM;
1195         }
1196
1197         dst = zs_map_object(zram->mem_pool, handle, ZS_MM_WO);
1198
1199         src = zstrm->buffer;
1200         if (comp_len == PAGE_SIZE)
1201                 src = kmap_atomic(page);
1202         memcpy(dst, src, comp_len);
1203         if (comp_len == PAGE_SIZE)
1204                 kunmap_atomic(src);
1205
1206         zcomp_stream_put(zram->comp);
1207         zs_unmap_object(zram->mem_pool, handle);
1208         atomic64_add(comp_len, &zram->stats.compr_data_size);
1209 out:
1210         /*
1211          * Free memory associated with this sector
1212          * before overwriting unused sectors.
1213          */
1214         zram_slot_lock(zram, index);
1215         zram_free_page(zram, index);
1216
1217         if (comp_len == PAGE_SIZE) {
1218                 zram_set_flag(zram, index, ZRAM_HUGE);
1219                 atomic64_inc(&zram->stats.huge_pages);
1220         }
1221
1222         if (flags) {
1223                 zram_set_flag(zram, index, flags);
1224                 zram_set_element(zram, index, element);
1225         }  else {
1226                 zram_set_handle(zram, index, handle);
1227                 zram_set_obj_size(zram, index, comp_len);
1228         }
1229         zram_slot_unlock(zram, index);
1230
1231         /* Update stats */
1232         atomic64_inc(&zram->stats.pages_stored);
1233         return ret;
1234 }
1235
1236 static int zram_bvec_write(struct zram *zram, struct bio_vec *bvec,
1237                                 u32 index, int offset, struct bio *bio)
1238 {
1239         int ret;
1240         struct page *page = NULL;
1241         void *src;
1242         struct bio_vec vec;
1243
1244         vec = *bvec;
1245         if (is_partial_io(bvec)) {
1246                 void *dst;
1247                 /*
1248                  * This is a partial IO. We need to read the full page
1249                  * before to write the changes.
1250                  */
1251                 page = alloc_page(GFP_NOIO|__GFP_HIGHMEM);
1252                 if (!page)
1253                         return -ENOMEM;
1254
1255                 ret = __zram_bvec_read(zram, page, index, bio, true);
1256                 if (ret)
1257                         goto out;
1258
1259                 src = kmap_atomic(bvec->bv_page);
1260                 dst = kmap_atomic(page);
1261                 memcpy(dst + offset, src + bvec->bv_offset, bvec->bv_len);
1262                 kunmap_atomic(dst);
1263                 kunmap_atomic(src);
1264
1265                 vec.bv_page = page;
1266                 vec.bv_len = PAGE_SIZE;
1267                 vec.bv_offset = 0;
1268         }
1269
1270         ret = __zram_bvec_write(zram, &vec, index, bio);
1271 out:
1272         if (is_partial_io(bvec))
1273                 __free_page(page);
1274         return ret;
1275 }
1276
1277 /*
1278  * zram_bio_discard - handler on discard request
1279  * @index: physical block index in PAGE_SIZE units
1280  * @offset: byte offset within physical block
1281  */
1282 static void zram_bio_discard(struct zram *zram, u32 index,
1283                              int offset, struct bio *bio)
1284 {
1285         size_t n = bio->bi_iter.bi_size;
1286
1287         /*
1288          * zram manages data in physical block size units. Because logical block
1289          * size isn't identical with physical block size on some arch, we
1290          * could get a discard request pointing to a specific offset within a
1291          * certain physical block.  Although we can handle this request by
1292          * reading that physiclal block and decompressing and partially zeroing
1293          * and re-compressing and then re-storing it, this isn't reasonable
1294          * because our intent with a discard request is to save memory.  So
1295          * skipping this logical block is appropriate here.
1296          */
1297         if (offset) {
1298                 if (n <= (PAGE_SIZE - offset))
1299                         return;
1300
1301                 n -= (PAGE_SIZE - offset);
1302                 index++;
1303         }
1304
1305         while (n >= PAGE_SIZE) {
1306                 zram_slot_lock(zram, index);
1307                 zram_free_page(zram, index);
1308                 zram_slot_unlock(zram, index);
1309                 atomic64_inc(&zram->stats.notify_free);
1310                 index++;
1311                 n -= PAGE_SIZE;
1312         }
1313 }
1314
1315 /*
1316  * Returns errno if it has some problem. Otherwise return 0 or 1.
1317  * Returns 0 if IO request was done synchronously
1318  * Returns 1 if IO request was successfully submitted.
1319  */
1320 static int zram_bvec_rw(struct zram *zram, struct bio_vec *bvec, u32 index,
1321                         int offset, unsigned int op, struct bio *bio)
1322 {
1323         unsigned long start_time = jiffies;
1324         struct request_queue *q = zram->disk->queue;
1325         int ret;
1326
1327         generic_start_io_acct(q, op, bvec->bv_len >> SECTOR_SHIFT,
1328                         &zram->disk->part0);
1329
1330         if (!op_is_write(op)) {
1331                 atomic64_inc(&zram->stats.num_reads);
1332                 ret = zram_bvec_read(zram, bvec, index, offset, bio);
1333                 flush_dcache_page(bvec->bv_page);
1334         } else {
1335                 atomic64_inc(&zram->stats.num_writes);
1336                 ret = zram_bvec_write(zram, bvec, index, offset, bio);
1337         }
1338
1339         generic_end_io_acct(q, op, &zram->disk->part0, start_time);
1340
1341         zram_slot_lock(zram, index);
1342         zram_accessed(zram, index);
1343         zram_slot_unlock(zram, index);
1344
1345         if (unlikely(ret < 0)) {
1346                 if (!op_is_write(op))
1347                         atomic64_inc(&zram->stats.failed_reads);
1348                 else
1349                         atomic64_inc(&zram->stats.failed_writes);
1350         }
1351
1352         return ret;
1353 }
1354
1355 static void __zram_make_request(struct zram *zram, struct bio *bio)
1356 {
1357         int offset;
1358         u32 index;
1359         struct bio_vec bvec;
1360         struct bvec_iter iter;
1361
1362         index = bio->bi_iter.bi_sector >> SECTORS_PER_PAGE_SHIFT;
1363         offset = (bio->bi_iter.bi_sector &
1364                   (SECTORS_PER_PAGE - 1)) << SECTOR_SHIFT;
1365
1366         switch (bio_op(bio)) {
1367         case REQ_OP_DISCARD:
1368         case REQ_OP_WRITE_ZEROES:
1369                 zram_bio_discard(zram, index, offset, bio);
1370                 bio_endio(bio);
1371                 return;
1372         default:
1373                 break;
1374         }
1375
1376         bio_for_each_segment(bvec, bio, iter) {
1377                 struct bio_vec bv = bvec;
1378                 unsigned int unwritten = bvec.bv_len;
1379
1380                 do {
1381                         bv.bv_len = min_t(unsigned int, PAGE_SIZE - offset,
1382                                                         unwritten);
1383                         if (zram_bvec_rw(zram, &bv, index, offset,
1384                                          bio_op(bio), bio) < 0)
1385                                 goto out;
1386
1387                         bv.bv_offset += bv.bv_len;
1388                         unwritten -= bv.bv_len;
1389
1390                         update_position(&index, &offset, &bv);
1391                 } while (unwritten);
1392         }
1393
1394         bio_endio(bio);
1395         return;
1396
1397 out:
1398         bio_io_error(bio);
1399 }
1400
1401 /*
1402  * Handler function for all zram I/O requests.
1403  */
1404 static blk_qc_t zram_make_request(struct request_queue *queue, struct bio *bio)
1405 {
1406         struct zram *zram = queue->queuedata;
1407
1408         if (!valid_io_request(zram, bio->bi_iter.bi_sector,
1409                                         bio->bi_iter.bi_size)) {
1410                 atomic64_inc(&zram->stats.invalid_io);
1411                 goto error;
1412         }
1413
1414         __zram_make_request(zram, bio);
1415         return BLK_QC_T_NONE;
1416
1417 error:
1418         bio_io_error(bio);
1419         return BLK_QC_T_NONE;
1420 }
1421
1422 static void zram_slot_free_notify(struct block_device *bdev,
1423                                 unsigned long index)
1424 {
1425         struct zram *zram;
1426
1427         zram = bdev->bd_disk->private_data;
1428
1429         atomic64_inc(&zram->stats.notify_free);
1430         if (!zram_slot_trylock(zram, index)) {
1431                 atomic64_inc(&zram->stats.miss_free);
1432                 return;
1433         }
1434
1435         zram_free_page(zram, index);
1436         zram_slot_unlock(zram, index);
1437 }
1438
1439 static int zram_rw_page(struct block_device *bdev, sector_t sector,
1440                        struct page *page, unsigned int op)
1441 {
1442         int offset, ret;
1443         u32 index;
1444         struct zram *zram;
1445         struct bio_vec bv;
1446
1447         if (PageTransHuge(page))
1448                 return -ENOTSUPP;
1449         zram = bdev->bd_disk->private_data;
1450
1451         if (!valid_io_request(zram, sector, PAGE_SIZE)) {
1452                 atomic64_inc(&zram->stats.invalid_io);
1453                 ret = -EINVAL;
1454                 goto out;
1455         }
1456
1457         index = sector >> SECTORS_PER_PAGE_SHIFT;
1458         offset = (sector & (SECTORS_PER_PAGE - 1)) << SECTOR_SHIFT;
1459
1460         bv.bv_page = page;
1461         bv.bv_len = PAGE_SIZE;
1462         bv.bv_offset = 0;
1463
1464         ret = zram_bvec_rw(zram, &bv, index, offset, op, NULL);
1465 out:
1466         /*
1467          * If I/O fails, just return error(ie, non-zero) without
1468          * calling page_endio.
1469          * It causes resubmit the I/O with bio request by upper functions
1470          * of rw_page(e.g., swap_readpage, __swap_writepage) and
1471          * bio->bi_end_io does things to handle the error
1472          * (e.g., SetPageError, set_page_dirty and extra works).
1473          */
1474         if (unlikely(ret < 0))
1475                 return ret;
1476
1477         switch (ret) {
1478         case 0:
1479                 page_endio(page, op_is_write(op), 0);
1480                 break;
1481         case 1:
1482                 ret = 0;
1483                 break;
1484         default:
1485                 WARN_ON(1);
1486         }
1487         return ret;
1488 }
1489
1490 static void zram_reset_device(struct zram *zram)
1491 {
1492         struct zcomp *comp;
1493         u64 disksize;
1494
1495         down_write(&zram->init_lock);
1496
1497         zram->limit_pages = 0;
1498
1499         if (!init_done(zram)) {
1500                 up_write(&zram->init_lock);
1501                 return;
1502         }
1503
1504         comp = zram->comp;
1505         disksize = zram->disksize;
1506         zram->disksize = 0;
1507
1508         set_capacity(zram->disk, 0);
1509         part_stat_set_all(&zram->disk->part0, 0);
1510
1511         up_write(&zram->init_lock);
1512         /* I/O operation under all of CPU are done so let's free */
1513         zram_meta_free(zram, disksize);
1514         memset(&zram->stats, 0, sizeof(zram->stats));
1515         zcomp_destroy(comp);
1516         reset_bdev(zram);
1517 }
1518
1519 static ssize_t disksize_store(struct device *dev,
1520                 struct device_attribute *attr, const char *buf, size_t len)
1521 {
1522         u64 disksize;
1523         struct zcomp *comp;
1524         struct zram *zram = dev_to_zram(dev);
1525         int err;
1526
1527         disksize = memparse(buf, NULL);
1528         if (!disksize)
1529                 return -EINVAL;
1530
1531         down_write(&zram->init_lock);
1532         if (init_done(zram)) {
1533                 pr_info("Cannot change disksize for initialized device\n");
1534                 err = -EBUSY;
1535                 goto out_unlock;
1536         }
1537
1538         disksize = PAGE_ALIGN(disksize);
1539         if (!zram_meta_alloc(zram, disksize)) {
1540                 err = -ENOMEM;
1541                 goto out_unlock;
1542         }
1543
1544         comp = zcomp_create(zram->compressor);
1545         if (IS_ERR(comp)) {
1546                 pr_err("Cannot initialise %s compressing backend\n",
1547                                 zram->compressor);
1548                 err = PTR_ERR(comp);
1549                 goto out_free_meta;
1550         }
1551
1552         zram->comp = comp;
1553         zram->disksize = disksize;
1554         set_capacity(zram->disk, zram->disksize >> SECTOR_SHIFT);
1555
1556         revalidate_disk(zram->disk);
1557         up_write(&zram->init_lock);
1558
1559         return len;
1560
1561 out_free_meta:
1562         zram_meta_free(zram, disksize);
1563 out_unlock:
1564         up_write(&zram->init_lock);
1565         return err;
1566 }
1567
1568 static ssize_t reset_store(struct device *dev,
1569                 struct device_attribute *attr, const char *buf, size_t len)
1570 {
1571         int ret;
1572         unsigned short do_reset;
1573         struct zram *zram;
1574         struct block_device *bdev;
1575
1576         ret = kstrtou16(buf, 10, &do_reset);
1577         if (ret)
1578                 return ret;
1579
1580         if (!do_reset)
1581                 return -EINVAL;
1582
1583         zram = dev_to_zram(dev);
1584         bdev = bdget_disk(zram->disk, 0);
1585         if (!bdev)
1586                 return -ENOMEM;
1587
1588         mutex_lock(&bdev->bd_mutex);
1589         /* Do not reset an active device or claimed device */
1590         if (bdev->bd_openers || zram->claim) {
1591                 mutex_unlock(&bdev->bd_mutex);
1592                 bdput(bdev);
1593                 return -EBUSY;
1594         }
1595
1596         /* From now on, anyone can't open /dev/zram[0-9] */
1597         zram->claim = true;
1598         mutex_unlock(&bdev->bd_mutex);
1599
1600         /* Make sure all the pending I/O are finished */
1601         fsync_bdev(bdev);
1602         zram_reset_device(zram);
1603         revalidate_disk(zram->disk);
1604         bdput(bdev);
1605
1606         mutex_lock(&bdev->bd_mutex);
1607         zram->claim = false;
1608         mutex_unlock(&bdev->bd_mutex);
1609
1610         return len;
1611 }
1612
1613 static int zram_open(struct block_device *bdev, fmode_t mode)
1614 {
1615         int ret = 0;
1616         struct zram *zram;
1617
1618         WARN_ON(!mutex_is_locked(&bdev->bd_mutex));
1619
1620         zram = bdev->bd_disk->private_data;
1621         /* zram was claimed to reset so open request fails */
1622         if (zram->claim)
1623                 ret = -EBUSY;
1624
1625         return ret;
1626 }
1627
1628 static const struct block_device_operations zram_devops = {
1629         .open = zram_open,
1630         .swap_slot_free_notify = zram_slot_free_notify,
1631         .rw_page = zram_rw_page,
1632         .owner = THIS_MODULE
1633 };
1634
1635 static DEVICE_ATTR_WO(compact);
1636 static DEVICE_ATTR_RW(disksize);
1637 static DEVICE_ATTR_RO(initstate);
1638 static DEVICE_ATTR_WO(reset);
1639 static DEVICE_ATTR_WO(mem_limit);
1640 static DEVICE_ATTR_WO(mem_used_max);
1641 static DEVICE_ATTR_WO(idle);
1642 static DEVICE_ATTR_RW(max_comp_streams);
1643 static DEVICE_ATTR_RW(comp_algorithm);
1644 #ifdef CONFIG_ZRAM_WRITEBACK
1645 static DEVICE_ATTR_RW(backing_dev);
1646 #endif
1647
1648 static struct attribute *zram_disk_attrs[] = {
1649         &dev_attr_disksize.attr,
1650         &dev_attr_initstate.attr,
1651         &dev_attr_reset.attr,
1652         &dev_attr_compact.attr,
1653         &dev_attr_mem_limit.attr,
1654         &dev_attr_mem_used_max.attr,
1655         &dev_attr_idle.attr,
1656         &dev_attr_max_comp_streams.attr,
1657         &dev_attr_comp_algorithm.attr,
1658 #ifdef CONFIG_ZRAM_WRITEBACK
1659         &dev_attr_backing_dev.attr,
1660 #endif
1661         &dev_attr_io_stat.attr,
1662         &dev_attr_mm_stat.attr,
1663         &dev_attr_debug_stat.attr,
1664         NULL,
1665 };
1666
1667 static const struct attribute_group zram_disk_attr_group = {
1668         .attrs = zram_disk_attrs,
1669 };
1670
1671 static const struct attribute_group *zram_disk_attr_groups[] = {
1672         &zram_disk_attr_group,
1673         NULL,
1674 };
1675
1676 /*
1677  * Allocate and initialize new zram device. the function returns
1678  * '>= 0' device_id upon success, and negative value otherwise.
1679  */
1680 static int zram_add(void)
1681 {
1682         struct zram *zram;
1683         struct request_queue *queue;
1684         int ret, device_id;
1685
1686         zram = kzalloc(sizeof(struct zram), GFP_KERNEL);
1687         if (!zram)
1688                 return -ENOMEM;
1689
1690         ret = idr_alloc(&zram_index_idr, zram, 0, 0, GFP_KERNEL);
1691         if (ret < 0)
1692                 goto out_free_dev;
1693         device_id = ret;
1694
1695         init_rwsem(&zram->init_lock);
1696
1697         queue = blk_alloc_queue(GFP_KERNEL);
1698         if (!queue) {
1699                 pr_err("Error allocating disk queue for device %d\n",
1700                         device_id);
1701                 ret = -ENOMEM;
1702                 goto out_free_idr;
1703         }
1704
1705         blk_queue_make_request(queue, zram_make_request);
1706
1707         /* gendisk structure */
1708         zram->disk = alloc_disk(1);
1709         if (!zram->disk) {
1710                 pr_err("Error allocating disk structure for device %d\n",
1711                         device_id);
1712                 ret = -ENOMEM;
1713                 goto out_free_queue;
1714         }
1715
1716         zram->disk->major = zram_major;
1717         zram->disk->first_minor = device_id;
1718         zram->disk->fops = &zram_devops;
1719         zram->disk->queue = queue;
1720         zram->disk->queue->queuedata = zram;
1721         zram->disk->private_data = zram;
1722         snprintf(zram->disk->disk_name, 16, "zram%d", device_id);
1723
1724         /* Actual capacity set using syfs (/sys/block/zram<id>/disksize */
1725         set_capacity(zram->disk, 0);
1726         /* zram devices sort of resembles non-rotational disks */
1727         blk_queue_flag_set(QUEUE_FLAG_NONROT, zram->disk->queue);
1728         blk_queue_flag_clear(QUEUE_FLAG_ADD_RANDOM, zram->disk->queue);
1729
1730         /*
1731          * To ensure that we always get PAGE_SIZE aligned
1732          * and n*PAGE_SIZED sized I/O requests.
1733          */
1734         blk_queue_physical_block_size(zram->disk->queue, PAGE_SIZE);
1735         blk_queue_logical_block_size(zram->disk->queue,
1736                                         ZRAM_LOGICAL_BLOCK_SIZE);
1737         blk_queue_io_min(zram->disk->queue, PAGE_SIZE);
1738         blk_queue_io_opt(zram->disk->queue, PAGE_SIZE);
1739         zram->disk->queue->limits.discard_granularity = PAGE_SIZE;
1740         blk_queue_max_discard_sectors(zram->disk->queue, UINT_MAX);
1741         blk_queue_flag_set(QUEUE_FLAG_DISCARD, zram->disk->queue);
1742
1743         /*
1744          * zram_bio_discard() will clear all logical blocks if logical block
1745          * size is identical with physical block size(PAGE_SIZE). But if it is
1746          * different, we will skip discarding some parts of logical blocks in
1747          * the part of the request range which isn't aligned to physical block
1748          * size.  So we can't ensure that all discarded logical blocks are
1749          * zeroed.
1750          */
1751         if (ZRAM_LOGICAL_BLOCK_SIZE == PAGE_SIZE)
1752                 blk_queue_max_write_zeroes_sectors(zram->disk->queue, UINT_MAX);
1753
1754         zram->disk->queue->backing_dev_info->capabilities |=
1755                         (BDI_CAP_STABLE_WRITES | BDI_CAP_SYNCHRONOUS_IO);
1756         device_add_disk(NULL, zram->disk, zram_disk_attr_groups);
1757
1758         strlcpy(zram->compressor, default_compressor, sizeof(zram->compressor));
1759
1760         zram_debugfs_register(zram);
1761         pr_info("Added device: %s\n", zram->disk->disk_name);
1762         return device_id;
1763
1764 out_free_queue:
1765         blk_cleanup_queue(queue);
1766 out_free_idr:
1767         idr_remove(&zram_index_idr, device_id);
1768 out_free_dev:
1769         kfree(zram);
1770         return ret;
1771 }
1772
1773 static int zram_remove(struct zram *zram)
1774 {
1775         struct block_device *bdev;
1776
1777         bdev = bdget_disk(zram->disk, 0);
1778         if (!bdev)
1779                 return -ENOMEM;
1780
1781         mutex_lock(&bdev->bd_mutex);
1782         if (bdev->bd_openers || zram->claim) {
1783                 mutex_unlock(&bdev->bd_mutex);
1784                 bdput(bdev);
1785                 return -EBUSY;
1786         }
1787
1788         zram->claim = true;
1789         mutex_unlock(&bdev->bd_mutex);
1790
1791         zram_debugfs_unregister(zram);
1792
1793         /* Make sure all the pending I/O are finished */
1794         fsync_bdev(bdev);
1795         zram_reset_device(zram);
1796         bdput(bdev);
1797
1798         pr_info("Removed device: %s\n", zram->disk->disk_name);
1799
1800         del_gendisk(zram->disk);
1801         blk_cleanup_queue(zram->disk->queue);
1802         put_disk(zram->disk);
1803         kfree(zram);
1804         return 0;
1805 }
1806
1807 /* zram-control sysfs attributes */
1808
1809 /*
1810  * NOTE: hot_add attribute is not the usual read-only sysfs attribute. In a
1811  * sense that reading from this file does alter the state of your system -- it
1812  * creates a new un-initialized zram device and returns back this device's
1813  * device_id (or an error code if it fails to create a new device).
1814  */
1815 static ssize_t hot_add_show(struct class *class,
1816                         struct class_attribute *attr,
1817                         char *buf)
1818 {
1819         int ret;
1820
1821         mutex_lock(&zram_index_mutex);
1822         ret = zram_add();
1823         mutex_unlock(&zram_index_mutex);
1824
1825         if (ret < 0)
1826                 return ret;
1827         return scnprintf(buf, PAGE_SIZE, "%d\n", ret);
1828 }
1829 static CLASS_ATTR_RO(hot_add);
1830
1831 static ssize_t hot_remove_store(struct class *class,
1832                         struct class_attribute *attr,
1833                         const char *buf,
1834                         size_t count)
1835 {
1836         struct zram *zram;
1837         int ret, dev_id;
1838
1839         /* dev_id is gendisk->first_minor, which is `int' */
1840         ret = kstrtoint(buf, 10, &dev_id);
1841         if (ret)
1842                 return ret;
1843         if (dev_id < 0)
1844                 return -EINVAL;
1845
1846         mutex_lock(&zram_index_mutex);
1847
1848         zram = idr_find(&zram_index_idr, dev_id);
1849         if (zram) {
1850                 ret = zram_remove(zram);
1851                 if (!ret)
1852                         idr_remove(&zram_index_idr, dev_id);
1853         } else {
1854                 ret = -ENODEV;
1855         }
1856
1857         mutex_unlock(&zram_index_mutex);
1858         return ret ? ret : count;
1859 }
1860 static CLASS_ATTR_WO(hot_remove);
1861
1862 static struct attribute *zram_control_class_attrs[] = {
1863         &class_attr_hot_add.attr,
1864         &class_attr_hot_remove.attr,
1865         NULL,
1866 };
1867 ATTRIBUTE_GROUPS(zram_control_class);
1868
1869 static struct class zram_control_class = {
1870         .name           = "zram-control",
1871         .owner          = THIS_MODULE,
1872         .class_groups   = zram_control_class_groups,
1873 };
1874
1875 static int zram_remove_cb(int id, void *ptr, void *data)
1876 {
1877         zram_remove(ptr);
1878         return 0;
1879 }
1880
1881 static void destroy_devices(void)
1882 {
1883         class_unregister(&zram_control_class);
1884         idr_for_each(&zram_index_idr, &zram_remove_cb, NULL);
1885         zram_debugfs_destroy();
1886         idr_destroy(&zram_index_idr);
1887         unregister_blkdev(zram_major, "zram");
1888         cpuhp_remove_multi_state(CPUHP_ZCOMP_PREPARE);
1889 }
1890
1891 static int __init zram_init(void)
1892 {
1893         int ret;
1894
1895         ret = cpuhp_setup_state_multi(CPUHP_ZCOMP_PREPARE, "block/zram:prepare",
1896                                       zcomp_cpu_up_prepare, zcomp_cpu_dead);
1897         if (ret < 0)
1898                 return ret;
1899
1900         ret = class_register(&zram_control_class);
1901         if (ret) {
1902                 pr_err("Unable to register zram-control class\n");
1903                 cpuhp_remove_multi_state(CPUHP_ZCOMP_PREPARE);
1904                 return ret;
1905         }
1906
1907         zram_debugfs_create();
1908         zram_major = register_blkdev(0, "zram");
1909         if (zram_major <= 0) {
1910                 pr_err("Unable to get major number\n");
1911                 class_unregister(&zram_control_class);
1912                 cpuhp_remove_multi_state(CPUHP_ZCOMP_PREPARE);
1913                 return -EBUSY;
1914         }
1915
1916         while (num_devices != 0) {
1917                 mutex_lock(&zram_index_mutex);
1918                 ret = zram_add();
1919                 mutex_unlock(&zram_index_mutex);
1920                 if (ret < 0)
1921                         goto out_error;
1922                 num_devices--;
1923         }
1924
1925         return 0;
1926
1927 out_error:
1928         destroy_devices();
1929         return ret;
1930 }
1931
1932 static void __exit zram_exit(void)
1933 {
1934         destroy_devices();
1935 }
1936
1937 module_init(zram_init);
1938 module_exit(zram_exit);
1939
1940 module_param(num_devices, uint, 0);
1941 MODULE_PARM_DESC(num_devices, "Number of pre-created zram devices");
1942
1943 MODULE_LICENSE("Dual BSD/GPL");
1944 MODULE_AUTHOR("Nitin Gupta <ngupta@vflare.org>");
1945 MODULE_DESCRIPTION("Compressed RAM Block Device");