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 #ifdef CONFIG_ZRAM_DEBUG
  19 #define DEBUG
  20 #endif
  21
  22 #include <linux/module.h>
  23 #include <linux/kernel.h>
  24 #include <linux/bio.h>
  25 #include <linux/bitops.h>
  26 #include <linux/blkdev.h>
  27 #include <linux/buffer_head.h>
  28 #include <linux/device.h>
  29 #include <linux/genhd.h>
  30 #include <linux/highmem.h>
  31 #include <linux/slab.h>
  32 #include <linux/string.h>
  33 #include <linux/vmalloc.h>
  34 #include <linux/err.h>
  35
  36 #include "zram_drv.h"
  37
  38 /* Globals */
  39 static int zram_major;
  40 static struct zram *zram_devices;
  41 static const char *default_compressor = "lzo";
  42
  43 /* Module params (documentation at end) */
  44 static unsigned int num_devices = 1;
  45
  46 #define ZRAM_ATTR_RO(name)                                              \
  47 static ssize_t zram_attr_##name##_show(struct device *d,                \
  48                                 struct device_attribute *attr, char *b) \
  49 {                                                                       \
  50         struct zram *zram = dev_to_zram(d);                             \
  51         return scnprintf(b, PAGE_SIZE, "%llu\n",                        \
  52                 (u64)atomic64_read(&zram->stats.name));                 \
  53 }                                                                       \
  54 static struct device_attribute dev_attr_##name =                        \
  55         __ATTR(name, S_IRUGO, zram_attr_##name##_show, NULL);
  56
  57 static inline int init_done(struct zram *zram)
  58 {
  59         return zram->meta != NULL;
  60 }
  61
  62 static inline struct zram *dev_to_zram(struct device *dev)
  63 {
  64         return (struct zram *)dev_to_disk(dev)->private_data;
  65 }
  66
  67 static ssize_t disksize_show(struct device *dev,
  68                 struct device_attribute *attr, char *buf)
  69 {
  70         struct zram *zram = dev_to_zram(dev);
  71
  72         return scnprintf(buf, PAGE_SIZE, "%llu\n", zram->disksize);
  73 }
  74
  75 static ssize_t initstate_show(struct device *dev,
  76                 struct device_attribute *attr, char *buf)
  77 {
  78         u32 val;
  79         struct zram *zram = dev_to_zram(dev);
  80
  81         down_read(&zram->init_lock);
  82         val = init_done(zram);
  83         up_read(&zram->init_lock);
  84
  85         return scnprintf(buf, PAGE_SIZE, "%u\n", val);
  86 }
  87
  88 static ssize_t orig_data_size_show(struct device *dev,
  89                 struct device_attribute *attr, char *buf)
  90 {
  91         struct zram *zram = dev_to_zram(dev);
  92
  93         return scnprintf(buf, PAGE_SIZE, "%llu\n",
  94                 (u64)(atomic64_read(&zram->stats.pages_stored)) << PAGE_SHIFT);
  95 }
  96
  97 static ssize_t mem_used_total_show(struct device *dev,
  98                 struct device_attribute *attr, char *buf)
  99 {
 100         u64 val = 0;
 101         struct zram *zram = dev_to_zram(dev);
 102
 103         down_read(&zram->init_lock);
 104         if (init_done(zram)) {
 105                 struct zram_meta *meta = zram->meta;
 106                 val = zs_get_total_pages(meta->mem_pool);
 107         }
 108         up_read(&zram->init_lock);
 109
 110         return scnprintf(buf, PAGE_SIZE, "%llu\n", val << PAGE_SHIFT);
 111 }
 112
 113 static ssize_t max_comp_streams_show(struct device *dev,
 114                 struct device_attribute *attr, char *buf)
 115 {
 116         int val;
 117         struct zram *zram = dev_to_zram(dev);
 118
 119         down_read(&zram->init_lock);
 120         val = zram->max_comp_streams;
 121         up_read(&zram->init_lock);
 122
 123         return scnprintf(buf, PAGE_SIZE, "%d\n", val);
 124 }
 125
 126 static ssize_t mem_limit_show(struct device *dev,
 127                 struct device_attribute *attr, char *buf)
 128 {
 129         u64 val;
 130         struct zram *zram = dev_to_zram(dev);
 131
 132         down_read(&zram->init_lock);
 133         val = zram->limit_pages;
 134         up_read(&zram->init_lock);
 135
 136         return scnprintf(buf, PAGE_SIZE, "%llu\n", val << PAGE_SHIFT);
 137 }
 138
 139 static ssize_t mem_limit_store(struct device *dev,
 140                 struct device_attribute *attr, const char *buf, size_t len)
 141 {
 142         u64 limit;
 143         char *tmp;
 144         struct zram *zram = dev_to_zram(dev);
 145
 146         limit = memparse(buf, &tmp);
 147         if (buf == tmp) /* no chars parsed, invalid input */
 148                 return -EINVAL;
 149
 150         down_write(&zram->init_lock);
 151         zram->limit_pages = PAGE_ALIGN(limit) >> PAGE_SHIFT;
 152         up_write(&zram->init_lock);
 153
 154         return len;
 155 }
 156
 157 static ssize_t mem_used_max_show(struct device *dev,
 158                 struct device_attribute *attr, char *buf)
 159 {
 160         u64 val = 0;
 161         struct zram *zram = dev_to_zram(dev);
 162
 163         down_read(&zram->init_lock);
 164         if (init_done(zram))
 165                 val = atomic_long_read(&zram->stats.max_used_pages);
 166         up_read(&zram->init_lock);
 167
 168         return scnprintf(buf, PAGE_SIZE, "%llu\n", val << PAGE_SHIFT);
 169 }
 170
 171 static ssize_t mem_used_max_store(struct device *dev,
 172                 struct device_attribute *attr, const char *buf, size_t len)
 173 {
 174         int err;
 175         unsigned long val;
 176         struct zram *zram = dev_to_zram(dev);
 177
 178         err = kstrtoul(buf, 10, &val);
 179         if (err || val != 0)
 180                 return -EINVAL;
 181
 182         down_read(&zram->init_lock);
 183         if (init_done(zram)) {
 184                 struct zram_meta *meta = zram->meta;
 185                 atomic_long_set(&zram->stats.max_used_pages,
 186                                 zs_get_total_pages(meta->mem_pool));
 187         }
 188         up_read(&zram->init_lock);
 189
 190         return len;
 191 }
 192
 193 static ssize_t max_comp_streams_store(struct device *dev,
 194                 struct device_attribute *attr, const char *buf, size_t len)
 195 {
 196         int num;
 197         struct zram *zram = dev_to_zram(dev);
 198         int ret;
 199
 200         ret = kstrtoint(buf, 0, &num);
 201         if (ret < 0)
 202                 return ret;
 203         if (num < 1)
 204                 return -EINVAL;
 205
 206         down_write(&zram->init_lock);
 207         if (init_done(zram)) {
 208                 if (!zcomp_set_max_streams(zram->comp, num)) {
 209                         pr_info("Cannot change max compression streams\n");
 210                         ret = -EINVAL;
 211                         goto out;
 212                 }
 213         }
 214
 215         zram->max_comp_streams = num;
 216         ret = len;
 217 out:
 218         up_write(&zram->init_lock);
 219         return ret;
 220 }
 221
 222 static ssize_t comp_algorithm_show(struct device *dev,
 223                 struct device_attribute *attr, char *buf)
 224 {
 225         size_t sz;
 226         struct zram *zram = dev_to_zram(dev);
 227
 228         down_read(&zram->init_lock);
 229         sz = zcomp_available_show(zram->compressor, buf);
 230         up_read(&zram->init_lock);
 231
 232         return sz;
 233 }
 234
 235 static ssize_t comp_algorithm_store(struct device *dev,
 236                 struct device_attribute *attr, const char *buf, size_t len)
 237 {
 238         struct zram *zram = dev_to_zram(dev);
 239         down_write(&zram->init_lock);
 240         if (init_done(zram)) {
 241                 up_write(&zram->init_lock);
 242                 pr_info("Can't change algorithm for initialized device\n");
 243                 return -EBUSY;
 244         }
 245         strlcpy(zram->compressor, buf, sizeof(zram->compressor));
 246         up_write(&zram->init_lock);
 247         return len;
 248 }
 249
 250 /* flag operations needs meta->tb_lock */
 251 static int zram_test_flag(struct zram_meta *meta, u32 index,
 252                         enum zram_pageflags flag)
 253 {
 254         return meta->table[index].value & BIT(flag);
 255 }
 256
 257 static void zram_set_flag(struct zram_meta *meta, u32 index,
 258                         enum zram_pageflags flag)
 259 {
 260         meta->table[index].value |= BIT(flag);
 261 }
 262
 263 static void zram_clear_flag(struct zram_meta *meta, u32 index,
 264                         enum zram_pageflags flag)
 265 {
 266         meta->table[index].value &= ~BIT(flag);
 267 }
 268
 269 static size_t zram_get_obj_size(struct zram_meta *meta, u32 index)
 270 {
 271         return meta->table[index].value & (BIT(ZRAM_FLAG_SHIFT) - 1);
 272 }
 273
 274 static void zram_set_obj_size(struct zram_meta *meta,
 275                                         u32 index, size_t size)
 276 {
 277         unsigned long flags = meta->table[index].value >> ZRAM_FLAG_SHIFT;
 278
 279         meta->table[index].value = (flags << ZRAM_FLAG_SHIFT) | size;
 280 }
 281
 282 static inline int is_partial_io(struct bio_vec *bvec)
 283 {
 284         return bvec->bv_len != PAGE_SIZE;
 285 }
 286
 287 /*
 288  * Check if request is within bounds and aligned on zram logical blocks.
 289  */
 290 static inline int valid_io_request(struct zram *zram, struct bio *bio)
 291 {
 292         u64 start, end, bound;
 293
 294         /* unaligned request */
 295         if (unlikely(bio->bi_iter.bi_sector &
 296                      (ZRAM_SECTOR_PER_LOGICAL_BLOCK - 1)))
 297                 return 0;
 298         if (unlikely(bio->bi_iter.bi_size & (ZRAM_LOGICAL_BLOCK_SIZE - 1)))
 299                 return 0;
 300
 301         start = bio->bi_iter.bi_sector;
 302         end = start + (bio->bi_iter.bi_size >> SECTOR_SHIFT);
 303         bound = zram->disksize >> SECTOR_SHIFT;
 304         /* out of range range */
 305         if (unlikely(start >= bound || end > bound || start > end))
 306                 return 0;
 307
 308         /* I/O request is valid */
 309         return 1;
 310 }
 311
 312 static void zram_meta_free(struct zram_meta *meta)
 313 {
 314         zs_destroy_pool(meta->mem_pool);
 315         vfree(meta->table);
 316         kfree(meta);
 317 }
 318
 319 static struct zram_meta *zram_meta_alloc(u64 disksize)
 320 {
 321         size_t num_pages;
 322         struct zram_meta *meta = kmalloc(sizeof(*meta), GFP_KERNEL);
 323         if (!meta)
 324                 goto out;
 325
 326         num_pages = disksize >> PAGE_SHIFT;
 327         meta->table = vzalloc(num_pages * sizeof(*meta->table));
 328         if (!meta->table) {
 329                 pr_err("Error allocating zram address table\n");
 330                 goto free_meta;
 331         }
 332
 333         meta->mem_pool = zs_create_pool(GFP_NOIO | __GFP_HIGHMEM);
 334         if (!meta->mem_pool) {
 335                 pr_err("Error creating memory pool\n");
 336                 goto free_table;
 337         }
 338
 339         return meta;
 340
 341 free_table:
 342         vfree(meta->table);
 343 free_meta:
 344         kfree(meta);
 345         meta = NULL;
 346 out:
 347         return meta;
 348 }
 349
 350 static void update_position(u32 *index, int *offset, struct bio_vec *bvec)
 351 {
 352         if (*offset + bvec->bv_len >= PAGE_SIZE)
 353                 (*index)++;
 354         *offset = (*offset + bvec->bv_len) % PAGE_SIZE;
 355 }
 356
 357 static int page_zero_filled(void *ptr)
 358 {
 359         unsigned int pos;
 360         unsigned long *page;
 361
 362         page = (unsigned long *)ptr;
 363
 364         for (pos = 0; pos != PAGE_SIZE / sizeof(*page); pos++) {
 365                 if (page[pos])
 366                         return 0;
 367         }
 368
 369         return 1;
 370 }
 371
 372 static void handle_zero_page(struct bio_vec *bvec)
 373 {
 374         struct page *page = bvec->bv_page;
 375         void *user_mem;
 376
 377         user_mem = kmap_atomic(page);
 378         if (is_partial_io(bvec))
 379                 memset(user_mem + bvec->bv_offset, 0, bvec->bv_len);
 380         else
 381                 clear_page(user_mem);
 382         kunmap_atomic(user_mem);
 383
 384         flush_dcache_page(page);
 385 }
 386
 387
 388 /*
 389  * To protect concurrent access to the same index entry,
 390  * caller should hold this table index entry's bit_spinlock to
 391  * indicate this index entry is accessing.
 392  */
 393 static void zram_free_page(struct zram *zram, size_t index)
 394 {
 395         struct zram_meta *meta = zram->meta;
 396         unsigned long handle = meta->table[index].handle;
 397
 398         if (unlikely(!handle)) {
 399                 /*
 400                  * No memory is allocated for zero filled pages.
 401                  * Simply clear zero page flag.
 402                  */
 403                 if (zram_test_flag(meta, index, ZRAM_ZERO)) {
 404                         zram_clear_flag(meta, index, ZRAM_ZERO);
 405                         atomic64_dec(&zram->stats.zero_pages);
 406                 }
 407                 return;
 408         }
 409
 410         zs_free(meta->mem_pool, handle);
 411
 412         atomic64_sub(zram_get_obj_size(meta, index),
 413                         &zram->stats.compr_data_size);
 414         atomic64_dec(&zram->stats.pages_stored);
 415
 416         meta->table[index].handle = 0;
 417         zram_set_obj_size(meta, index, 0);
 418 }
 419
 420 static int zram_decompress_page(struct zram *zram, char *mem, u32 index)
 421 {
 422         int ret = 0;
 423         unsigned char *cmem;
 424         struct zram_meta *meta = zram->meta;
 425         unsigned long handle;
 426         size_t size;
 427
 428         bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
 429         handle = meta->table[index].handle;
 430         size = zram_get_obj_size(meta, index);
 431
 432         if (!handle || zram_test_flag(meta, index, ZRAM_ZERO)) {
 433                 bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 434                 clear_page(mem);
 435                 return 0;
 436         }
 437
 438         cmem = zs_map_object(meta->mem_pool, handle, ZS_MM_RO);
 439         if (size == PAGE_SIZE)
 440                 copy_page(mem, cmem);
 441         else
 442                 ret = zcomp_decompress(zram->comp, cmem, size, mem);
 443         zs_unmap_object(meta->mem_pool, handle);
 444         bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 445
 446         /* Should NEVER happen. Return bio error if it does. */
 447         if (unlikely(ret)) {
 448                 pr_err("Decompression failed! err=%d, page=%u\n", ret, index);
 449                 return ret;
 450         }
 451
 452         return 0;
 453 }
 454
 455 static int zram_bvec_read(struct zram *zram, struct bio_vec *bvec,
 456                           u32 index, int offset, struct bio *bio)
 457 {
 458         int ret;
 459         struct page *page;
 460         unsigned char *user_mem, *uncmem = NULL;
 461         struct zram_meta *meta = zram->meta;
 462         page = bvec->bv_page;
 463
 464         bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
 465         if (unlikely(!meta->table[index].handle) ||
 466                         zram_test_flag(meta, index, ZRAM_ZERO)) {
 467                 bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 468                 handle_zero_page(bvec);
 469                 return 0;
 470         }
 471         bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 472
 473         if (is_partial_io(bvec))
 474                 /* Use  a temporary buffer to decompress the page */
 475                 uncmem = kmalloc(PAGE_SIZE, GFP_NOIO);
 476
 477         user_mem = kmap_atomic(page);
 478         if (!is_partial_io(bvec))
 479                 uncmem = user_mem;
 480
 481         if (!uncmem) {
 482                 pr_info("Unable to allocate temp memory\n");
 483                 ret = -ENOMEM;
 484                 goto out_cleanup;
 485         }
 486
 487         ret = zram_decompress_page(zram, uncmem, index);
 488         /* Should NEVER happen. Return bio error if it does. */
 489         if (unlikely(ret))
 490                 goto out_cleanup;
 491
 492         if (is_partial_io(bvec))
 493                 memcpy(user_mem + bvec->bv_offset, uncmem + offset,
 494                                 bvec->bv_len);
 495
 496         flush_dcache_page(page);
 497         ret = 0;
 498 out_cleanup:
 499         kunmap_atomic(user_mem);
 500         if (is_partial_io(bvec))
 501                 kfree(uncmem);
 502         return ret;
 503 }
 504
 505 static inline void update_used_max(struct zram *zram,
 506                                         const unsigned long pages)
 507 {
 508         int old_max, cur_max;
 509
 510         old_max = atomic_long_read(&zram->stats.max_used_pages);
 511
 512         do {
 513                 cur_max = old_max;
 514                 if (pages > cur_max)
 515                         old_max = atomic_long_cmpxchg(
 516                                 &zram->stats.max_used_pages, cur_max, pages);
 517         } while (old_max != cur_max);
 518 }
 519
 520 static int zram_bvec_write(struct zram *zram, struct bio_vec *bvec, u32 index,
 521                            int offset)
 522 {
 523         int ret = 0;
 524         size_t clen;
 525         unsigned long handle;
 526         struct page *page;
 527         unsigned char *user_mem, *cmem, *src, *uncmem = NULL;
 528         struct zram_meta *meta = zram->meta;
 529         struct zcomp_strm *zstrm;
 530         bool locked = false;
 531         unsigned long alloced_pages;
 532
 533         page = bvec->bv_page;
 534         if (is_partial_io(bvec)) {
 535                 /*
 536                  * This is a partial IO. We need to read the full page
 537                  * before to write the changes.
 538                  */
 539                 uncmem = kmalloc(PAGE_SIZE, GFP_NOIO);
 540                 if (!uncmem) {
 541                         ret = -ENOMEM;
 542                         goto out;
 543                 }
 544                 ret = zram_decompress_page(zram, uncmem, index);
 545                 if (ret)
 546                         goto out;
 547         }
 548
 549         zstrm = zcomp_strm_find(zram->comp);
 550         locked = true;
 551         user_mem = kmap_atomic(page);
 552
 553         if (is_partial_io(bvec)) {
 554                 memcpy(uncmem + offset, user_mem + bvec->bv_offset,
 555                        bvec->bv_len);
 556                 kunmap_atomic(user_mem);
 557                 user_mem = NULL;
 558         } else {
 559                 uncmem = user_mem;
 560         }
 561
 562         if (page_zero_filled(uncmem)) {
 563                 if (user_mem)
 564                         kunmap_atomic(user_mem);
 565                 /* Free memory associated with this sector now. */
 566                 bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
 567                 zram_free_page(zram, index);
 568                 zram_set_flag(meta, index, ZRAM_ZERO);
 569                 bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 570
 571                 atomic64_inc(&zram->stats.zero_pages);
 572                 ret = 0;
 573                 goto out;
 574         }
 575
 576         ret = zcomp_compress(zram->comp, zstrm, uncmem, &clen);
 577         if (!is_partial_io(bvec)) {
 578                 kunmap_atomic(user_mem);
 579                 user_mem = NULL;
 580                 uncmem = NULL;
 581         }
 582
 583         if (unlikely(ret)) {
 584                 pr_err("Compression failed! err=%d\n", ret);
 585                 goto out;
 586         }
 587         src = zstrm->buffer;
 588         if (unlikely(clen > max_zpage_size)) {
 589                 clen = PAGE_SIZE;
 590                 if (is_partial_io(bvec))
 591                         src = uncmem;
 592         }
 593
 594         handle = zs_malloc(meta->mem_pool, clen);
 595         if (!handle) {
 596                 pr_info("Error allocating memory for compressed page: %u, size=%zu\n",
 597                         index, clen);
 598                 ret = -ENOMEM;
 599                 goto out;
 600         }
 601
 602         alloced_pages = zs_get_total_pages(meta->mem_pool);
 603         if (zram->limit_pages && alloced_pages > zram->limit_pages) {
 604                 zs_free(meta->mem_pool, handle);
 605                 ret = -ENOMEM;
 606                 goto out;
 607         }
 608
 609         update_used_max(zram, alloced_pages);
 610
 611         cmem = zs_map_object(meta->mem_pool, handle, ZS_MM_WO);
 612
 613         if ((clen == PAGE_SIZE) && !is_partial_io(bvec)) {
 614                 src = kmap_atomic(page);
 615                 copy_page(cmem, src);
 616                 kunmap_atomic(src);
 617         } else {
 618                 memcpy(cmem, src, clen);
 619         }
 620
 621         zcomp_strm_release(zram->comp, zstrm);
 622         locked = false;
 623         zs_unmap_object(meta->mem_pool, handle);
 624
 625         /*
 626          * Free memory associated with this sector
 627          * before overwriting unused sectors.
 628          */
 629         bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
 630         zram_free_page(zram, index);
 631
 632         meta->table[index].handle = handle;
 633         zram_set_obj_size(meta, index, clen);
 634         bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 635
 636         /* Update stats */
 637         atomic64_add(clen, &zram->stats.compr_data_size);
 638         atomic64_inc(&zram->stats.pages_stored);
 639 out:
 640         if (locked)
 641                 zcomp_strm_release(zram->comp, zstrm);
 642         if (is_partial_io(bvec))
 643                 kfree(uncmem);
 644         return ret;
 645 }
 646
 647 static int zram_bvec_rw(struct zram *zram, struct bio_vec *bvec, u32 index,
 648                         int offset, struct bio *bio)
 649 {
 650         int ret;
 651         int rw = bio_data_dir(bio);
 652
 653         if (rw == READ) {
 654                 atomic64_inc(&zram->stats.num_reads);
 655                 ret = zram_bvec_read(zram, bvec, index, offset, bio);
 656         } else {
 657                 atomic64_inc(&zram->stats.num_writes);
 658                 ret = zram_bvec_write(zram, bvec, index, offset);
 659         }
 660
 661         if (unlikely(ret)) {
 662                 if (rw == READ)
 663                         atomic64_inc(&zram->stats.failed_reads);
 664                 else
 665                         atomic64_inc(&zram->stats.failed_writes);
 666         }
 667
 668         return ret;
 669 }
 670
 671 /*
 672  * zram_bio_discard - handler on discard request
 673  * @index: physical block index in PAGE_SIZE units
 674  * @offset: byte offset within physical block
 675  */
 676 static void zram_bio_discard(struct zram *zram, u32 index,
 677                              int offset, struct bio *bio)
 678 {
 679         size_t n = bio->bi_iter.bi_size;
 680         struct zram_meta *meta = zram->meta;
 681
 682         /*
 683          * zram manages data in physical block size units. Because logical block
 684          * size isn't identical with physical block size on some arch, we
 685          * could get a discard request pointing to a specific offset within a
 686          * certain physical block.  Although we can handle this request by
 687          * reading that physiclal block and decompressing and partially zeroing
 688          * and re-compressing and then re-storing it, this isn't reasonable
 689          * because our intent with a discard request is to save memory.  So
 690          * skipping this logical block is appropriate here.
 691          */
 692         if (offset) {
 693                 if (n <= (PAGE_SIZE - offset))
 694                         return;
 695
 696                 n -= (PAGE_SIZE - offset);
 697                 index++;
 698         }
 699
 700         while (n >= PAGE_SIZE) {
 701                 bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
 702                 zram_free_page(zram, index);
 703                 bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 704                 atomic64_inc(&zram->stats.notify_free);
 705                 index++;
 706                 n -= PAGE_SIZE;
 707         }
 708 }
 709
 710 static void zram_reset_device(struct zram *zram, bool reset_capacity)
 711 {
 712         size_t index;
 713         struct zram_meta *meta;
 714
 715         down_write(&zram->init_lock);
 716
 717         zram->limit_pages = 0;
 718
 719         if (!init_done(zram)) {
 720                 up_write(&zram->init_lock);
 721                 return;
 722         }
 723
 724         meta = zram->meta;
 725         /* Free all pages that are still in this zram device */
 726         for (index = 0; index < zram->disksize >> PAGE_SHIFT; index++) {
 727                 unsigned long handle = meta->table[index].handle;
 728                 if (!handle)
 729                         continue;
 730
 731                 zs_free(meta->mem_pool, handle);
 732         }
 733
 734         zcomp_destroy(zram->comp);
 735         zram->max_comp_streams = 1;
 736
 737         zram_meta_free(zram->meta);
 738         zram->meta = NULL;
 739         /* Reset stats */
 740         memset(&zram->stats, 0, sizeof(zram->stats));
 741
 742         zram->disksize = 0;
 743         if (reset_capacity)
 744                 set_capacity(zram->disk, 0);
 745
 746         up_write(&zram->init_lock);
 747
 748         /*
 749          * Revalidate disk out of the init_lock to avoid lockdep splat.
 750          * It's okay because disk's capacity is protected by init_lock
 751          * so that revalidate_disk always sees up-to-date capacity.
 752          */
 753         if (reset_capacity)
 754                 revalidate_disk(zram->disk);
 755 }
 756
 757 static ssize_t disksize_store(struct device *dev,
 758                 struct device_attribute *attr, const char *buf, size_t len)
 759 {
 760         u64 disksize;
 761         struct zcomp *comp;
 762         struct zram_meta *meta;
 763         struct zram *zram = dev_to_zram(dev);
 764         int err;
 765
 766         disksize = memparse(buf, NULL);
 767         if (!disksize)
 768                 return -EINVAL;
 769
 770         disksize = PAGE_ALIGN(disksize);
 771         meta = zram_meta_alloc(disksize);
 772         if (!meta)
 773                 return -ENOMEM;
 774
 775         comp = zcomp_create(zram->compressor, zram->max_comp_streams);
 776         if (IS_ERR(comp)) {
 777                 pr_info("Cannot initialise %s compressing backend\n",
 778                                 zram->compressor);
 779                 err = PTR_ERR(comp);
 780                 goto out_free_meta;
 781         }
 782
 783         down_write(&zram->init_lock);
 784         if (init_done(zram)) {
 785                 pr_info("Cannot change disksize for initialized device\n");
 786                 err = -EBUSY;
 787                 goto out_destroy_comp;
 788         }
 789
 790         zram->meta = meta;
 791         zram->comp = comp;
 792         zram->disksize = disksize;
 793         set_capacity(zram->disk, zram->disksize >> SECTOR_SHIFT);
 794         up_write(&zram->init_lock);
 795
 796         /*
 797          * Revalidate disk out of the init_lock to avoid lockdep splat.
 798          * It's okay because disk's capacity is protected by init_lock
 799          * so that revalidate_disk always sees up-to-date capacity.
 800          */
 801         revalidate_disk(zram->disk);
 802
 803         return len;
 804
 805 out_destroy_comp:
 806         up_write(&zram->init_lock);
 807         zcomp_destroy(comp);
 808 out_free_meta:
 809         zram_meta_free(meta);
 810         return err;
 811 }
 812
 813 static ssize_t reset_store(struct device *dev,
 814                 struct device_attribute *attr, const char *buf, size_t len)
 815 {
 816         int ret;
 817         unsigned short do_reset;
 818         struct zram *zram;
 819         struct block_device *bdev;
 820
 821         zram = dev_to_zram(dev);
 822         bdev = bdget_disk(zram->disk, 0);
 823
 824         if (!bdev)
 825                 return -ENOMEM;
 826
 827         /* Do not reset an active device! */
 828         if (bdev->bd_holders) {
 829                 ret = -EBUSY;
 830                 goto out;
 831         }
 832
 833         ret = kstrtou16(buf, 10, &do_reset);
 834         if (ret)
 835                 goto out;
 836
 837         if (!do_reset) {
 838                 ret = -EINVAL;
 839                 goto out;
 840         }
 841
 842         /* Make sure all pending I/O is finished */
 843         fsync_bdev(bdev);
 844         bdput(bdev);
 845
 846         zram_reset_device(zram, true);
 847         return len;
 848
 849 out:
 850         bdput(bdev);
 851         return ret;
 852 }
 853
 854 static void __zram_make_request(struct zram *zram, struct bio *bio)
 855 {
 856         int offset;
 857         u32 index;
 858         struct bio_vec bvec;
 859         struct bvec_iter iter;
 860
 861         index = bio->bi_iter.bi_sector >> SECTORS_PER_PAGE_SHIFT;
 862         offset = (bio->bi_iter.bi_sector &
 863                   (SECTORS_PER_PAGE - 1)) << SECTOR_SHIFT;
 864
 865         if (unlikely(bio->bi_rw & REQ_DISCARD)) {
 866                 zram_bio_discard(zram, index, offset, bio);
 867                 bio_endio(bio, 0);
 868                 return;
 869         }
 870
 871         bio_for_each_segment(bvec, bio, iter) {
 872                 int max_transfer_size = PAGE_SIZE - offset;
 873
 874                 if (bvec.bv_len > max_transfer_size) {
 875                         /*
 876                          * zram_bvec_rw() can only make operation on a single
 877                          * zram page. Split the bio vector.
 878                          */
 879                         struct bio_vec bv;
 880
 881                         bv.bv_page = bvec.bv_page;
 882                         bv.bv_len = max_transfer_size;
 883                         bv.bv_offset = bvec.bv_offset;
 884
 885                         if (zram_bvec_rw(zram, &bv, index, offset, bio) < 0)
 886                                 goto out;
 887
 888                         bv.bv_len = bvec.bv_len - max_transfer_size;
 889                         bv.bv_offset += max_transfer_size;
 890                         if (zram_bvec_rw(zram, &bv, index + 1, 0, bio) < 0)
 891                                 goto out;
 892                 } else
 893                         if (zram_bvec_rw(zram, &bvec, index, offset, bio) < 0)
 894                                 goto out;
 895
 896                 update_position(&index, &offset, &bvec);
 897         }
 898
 899         set_bit(BIO_UPTODATE, &bio->bi_flags);
 900         bio_endio(bio, 0);
 901         return;
 902
 903 out:
 904         bio_io_error(bio);
 905 }
 906
 907 /*
 908  * Handler function for all zram I/O requests.
 909  */
 910 static void zram_make_request(struct request_queue *queue, struct bio *bio)
 911 {
 912         struct zram *zram = queue->queuedata;
 913
 914         down_read(&zram->init_lock);
 915         if (unlikely(!init_done(zram)))
 916                 goto error;
 917
 918         if (!valid_io_request(zram, bio)) {
 919                 atomic64_inc(&zram->stats.invalid_io);
 920                 goto error;
 921         }
 922
 923         __zram_make_request(zram, bio);
 924         up_read(&zram->init_lock);
 925
 926         return;
 927
 928 error:
 929         up_read(&zram->init_lock);
 930         bio_io_error(bio);
 931 }
 932
 933 static void zram_slot_free_notify(struct block_device *bdev,
 934                                 unsigned long index)
 935 {
 936         struct zram *zram;
 937         struct zram_meta *meta;
 938
 939         zram = bdev->bd_disk->private_data;
 940         meta = zram->meta;
 941
 942         bit_spin_lock(ZRAM_ACCESS, &meta->table[index].value);
 943         zram_free_page(zram, index);
 944         bit_spin_unlock(ZRAM_ACCESS, &meta->table[index].value);
 945         atomic64_inc(&zram->stats.notify_free);
 946 }
 947
 948 static const struct block_device_operations zram_devops = {
 949         .swap_slot_free_notify = zram_slot_free_notify,
 950         .owner = THIS_MODULE
 951 };
 952
 953 static DEVICE_ATTR(disksize, S_IRUGO | S_IWUSR,
 954                 disksize_show, disksize_store);
 955 static DEVICE_ATTR(initstate, S_IRUGO, initstate_show, NULL);
 956 static DEVICE_ATTR(reset, S_IWUSR, NULL, reset_store);
 957 static DEVICE_ATTR(orig_data_size, S_IRUGO, orig_data_size_show, NULL);
 958 static DEVICE_ATTR(mem_used_total, S_IRUGO, mem_used_total_show, NULL);
 959 static DEVICE_ATTR(mem_limit, S_IRUGO | S_IWUSR, mem_limit_show,
 960                 mem_limit_store);
 961 static DEVICE_ATTR(mem_used_max, S_IRUGO | S_IWUSR, mem_used_max_show,
 962                 mem_used_max_store);
 963 static DEVICE_ATTR(max_comp_streams, S_IRUGO | S_IWUSR,
 964                 max_comp_streams_show, max_comp_streams_store);
 965 static DEVICE_ATTR(comp_algorithm, S_IRUGO | S_IWUSR,
 966                 comp_algorithm_show, comp_algorithm_store);
 967
 968 ZRAM_ATTR_RO(num_reads);
 969 ZRAM_ATTR_RO(num_writes);
 970 ZRAM_ATTR_RO(failed_reads);
 971 ZRAM_ATTR_RO(failed_writes);
 972 ZRAM_ATTR_RO(invalid_io);
 973 ZRAM_ATTR_RO(notify_free);
 974 ZRAM_ATTR_RO(zero_pages);
 975 ZRAM_ATTR_RO(compr_data_size);
 976
 977 static struct attribute *zram_disk_attrs[] = {
 978         &dev_attr_disksize.attr,
 979         &dev_attr_initstate.attr,
 980         &dev_attr_reset.attr,
 981         &dev_attr_num_reads.attr,
 982         &dev_attr_num_writes.attr,
 983         &dev_attr_failed_reads.attr,
 984         &dev_attr_failed_writes.attr,
 985         &dev_attr_invalid_io.attr,
 986         &dev_attr_notify_free.attr,
 987         &dev_attr_zero_pages.attr,
 988         &dev_attr_orig_data_size.attr,
 989         &dev_attr_compr_data_size.attr,
 990         &dev_attr_mem_used_total.attr,
 991         &dev_attr_mem_limit.attr,
 992         &dev_attr_mem_used_max.attr,
 993         &dev_attr_max_comp_streams.attr,
 994         &dev_attr_comp_algorithm.attr,
 995         NULL,
 996 };
 997
 998 static struct attribute_group zram_disk_attr_group = {
 999         .attrs = zram_disk_attrs,
1000 };
1001
1002 static int create_device(struct zram *zram, int device_id)
1003 {
1004         int ret = -ENOMEM;
1005
1006         init_rwsem(&zram->init_lock);
1007
1008         zram->queue = blk_alloc_queue(GFP_KERNEL);
1009         if (!zram->queue) {
1010                 pr_err("Error allocating disk queue for device %d\n",
1011                         device_id);
1012                 goto out;
1013         }
1014
1015         blk_queue_make_request(zram->queue, zram_make_request);
1016         zram->queue->queuedata = zram;
1017
1018          /* gendisk structure */
1019         zram->disk = alloc_disk(1);
1020         if (!zram->disk) {
1021                 pr_warn("Error allocating disk structure for device %d\n",
1022                         device_id);
1023                 goto out_free_queue;
1024         }
1025
1026         zram->disk->major = zram_major;
1027         zram->disk->first_minor = device_id;
1028         zram->disk->fops = &zram_devops;
1029         zram->disk->queue = zram->queue;
1030         zram->disk->private_data = zram;
1031         snprintf(zram->disk->disk_name, 16, "zram%d", device_id);
1032
1033         /* Actual capacity set using syfs (/sys/block/zram<id>/disksize */
1034         set_capacity(zram->disk, 0);
1035         /* zram devices sort of resembles non-rotational disks */
1036         queue_flag_set_unlocked(QUEUE_FLAG_NONROT, zram->disk->queue);
1037         queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, zram->disk->queue);
1038         /*
1039          * To ensure that we always get PAGE_SIZE aligned
1040          * and n*PAGE_SIZED sized I/O requests.
1041          */
1042         blk_queue_physical_block_size(zram->disk->queue, PAGE_SIZE);
1043         blk_queue_logical_block_size(zram->disk->queue,
1044                                         ZRAM_LOGICAL_BLOCK_SIZE);
1045         blk_queue_io_min(zram->disk->queue, PAGE_SIZE);
1046         blk_queue_io_opt(zram->disk->queue, PAGE_SIZE);
1047         zram->disk->queue->limits.discard_granularity = PAGE_SIZE;
1048         zram->disk->queue->limits.max_discard_sectors = UINT_MAX;
1049         /*
1050          * zram_bio_discard() will clear all logical blocks if logical block
1051          * size is identical with physical block size(PAGE_SIZE). But if it is
1052          * different, we will skip discarding some parts of logical blocks in
1053          * the part of the request range which isn't aligned to physical block
1054          * size.  So we can't ensure that all discarded logical blocks are
1055          * zeroed.
1056          */
1057         if (ZRAM_LOGICAL_BLOCK_SIZE == PAGE_SIZE)
1058                 zram->disk->queue->limits.discard_zeroes_data = 1;
1059         else
1060                 zram->disk->queue->limits.discard_zeroes_data = 0;
1061         queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, zram->disk->queue);
1062
1063         add_disk(zram->disk);
1064
1065         ret = sysfs_create_group(&disk_to_dev(zram->disk)->kobj,
1066                                 &zram_disk_attr_group);
1067         if (ret < 0) {
1068                 pr_warn("Error creating sysfs group");
1069                 goto out_free_disk;
1070         }
1071         strlcpy(zram->compressor, default_compressor, sizeof(zram->compressor));
1072         zram->meta = NULL;
1073         zram->max_comp_streams = 1;
1074         return 0;
1075
1076 out_free_disk:
1077         del_gendisk(zram->disk);
1078         put_disk(zram->disk);
1079 out_free_queue:
1080         blk_cleanup_queue(zram->queue);
1081 out:
1082         return ret;
1083 }
1084
1085 static void destroy_device(struct zram *zram)
1086 {
1087         sysfs_remove_group(&disk_to_dev(zram->disk)->kobj,
1088                         &zram_disk_attr_group);
1089
1090         del_gendisk(zram->disk);
1091         put_disk(zram->disk);
1092
1093         blk_cleanup_queue(zram->queue);
1094 }
1095
1096 static int __init zram_init(void)
1097 {
1098         int ret, dev_id;
1099
1100         if (num_devices > max_num_devices) {
1101                 pr_warn("Invalid value for num_devices: %u\n",
1102                                 num_devices);
1103                 ret = -EINVAL;
1104                 goto out;
1105         }
1106
1107         zram_major = register_blkdev(0, "zram");
1108         if (zram_major <= 0) {
1109                 pr_warn("Unable to get major number\n");
1110                 ret = -EBUSY;
1111                 goto out;
1112         }
1113
1114         /* Allocate the device array and initialize each one */
1115         zram_devices = kzalloc(num_devices * sizeof(struct zram), GFP_KERNEL);
1116         if (!zram_devices) {
1117                 ret = -ENOMEM;
1118                 goto unregister;
1119         }
1120
1121         for (dev_id = 0; dev_id < num_devices; dev_id++) {
1122                 ret = create_device(&zram_devices[dev_id], dev_id);
1123                 if (ret)
1124                         goto free_devices;
1125         }
1126
1127         pr_info("Created %u device(s) ...\n", num_devices);
1128
1129         return 0;
1130
1131 free_devices:
1132         while (dev_id)
1133                 destroy_device(&zram_devices[--dev_id]);
1134         kfree(zram_devices);
1135 unregister:
1136         unregister_blkdev(zram_major, "zram");
1137 out:
1138         return ret;
1139 }
1140
1141 static void __exit zram_exit(void)
1142 {
1143         int i;
1144         struct zram *zram;
1145
1146         for (i = 0; i < num_devices; i++) {
1147                 zram = &zram_devices[i];
1148
1149                 destroy_device(zram);
1150                 /*
1151                  * Shouldn't access zram->disk after destroy_device
1152                  * because destroy_device already released zram->disk.
1153                  */
1154                 zram_reset_device(zram, false);
1155         }
1156
1157         unregister_blkdev(zram_major, "zram");
1158
1159         kfree(zram_devices);
1160         pr_debug("Cleanup done!\n");
1161 }
1162
1163 module_init(zram_init);
1164 module_exit(zram_exit);
1165
1166 module_param(num_devices, uint, 0);
1167 MODULE_PARM_DESC(num_devices, "Number of zram devices");
1168
1169 MODULE_LICENSE("Dual BSD/GPL");
1170 MODULE_AUTHOR("Nitin Gupta <ngupta@vflare.org>");
1171 MODULE_DESCRIPTION("Compressed RAM Block Device");