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/highmem.h>
  26 #include <linux/slab.h>
  27 #include <linux/backing-dev.h>
  28 #include <linux/string.h>
  29 #include <linux/vmalloc.h>
  30 #include <linux/err.h>
  31 #include <linux/idr.h>
  32 #include <linux/sysfs.h>
  33 #include <linux/debugfs.h>
  34 #include <linux/cpuhotplug.h>
  35 #include <linux/part_stat.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 = CONFIG_ZRAM_DEF_COMP;
  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 const struct block_device_operations zram_devops;
  55
  56 static void zram_free_page(struct zram *zram, size_t index);
  57 static int zram_read_page(struct zram *zram, struct page *page, u32 index,
  58                           struct bio *parent);
  59
  60 static int zram_slot_trylock(struct zram *zram, u32 index)
  61 {
  62         return bit_spin_trylock(ZRAM_LOCK, &zram->table[index].flags);
  63 }
  64
  65 static void zram_slot_lock(struct zram *zram, u32 index)
  66 {
  67         bit_spin_lock(ZRAM_LOCK, &zram->table[index].flags);
  68 }
  69
  70 static void zram_slot_unlock(struct zram *zram, u32 index)
  71 {
  72         bit_spin_unlock(ZRAM_LOCK, &zram->table[index].flags);
  73 }
  74
  75 static inline bool init_done(struct zram *zram)
  76 {
  77         return zram->disksize;
  78 }
  79
  80 static inline struct zram *dev_to_zram(struct device *dev)
  81 {
  82         return (struct zram *)dev_to_disk(dev)->private_data;
  83 }
  84
  85 static unsigned long zram_get_handle(struct zram *zram, u32 index)
  86 {
  87         return zram->table[index].handle;
  88 }
  89
  90 static void zram_set_handle(struct zram *zram, u32 index, unsigned long handle)
  91 {
  92         zram->table[index].handle = handle;
  93 }
  94
  95 /* flag operations require table entry bit_spin_lock() being held */
  96 static bool zram_test_flag(struct zram *zram, u32 index,
  97                         enum zram_pageflags flag)
  98 {
  99         return zram->table[index].flags & BIT(flag);
 100 }
 101
 102 static void zram_set_flag(struct zram *zram, u32 index,
 103                         enum zram_pageflags flag)
 104 {
 105         zram->table[index].flags |= BIT(flag);
 106 }
 107
 108 static void zram_clear_flag(struct zram *zram, u32 index,
 109                         enum zram_pageflags flag)
 110 {
 111         zram->table[index].flags &= ~BIT(flag);
 112 }
 113
 114 static inline void zram_set_element(struct zram *zram, u32 index,
 115                         unsigned long element)
 116 {
 117         zram->table[index].element = element;
 118 }
 119
 120 static unsigned long zram_get_element(struct zram *zram, u32 index)
 121 {
 122         return zram->table[index].element;
 123 }
 124
 125 static size_t zram_get_obj_size(struct zram *zram, u32 index)
 126 {
 127         return zram->table[index].flags & (BIT(ZRAM_FLAG_SHIFT) - 1);
 128 }
 129
 130 static void zram_set_obj_size(struct zram *zram,
 131                                         u32 index, size_t size)
 132 {
 133         unsigned long flags = zram->table[index].flags >> ZRAM_FLAG_SHIFT;
 134
 135         zram->table[index].flags = (flags << ZRAM_FLAG_SHIFT) | size;
 136 }
 137
 138 static inline bool zram_allocated(struct zram *zram, u32 index)
 139 {
 140         return zram_get_obj_size(zram, index) ||
 141                         zram_test_flag(zram, index, ZRAM_SAME) ||
 142                         zram_test_flag(zram, index, ZRAM_WB);
 143 }
 144
 145 #if PAGE_SIZE != 4096
 146 static inline bool is_partial_io(struct bio_vec *bvec)
 147 {
 148         return bvec->bv_len != PAGE_SIZE;
 149 }
 150 #define ZRAM_PARTIAL_IO         1
 151 #else
 152 static inline bool is_partial_io(struct bio_vec *bvec)
 153 {
 154         return false;
 155 }
 156 #endif
 157
 158 static inline void zram_set_priority(struct zram *zram, u32 index, u32 prio)
 159 {
 160         prio &= ZRAM_COMP_PRIORITY_MASK;
 161         /*
 162          * Clear previous priority value first, in case if we recompress
 163          * further an already recompressed page
 164          */
 165         zram->table[index].flags &= ~(ZRAM_COMP_PRIORITY_MASK <<
 166                                       ZRAM_COMP_PRIORITY_BIT1);
 167         zram->table[index].flags |= (prio << ZRAM_COMP_PRIORITY_BIT1);
 168 }
 169
 170 static inline u32 zram_get_priority(struct zram *zram, u32 index)
 171 {
 172         u32 prio = zram->table[index].flags >> ZRAM_COMP_PRIORITY_BIT1;
 173
 174         return prio & ZRAM_COMP_PRIORITY_MASK;
 175 }
 176
 177 static inline void update_used_max(struct zram *zram,
 178                                         const unsigned long pages)
 179 {
 180         unsigned long cur_max = atomic_long_read(&zram->stats.max_used_pages);
 181
 182         do {
 183                 if (cur_max >= pages)
 184                         return;
 185         } while (!atomic_long_try_cmpxchg(&zram->stats.max_used_pages,
 186                                           &cur_max, pages));
 187 }
 188
 189 static inline void zram_fill_page(void *ptr, unsigned long len,
 190                                         unsigned long value)
 191 {
 192         WARN_ON_ONCE(!IS_ALIGNED(len, sizeof(unsigned long)));
 193         memset_l(ptr, value, len / sizeof(unsigned long));
 194 }
 195
 196 static bool page_same_filled(void *ptr, unsigned long *element)
 197 {
 198         unsigned long *page;
 199         unsigned long val;
 200         unsigned int pos, last_pos = PAGE_SIZE / sizeof(*page) - 1;
 201
 202         page = (unsigned long *)ptr;
 203         val = page[0];
 204
 205         if (val != page[last_pos])
 206                 return false;
 207
 208         for (pos = 1; pos < last_pos; pos++) {
 209                 if (val != page[pos])
 210                         return false;
 211         }
 212
 213         *element = val;
 214
 215         return true;
 216 }
 217
 218 static ssize_t initstate_show(struct device *dev,
 219                 struct device_attribute *attr, char *buf)
 220 {
 221         u32 val;
 222         struct zram *zram = dev_to_zram(dev);
 223
 224         down_read(&zram->init_lock);
 225         val = init_done(zram);
 226         up_read(&zram->init_lock);
 227
 228         return scnprintf(buf, PAGE_SIZE, "%u\n", val);
 229 }
 230
 231 static ssize_t disksize_show(struct device *dev,
 232                 struct device_attribute *attr, char *buf)
 233 {
 234         struct zram *zram = dev_to_zram(dev);
 235
 236         return scnprintf(buf, PAGE_SIZE, "%llu\n", zram->disksize);
 237 }
 238
 239 static ssize_t mem_limit_store(struct device *dev,
 240                 struct device_attribute *attr, const char *buf, size_t len)
 241 {
 242         u64 limit;
 243         char *tmp;
 244         struct zram *zram = dev_to_zram(dev);
 245
 246         limit = memparse(buf, &tmp);
 247         if (buf == tmp) /* no chars parsed, invalid input */
 248                 return -EINVAL;
 249
 250         down_write(&zram->init_lock);
 251         zram->limit_pages = PAGE_ALIGN(limit) >> PAGE_SHIFT;
 252         up_write(&zram->init_lock);
 253
 254         return len;
 255 }
 256
 257 static ssize_t mem_used_max_store(struct device *dev,
 258                 struct device_attribute *attr, const char *buf, size_t len)
 259 {
 260         int err;
 261         unsigned long val;
 262         struct zram *zram = dev_to_zram(dev);
 263
 264         err = kstrtoul(buf, 10, &val);
 265         if (err || val != 0)
 266                 return -EINVAL;
 267
 268         down_read(&zram->init_lock);
 269         if (init_done(zram)) {
 270                 atomic_long_set(&zram->stats.max_used_pages,
 271                                 zs_get_total_pages(zram->mem_pool));
 272         }
 273         up_read(&zram->init_lock);
 274
 275         return len;
 276 }
 277
 278 /*
 279  * Mark all pages which are older than or equal to cutoff as IDLE.
 280  * Callers should hold the zram init lock in read mode
 281  */
 282 static void mark_idle(struct zram *zram, ktime_t cutoff)
 283 {
 284         int is_idle = 1;
 285         unsigned long nr_pages = zram->disksize >> PAGE_SHIFT;
 286         int index;
 287
 288         for (index = 0; index < nr_pages; index++) {
 289                 /*
 290                  * Do not mark ZRAM_UNDER_WB slot as ZRAM_IDLE to close race.
 291                  * See the comment in writeback_store.
 292                  */
 293                 zram_slot_lock(zram, index);
 294                 if (zram_allocated(zram, index) &&
 295                                 !zram_test_flag(zram, index, ZRAM_UNDER_WB)) {
 296 #ifdef CONFIG_ZRAM_MEMORY_TRACKING
 297                         is_idle = !cutoff || ktime_after(cutoff, zram->table[index].ac_time);
 298 #endif
 299                         if (is_idle)
 300                                 zram_set_flag(zram, index, ZRAM_IDLE);
 301                 }
 302                 zram_slot_unlock(zram, index);
 303         }
 304 }
 305
 306 static ssize_t idle_store(struct device *dev,
 307                 struct device_attribute *attr, const char *buf, size_t len)
 308 {
 309         struct zram *zram = dev_to_zram(dev);
 310         ktime_t cutoff_time = 0;
 311         ssize_t rv = -EINVAL;
 312
 313         if (!sysfs_streq(buf, "all")) {
 314                 /*
 315                  * If it did not parse as 'all' try to treat it as an integer
 316                  * when we have memory tracking enabled.
 317                  */
 318                 u64 age_sec;
 319
 320                 if (IS_ENABLED(CONFIG_ZRAM_MEMORY_TRACKING) && !kstrtoull(buf, 0, &age_sec))
 321                         cutoff_time = ktime_sub(ktime_get_boottime(),
 322                                         ns_to_ktime(age_sec * NSEC_PER_SEC));
 323                 else
 324                         goto out;
 325         }
 326
 327         down_read(&zram->init_lock);
 328         if (!init_done(zram))
 329                 goto out_unlock;
 330
 331         /*
 332          * A cutoff_time of 0 marks everything as idle, this is the
 333          * "all" behavior.
 334          */
 335         mark_idle(zram, cutoff_time);
 336         rv = len;
 337
 338 out_unlock:
 339         up_read(&zram->init_lock);
 340 out:
 341         return rv;
 342 }
 343
 344 #ifdef CONFIG_ZRAM_WRITEBACK
 345 static ssize_t writeback_limit_enable_store(struct device *dev,
 346                 struct device_attribute *attr, const char *buf, size_t len)
 347 {
 348         struct zram *zram = dev_to_zram(dev);
 349         u64 val;
 350         ssize_t ret = -EINVAL;
 351
 352         if (kstrtoull(buf, 10, &val))
 353                 return ret;
 354
 355         down_read(&zram->init_lock);
 356         spin_lock(&zram->wb_limit_lock);
 357         zram->wb_limit_enable = val;
 358         spin_unlock(&zram->wb_limit_lock);
 359         up_read(&zram->init_lock);
 360         ret = len;
 361
 362         return ret;
 363 }
 364
 365 static ssize_t writeback_limit_enable_show(struct device *dev,
 366                 struct device_attribute *attr, char *buf)
 367 {
 368         bool val;
 369         struct zram *zram = dev_to_zram(dev);
 370
 371         down_read(&zram->init_lock);
 372         spin_lock(&zram->wb_limit_lock);
 373         val = zram->wb_limit_enable;
 374         spin_unlock(&zram->wb_limit_lock);
 375         up_read(&zram->init_lock);
 376
 377         return scnprintf(buf, PAGE_SIZE, "%d\n", val);
 378 }
 379
 380 static ssize_t writeback_limit_store(struct device *dev,
 381                 struct device_attribute *attr, const char *buf, size_t len)
 382 {
 383         struct zram *zram = dev_to_zram(dev);
 384         u64 val;
 385         ssize_t ret = -EINVAL;
 386
 387         if (kstrtoull(buf, 10, &val))
 388                 return ret;
 389
 390         down_read(&zram->init_lock);
 391         spin_lock(&zram->wb_limit_lock);
 392         zram->bd_wb_limit = val;
 393         spin_unlock(&zram->wb_limit_lock);
 394         up_read(&zram->init_lock);
 395         ret = len;
 396
 397         return ret;
 398 }
 399
 400 static ssize_t writeback_limit_show(struct device *dev,
 401                 struct device_attribute *attr, char *buf)
 402 {
 403         u64 val;
 404         struct zram *zram = dev_to_zram(dev);
 405
 406         down_read(&zram->init_lock);
 407         spin_lock(&zram->wb_limit_lock);
 408         val = zram->bd_wb_limit;
 409         spin_unlock(&zram->wb_limit_lock);
 410         up_read(&zram->init_lock);
 411
 412         return scnprintf(buf, PAGE_SIZE, "%llu\n", val);
 413 }
 414
 415 static void reset_bdev(struct zram *zram)
 416 {
 417         struct block_device *bdev;
 418
 419         if (!zram->backing_dev)
 420                 return;
 421
 422         bdev = zram->bdev;
 423         blkdev_put(bdev, zram);
 424         /* hope filp_close flush all of IO */
 425         filp_close(zram->backing_dev, NULL);
 426         zram->backing_dev = NULL;
 427         zram->bdev = NULL;
 428         zram->disk->fops = &zram_devops;
 429         kvfree(zram->bitmap);
 430         zram->bitmap = NULL;
 431 }
 432
 433 static ssize_t backing_dev_show(struct device *dev,
 434                 struct device_attribute *attr, char *buf)
 435 {
 436         struct file *file;
 437         struct zram *zram = dev_to_zram(dev);
 438         char *p;
 439         ssize_t ret;
 440
 441         down_read(&zram->init_lock);
 442         file = zram->backing_dev;
 443         if (!file) {
 444                 memcpy(buf, "none\n", 5);
 445                 up_read(&zram->init_lock);
 446                 return 5;
 447         }
 448
 449         p = file_path(file, buf, PAGE_SIZE - 1);
 450         if (IS_ERR(p)) {
 451                 ret = PTR_ERR(p);
 452                 goto out;
 453         }
 454
 455         ret = strlen(p);
 456         memmove(buf, p, ret);
 457         buf[ret++] = '\n';
 458 out:
 459         up_read(&zram->init_lock);
 460         return ret;
 461 }
 462
 463 static ssize_t backing_dev_store(struct device *dev,
 464                 struct device_attribute *attr, const char *buf, size_t len)
 465 {
 466         char *file_name;
 467         size_t sz;
 468         struct file *backing_dev = NULL;
 469         struct inode *inode;
 470         struct address_space *mapping;
 471         unsigned int bitmap_sz;
 472         unsigned long nr_pages, *bitmap = NULL;
 473         struct block_device *bdev = NULL;
 474         int err;
 475         struct zram *zram = dev_to_zram(dev);
 476
 477         file_name = kmalloc(PATH_MAX, GFP_KERNEL);
 478         if (!file_name)
 479                 return -ENOMEM;
 480
 481         down_write(&zram->init_lock);
 482         if (init_done(zram)) {
 483                 pr_info("Can't setup backing device for initialized device\n");
 484                 err = -EBUSY;
 485                 goto out;
 486         }
 487
 488         strscpy(file_name, buf, PATH_MAX);
 489         /* ignore trailing newline */
 490         sz = strlen(file_name);
 491         if (sz > 0 && file_name[sz - 1] == '\n')
 492                 file_name[sz - 1] = 0x00;
 493
 494         backing_dev = filp_open(file_name, O_RDWR|O_LARGEFILE, 0);
 495         if (IS_ERR(backing_dev)) {
 496                 err = PTR_ERR(backing_dev);
 497                 backing_dev = NULL;
 498                 goto out;
 499         }
 500
 501         mapping = backing_dev->f_mapping;
 502         inode = mapping->host;
 503
 504         /* Support only block device in this moment */
 505         if (!S_ISBLK(inode->i_mode)) {
 506                 err = -ENOTBLK;
 507                 goto out;
 508         }
 509
 510         bdev = blkdev_get_by_dev(inode->i_rdev, BLK_OPEN_READ | BLK_OPEN_WRITE,
 511                                  zram, NULL);
 512         if (IS_ERR(bdev)) {
 513                 err = PTR_ERR(bdev);
 514                 bdev = NULL;
 515                 goto out;
 516         }
 517
 518         nr_pages = i_size_read(inode) >> PAGE_SHIFT;
 519         bitmap_sz = BITS_TO_LONGS(nr_pages) * sizeof(long);
 520         bitmap = kvzalloc(bitmap_sz, GFP_KERNEL);
 521         if (!bitmap) {
 522                 err = -ENOMEM;
 523                 goto out;
 524         }
 525
 526         reset_bdev(zram);
 527
 528         zram->bdev = bdev;
 529         zram->backing_dev = backing_dev;
 530         zram->bitmap = bitmap;
 531         zram->nr_pages = nr_pages;
 532         up_write(&zram->init_lock);
 533
 534         pr_info("setup backing device %s\n", file_name);
 535         kfree(file_name);
 536
 537         return len;
 538 out:
 539         kvfree(bitmap);
 540
 541         if (bdev)
 542                 blkdev_put(bdev, zram);
 543
 544         if (backing_dev)
 545                 filp_close(backing_dev, NULL);
 546
 547         up_write(&zram->init_lock);
 548
 549         kfree(file_name);
 550
 551         return err;
 552 }
 553
 554 static unsigned long alloc_block_bdev(struct zram *zram)
 555 {
 556         unsigned long blk_idx = 1;
 557 retry:
 558         /* skip 0 bit to confuse zram.handle = 0 */
 559         blk_idx = find_next_zero_bit(zram->bitmap, zram->nr_pages, blk_idx);
 560         if (blk_idx == zram->nr_pages)
 561                 return 0;
 562
 563         if (test_and_set_bit(blk_idx, zram->bitmap))
 564                 goto retry;
 565
 566         atomic64_inc(&zram->stats.bd_count);
 567         return blk_idx;
 568 }
 569
 570 static void free_block_bdev(struct zram *zram, unsigned long blk_idx)
 571 {
 572         int was_set;
 573
 574         was_set = test_and_clear_bit(blk_idx, zram->bitmap);
 575         WARN_ON_ONCE(!was_set);
 576         atomic64_dec(&zram->stats.bd_count);
 577 }
 578
 579 static void read_from_bdev_async(struct zram *zram, struct page *page,
 580                         unsigned long entry, struct bio *parent)
 581 {
 582         struct bio *bio;
 583
 584         bio = bio_alloc(zram->bdev, 1, parent->bi_opf, GFP_NOIO);
 585         bio->bi_iter.bi_sector = entry * (PAGE_SIZE >> 9);
 586         __bio_add_page(bio, page, PAGE_SIZE, 0);
 587         bio_chain(bio, parent);
 588         submit_bio(bio);
 589 }
 590
 591 #define PAGE_WB_SIG "page_index="
 592
 593 #define PAGE_WRITEBACK                  0
 594 #define HUGE_WRITEBACK                  (1<<0)
 595 #define IDLE_WRITEBACK                  (1<<1)
 596 #define INCOMPRESSIBLE_WRITEBACK        (1<<2)
 597
 598 static ssize_t writeback_store(struct device *dev,
 599                 struct device_attribute *attr, const char *buf, size_t len)
 600 {
 601         struct zram *zram = dev_to_zram(dev);
 602         unsigned long nr_pages = zram->disksize >> PAGE_SHIFT;
 603         unsigned long index = 0;
 604         struct bio bio;
 605         struct bio_vec bio_vec;
 606         struct page *page;
 607         ssize_t ret = len;
 608         int mode, err;
 609         unsigned long blk_idx = 0;
 610
 611         if (sysfs_streq(buf, "idle"))
 612                 mode = IDLE_WRITEBACK;
 613         else if (sysfs_streq(buf, "huge"))
 614                 mode = HUGE_WRITEBACK;
 615         else if (sysfs_streq(buf, "huge_idle"))
 616                 mode = IDLE_WRITEBACK | HUGE_WRITEBACK;
 617         else if (sysfs_streq(buf, "incompressible"))
 618                 mode = INCOMPRESSIBLE_WRITEBACK;
 619         else {
 620                 if (strncmp(buf, PAGE_WB_SIG, sizeof(PAGE_WB_SIG) - 1))
 621                         return -EINVAL;
 622
 623                 if (kstrtol(buf + sizeof(PAGE_WB_SIG) - 1, 10, &index) ||
 624                                 index >= nr_pages)
 625                         return -EINVAL;
 626
 627                 nr_pages = 1;
 628                 mode = PAGE_WRITEBACK;
 629         }
 630
 631         down_read(&zram->init_lock);
 632         if (!init_done(zram)) {
 633                 ret = -EINVAL;
 634                 goto release_init_lock;
 635         }
 636
 637         if (!zram->backing_dev) {
 638                 ret = -ENODEV;
 639                 goto release_init_lock;
 640         }
 641
 642         page = alloc_page(GFP_KERNEL);
 643         if (!page) {
 644                 ret = -ENOMEM;
 645                 goto release_init_lock;
 646         }
 647
 648         for (; nr_pages != 0; index++, nr_pages--) {
 649                 spin_lock(&zram->wb_limit_lock);
 650                 if (zram->wb_limit_enable && !zram->bd_wb_limit) {
 651                         spin_unlock(&zram->wb_limit_lock);
 652                         ret = -EIO;
 653                         break;
 654                 }
 655                 spin_unlock(&zram->wb_limit_lock);
 656
 657                 if (!blk_idx) {
 658                         blk_idx = alloc_block_bdev(zram);
 659                         if (!blk_idx) {
 660                                 ret = -ENOSPC;
 661                                 break;
 662                         }
 663                 }
 664
 665                 zram_slot_lock(zram, index);
 666                 if (!zram_allocated(zram, index))
 667                         goto next;
 668
 669                 if (zram_test_flag(zram, index, ZRAM_WB) ||
 670                                 zram_test_flag(zram, index, ZRAM_SAME) ||
 671                                 zram_test_flag(zram, index, ZRAM_UNDER_WB))
 672                         goto next;
 673
 674                 if (mode & IDLE_WRITEBACK &&
 675                     !zram_test_flag(zram, index, ZRAM_IDLE))
 676                         goto next;
 677                 if (mode & HUGE_WRITEBACK &&
 678                     !zram_test_flag(zram, index, ZRAM_HUGE))
 679                         goto next;
 680                 if (mode & INCOMPRESSIBLE_WRITEBACK &&
 681                     !zram_test_flag(zram, index, ZRAM_INCOMPRESSIBLE))
 682                         goto next;
 683
 684                 /*
 685                  * Clearing ZRAM_UNDER_WB is duty of caller.
 686                  * IOW, zram_free_page never clear it.
 687                  */
 688                 zram_set_flag(zram, index, ZRAM_UNDER_WB);
 689                 /* Need for hugepage writeback racing */
 690                 zram_set_flag(zram, index, ZRAM_IDLE);
 691                 zram_slot_unlock(zram, index);
 692                 if (zram_read_page(zram, page, index, NULL)) {
 693                         zram_slot_lock(zram, index);
 694                         zram_clear_flag(zram, index, ZRAM_UNDER_WB);
 695                         zram_clear_flag(zram, index, ZRAM_IDLE);
 696                         zram_slot_unlock(zram, index);
 697                         continue;
 698                 }
 699
 700                 bio_init(&bio, zram->bdev, &bio_vec, 1,
 701                          REQ_OP_WRITE | REQ_SYNC);
 702                 bio.bi_iter.bi_sector = blk_idx * (PAGE_SIZE >> 9);
 703                 __bio_add_page(&bio, page, PAGE_SIZE, 0);
 704
 705                 /*
 706                  * XXX: A single page IO would be inefficient for write
 707                  * but it would be not bad as starter.
 708                  */
 709                 err = submit_bio_wait(&bio);
 710                 if (err) {
 711                         zram_slot_lock(zram, index);
 712                         zram_clear_flag(zram, index, ZRAM_UNDER_WB);
 713                         zram_clear_flag(zram, index, ZRAM_IDLE);
 714                         zram_slot_unlock(zram, index);
 715                         /*
 716                          * BIO errors are not fatal, we continue and simply
 717                          * attempt to writeback the remaining objects (pages).
 718                          * At the same time we need to signal user-space that
 719                          * some writes (at least one, but also could be all of
 720                          * them) were not successful and we do so by returning
 721                          * the most recent BIO error.
 722                          */
 723                         ret = err;
 724                         continue;
 725                 }
 726
 727                 atomic64_inc(&zram->stats.bd_writes);
 728                 /*
 729                  * We released zram_slot_lock so need to check if the slot was
 730                  * changed. If there is freeing for the slot, we can catch it
 731                  * easily by zram_allocated.
 732                  * A subtle case is the slot is freed/reallocated/marked as
 733                  * ZRAM_IDLE again. To close the race, idle_store doesn't
 734                  * mark ZRAM_IDLE once it found the slot was ZRAM_UNDER_WB.
 735                  * Thus, we could close the race by checking ZRAM_IDLE bit.
 736                  */
 737                 zram_slot_lock(zram, index);
 738                 if (!zram_allocated(zram, index) ||
 739                           !zram_test_flag(zram, index, ZRAM_IDLE)) {
 740                         zram_clear_flag(zram, index, ZRAM_UNDER_WB);
 741                         zram_clear_flag(zram, index, ZRAM_IDLE);
 742                         goto next;
 743                 }
 744
 745                 zram_free_page(zram, index);
 746                 zram_clear_flag(zram, index, ZRAM_UNDER_WB);
 747                 zram_set_flag(zram, index, ZRAM_WB);
 748                 zram_set_element(zram, index, blk_idx);
 749                 blk_idx = 0;
 750                 atomic64_inc(&zram->stats.pages_stored);
 751                 spin_lock(&zram->wb_limit_lock);
 752                 if (zram->wb_limit_enable && zram->bd_wb_limit > 0)
 753                         zram->bd_wb_limit -=  1UL << (PAGE_SHIFT - 12);
 754                 spin_unlock(&zram->wb_limit_lock);
 755 next:
 756                 zram_slot_unlock(zram, index);
 757         }
 758
 759         if (blk_idx)
 760                 free_block_bdev(zram, blk_idx);
 761         __free_page(page);
 762 release_init_lock:
 763         up_read(&zram->init_lock);
 764
 765         return ret;
 766 }
 767
 768 struct zram_work {
 769         struct work_struct work;
 770         struct zram *zram;
 771         unsigned long entry;
 772         struct page *page;
 773         int error;
 774 };
 775
 776 static void zram_sync_read(struct work_struct *work)
 777 {
 778         struct zram_work *zw = container_of(work, struct zram_work, work);
 779         struct bio_vec bv;
 780         struct bio bio;
 781
 782         bio_init(&bio, zw->zram->bdev, &bv, 1, REQ_OP_READ);
 783         bio.bi_iter.bi_sector = zw->entry * (PAGE_SIZE >> 9);
 784         __bio_add_page(&bio, zw->page, PAGE_SIZE, 0);
 785         zw->error = submit_bio_wait(&bio);
 786 }
 787
 788 /*
 789  * Block layer want one ->submit_bio to be active at a time, so if we use
 790  * chained IO with parent IO in same context, it's a deadlock. To avoid that,
 791  * use a worker thread context.
 792  */
 793 static int read_from_bdev_sync(struct zram *zram, struct page *page,
 794                                 unsigned long entry)
 795 {
 796         struct zram_work work;
 797
 798         work.page = page;
 799         work.zram = zram;
 800         work.entry = entry;
 801
 802         INIT_WORK_ONSTACK(&work.work, zram_sync_read);
 803         queue_work(system_unbound_wq, &work.work);
 804         flush_work(&work.work);
 805         destroy_work_on_stack(&work.work);
 806
 807         return work.error;
 808 }
 809
 810 static int read_from_bdev(struct zram *zram, struct page *page,
 811                         unsigned long entry, struct bio *parent)
 812 {
 813         atomic64_inc(&zram->stats.bd_reads);
 814         if (!parent) {
 815                 if (WARN_ON_ONCE(!IS_ENABLED(ZRAM_PARTIAL_IO)))
 816                         return -EIO;
 817                 return read_from_bdev_sync(zram, page, entry);
 818         }
 819         read_from_bdev_async(zram, page, entry, parent);
 820         return 0;
 821 }
 822 #else
 823 static inline void reset_bdev(struct zram *zram) {};
 824 static int read_from_bdev(struct zram *zram, struct page *page,
 825                         unsigned long entry, struct bio *parent)
 826 {
 827         return -EIO;
 828 }
 829
 830 static void free_block_bdev(struct zram *zram, unsigned long blk_idx) {};
 831 #endif
 832
 833 #ifdef CONFIG_ZRAM_MEMORY_TRACKING
 834
 835 static struct dentry *zram_debugfs_root;
 836
 837 static void zram_debugfs_create(void)
 838 {
 839         zram_debugfs_root = debugfs_create_dir("zram", NULL);
 840 }
 841
 842 static void zram_debugfs_destroy(void)
 843 {
 844         debugfs_remove_recursive(zram_debugfs_root);
 845 }
 846
 847 static void zram_accessed(struct zram *zram, u32 index)
 848 {
 849         zram_clear_flag(zram, index, ZRAM_IDLE);
 850         zram->table[index].ac_time = ktime_get_boottime();
 851 }
 852
 853 static ssize_t read_block_state(struct file *file, char __user *buf,
 854                                 size_t count, loff_t *ppos)
 855 {
 856         char *kbuf;
 857         ssize_t index, written = 0;
 858         struct zram *zram = file->private_data;
 859         unsigned long nr_pages = zram->disksize >> PAGE_SHIFT;
 860         struct timespec64 ts;
 861
 862         kbuf = kvmalloc(count, GFP_KERNEL);
 863         if (!kbuf)
 864                 return -ENOMEM;
 865
 866         down_read(&zram->init_lock);
 867         if (!init_done(zram)) {
 868                 up_read(&zram->init_lock);
 869                 kvfree(kbuf);
 870                 return -EINVAL;
 871         }
 872
 873         for (index = *ppos; index < nr_pages; index++) {
 874                 int copied;
 875
 876                 zram_slot_lock(zram, index);
 877                 if (!zram_allocated(zram, index))
 878                         goto next;
 879
 880                 ts = ktime_to_timespec64(zram->table[index].ac_time);
 881                 copied = snprintf(kbuf + written, count,
 882                         "%12zd %12lld.%06lu %c%c%c%c%c%c\n",
 883                         index, (s64)ts.tv_sec,
 884                         ts.tv_nsec / NSEC_PER_USEC,
 885                         zram_test_flag(zram, index, ZRAM_SAME) ? 's' : '.',
 886                         zram_test_flag(zram, index, ZRAM_WB) ? 'w' : '.',
 887                         zram_test_flag(zram, index, ZRAM_HUGE) ? 'h' : '.',
 888                         zram_test_flag(zram, index, ZRAM_IDLE) ? 'i' : '.',
 889                         zram_get_priority(zram, index) ? 'r' : '.',
 890                         zram_test_flag(zram, index,
 891                                        ZRAM_INCOMPRESSIBLE) ? 'n' : '.');
 892
 893                 if (count <= copied) {
 894                         zram_slot_unlock(zram, index);
 895                         break;
 896                 }
 897                 written += copied;
 898                 count -= copied;
 899 next:
 900                 zram_slot_unlock(zram, index);
 901                 *ppos += 1;
 902         }
 903
 904         up_read(&zram->init_lock);
 905         if (copy_to_user(buf, kbuf, written))
 906                 written = -EFAULT;
 907         kvfree(kbuf);
 908
 909         return written;
 910 }
 911
 912 static const struct file_operations proc_zram_block_state_op = {
 913         .open = simple_open,
 914         .read = read_block_state,
 915         .llseek = default_llseek,
 916 };
 917
 918 static void zram_debugfs_register(struct zram *zram)
 919 {
 920         if (!zram_debugfs_root)
 921                 return;
 922
 923         zram->debugfs_dir = debugfs_create_dir(zram->disk->disk_name,
 924                                                 zram_debugfs_root);
 925         debugfs_create_file("block_state", 0400, zram->debugfs_dir,
 926                                 zram, &proc_zram_block_state_op);
 927 }
 928
 929 static void zram_debugfs_unregister(struct zram *zram)
 930 {
 931         debugfs_remove_recursive(zram->debugfs_dir);
 932 }
 933 #else
 934 static void zram_debugfs_create(void) {};
 935 static void zram_debugfs_destroy(void) {};
 936 static void zram_accessed(struct zram *zram, u32 index)
 937 {
 938         zram_clear_flag(zram, index, ZRAM_IDLE);
 939 };
 940 static void zram_debugfs_register(struct zram *zram) {};
 941 static void zram_debugfs_unregister(struct zram *zram) {};
 942 #endif
 943
 944 /*
 945  * We switched to per-cpu streams and this attr is not needed anymore.
 946  * However, we will keep it around for some time, because:
 947  * a) we may revert per-cpu streams in the future
 948  * b) it's visible to user space and we need to follow our 2 years
 949  *    retirement rule; but we already have a number of 'soon to be
 950  *    altered' attrs, so max_comp_streams need to wait for the next
 951  *    layoff cycle.
 952  */
 953 static ssize_t max_comp_streams_show(struct device *dev,
 954                 struct device_attribute *attr, char *buf)
 955 {
 956         return scnprintf(buf, PAGE_SIZE, "%d\n", num_online_cpus());
 957 }
 958
 959 static ssize_t max_comp_streams_store(struct device *dev,
 960                 struct device_attribute *attr, const char *buf, size_t len)
 961 {
 962         return len;
 963 }
 964
 965 static void comp_algorithm_set(struct zram *zram, u32 prio, const char *alg)
 966 {
 967         /* Do not free statically defined compression algorithms */
 968         if (zram->comp_algs[prio] != default_compressor)
 969                 kfree(zram->comp_algs[prio]);
 970
 971         zram->comp_algs[prio] = alg;
 972 }
 973
 974 static ssize_t __comp_algorithm_show(struct zram *zram, u32 prio, char *buf)
 975 {
 976         ssize_t sz;
 977
 978         down_read(&zram->init_lock);
 979         sz = zcomp_available_show(zram->comp_algs[prio], buf);
 980         up_read(&zram->init_lock);
 981
 982         return sz;
 983 }
 984
 985 static int __comp_algorithm_store(struct zram *zram, u32 prio, const char *buf)
 986 {
 987         char *compressor;
 988         size_t sz;
 989
 990         sz = strlen(buf);
 991         if (sz >= CRYPTO_MAX_ALG_NAME)
 992                 return -E2BIG;
 993
 994         compressor = kstrdup(buf, GFP_KERNEL);
 995         if (!compressor)
 996                 return -ENOMEM;
 997
 998         /* ignore trailing newline */
 999         if (sz > 0 && compressor[sz - 1] == '\n')
1000                 compressor[sz - 1] = 0x00;
1001
1002         if (!zcomp_available_algorithm(compressor)) {
1003                 kfree(compressor);
1004                 return -EINVAL;
1005         }
1006
1007         down_write(&zram->init_lock);
1008         if (init_done(zram)) {
1009                 up_write(&zram->init_lock);
1010                 kfree(compressor);
1011                 pr_info("Can't change algorithm for initialized device\n");
1012                 return -EBUSY;
1013         }
1014
1015         comp_algorithm_set(zram, prio, compressor);
1016         up_write(&zram->init_lock);
1017         return 0;
1018 }
1019
1020 static ssize_t comp_algorithm_show(struct device *dev,
1021                                    struct device_attribute *attr,
1022                                    char *buf)
1023 {
1024         struct zram *zram = dev_to_zram(dev);
1025
1026         return __comp_algorithm_show(zram, ZRAM_PRIMARY_COMP, buf);
1027 }
1028
1029 static ssize_t comp_algorithm_store(struct device *dev,
1030                                     struct device_attribute *attr,
1031                                     const char *buf,
1032                                     size_t len)
1033 {
1034         struct zram *zram = dev_to_zram(dev);
1035         int ret;
1036
1037         ret = __comp_algorithm_store(zram, ZRAM_PRIMARY_COMP, buf);
1038         return ret ? ret : len;
1039 }
1040
1041 #ifdef CONFIG_ZRAM_MULTI_COMP
1042 static ssize_t recomp_algorithm_show(struct device *dev,
1043                                      struct device_attribute *attr,
1044                                      char *buf)
1045 {
1046         struct zram *zram = dev_to_zram(dev);
1047         ssize_t sz = 0;
1048         u32 prio;
1049
1050         for (prio = ZRAM_SECONDARY_COMP; prio < ZRAM_MAX_COMPS; prio++) {
1051                 if (!zram->comp_algs[prio])
1052                         continue;
1053
1054                 sz += scnprintf(buf + sz, PAGE_SIZE - sz - 2, "#%d: ", prio);
1055                 sz += __comp_algorithm_show(zram, prio, buf + sz);
1056         }
1057
1058         return sz;
1059 }
1060
1061 static ssize_t recomp_algorithm_store(struct device *dev,
1062                                       struct device_attribute *attr,
1063                                       const char *buf,
1064                                       size_t len)
1065 {
1066         struct zram *zram = dev_to_zram(dev);
1067         int prio = ZRAM_SECONDARY_COMP;
1068         char *args, *param, *val;
1069         char *alg = NULL;
1070         int ret;
1071
1072         args = skip_spaces(buf);
1073         while (*args) {
1074                 args = next_arg(args, &param, &val);
1075
1076                 if (!val || !*val)
1077                         return -EINVAL;
1078
1079                 if (!strcmp(param, "algo")) {
1080                         alg = val;
1081                         continue;
1082                 }
1083
1084                 if (!strcmp(param, "priority")) {
1085                         ret = kstrtoint(val, 10, &prio);
1086                         if (ret)
1087                                 return ret;
1088                         continue;
1089                 }
1090         }
1091
1092         if (!alg)
1093                 return -EINVAL;
1094
1095         if (prio < ZRAM_SECONDARY_COMP || prio >= ZRAM_MAX_COMPS)
1096                 return -EINVAL;
1097
1098         ret = __comp_algorithm_store(zram, prio, alg);
1099         return ret ? ret : len;
1100 }
1101 #endif
1102
1103 static ssize_t compact_store(struct device *dev,
1104                 struct device_attribute *attr, const char *buf, size_t len)
1105 {
1106         struct zram *zram = dev_to_zram(dev);
1107
1108         down_read(&zram->init_lock);
1109         if (!init_done(zram)) {
1110                 up_read(&zram->init_lock);
1111                 return -EINVAL;
1112         }
1113
1114         zs_compact(zram->mem_pool);
1115         up_read(&zram->init_lock);
1116
1117         return len;
1118 }
1119
1120 static ssize_t io_stat_show(struct device *dev,
1121                 struct device_attribute *attr, char *buf)
1122 {
1123         struct zram *zram = dev_to_zram(dev);
1124         ssize_t ret;
1125
1126         down_read(&zram->init_lock);
1127         ret = scnprintf(buf, PAGE_SIZE,
1128                         "%8llu %8llu 0 %8llu\n",
1129                         (u64)atomic64_read(&zram->stats.failed_reads),
1130                         (u64)atomic64_read(&zram->stats.failed_writes),
1131                         (u64)atomic64_read(&zram->stats.notify_free));
1132         up_read(&zram->init_lock);
1133
1134         return ret;
1135 }
1136
1137 static ssize_t mm_stat_show(struct device *dev,
1138                 struct device_attribute *attr, char *buf)
1139 {
1140         struct zram *zram = dev_to_zram(dev);
1141         struct zs_pool_stats pool_stats;
1142         u64 orig_size, mem_used = 0;
1143         long max_used;
1144         ssize_t ret;
1145
1146         memset(&pool_stats, 0x00, sizeof(struct zs_pool_stats));
1147
1148         down_read(&zram->init_lock);
1149         if (init_done(zram)) {
1150                 mem_used = zs_get_total_pages(zram->mem_pool);
1151                 zs_pool_stats(zram->mem_pool, &pool_stats);
1152         }
1153
1154         orig_size = atomic64_read(&zram->stats.pages_stored);
1155         max_used = atomic_long_read(&zram->stats.max_used_pages);
1156
1157         ret = scnprintf(buf, PAGE_SIZE,
1158                         "%8llu %8llu %8llu %8lu %8ld %8llu %8lu %8llu %8llu\n",
1159                         orig_size << PAGE_SHIFT,
1160                         (u64)atomic64_read(&zram->stats.compr_data_size),
1161                         mem_used << PAGE_SHIFT,
1162                         zram->limit_pages << PAGE_SHIFT,
1163                         max_used << PAGE_SHIFT,
1164                         (u64)atomic64_read(&zram->stats.same_pages),
1165                         atomic_long_read(&pool_stats.pages_compacted),
1166                         (u64)atomic64_read(&zram->stats.huge_pages),
1167                         (u64)atomic64_read(&zram->stats.huge_pages_since));
1168         up_read(&zram->init_lock);
1169
1170         return ret;
1171 }
1172
1173 #ifdef CONFIG_ZRAM_WRITEBACK
1174 #define FOUR_K(x) ((x) * (1 << (PAGE_SHIFT - 12)))
1175 static ssize_t bd_stat_show(struct device *dev,
1176                 struct device_attribute *attr, char *buf)
1177 {
1178         struct zram *zram = dev_to_zram(dev);
1179         ssize_t ret;
1180
1181         down_read(&zram->init_lock);
1182         ret = scnprintf(buf, PAGE_SIZE,
1183                 "%8llu %8llu %8llu\n",
1184                         FOUR_K((u64)atomic64_read(&zram->stats.bd_count)),
1185                         FOUR_K((u64)atomic64_read(&zram->stats.bd_reads)),
1186                         FOUR_K((u64)atomic64_read(&zram->stats.bd_writes)));
1187         up_read(&zram->init_lock);
1188
1189         return ret;
1190 }
1191 #endif
1192
1193 static ssize_t debug_stat_show(struct device *dev,
1194                 struct device_attribute *attr, char *buf)
1195 {
1196         int version = 1;
1197         struct zram *zram = dev_to_zram(dev);
1198         ssize_t ret;
1199
1200         down_read(&zram->init_lock);
1201         ret = scnprintf(buf, PAGE_SIZE,
1202                         "version: %d\n%8llu %8llu\n",
1203                         version,
1204                         (u64)atomic64_read(&zram->stats.writestall),
1205                         (u64)atomic64_read(&zram->stats.miss_free));
1206         up_read(&zram->init_lock);
1207
1208         return ret;
1209 }
1210
1211 static DEVICE_ATTR_RO(io_stat);
1212 static DEVICE_ATTR_RO(mm_stat);
1213 #ifdef CONFIG_ZRAM_WRITEBACK
1214 static DEVICE_ATTR_RO(bd_stat);
1215 #endif
1216 static DEVICE_ATTR_RO(debug_stat);
1217
1218 static void zram_meta_free(struct zram *zram, u64 disksize)
1219 {
1220         size_t num_pages = disksize >> PAGE_SHIFT;
1221         size_t index;
1222
1223         /* Free all pages that are still in this zram device */
1224         for (index = 0; index < num_pages; index++)
1225                 zram_free_page(zram, index);
1226
1227         zs_destroy_pool(zram->mem_pool);
1228         vfree(zram->table);
1229 }
1230
1231 static bool zram_meta_alloc(struct zram *zram, u64 disksize)
1232 {
1233         size_t num_pages;
1234
1235         num_pages = disksize >> PAGE_SHIFT;
1236         zram->table = vzalloc(array_size(num_pages, sizeof(*zram->table)));
1237         if (!zram->table)
1238                 return false;
1239
1240         zram->mem_pool = zs_create_pool(zram->disk->disk_name);
1241         if (!zram->mem_pool) {
1242                 vfree(zram->table);
1243                 return false;
1244         }
1245
1246         if (!huge_class_size)
1247                 huge_class_size = zs_huge_class_size(zram->mem_pool);
1248         return true;
1249 }
1250
1251 /*
1252  * To protect concurrent access to the same index entry,
1253  * caller should hold this table index entry's bit_spinlock to
1254  * indicate this index entry is accessing.
1255  */
1256 static void zram_free_page(struct zram *zram, size_t index)
1257 {
1258         unsigned long handle;
1259
1260 #ifdef CONFIG_ZRAM_MEMORY_TRACKING
1261         zram->table[index].ac_time = 0;
1262 #endif
1263         if (zram_test_flag(zram, index, ZRAM_IDLE))
1264                 zram_clear_flag(zram, index, ZRAM_IDLE);
1265
1266         if (zram_test_flag(zram, index, ZRAM_HUGE)) {
1267                 zram_clear_flag(zram, index, ZRAM_HUGE);
1268                 atomic64_dec(&zram->stats.huge_pages);
1269         }
1270
1271         if (zram_test_flag(zram, index, ZRAM_INCOMPRESSIBLE))
1272                 zram_clear_flag(zram, index, ZRAM_INCOMPRESSIBLE);
1273
1274         zram_set_priority(zram, index, 0);
1275
1276         if (zram_test_flag(zram, index, ZRAM_WB)) {
1277                 zram_clear_flag(zram, index, ZRAM_WB);
1278                 free_block_bdev(zram, zram_get_element(zram, index));
1279                 goto out;
1280         }
1281
1282         /*
1283          * No memory is allocated for same element filled pages.
1284          * Simply clear same page flag.
1285          */
1286         if (zram_test_flag(zram, index, ZRAM_SAME)) {
1287                 zram_clear_flag(zram, index, ZRAM_SAME);
1288                 atomic64_dec(&zram->stats.same_pages);
1289                 goto out;
1290         }
1291
1292         handle = zram_get_handle(zram, index);
1293         if (!handle)
1294                 return;
1295
1296         zs_free(zram->mem_pool, handle);
1297
1298         atomic64_sub(zram_get_obj_size(zram, index),
1299                         &zram->stats.compr_data_size);
1300 out:
1301         atomic64_dec(&zram->stats.pages_stored);
1302         zram_set_handle(zram, index, 0);
1303         zram_set_obj_size(zram, index, 0);
1304         WARN_ON_ONCE(zram->table[index].flags &
1305                 ~(1UL << ZRAM_LOCK | 1UL << ZRAM_UNDER_WB));
1306 }
1307
1308 /*
1309  * Reads (decompresses if needed) a page from zspool (zsmalloc).
1310  * Corresponding ZRAM slot should be locked.
1311  */
1312 static int zram_read_from_zspool(struct zram *zram, struct page *page,
1313                                  u32 index)
1314 {
1315         struct zcomp_strm *zstrm;
1316         unsigned long handle;
1317         unsigned int size;
1318         void *src, *dst;
1319         u32 prio;
1320         int ret;
1321
1322         handle = zram_get_handle(zram, index);
1323         if (!handle || zram_test_flag(zram, index, ZRAM_SAME)) {
1324                 unsigned long value;
1325                 void *mem;
1326
1327                 value = handle ? zram_get_element(zram, index) : 0;
1328                 mem = kmap_atomic(page);
1329                 zram_fill_page(mem, PAGE_SIZE, value);
1330                 kunmap_atomic(mem);
1331                 return 0;
1332         }
1333
1334         size = zram_get_obj_size(zram, index);
1335
1336         if (size != PAGE_SIZE) {
1337                 prio = zram_get_priority(zram, index);
1338                 zstrm = zcomp_stream_get(zram->comps[prio]);
1339         }
1340
1341         src = zs_map_object(zram->mem_pool, handle, ZS_MM_RO);
1342         if (size == PAGE_SIZE) {
1343                 dst = kmap_atomic(page);
1344                 memcpy(dst, src, PAGE_SIZE);
1345                 kunmap_atomic(dst);
1346                 ret = 0;
1347         } else {
1348                 dst = kmap_atomic(page);
1349                 ret = zcomp_decompress(zstrm, src, size, dst);
1350                 kunmap_atomic(dst);
1351                 zcomp_stream_put(zram->comps[prio]);
1352         }
1353         zs_unmap_object(zram->mem_pool, handle);
1354         return ret;
1355 }
1356
1357 static int zram_read_page(struct zram *zram, struct page *page, u32 index,
1358                           struct bio *parent)
1359 {
1360         int ret;
1361
1362         zram_slot_lock(zram, index);
1363         if (!zram_test_flag(zram, index, ZRAM_WB)) {
1364                 /* Slot should be locked through out the function call */
1365                 ret = zram_read_from_zspool(zram, page, index);
1366                 zram_slot_unlock(zram, index);
1367         } else {
1368                 /*
1369                  * The slot should be unlocked before reading from the backing
1370                  * device.
1371                  */
1372                 zram_slot_unlock(zram, index);
1373
1374                 ret = read_from_bdev(zram, page, zram_get_element(zram, index),
1375                                      parent);
1376         }
1377
1378         /* Should NEVER happen. Return bio error if it does. */
1379         if (WARN_ON(ret < 0))
1380                 pr_err("Decompression failed! err=%d, page=%u\n", ret, index);
1381
1382         return ret;
1383 }
1384
1385 /*
1386  * Use a temporary buffer to decompress the page, as the decompressor
1387  * always expects a full page for the output.
1388  */
1389 static int zram_bvec_read_partial(struct zram *zram, struct bio_vec *bvec,
1390                                   u32 index, int offset)
1391 {
1392         struct page *page = alloc_page(GFP_NOIO);
1393         int ret;
1394
1395         if (!page)
1396                 return -ENOMEM;
1397         ret = zram_read_page(zram, page, index, NULL);
1398         if (likely(!ret))
1399                 memcpy_to_bvec(bvec, page_address(page) + offset);
1400         __free_page(page);
1401         return ret;
1402 }
1403
1404 static int zram_bvec_read(struct zram *zram, struct bio_vec *bvec,
1405                           u32 index, int offset, struct bio *bio)
1406 {
1407         if (is_partial_io(bvec))
1408                 return zram_bvec_read_partial(zram, bvec, index, offset);
1409         return zram_read_page(zram, bvec->bv_page, index, bio);
1410 }
1411
1412 static int zram_write_page(struct zram *zram, struct page *page, u32 index)
1413 {
1414         int ret = 0;
1415         unsigned long alloced_pages;
1416         unsigned long handle = -ENOMEM;
1417         unsigned int comp_len = 0;
1418         void *src, *dst, *mem;
1419         struct zcomp_strm *zstrm;
1420         unsigned long element = 0;
1421         enum zram_pageflags flags = 0;
1422
1423         mem = kmap_atomic(page);
1424         if (page_same_filled(mem, &element)) {
1425                 kunmap_atomic(mem);
1426                 /* Free memory associated with this sector now. */
1427                 flags = ZRAM_SAME;
1428                 atomic64_inc(&zram->stats.same_pages);
1429                 goto out;
1430         }
1431         kunmap_atomic(mem);
1432
1433 compress_again:
1434         zstrm = zcomp_stream_get(zram->comps[ZRAM_PRIMARY_COMP]);
1435         src = kmap_atomic(page);
1436         ret = zcomp_compress(zstrm, src, &comp_len);
1437         kunmap_atomic(src);
1438
1439         if (unlikely(ret)) {
1440                 zcomp_stream_put(zram->comps[ZRAM_PRIMARY_COMP]);
1441                 pr_err("Compression failed! err=%d\n", ret);
1442                 zs_free(zram->mem_pool, handle);
1443                 return ret;
1444         }
1445
1446         if (comp_len >= huge_class_size)
1447                 comp_len = PAGE_SIZE;
1448         /*
1449          * handle allocation has 2 paths:
1450          * a) fast path is executed with preemption disabled (for
1451          *  per-cpu streams) and has __GFP_DIRECT_RECLAIM bit clear,
1452          *  since we can't sleep;
1453          * b) slow path enables preemption and attempts to allocate
1454          *  the page with __GFP_DIRECT_RECLAIM bit set. we have to
1455          *  put per-cpu compression stream and, thus, to re-do
1456          *  the compression once handle is allocated.
1457          *
1458          * if we have a 'non-null' handle here then we are coming
1459          * from the slow path and handle has already been allocated.
1460          */
1461         if (IS_ERR_VALUE(handle))
1462                 handle = zs_malloc(zram->mem_pool, comp_len,
1463                                 __GFP_KSWAPD_RECLAIM |
1464                                 __GFP_NOWARN |
1465                                 __GFP_HIGHMEM |
1466                                 __GFP_MOVABLE);
1467         if (IS_ERR_VALUE(handle)) {
1468                 zcomp_stream_put(zram->comps[ZRAM_PRIMARY_COMP]);
1469                 atomic64_inc(&zram->stats.writestall);
1470                 handle = zs_malloc(zram->mem_pool, comp_len,
1471                                 GFP_NOIO | __GFP_HIGHMEM |
1472                                 __GFP_MOVABLE);
1473                 if (IS_ERR_VALUE(handle))
1474                         return PTR_ERR((void *)handle);
1475
1476                 if (comp_len != PAGE_SIZE)
1477                         goto compress_again;
1478                 /*
1479                  * If the page is not compressible, you need to acquire the
1480                  * lock and execute the code below. The zcomp_stream_get()
1481                  * call is needed to disable the cpu hotplug and grab the
1482                  * zstrm buffer back. It is necessary that the dereferencing
1483                  * of the zstrm variable below occurs correctly.
1484                  */
1485                 zstrm = zcomp_stream_get(zram->comps[ZRAM_PRIMARY_COMP]);
1486         }
1487
1488         alloced_pages = zs_get_total_pages(zram->mem_pool);
1489         update_used_max(zram, alloced_pages);
1490
1491         if (zram->limit_pages && alloced_pages > zram->limit_pages) {
1492                 zcomp_stream_put(zram->comps[ZRAM_PRIMARY_COMP]);
1493                 zs_free(zram->mem_pool, handle);
1494                 return -ENOMEM;
1495         }
1496
1497         dst = zs_map_object(zram->mem_pool, handle, ZS_MM_WO);
1498
1499         src = zstrm->buffer;
1500         if (comp_len == PAGE_SIZE)
1501                 src = kmap_atomic(page);
1502         memcpy(dst, src, comp_len);
1503         if (comp_len == PAGE_SIZE)
1504                 kunmap_atomic(src);
1505
1506         zcomp_stream_put(zram->comps[ZRAM_PRIMARY_COMP]);
1507         zs_unmap_object(zram->mem_pool, handle);
1508         atomic64_add(comp_len, &zram->stats.compr_data_size);
1509 out:
1510         /*
1511          * Free memory associated with this sector
1512          * before overwriting unused sectors.
1513          */
1514         zram_slot_lock(zram, index);
1515         zram_free_page(zram, index);
1516
1517         if (comp_len == PAGE_SIZE) {
1518                 zram_set_flag(zram, index, ZRAM_HUGE);
1519                 atomic64_inc(&zram->stats.huge_pages);
1520                 atomic64_inc(&zram->stats.huge_pages_since);
1521         }
1522
1523         if (flags) {
1524                 zram_set_flag(zram, index, flags);
1525                 zram_set_element(zram, index, element);
1526         }  else {
1527                 zram_set_handle(zram, index, handle);
1528                 zram_set_obj_size(zram, index, comp_len);
1529         }
1530         zram_slot_unlock(zram, index);
1531
1532         /* Update stats */
1533         atomic64_inc(&zram->stats.pages_stored);
1534         return ret;
1535 }
1536
1537 /*
1538  * This is a partial IO. Read the full page before writing the changes.
1539  */
1540 static int zram_bvec_write_partial(struct zram *zram, struct bio_vec *bvec,
1541                                    u32 index, int offset, struct bio *bio)
1542 {
1543         struct page *page = alloc_page(GFP_NOIO);
1544         int ret;
1545
1546         if (!page)
1547                 return -ENOMEM;
1548
1549         ret = zram_read_page(zram, page, index, bio);
1550         if (!ret) {
1551                 memcpy_from_bvec(page_address(page) + offset, bvec);
1552                 ret = zram_write_page(zram, page, index);
1553         }
1554         __free_page(page);
1555         return ret;
1556 }
1557
1558 static int zram_bvec_write(struct zram *zram, struct bio_vec *bvec,
1559                            u32 index, int offset, struct bio *bio)
1560 {
1561         if (is_partial_io(bvec))
1562                 return zram_bvec_write_partial(zram, bvec, index, offset, bio);
1563         return zram_write_page(zram, bvec->bv_page, index);
1564 }
1565
1566 #ifdef CONFIG_ZRAM_MULTI_COMP
1567 /*
1568  * This function will decompress (unless it's ZRAM_HUGE) the page and then
1569  * attempt to compress it using provided compression algorithm priority
1570  * (which is potentially more effective).
1571  *
1572  * Corresponding ZRAM slot should be locked.
1573  */
1574 static int zram_recompress(struct zram *zram, u32 index, struct page *page,
1575                            u32 threshold, u32 prio, u32 prio_max)
1576 {
1577         struct zcomp_strm *zstrm = NULL;
1578         unsigned long handle_old;
1579         unsigned long handle_new;
1580         unsigned int comp_len_old;
1581         unsigned int comp_len_new;
1582         unsigned int class_index_old;
1583         unsigned int class_index_new;
1584         u32 num_recomps = 0;
1585         void *src, *dst;
1586         int ret;
1587
1588         handle_old = zram_get_handle(zram, index);
1589         if (!handle_old)
1590                 return -EINVAL;
1591
1592         comp_len_old = zram_get_obj_size(zram, index);
1593         /*
1594          * Do not recompress objects that are already "small enough".
1595          */
1596         if (comp_len_old < threshold)
1597                 return 0;
1598
1599         ret = zram_read_from_zspool(zram, page, index);
1600         if (ret)
1601                 return ret;
1602
1603         class_index_old = zs_lookup_class_index(zram->mem_pool, comp_len_old);
1604         /*
1605          * Iterate the secondary comp algorithms list (in order of priority)
1606          * and try to recompress the page.
1607          */
1608         for (; prio < prio_max; prio++) {
1609                 if (!zram->comps[prio])
1610                         continue;
1611
1612                 /*
1613                  * Skip if the object is already re-compressed with a higher
1614                  * priority algorithm (or same algorithm).
1615                  */
1616                 if (prio <= zram_get_priority(zram, index))
1617                         continue;
1618
1619                 num_recomps++;
1620                 zstrm = zcomp_stream_get(zram->comps[prio]);
1621                 src = kmap_atomic(page);
1622                 ret = zcomp_compress(zstrm, src, &comp_len_new);
1623                 kunmap_atomic(src);
1624
1625                 if (ret) {
1626                         zcomp_stream_put(zram->comps[prio]);
1627                         return ret;
1628                 }
1629
1630                 class_index_new = zs_lookup_class_index(zram->mem_pool,
1631                                                         comp_len_new);
1632
1633                 /* Continue until we make progress */
1634                 if (class_index_new >= class_index_old ||
1635                     (threshold && comp_len_new >= threshold)) {
1636                         zcomp_stream_put(zram->comps[prio]);
1637                         continue;
1638                 }
1639
1640                 /* Recompression was successful so break out */
1641                 break;
1642         }
1643
1644         /*
1645          * We did not try to recompress, e.g. when we have only one
1646          * secondary algorithm and the page is already recompressed
1647          * using that algorithm
1648          */
1649         if (!zstrm)
1650                 return 0;
1651
1652         if (class_index_new >= class_index_old) {
1653                 /*
1654                  * Secondary algorithms failed to re-compress the page
1655                  * in a way that would save memory, mark the object as
1656                  * incompressible so that we will not try to compress
1657                  * it again.
1658                  *
1659                  * We need to make sure that all secondary algorithms have
1660                  * failed, so we test if the number of recompressions matches
1661                  * the number of active secondary algorithms.
1662                  */
1663                 if (num_recomps == zram->num_active_comps - 1)
1664                         zram_set_flag(zram, index, ZRAM_INCOMPRESSIBLE);
1665                 return 0;
1666         }
1667
1668         /* Successful recompression but above threshold */
1669         if (threshold && comp_len_new >= threshold)
1670                 return 0;
1671
1672         /*
1673          * No direct reclaim (slow path) for handle allocation and no
1674          * re-compression attempt (unlike in zram_write_bvec()) since
1675          * we already have stored that object in zsmalloc. If we cannot
1676          * alloc memory for recompressed object then we bail out and
1677          * simply keep the old (existing) object in zsmalloc.
1678          */
1679         handle_new = zs_malloc(zram->mem_pool, comp_len_new,
1680                                __GFP_KSWAPD_RECLAIM |
1681                                __GFP_NOWARN |
1682                                __GFP_HIGHMEM |
1683                                __GFP_MOVABLE);
1684         if (IS_ERR_VALUE(handle_new)) {
1685                 zcomp_stream_put(zram->comps[prio]);
1686                 return PTR_ERR((void *)handle_new);
1687         }
1688
1689         dst = zs_map_object(zram->mem_pool, handle_new, ZS_MM_WO);
1690         memcpy(dst, zstrm->buffer, comp_len_new);
1691         zcomp_stream_put(zram->comps[prio]);
1692
1693         zs_unmap_object(zram->mem_pool, handle_new);
1694
1695         zram_free_page(zram, index);
1696         zram_set_handle(zram, index, handle_new);
1697         zram_set_obj_size(zram, index, comp_len_new);
1698         zram_set_priority(zram, index, prio);
1699
1700         atomic64_add(comp_len_new, &zram->stats.compr_data_size);
1701         atomic64_inc(&zram->stats.pages_stored);
1702
1703         return 0;
1704 }
1705
1706 #define RECOMPRESS_IDLE         (1 << 0)
1707 #define RECOMPRESS_HUGE         (1 << 1)
1708
1709 static ssize_t recompress_store(struct device *dev,
1710                                 struct device_attribute *attr,
1711                                 const char *buf, size_t len)
1712 {
1713         u32 prio = ZRAM_SECONDARY_COMP, prio_max = ZRAM_MAX_COMPS;
1714         struct zram *zram = dev_to_zram(dev);
1715         unsigned long nr_pages = zram->disksize >> PAGE_SHIFT;
1716         char *args, *param, *val, *algo = NULL;
1717         u32 mode = 0, threshold = 0;
1718         unsigned long index;
1719         struct page *page;
1720         ssize_t ret;
1721
1722         args = skip_spaces(buf);
1723         while (*args) {
1724                 args = next_arg(args, &param, &val);
1725
1726                 if (!val || !*val)
1727                         return -EINVAL;
1728
1729                 if (!strcmp(param, "type")) {
1730                         if (!strcmp(val, "idle"))
1731                                 mode = RECOMPRESS_IDLE;
1732                         if (!strcmp(val, "huge"))
1733                                 mode = RECOMPRESS_HUGE;
1734                         if (!strcmp(val, "huge_idle"))
1735                                 mode = RECOMPRESS_IDLE | RECOMPRESS_HUGE;
1736                         continue;
1737                 }
1738
1739                 if (!strcmp(param, "threshold")) {
1740                         /*
1741                          * We will re-compress only idle objects equal or
1742                          * greater in size than watermark.
1743                          */
1744                         ret = kstrtouint(val, 10, &threshold);
1745                         if (ret)
1746                                 return ret;
1747                         continue;
1748                 }
1749
1750                 if (!strcmp(param, "algo")) {
1751                         algo = val;
1752                         continue;
1753                 }
1754         }
1755
1756         if (threshold >= huge_class_size)
1757                 return -EINVAL;
1758
1759         down_read(&zram->init_lock);
1760         if (!init_done(zram)) {
1761                 ret = -EINVAL;
1762                 goto release_init_lock;
1763         }
1764
1765         if (algo) {
1766                 bool found = false;
1767
1768                 for (; prio < ZRAM_MAX_COMPS; prio++) {
1769                         if (!zram->comp_algs[prio])
1770                                 continue;
1771
1772                         if (!strcmp(zram->comp_algs[prio], algo)) {
1773                                 prio_max = min(prio + 1, ZRAM_MAX_COMPS);
1774                                 found = true;
1775                                 break;
1776                         }
1777                 }
1778
1779                 if (!found) {
1780                         ret = -EINVAL;
1781                         goto release_init_lock;
1782                 }
1783         }
1784
1785         page = alloc_page(GFP_KERNEL);
1786         if (!page) {
1787                 ret = -ENOMEM;
1788                 goto release_init_lock;
1789         }
1790
1791         ret = len;
1792         for (index = 0; index < nr_pages; index++) {
1793                 int err = 0;
1794
1795                 zram_slot_lock(zram, index);
1796
1797                 if (!zram_allocated(zram, index))
1798                         goto next;
1799
1800                 if (mode & RECOMPRESS_IDLE &&
1801                     !zram_test_flag(zram, index, ZRAM_IDLE))
1802                         goto next;
1803
1804                 if (mode & RECOMPRESS_HUGE &&
1805                     !zram_test_flag(zram, index, ZRAM_HUGE))
1806                         goto next;
1807
1808                 if (zram_test_flag(zram, index, ZRAM_WB) ||
1809                     zram_test_flag(zram, index, ZRAM_UNDER_WB) ||
1810                     zram_test_flag(zram, index, ZRAM_SAME) ||
1811                     zram_test_flag(zram, index, ZRAM_INCOMPRESSIBLE))
1812                         goto next;
1813
1814                 err = zram_recompress(zram, index, page, threshold,
1815                                       prio, prio_max);
1816 next:
1817                 zram_slot_unlock(zram, index);
1818                 if (err) {
1819                         ret = err;
1820                         break;
1821                 }
1822
1823                 cond_resched();
1824         }
1825
1826         __free_page(page);
1827
1828 release_init_lock:
1829         up_read(&zram->init_lock);
1830         return ret;
1831 }
1832 #endif
1833
1834 static void zram_bio_discard(struct zram *zram, struct bio *bio)
1835 {
1836         size_t n = bio->bi_iter.bi_size;
1837         u32 index = bio->bi_iter.bi_sector >> SECTORS_PER_PAGE_SHIFT;
1838         u32 offset = (bio->bi_iter.bi_sector & (SECTORS_PER_PAGE - 1)) <<
1839                         SECTOR_SHIFT;
1840
1841         /*
1842          * zram manages data in physical block size units. Because logical block
1843          * size isn't identical with physical block size on some arch, we
1844          * could get a discard request pointing to a specific offset within a
1845          * certain physical block.  Although we can handle this request by
1846          * reading that physiclal block and decompressing and partially zeroing
1847          * and re-compressing and then re-storing it, this isn't reasonable
1848          * because our intent with a discard request is to save memory.  So
1849          * skipping this logical block is appropriate here.
1850          */
1851         if (offset) {
1852                 if (n <= (PAGE_SIZE - offset))
1853                         return;
1854
1855                 n -= (PAGE_SIZE - offset);
1856                 index++;
1857         }
1858
1859         while (n >= PAGE_SIZE) {
1860                 zram_slot_lock(zram, index);
1861                 zram_free_page(zram, index);
1862                 zram_slot_unlock(zram, index);
1863                 atomic64_inc(&zram->stats.notify_free);
1864                 index++;
1865                 n -= PAGE_SIZE;
1866         }
1867
1868         bio_endio(bio);
1869 }
1870
1871 static void zram_bio_read(struct zram *zram, struct bio *bio)
1872 {
1873         unsigned long start_time = bio_start_io_acct(bio);
1874         struct bvec_iter iter = bio->bi_iter;
1875
1876         do {
1877                 u32 index = iter.bi_sector >> SECTORS_PER_PAGE_SHIFT;
1878                 u32 offset = (iter.bi_sector & (SECTORS_PER_PAGE - 1)) <<
1879                                 SECTOR_SHIFT;
1880                 struct bio_vec bv = bio_iter_iovec(bio, iter);
1881
1882                 bv.bv_len = min_t(u32, bv.bv_len, PAGE_SIZE - offset);
1883
1884                 if (zram_bvec_read(zram, &bv, index, offset, bio) < 0) {
1885                         atomic64_inc(&zram->stats.failed_reads);
1886                         bio->bi_status = BLK_STS_IOERR;
1887                         break;
1888                 }
1889                 flush_dcache_page(bv.bv_page);
1890
1891                 zram_slot_lock(zram, index);
1892                 zram_accessed(zram, index);
1893                 zram_slot_unlock(zram, index);
1894
1895                 bio_advance_iter_single(bio, &iter, bv.bv_len);
1896         } while (iter.bi_size);
1897
1898         bio_end_io_acct(bio, start_time);
1899         bio_endio(bio);
1900 }
1901
1902 static void zram_bio_write(struct zram *zram, struct bio *bio)
1903 {
1904         unsigned long start_time = bio_start_io_acct(bio);
1905         struct bvec_iter iter = bio->bi_iter;
1906
1907         do {
1908                 u32 index = iter.bi_sector >> SECTORS_PER_PAGE_SHIFT;
1909                 u32 offset = (iter.bi_sector & (SECTORS_PER_PAGE - 1)) <<
1910                                 SECTOR_SHIFT;
1911                 struct bio_vec bv = bio_iter_iovec(bio, iter);
1912
1913                 bv.bv_len = min_t(u32, bv.bv_len, PAGE_SIZE - offset);
1914
1915                 if (zram_bvec_write(zram, &bv, index, offset, bio) < 0) {
1916                         atomic64_inc(&zram->stats.failed_writes);
1917                         bio->bi_status = BLK_STS_IOERR;
1918                         break;
1919                 }
1920
1921                 zram_slot_lock(zram, index);
1922                 zram_accessed(zram, index);
1923                 zram_slot_unlock(zram, index);
1924
1925                 bio_advance_iter_single(bio, &iter, bv.bv_len);
1926         } while (iter.bi_size);
1927
1928         bio_end_io_acct(bio, start_time);
1929         bio_endio(bio);
1930 }
1931
1932 /*
1933  * Handler function for all zram I/O requests.
1934  */
1935 static void zram_submit_bio(struct bio *bio)
1936 {
1937         struct zram *zram = bio->bi_bdev->bd_disk->private_data;
1938
1939         switch (bio_op(bio)) {
1940         case REQ_OP_READ:
1941                 zram_bio_read(zram, bio);
1942                 break;
1943         case REQ_OP_WRITE:
1944                 zram_bio_write(zram, bio);
1945                 break;
1946         case REQ_OP_DISCARD:
1947         case REQ_OP_WRITE_ZEROES:
1948                 zram_bio_discard(zram, bio);
1949                 break;
1950         default:
1951                 WARN_ON_ONCE(1);
1952                 bio_endio(bio);
1953         }
1954 }
1955
1956 static void zram_slot_free_notify(struct block_device *bdev,
1957                                 unsigned long index)
1958 {
1959         struct zram *zram;
1960
1961         zram = bdev->bd_disk->private_data;
1962
1963         atomic64_inc(&zram->stats.notify_free);
1964         if (!zram_slot_trylock(zram, index)) {
1965                 atomic64_inc(&zram->stats.miss_free);
1966                 return;
1967         }
1968
1969         zram_free_page(zram, index);
1970         zram_slot_unlock(zram, index);
1971 }
1972
1973 static void zram_destroy_comps(struct zram *zram)
1974 {
1975         u32 prio;
1976
1977         for (prio = 0; prio < ZRAM_MAX_COMPS; prio++) {
1978                 struct zcomp *comp = zram->comps[prio];
1979
1980                 zram->comps[prio] = NULL;
1981                 if (!comp)
1982                         continue;
1983                 zcomp_destroy(comp);
1984                 zram->num_active_comps--;
1985         }
1986 }
1987
1988 static void zram_reset_device(struct zram *zram)
1989 {
1990         down_write(&zram->init_lock);
1991
1992         zram->limit_pages = 0;
1993
1994         if (!init_done(zram)) {
1995                 up_write(&zram->init_lock);
1996                 return;
1997         }
1998
1999         set_capacity_and_notify(zram->disk, 0);
2000         part_stat_set_all(zram->disk->part0, 0);
2001
2002         /* I/O operation under all of CPU are done so let's free */
2003         zram_meta_free(zram, zram->disksize);
2004         zram->disksize = 0;
2005         zram_destroy_comps(zram);
2006         memset(&zram->stats, 0, sizeof(zram->stats));
2007         reset_bdev(zram);
2008
2009         comp_algorithm_set(zram, ZRAM_PRIMARY_COMP, default_compressor);
2010         up_write(&zram->init_lock);
2011 }
2012
2013 static ssize_t disksize_store(struct device *dev,
2014                 struct device_attribute *attr, const char *buf, size_t len)
2015 {
2016         u64 disksize;
2017         struct zcomp *comp;
2018         struct zram *zram = dev_to_zram(dev);
2019         int err;
2020         u32 prio;
2021
2022         disksize = memparse(buf, NULL);
2023         if (!disksize)
2024                 return -EINVAL;
2025
2026         down_write(&zram->init_lock);
2027         if (init_done(zram)) {
2028                 pr_info("Cannot change disksize for initialized device\n");
2029                 err = -EBUSY;
2030                 goto out_unlock;
2031         }
2032
2033         disksize = PAGE_ALIGN(disksize);
2034         if (!zram_meta_alloc(zram, disksize)) {
2035                 err = -ENOMEM;
2036                 goto out_unlock;
2037         }
2038
2039         for (prio = 0; prio < ZRAM_MAX_COMPS; prio++) {
2040                 if (!zram->comp_algs[prio])
2041                         continue;
2042
2043                 comp = zcomp_create(zram->comp_algs[prio]);
2044                 if (IS_ERR(comp)) {
2045                         pr_err("Cannot initialise %s compressing backend\n",
2046                                zram->comp_algs[prio]);
2047                         err = PTR_ERR(comp);
2048                         goto out_free_comps;
2049                 }
2050
2051                 zram->comps[prio] = comp;
2052                 zram->num_active_comps++;
2053         }
2054         zram->disksize = disksize;
2055         set_capacity_and_notify(zram->disk, zram->disksize >> SECTOR_SHIFT);
2056         up_write(&zram->init_lock);
2057
2058         return len;
2059
2060 out_free_comps:
2061         zram_destroy_comps(zram);
2062         zram_meta_free(zram, disksize);
2063 out_unlock:
2064         up_write(&zram->init_lock);
2065         return err;
2066 }
2067
2068 static ssize_t reset_store(struct device *dev,
2069                 struct device_attribute *attr, const char *buf, size_t len)
2070 {
2071         int ret;
2072         unsigned short do_reset;
2073         struct zram *zram;
2074         struct gendisk *disk;
2075
2076         ret = kstrtou16(buf, 10, &do_reset);
2077         if (ret)
2078                 return ret;
2079
2080         if (!do_reset)
2081                 return -EINVAL;
2082
2083         zram = dev_to_zram(dev);
2084         disk = zram->disk;
2085
2086         mutex_lock(&disk->open_mutex);
2087         /* Do not reset an active device or claimed device */
2088         if (disk_openers(disk) || zram->claim) {
2089                 mutex_unlock(&disk->open_mutex);
2090                 return -EBUSY;
2091         }
2092
2093         /* From now on, anyone can't open /dev/zram[0-9] */
2094         zram->claim = true;
2095         mutex_unlock(&disk->open_mutex);
2096
2097         /* Make sure all the pending I/O are finished */
2098         sync_blockdev(disk->part0);
2099         zram_reset_device(zram);
2100
2101         mutex_lock(&disk->open_mutex);
2102         zram->claim = false;
2103         mutex_unlock(&disk->open_mutex);
2104
2105         return len;
2106 }
2107
2108 static int zram_open(struct gendisk *disk, blk_mode_t mode)
2109 {
2110         struct zram *zram = disk->private_data;
2111
2112         WARN_ON(!mutex_is_locked(&disk->open_mutex));
2113
2114         /* zram was claimed to reset so open request fails */
2115         if (zram->claim)
2116                 return -EBUSY;
2117         return 0;
2118 }
2119
2120 static const struct block_device_operations zram_devops = {
2121         .open = zram_open,
2122         .submit_bio = zram_submit_bio,
2123         .swap_slot_free_notify = zram_slot_free_notify,
2124         .owner = THIS_MODULE
2125 };
2126
2127 static DEVICE_ATTR_WO(compact);
2128 static DEVICE_ATTR_RW(disksize);
2129 static DEVICE_ATTR_RO(initstate);
2130 static DEVICE_ATTR_WO(reset);
2131 static DEVICE_ATTR_WO(mem_limit);
2132 static DEVICE_ATTR_WO(mem_used_max);
2133 static DEVICE_ATTR_WO(idle);
2134 static DEVICE_ATTR_RW(max_comp_streams);
2135 static DEVICE_ATTR_RW(comp_algorithm);
2136 #ifdef CONFIG_ZRAM_WRITEBACK
2137 static DEVICE_ATTR_RW(backing_dev);
2138 static DEVICE_ATTR_WO(writeback);
2139 static DEVICE_ATTR_RW(writeback_limit);
2140 static DEVICE_ATTR_RW(writeback_limit_enable);
2141 #endif
2142 #ifdef CONFIG_ZRAM_MULTI_COMP
2143 static DEVICE_ATTR_RW(recomp_algorithm);
2144 static DEVICE_ATTR_WO(recompress);
2145 #endif
2146
2147 static struct attribute *zram_disk_attrs[] = {
2148         &dev_attr_disksize.attr,
2149         &dev_attr_initstate.attr,
2150         &dev_attr_reset.attr,
2151         &dev_attr_compact.attr,
2152         &dev_attr_mem_limit.attr,
2153         &dev_attr_mem_used_max.attr,
2154         &dev_attr_idle.attr,
2155         &dev_attr_max_comp_streams.attr,
2156         &dev_attr_comp_algorithm.attr,
2157 #ifdef CONFIG_ZRAM_WRITEBACK
2158         &dev_attr_backing_dev.attr,
2159         &dev_attr_writeback.attr,
2160         &dev_attr_writeback_limit.attr,
2161         &dev_attr_writeback_limit_enable.attr,
2162 #endif
2163         &dev_attr_io_stat.attr,
2164         &dev_attr_mm_stat.attr,
2165 #ifdef CONFIG_ZRAM_WRITEBACK
2166         &dev_attr_bd_stat.attr,
2167 #endif
2168         &dev_attr_debug_stat.attr,
2169 #ifdef CONFIG_ZRAM_MULTI_COMP
2170         &dev_attr_recomp_algorithm.attr,
2171         &dev_attr_recompress.attr,
2172 #endif
2173         NULL,
2174 };
2175
2176 ATTRIBUTE_GROUPS(zram_disk);
2177
2178 /*
2179  * Allocate and initialize new zram device. the function returns
2180  * '>= 0' device_id upon success, and negative value otherwise.
2181  */
2182 static int zram_add(void)
2183 {
2184         struct zram *zram;
2185         int ret, device_id;
2186
2187         zram = kzalloc(sizeof(struct zram), GFP_KERNEL);
2188         if (!zram)
2189                 return -ENOMEM;
2190
2191         ret = idr_alloc(&zram_index_idr, zram, 0, 0, GFP_KERNEL);
2192         if (ret < 0)
2193                 goto out_free_dev;
2194         device_id = ret;
2195
2196         init_rwsem(&zram->init_lock);
2197 #ifdef CONFIG_ZRAM_WRITEBACK
2198         spin_lock_init(&zram->wb_limit_lock);
2199 #endif
2200
2201         /* gendisk structure */
2202         zram->disk = blk_alloc_disk(NUMA_NO_NODE);
2203         if (!zram->disk) {
2204                 pr_err("Error allocating disk structure for device %d\n",
2205                         device_id);
2206                 ret = -ENOMEM;
2207                 goto out_free_idr;
2208         }
2209
2210         zram->disk->major = zram_major;
2211         zram->disk->first_minor = device_id;
2212         zram->disk->minors = 1;
2213         zram->disk->flags |= GENHD_FL_NO_PART;
2214         zram->disk->fops = &zram_devops;
2215         zram->disk->private_data = zram;
2216         snprintf(zram->disk->disk_name, 16, "zram%d", device_id);
2217
2218         /* Actual capacity set using sysfs (/sys/block/zram<id>/disksize */
2219         set_capacity(zram->disk, 0);
2220         /* zram devices sort of resembles non-rotational disks */
2221         blk_queue_flag_set(QUEUE_FLAG_NONROT, zram->disk->queue);
2222         blk_queue_flag_set(QUEUE_FLAG_SYNCHRONOUS, zram->disk->queue);
2223
2224         /*
2225          * To ensure that we always get PAGE_SIZE aligned
2226          * and n*PAGE_SIZED sized I/O requests.
2227          */
2228         blk_queue_physical_block_size(zram->disk->queue, PAGE_SIZE);
2229         blk_queue_logical_block_size(zram->disk->queue,
2230                                         ZRAM_LOGICAL_BLOCK_SIZE);
2231         blk_queue_io_min(zram->disk->queue, PAGE_SIZE);
2232         blk_queue_io_opt(zram->disk->queue, PAGE_SIZE);
2233         zram->disk->queue->limits.discard_granularity = PAGE_SIZE;
2234         blk_queue_max_discard_sectors(zram->disk->queue, UINT_MAX);
2235
2236         /*
2237          * zram_bio_discard() will clear all logical blocks if logical block
2238          * size is identical with physical block size(PAGE_SIZE). But if it is
2239          * different, we will skip discarding some parts of logical blocks in
2240          * the part of the request range which isn't aligned to physical block
2241          * size.  So we can't ensure that all discarded logical blocks are
2242          * zeroed.
2243          */
2244         if (ZRAM_LOGICAL_BLOCK_SIZE == PAGE_SIZE)
2245                 blk_queue_max_write_zeroes_sectors(zram->disk->queue, UINT_MAX);
2246
2247         blk_queue_flag_set(QUEUE_FLAG_STABLE_WRITES, zram->disk->queue);
2248         ret = device_add_disk(NULL, zram->disk, zram_disk_groups);
2249         if (ret)
2250                 goto out_cleanup_disk;
2251
2252         comp_algorithm_set(zram, ZRAM_PRIMARY_COMP, default_compressor);
2253
2254         zram_debugfs_register(zram);
2255         pr_info("Added device: %s\n", zram->disk->disk_name);
2256         return device_id;
2257
2258 out_cleanup_disk:
2259         put_disk(zram->disk);
2260 out_free_idr:
2261         idr_remove(&zram_index_idr, device_id);
2262 out_free_dev:
2263         kfree(zram);
2264         return ret;
2265 }
2266
2267 static int zram_remove(struct zram *zram)
2268 {
2269         bool claimed;
2270
2271         mutex_lock(&zram->disk->open_mutex);
2272         if (disk_openers(zram->disk)) {
2273                 mutex_unlock(&zram->disk->open_mutex);
2274                 return -EBUSY;
2275         }
2276
2277         claimed = zram->claim;
2278         if (!claimed)
2279                 zram->claim = true;
2280         mutex_unlock(&zram->disk->open_mutex);
2281
2282         zram_debugfs_unregister(zram);
2283
2284         if (claimed) {
2285                 /*
2286                  * If we were claimed by reset_store(), del_gendisk() will
2287                  * wait until reset_store() is done, so nothing need to do.
2288                  */
2289                 ;
2290         } else {
2291                 /* Make sure all the pending I/O are finished */
2292                 sync_blockdev(zram->disk->part0);
2293                 zram_reset_device(zram);
2294         }
2295
2296         pr_info("Removed device: %s\n", zram->disk->disk_name);
2297
2298         del_gendisk(zram->disk);
2299
2300         /* del_gendisk drains pending reset_store */
2301         WARN_ON_ONCE(claimed && zram->claim);
2302
2303         /*
2304          * disksize_store() may be called in between zram_reset_device()
2305          * and del_gendisk(), so run the last reset to avoid leaking
2306          * anything allocated with disksize_store()
2307          */
2308         zram_reset_device(zram);
2309
2310         put_disk(zram->disk);
2311         kfree(zram);
2312         return 0;
2313 }
2314
2315 /* zram-control sysfs attributes */
2316
2317 /*
2318  * NOTE: hot_add attribute is not the usual read-only sysfs attribute. In a
2319  * sense that reading from this file does alter the state of your system -- it
2320  * creates a new un-initialized zram device and returns back this device's
2321  * device_id (or an error code if it fails to create a new device).
2322  */
2323 static ssize_t hot_add_show(const struct class *class,
2324                         const struct class_attribute *attr,
2325                         char *buf)
2326 {
2327         int ret;
2328
2329         mutex_lock(&zram_index_mutex);
2330         ret = zram_add();
2331         mutex_unlock(&zram_index_mutex);
2332
2333         if (ret < 0)
2334                 return ret;
2335         return scnprintf(buf, PAGE_SIZE, "%d\n", ret);
2336 }
2337 /* This attribute must be set to 0400, so CLASS_ATTR_RO() can not be used */
2338 static struct class_attribute class_attr_hot_add =
2339         __ATTR(hot_add, 0400, hot_add_show, NULL);
2340
2341 static ssize_t hot_remove_store(const struct class *class,
2342                         const struct class_attribute *attr,
2343                         const char *buf,
2344                         size_t count)
2345 {
2346         struct zram *zram;
2347         int ret, dev_id;
2348
2349         /* dev_id is gendisk->first_minor, which is `int' */
2350         ret = kstrtoint(buf, 10, &dev_id);
2351         if (ret)
2352                 return ret;
2353         if (dev_id < 0)
2354                 return -EINVAL;
2355
2356         mutex_lock(&zram_index_mutex);
2357
2358         zram = idr_find(&zram_index_idr, dev_id);
2359         if (zram) {
2360                 ret = zram_remove(zram);
2361                 if (!ret)
2362                         idr_remove(&zram_index_idr, dev_id);
2363         } else {
2364                 ret = -ENODEV;
2365         }
2366
2367         mutex_unlock(&zram_index_mutex);
2368         return ret ? ret : count;
2369 }
2370 static CLASS_ATTR_WO(hot_remove);
2371
2372 static struct attribute *zram_control_class_attrs[] = {
2373         &class_attr_hot_add.attr,
2374         &class_attr_hot_remove.attr,
2375         NULL,
2376 };
2377 ATTRIBUTE_GROUPS(zram_control_class);
2378
2379 static struct class zram_control_class = {
2380         .name           = "zram-control",
2381         .class_groups   = zram_control_class_groups,
2382 };
2383
2384 static int zram_remove_cb(int id, void *ptr, void *data)
2385 {
2386         WARN_ON_ONCE(zram_remove(ptr));
2387         return 0;
2388 }
2389
2390 static void destroy_devices(void)
2391 {
2392         class_unregister(&zram_control_class);
2393         idr_for_each(&zram_index_idr, &zram_remove_cb, NULL);
2394         zram_debugfs_destroy();
2395         idr_destroy(&zram_index_idr);
2396         unregister_blkdev(zram_major, "zram");
2397         cpuhp_remove_multi_state(CPUHP_ZCOMP_PREPARE);
2398 }
2399
2400 static int __init zram_init(void)
2401 {
2402         int ret;
2403
2404         BUILD_BUG_ON(__NR_ZRAM_PAGEFLAGS > BITS_PER_LONG);
2405
2406         ret = cpuhp_setup_state_multi(CPUHP_ZCOMP_PREPARE, "block/zram:prepare",
2407                                       zcomp_cpu_up_prepare, zcomp_cpu_dead);
2408         if (ret < 0)
2409                 return ret;
2410
2411         ret = class_register(&zram_control_class);
2412         if (ret) {
2413                 pr_err("Unable to register zram-control class\n");
2414                 cpuhp_remove_multi_state(CPUHP_ZCOMP_PREPARE);
2415                 return ret;
2416         }
2417
2418         zram_debugfs_create();
2419         zram_major = register_blkdev(0, "zram");
2420         if (zram_major <= 0) {
2421                 pr_err("Unable to get major number\n");
2422                 class_unregister(&zram_control_class);
2423                 cpuhp_remove_multi_state(CPUHP_ZCOMP_PREPARE);
2424                 return -EBUSY;
2425         }
2426
2427         while (num_devices != 0) {
2428                 mutex_lock(&zram_index_mutex);
2429                 ret = zram_add();
2430                 mutex_unlock(&zram_index_mutex);
2431                 if (ret < 0)
2432                         goto out_error;
2433                 num_devices--;
2434         }
2435
2436         return 0;
2437
2438 out_error:
2439         destroy_devices();
2440         return ret;
2441 }
2442
2443 static void __exit zram_exit(void)
2444 {
2445         destroy_devices();
2446 }
2447
2448 module_init(zram_init);
2449 module_exit(zram_exit);
2450
2451 module_param(num_devices, uint, 0);
2452 MODULE_PARM_DESC(num_devices, "Number of pre-created zram devices");
2453
2454 MODULE_LICENSE("Dual BSD/GPL");
2455 MODULE_AUTHOR("Nitin Gupta <ngupta@vflare.org>");
2456 MODULE_DESCRIPTION("Compressed RAM Block Device");