drivers/md/dm-integrity.c

   1 /*
   2  * Copyright (C) 2016-2017 Red Hat, Inc. All rights reserved.
   3  * Copyright (C) 2016-2017 Milan Broz
   4  * Copyright (C) 2016-2017 Mikulas Patocka
   5  *
   6  * This file is released under the GPL.
   7  */
   8
   9 #include <linux/compiler.h>
  10 #include <linux/module.h>
  11 #include <linux/device-mapper.h>
  12 #include <linux/dm-io.h>
  13 #include <linux/vmalloc.h>
  14 #include <linux/sort.h>
  15 #include <linux/rbtree.h>
  16 #include <linux/delay.h>
  17 #include <linux/random.h>
  18 #include <crypto/hash.h>
  19 #include <crypto/skcipher.h>
  20 #include <linux/async_tx.h>
  21 #include <linux/dm-bufio.h>
  22
  23 #define DM_MSG_PREFIX "integrity"
  24
  25 #define DEFAULT_INTERLEAVE_SECTORS      32768
  26 #define DEFAULT_JOURNAL_SIZE_FACTOR     7
  27 #define DEFAULT_BUFFER_SECTORS          128
  28 #define DEFAULT_JOURNAL_WATERMARK       50
  29 #define DEFAULT_SYNC_MSEC               10000
  30 #define DEFAULT_MAX_JOURNAL_SECTORS     131072
  31 #define MIN_LOG2_INTERLEAVE_SECTORS     3
  32 #define MAX_LOG2_INTERLEAVE_SECTORS     31
  33 #define METADATA_WORKQUEUE_MAX_ACTIVE   16
  34 #define RECALC_SECTORS                  8192
  35 #define RECALC_WRITE_SUPER              16
  36
  37 /*
  38  * Warning - DEBUG_PRINT prints security-sensitive data to the log,
  39  * so it should not be enabled in the official kernel
  40  */
  41 //#define DEBUG_PRINT
  42 //#define INTERNAL_VERIFY
  43
  44 /*
  45  * On disk structures
  46  */
  47
  48 #define SB_MAGIC                        "integrt"
  49 #define SB_VERSION_1                    1
  50 #define SB_VERSION_2                    2
  51 #define SB_SECTORS                      8
  52 #define MAX_SECTORS_PER_BLOCK           8
  53
  54 struct superblock {
  55         __u8 magic[8];
  56         __u8 version;
  57         __u8 log2_interleave_sectors;
  58         __u16 integrity_tag_size;
  59         __u32 journal_sections;
  60         __u64 provided_data_sectors;    /* userspace uses this value */
  61         __u32 flags;
  62         __u8 log2_sectors_per_block;
  63         __u8 pad[3];
  64         __u64 recalc_sector;
  65 };
  66
  67 #define SB_FLAG_HAVE_JOURNAL_MAC        0x1
  68 #define SB_FLAG_RECALCULATING           0x2
  69
  70 #define JOURNAL_ENTRY_ROUNDUP           8
  71
  72 typedef __u64 commit_id_t;
  73 #define JOURNAL_MAC_PER_SECTOR          8
  74
  75 struct journal_entry {
  76         union {
  77                 struct {
  78                         __u32 sector_lo;
  79                         __u32 sector_hi;
  80                 } s;
  81                 __u64 sector;
  82         } u;
  83         commit_id_t last_bytes[0];
  84         /* __u8 tag[0]; */
  85 };
  86
  87 #define journal_entry_tag(ic, je)               ((__u8 *)&(je)->last_bytes[(ic)->sectors_per_block])
  88
  89 #if BITS_PER_LONG == 64
  90 #define journal_entry_set_sector(je, x)         do { smp_wmb(); WRITE_ONCE((je)->u.sector, cpu_to_le64(x)); } while (0)
  91 #define journal_entry_get_sector(je)            le64_to_cpu((je)->u.sector)
  92 #elif defined(CONFIG_LBDAF)
  93 #define journal_entry_set_sector(je, x)         do { (je)->u.s.sector_lo = cpu_to_le32(x); smp_wmb(); WRITE_ONCE((je)->u.s.sector_hi, cpu_to_le32((x) >> 32)); } while (0)
  94 #define journal_entry_get_sector(je)            le64_to_cpu((je)->u.sector)
  95 #else
  96 #define journal_entry_set_sector(je, x)         do { (je)->u.s.sector_lo = cpu_to_le32(x); smp_wmb(); WRITE_ONCE((je)->u.s.sector_hi, cpu_to_le32(0)); } while (0)
  97 #define journal_entry_get_sector(je)            le32_to_cpu((je)->u.s.sector_lo)
  98 #endif
  99 #define journal_entry_is_unused(je)             ((je)->u.s.sector_hi == cpu_to_le32(-1))
 100 #define journal_entry_set_unused(je)            do { ((je)->u.s.sector_hi = cpu_to_le32(-1)); } while (0)
 101 #define journal_entry_is_inprogress(je)         ((je)->u.s.sector_hi == cpu_to_le32(-2))
 102 #define journal_entry_set_inprogress(je)        do { ((je)->u.s.sector_hi = cpu_to_le32(-2)); } while (0)
 103
 104 #define JOURNAL_BLOCK_SECTORS           8
 105 #define JOURNAL_SECTOR_DATA             ((1 << SECTOR_SHIFT) - sizeof(commit_id_t))
 106 #define JOURNAL_MAC_SIZE                (JOURNAL_MAC_PER_SECTOR * JOURNAL_BLOCK_SECTORS)
 107
 108 struct journal_sector {
 109         __u8 entries[JOURNAL_SECTOR_DATA - JOURNAL_MAC_PER_SECTOR];
 110         __u8 mac[JOURNAL_MAC_PER_SECTOR];
 111         commit_id_t commit_id;
 112 };
 113
 114 #define MAX_TAG_SIZE                    (JOURNAL_SECTOR_DATA - JOURNAL_MAC_PER_SECTOR - offsetof(struct journal_entry, last_bytes[MAX_SECTORS_PER_BLOCK]))
 115
 116 #define METADATA_PADDING_SECTORS        8
 117
 118 #define N_COMMIT_IDS                    4
 119
 120 static unsigned char prev_commit_seq(unsigned char seq)
 121 {
 122         return (seq + N_COMMIT_IDS - 1) % N_COMMIT_IDS;
 123 }
 124
 125 static unsigned char next_commit_seq(unsigned char seq)
 126 {
 127         return (seq + 1) % N_COMMIT_IDS;
 128 }
 129
 130 /*
 131  * In-memory structures
 132  */
 133
 134 struct journal_node {
 135         struct rb_node node;
 136         sector_t sector;
 137 };
 138
 139 struct alg_spec {
 140         char *alg_string;
 141         char *key_string;
 142         __u8 *key;
 143         unsigned key_size;
 144 };
 145
 146 struct dm_integrity_c {
 147         struct dm_dev *dev;
 148         struct dm_dev *meta_dev;
 149         unsigned tag_size;
 150         __s8 log2_tag_size;
 151         sector_t start;
 152         mempool_t journal_io_mempool;
 153         struct dm_io_client *io;
 154         struct dm_bufio_client *bufio;
 155         struct workqueue_struct *metadata_wq;
 156         struct superblock *sb;
 157         unsigned journal_pages;
 158         struct page_list *journal;
 159         struct page_list *journal_io;
 160         struct page_list *journal_xor;
 161
 162         struct crypto_skcipher *journal_crypt;
 163         struct scatterlist **journal_scatterlist;
 164         struct scatterlist **journal_io_scatterlist;
 165         struct skcipher_request **sk_requests;
 166
 167         struct crypto_shash *journal_mac;
 168
 169         struct journal_node *journal_tree;
 170         struct rb_root journal_tree_root;
 171
 172         sector_t provided_data_sectors;
 173
 174         unsigned short journal_entry_size;
 175         unsigned char journal_entries_per_sector;
 176         unsigned char journal_section_entries;
 177         unsigned short journal_section_sectors;
 178         unsigned journal_sections;
 179         unsigned journal_entries;
 180         sector_t data_device_sectors;
 181         sector_t meta_device_sectors;
 182         unsigned initial_sectors;
 183         unsigned metadata_run;
 184         __s8 log2_metadata_run;
 185         __u8 log2_buffer_sectors;
 186         __u8 sectors_per_block;
 187
 188         unsigned char mode;
 189         int suspending;
 190
 191         int failed;
 192
 193         struct crypto_shash *internal_hash;
 194
 195         /* these variables are locked with endio_wait.lock */
 196         struct rb_root in_progress;
 197         struct list_head wait_list;
 198         wait_queue_head_t endio_wait;
 199         struct workqueue_struct *wait_wq;
 200
 201         unsigned char commit_seq;
 202         commit_id_t commit_ids[N_COMMIT_IDS];
 203
 204         unsigned committed_section;
 205         unsigned n_committed_sections;
 206
 207         unsigned uncommitted_section;
 208         unsigned n_uncommitted_sections;
 209
 210         unsigned free_section;
 211         unsigned char free_section_entry;
 212         unsigned free_sectors;
 213
 214         unsigned free_sectors_threshold;
 215
 216         struct workqueue_struct *commit_wq;
 217         struct work_struct commit_work;
 218
 219         struct workqueue_struct *writer_wq;
 220         struct work_struct writer_work;
 221
 222         struct workqueue_struct *recalc_wq;
 223         struct work_struct recalc_work;
 224         u8 *recalc_buffer;
 225         u8 *recalc_tags;
 226
 227         struct bio_list flush_bio_list;
 228
 229         unsigned long autocommit_jiffies;
 230         struct timer_list autocommit_timer;
 231         unsigned autocommit_msec;
 232
 233         wait_queue_head_t copy_to_journal_wait;
 234
 235         struct completion crypto_backoff;
 236
 237         bool journal_uptodate;
 238         bool just_formatted;
 239
 240         struct alg_spec internal_hash_alg;
 241         struct alg_spec journal_crypt_alg;
 242         struct alg_spec journal_mac_alg;
 243
 244         atomic64_t number_of_mismatches;
 245 };
 246
 247 struct dm_integrity_range {
 248         sector_t logical_sector;
 249         unsigned n_sectors;
 250         bool waiting;
 251         union {
 252                 struct rb_node node;
 253                 struct {
 254                         struct task_struct *task;
 255                         struct list_head wait_entry;
 256                 };
 257         };
 258 };
 259
 260 struct dm_integrity_io {
 261         struct work_struct work;
 262
 263         struct dm_integrity_c *ic;
 264         bool write;
 265         bool fua;
 266
 267         struct dm_integrity_range range;
 268
 269         sector_t metadata_block;
 270         unsigned metadata_offset;
 271
 272         atomic_t in_flight;
 273         blk_status_t bi_status;
 274
 275         struct completion *completion;
 276
 277         struct gendisk *orig_bi_disk;
 278         u8 orig_bi_partno;
 279         bio_end_io_t *orig_bi_end_io;
 280         struct bio_integrity_payload *orig_bi_integrity;
 281         struct bvec_iter orig_bi_iter;
 282 };
 283
 284 struct journal_completion {
 285         struct dm_integrity_c *ic;
 286         atomic_t in_flight;
 287         struct completion comp;
 288 };
 289
 290 struct journal_io {
 291         struct dm_integrity_range range;
 292         struct journal_completion *comp;
 293 };
 294
 295 static struct kmem_cache *journal_io_cache;
 296
 297 #define JOURNAL_IO_MEMPOOL      32
 298
 299 #ifdef DEBUG_PRINT
 300 #define DEBUG_print(x, ...)     printk(KERN_DEBUG x, ##__VA_ARGS__)
 301 static void __DEBUG_bytes(__u8 *bytes, size_t len, const char *msg, ...)
 302 {
 303         va_list args;
 304         va_start(args, msg);
 305         vprintk(msg, args);
 306         va_end(args);
 307         if (len)
 308                 pr_cont(":");
 309         while (len) {
 310                 pr_cont(" %02x", *bytes);
 311                 bytes++;
 312                 len--;
 313         }
 314         pr_cont("\n");
 315 }
 316 #define DEBUG_bytes(bytes, len, msg, ...)       __DEBUG_bytes(bytes, len, KERN_DEBUG msg, ##__VA_ARGS__)
 317 #else
 318 #define DEBUG_print(x, ...)                     do { } while (0)
 319 #define DEBUG_bytes(bytes, len, msg, ...)       do { } while (0)
 320 #endif
 321
 322 /*
 323  * DM Integrity profile, protection is performed layer above (dm-crypt)
 324  */
 325 static const struct blk_integrity_profile dm_integrity_profile = {
 326         .name                   = "DM-DIF-EXT-TAG",
 327         .generate_fn            = NULL,
 328         .verify_fn              = NULL,
 329 };
 330
 331 static void dm_integrity_map_continue(struct dm_integrity_io *dio, bool from_map);
 332 static void integrity_bio_wait(struct work_struct *w);
 333 static void dm_integrity_dtr(struct dm_target *ti);
 334
 335 static void dm_integrity_io_error(struct dm_integrity_c *ic, const char *msg, int err)
 336 {
 337         if (err == -EILSEQ)
 338                 atomic64_inc(&ic->number_of_mismatches);
 339         if (!cmpxchg(&ic->failed, 0, err))
 340                 DMERR("Error on %s: %d", msg, err);
 341 }
 342
 343 static int dm_integrity_failed(struct dm_integrity_c *ic)
 344 {
 345         return READ_ONCE(ic->failed);
 346 }
 347
 348 static commit_id_t dm_integrity_commit_id(struct dm_integrity_c *ic, unsigned i,
 349                                           unsigned j, unsigned char seq)
 350 {
 351         /*
 352          * Xor the number with section and sector, so that if a piece of
 353          * journal is written at wrong place, it is detected.
 354          */
 355         return ic->commit_ids[seq] ^ cpu_to_le64(((__u64)i << 32) ^ j);
 356 }
 357
 358 static void get_area_and_offset(struct dm_integrity_c *ic, sector_t data_sector,
 359                                 sector_t *area, sector_t *offset)
 360 {
 361         if (!ic->meta_dev) {
 362                 __u8 log2_interleave_sectors = ic->sb->log2_interleave_sectors;
 363                 *area = data_sector >> log2_interleave_sectors;
 364                 *offset = (unsigned)data_sector & ((1U << log2_interleave_sectors) - 1);
 365         } else {
 366                 *area = 0;
 367                 *offset = data_sector;
 368         }
 369 }
 370
 371 #define sector_to_block(ic, n)                                          \
 372 do {                                                                    \
 373         BUG_ON((n) & (unsigned)((ic)->sectors_per_block - 1));          \
 374         (n) >>= (ic)->sb->log2_sectors_per_block;                       \
 375 } while (0)
 376
 377 static __u64 get_metadata_sector_and_offset(struct dm_integrity_c *ic, sector_t area,
 378                                             sector_t offset, unsigned *metadata_offset)
 379 {
 380         __u64 ms;
 381         unsigned mo;
 382
 383         ms = area << ic->sb->log2_interleave_sectors;
 384         if (likely(ic->log2_metadata_run >= 0))
 385                 ms += area << ic->log2_metadata_run;
 386         else
 387                 ms += area * ic->metadata_run;
 388         ms >>= ic->log2_buffer_sectors;
 389
 390         sector_to_block(ic, offset);
 391
 392         if (likely(ic->log2_tag_size >= 0)) {
 393                 ms += offset >> (SECTOR_SHIFT + ic->log2_buffer_sectors - ic->log2_tag_size);
 394                 mo = (offset << ic->log2_tag_size) & ((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - 1);
 395         } else {
 396                 ms += (__u64)offset * ic->tag_size >> (SECTOR_SHIFT + ic->log2_buffer_sectors);
 397                 mo = (offset * ic->tag_size) & ((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - 1);
 398         }
 399         *metadata_offset = mo;
 400         return ms;
 401 }
 402
 403 static sector_t get_data_sector(struct dm_integrity_c *ic, sector_t area, sector_t offset)
 404 {
 405         sector_t result;
 406
 407         if (ic->meta_dev)
 408                 return offset;
 409
 410         result = area << ic->sb->log2_interleave_sectors;
 411         if (likely(ic->log2_metadata_run >= 0))
 412                 result += (area + 1) << ic->log2_metadata_run;
 413         else
 414                 result += (area + 1) * ic->metadata_run;
 415
 416         result += (sector_t)ic->initial_sectors + offset;
 417         result += ic->start;
 418
 419         return result;
 420 }
 421
 422 static void wraparound_section(struct dm_integrity_c *ic, unsigned *sec_ptr)
 423 {
 424         if (unlikely(*sec_ptr >= ic->journal_sections))
 425                 *sec_ptr -= ic->journal_sections;
 426 }
 427
 428 static void sb_set_version(struct dm_integrity_c *ic)
 429 {
 430         if (ic->meta_dev || ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
 431                 ic->sb->version = SB_VERSION_2;
 432         else
 433                 ic->sb->version = SB_VERSION_1;
 434 }
 435
 436 static int sync_rw_sb(struct dm_integrity_c *ic, int op, int op_flags)
 437 {
 438         struct dm_io_request io_req;
 439         struct dm_io_region io_loc;
 440
 441         io_req.bi_op = op;
 442         io_req.bi_op_flags = op_flags;
 443         io_req.mem.type = DM_IO_KMEM;
 444         io_req.mem.ptr.addr = ic->sb;
 445         io_req.notify.fn = NULL;
 446         io_req.client = ic->io;
 447         io_loc.bdev = ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev;
 448         io_loc.sector = ic->start;
 449         io_loc.count = SB_SECTORS;
 450
 451         return dm_io(&io_req, 1, &io_loc, NULL);
 452 }
 453
 454 static void access_journal_check(struct dm_integrity_c *ic, unsigned section, unsigned offset,
 455                                  bool e, const char *function)
 456 {
 457 #if defined(CONFIG_DM_DEBUG) || defined(INTERNAL_VERIFY)
 458         unsigned limit = e ? ic->journal_section_entries : ic->journal_section_sectors;
 459
 460         if (unlikely(section >= ic->journal_sections) ||
 461             unlikely(offset >= limit)) {
 462                 printk(KERN_CRIT "%s: invalid access at (%u,%u), limit (%u,%u)\n",
 463                         function, section, offset, ic->journal_sections, limit);
 464                 BUG();
 465         }
 466 #endif
 467 }
 468
 469 static void page_list_location(struct dm_integrity_c *ic, unsigned section, unsigned offset,
 470                                unsigned *pl_index, unsigned *pl_offset)
 471 {
 472         unsigned sector;
 473
 474         access_journal_check(ic, section, offset, false, "page_list_location");
 475
 476         sector = section * ic->journal_section_sectors + offset;
 477
 478         *pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
 479         *pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
 480 }
 481
 482 static struct journal_sector *access_page_list(struct dm_integrity_c *ic, struct page_list *pl,
 483                                                unsigned section, unsigned offset, unsigned *n_sectors)
 484 {
 485         unsigned pl_index, pl_offset;
 486         char *va;
 487
 488         page_list_location(ic, section, offset, &pl_index, &pl_offset);
 489
 490         if (n_sectors)
 491                 *n_sectors = (PAGE_SIZE - pl_offset) >> SECTOR_SHIFT;
 492
 493         va = lowmem_page_address(pl[pl_index].page);
 494
 495         return (struct journal_sector *)(va + pl_offset);
 496 }
 497
 498 static struct journal_sector *access_journal(struct dm_integrity_c *ic, unsigned section, unsigned offset)
 499 {
 500         return access_page_list(ic, ic->journal, section, offset, NULL);
 501 }
 502
 503 static struct journal_entry *access_journal_entry(struct dm_integrity_c *ic, unsigned section, unsigned n)
 504 {
 505         unsigned rel_sector, offset;
 506         struct journal_sector *js;
 507
 508         access_journal_check(ic, section, n, true, "access_journal_entry");
 509
 510         rel_sector = n % JOURNAL_BLOCK_SECTORS;
 511         offset = n / JOURNAL_BLOCK_SECTORS;
 512
 513         js = access_journal(ic, section, rel_sector);
 514         return (struct journal_entry *)((char *)js + offset * ic->journal_entry_size);
 515 }
 516
 517 static struct journal_sector *access_journal_data(struct dm_integrity_c *ic, unsigned section, unsigned n)
 518 {
 519         n <<= ic->sb->log2_sectors_per_block;
 520
 521         n += JOURNAL_BLOCK_SECTORS;
 522
 523         access_journal_check(ic, section, n, false, "access_journal_data");
 524
 525         return access_journal(ic, section, n);
 526 }
 527
 528 static void section_mac(struct dm_integrity_c *ic, unsigned section, __u8 result[JOURNAL_MAC_SIZE])
 529 {
 530         SHASH_DESC_ON_STACK(desc, ic->journal_mac);
 531         int r;
 532         unsigned j, size;
 533
 534         desc->tfm = ic->journal_mac;
 535         desc->flags = 0;
 536
 537         r = crypto_shash_init(desc);
 538         if (unlikely(r)) {
 539                 dm_integrity_io_error(ic, "crypto_shash_init", r);
 540                 goto err;
 541         }
 542
 543         for (j = 0; j < ic->journal_section_entries; j++) {
 544                 struct journal_entry *je = access_journal_entry(ic, section, j);
 545                 r = crypto_shash_update(desc, (__u8 *)&je->u.sector, sizeof je->u.sector);
 546                 if (unlikely(r)) {
 547                         dm_integrity_io_error(ic, "crypto_shash_update", r);
 548                         goto err;
 549                 }
 550         }
 551
 552         size = crypto_shash_digestsize(ic->journal_mac);
 553
 554         if (likely(size <= JOURNAL_MAC_SIZE)) {
 555                 r = crypto_shash_final(desc, result);
 556                 if (unlikely(r)) {
 557                         dm_integrity_io_error(ic, "crypto_shash_final", r);
 558                         goto err;
 559                 }
 560                 memset(result + size, 0, JOURNAL_MAC_SIZE - size);
 561         } else {
 562                 __u8 digest[HASH_MAX_DIGESTSIZE];
 563
 564                 if (WARN_ON(size > sizeof(digest))) {
 565                         dm_integrity_io_error(ic, "digest_size", -EINVAL);
 566                         goto err;
 567                 }
 568                 r = crypto_shash_final(desc, digest);
 569                 if (unlikely(r)) {
 570                         dm_integrity_io_error(ic, "crypto_shash_final", r);
 571                         goto err;
 572                 }
 573                 memcpy(result, digest, JOURNAL_MAC_SIZE);
 574         }
 575
 576         return;
 577 err:
 578         memset(result, 0, JOURNAL_MAC_SIZE);
 579 }
 580
 581 static void rw_section_mac(struct dm_integrity_c *ic, unsigned section, bool wr)
 582 {
 583         __u8 result[JOURNAL_MAC_SIZE];
 584         unsigned j;
 585
 586         if (!ic->journal_mac)
 587                 return;
 588
 589         section_mac(ic, section, result);
 590
 591         for (j = 0; j < JOURNAL_BLOCK_SECTORS; j++) {
 592                 struct journal_sector *js = access_journal(ic, section, j);
 593
 594                 if (likely(wr))
 595                         memcpy(&js->mac, result + (j * JOURNAL_MAC_PER_SECTOR), JOURNAL_MAC_PER_SECTOR);
 596                 else {
 597                         if (memcmp(&js->mac, result + (j * JOURNAL_MAC_PER_SECTOR), JOURNAL_MAC_PER_SECTOR))
 598                                 dm_integrity_io_error(ic, "journal mac", -EILSEQ);
 599                 }
 600         }
 601 }
 602
 603 static void complete_journal_op(void *context)
 604 {
 605         struct journal_completion *comp = context;
 606         BUG_ON(!atomic_read(&comp->in_flight));
 607         if (likely(atomic_dec_and_test(&comp->in_flight)))
 608                 complete(&comp->comp);
 609 }
 610
 611 static void xor_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
 612                         unsigned n_sections, struct journal_completion *comp)
 613 {
 614         struct async_submit_ctl submit;
 615         size_t n_bytes = (size_t)(n_sections * ic->journal_section_sectors) << SECTOR_SHIFT;
 616         unsigned pl_index, pl_offset, section_index;
 617         struct page_list *source_pl, *target_pl;
 618
 619         if (likely(encrypt)) {
 620                 source_pl = ic->journal;
 621                 target_pl = ic->journal_io;
 622         } else {
 623                 source_pl = ic->journal_io;
 624                 target_pl = ic->journal;
 625         }
 626
 627         page_list_location(ic, section, 0, &pl_index, &pl_offset);
 628
 629         atomic_add(roundup(pl_offset + n_bytes, PAGE_SIZE) >> PAGE_SHIFT, &comp->in_flight);
 630
 631         init_async_submit(&submit, ASYNC_TX_XOR_ZERO_DST, NULL, complete_journal_op, comp, NULL);
 632
 633         section_index = pl_index;
 634
 635         do {
 636                 size_t this_step;
 637                 struct page *src_pages[2];
 638                 struct page *dst_page;
 639
 640                 while (unlikely(pl_index == section_index)) {
 641                         unsigned dummy;
 642                         if (likely(encrypt))
 643                                 rw_section_mac(ic, section, true);
 644                         section++;
 645                         n_sections--;
 646                         if (!n_sections)
 647                                 break;
 648                         page_list_location(ic, section, 0, &section_index, &dummy);
 649                 }
 650
 651                 this_step = min(n_bytes, (size_t)PAGE_SIZE - pl_offset);
 652                 dst_page = target_pl[pl_index].page;
 653                 src_pages[0] = source_pl[pl_index].page;
 654                 src_pages[1] = ic->journal_xor[pl_index].page;
 655
 656                 async_xor(dst_page, src_pages, pl_offset, 2, this_step, &submit);
 657
 658                 pl_index++;
 659                 pl_offset = 0;
 660                 n_bytes -= this_step;
 661         } while (n_bytes);
 662
 663         BUG_ON(n_sections);
 664
 665         async_tx_issue_pending_all();
 666 }
 667
 668 static void complete_journal_encrypt(struct crypto_async_request *req, int err)
 669 {
 670         struct journal_completion *comp = req->data;
 671         if (unlikely(err)) {
 672                 if (likely(err == -EINPROGRESS)) {
 673                         complete(&comp->ic->crypto_backoff);
 674                         return;
 675                 }
 676                 dm_integrity_io_error(comp->ic, "asynchronous encrypt", err);
 677         }
 678         complete_journal_op(comp);
 679 }
 680
 681 static bool do_crypt(bool encrypt, struct skcipher_request *req, struct journal_completion *comp)
 682 {
 683         int r;
 684         skcipher_request_set_callback(req, CRYPTO_TFM_REQ_MAY_BACKLOG,
 685                                       complete_journal_encrypt, comp);
 686         if (likely(encrypt))
 687                 r = crypto_skcipher_encrypt(req);
 688         else
 689                 r = crypto_skcipher_decrypt(req);
 690         if (likely(!r))
 691                 return false;
 692         if (likely(r == -EINPROGRESS))
 693                 return true;
 694         if (likely(r == -EBUSY)) {
 695                 wait_for_completion(&comp->ic->crypto_backoff);
 696                 reinit_completion(&comp->ic->crypto_backoff);
 697                 return true;
 698         }
 699         dm_integrity_io_error(comp->ic, "encrypt", r);
 700         return false;
 701 }
 702
 703 static void crypt_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
 704                           unsigned n_sections, struct journal_completion *comp)
 705 {
 706         struct scatterlist **source_sg;
 707         struct scatterlist **target_sg;
 708
 709         atomic_add(2, &comp->in_flight);
 710
 711         if (likely(encrypt)) {
 712                 source_sg = ic->journal_scatterlist;
 713                 target_sg = ic->journal_io_scatterlist;
 714         } else {
 715                 source_sg = ic->journal_io_scatterlist;
 716                 target_sg = ic->journal_scatterlist;
 717         }
 718
 719         do {
 720                 struct skcipher_request *req;
 721                 unsigned ivsize;
 722                 char *iv;
 723
 724                 if (likely(encrypt))
 725                         rw_section_mac(ic, section, true);
 726
 727                 req = ic->sk_requests[section];
 728                 ivsize = crypto_skcipher_ivsize(ic->journal_crypt);
 729                 iv = req->iv;
 730
 731                 memcpy(iv, iv + ivsize, ivsize);
 732
 733                 req->src = source_sg[section];
 734                 req->dst = target_sg[section];
 735
 736                 if (unlikely(do_crypt(encrypt, req, comp)))
 737                         atomic_inc(&comp->in_flight);
 738
 739                 section++;
 740                 n_sections--;
 741         } while (n_sections);
 742
 743         atomic_dec(&comp->in_flight);
 744         complete_journal_op(comp);
 745 }
 746
 747 static void encrypt_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
 748                             unsigned n_sections, struct journal_completion *comp)
 749 {
 750         if (ic->journal_xor)
 751                 return xor_journal(ic, encrypt, section, n_sections, comp);
 752         else
 753                 return crypt_journal(ic, encrypt, section, n_sections, comp);
 754 }
 755
 756 static void complete_journal_io(unsigned long error, void *context)
 757 {
 758         struct journal_completion *comp = context;
 759         if (unlikely(error != 0))
 760                 dm_integrity_io_error(comp->ic, "writing journal", -EIO);
 761         complete_journal_op(comp);
 762 }
 763
 764 static void rw_journal(struct dm_integrity_c *ic, int op, int op_flags, unsigned section,
 765                        unsigned n_sections, struct journal_completion *comp)
 766 {
 767         struct dm_io_request io_req;
 768         struct dm_io_region io_loc;
 769         unsigned sector, n_sectors, pl_index, pl_offset;
 770         int r;
 771
 772         if (unlikely(dm_integrity_failed(ic))) {
 773                 if (comp)
 774                         complete_journal_io(-1UL, comp);
 775                 return;
 776         }
 777
 778         sector = section * ic->journal_section_sectors;
 779         n_sectors = n_sections * ic->journal_section_sectors;
 780
 781         pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
 782         pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
 783
 784         io_req.bi_op = op;
 785         io_req.bi_op_flags = op_flags;
 786         io_req.mem.type = DM_IO_PAGE_LIST;
 787         if (ic->journal_io)
 788                 io_req.mem.ptr.pl = &ic->journal_io[pl_index];
 789         else
 790                 io_req.mem.ptr.pl = &ic->journal[pl_index];
 791         io_req.mem.offset = pl_offset;
 792         if (likely(comp != NULL)) {
 793                 io_req.notify.fn = complete_journal_io;
 794                 io_req.notify.context = comp;
 795         } else {
 796                 io_req.notify.fn = NULL;
 797         }
 798         io_req.client = ic->io;
 799         io_loc.bdev = ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev;
 800         io_loc.sector = ic->start + SB_SECTORS + sector;
 801         io_loc.count = n_sectors;
 802
 803         r = dm_io(&io_req, 1, &io_loc, NULL);
 804         if (unlikely(r)) {
 805                 dm_integrity_io_error(ic, op == REQ_OP_READ ? "reading journal" : "writing journal", r);
 806                 if (comp) {
 807                         WARN_ONCE(1, "asynchronous dm_io failed: %d", r);
 808                         complete_journal_io(-1UL, comp);
 809                 }
 810         }
 811 }
 812
 813 static void write_journal(struct dm_integrity_c *ic, unsigned commit_start, unsigned commit_sections)
 814 {
 815         struct journal_completion io_comp;
 816         struct journal_completion crypt_comp_1;
 817         struct journal_completion crypt_comp_2;
 818         unsigned i;
 819
 820         io_comp.ic = ic;
 821         init_completion(&io_comp.comp);
 822
 823         if (commit_start + commit_sections <= ic->journal_sections) {
 824                 io_comp.in_flight = (atomic_t)ATOMIC_INIT(1);
 825                 if (ic->journal_io) {
 826                         crypt_comp_1.ic = ic;
 827                         init_completion(&crypt_comp_1.comp);
 828                         crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
 829                         encrypt_journal(ic, true, commit_start, commit_sections, &crypt_comp_1);
 830                         wait_for_completion_io(&crypt_comp_1.comp);
 831                 } else {
 832                         for (i = 0; i < commit_sections; i++)
 833                                 rw_section_mac(ic, commit_start + i, true);
 834                 }
 835                 rw_journal(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, commit_start,
 836                            commit_sections, &io_comp);
 837         } else {
 838                 unsigned to_end;
 839                 io_comp.in_flight = (atomic_t)ATOMIC_INIT(2);
 840                 to_end = ic->journal_sections - commit_start;
 841                 if (ic->journal_io) {
 842                         crypt_comp_1.ic = ic;
 843                         init_completion(&crypt_comp_1.comp);
 844                         crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
 845                         encrypt_journal(ic, true, commit_start, to_end, &crypt_comp_1);
 846                         if (try_wait_for_completion(&crypt_comp_1.comp)) {
 847                                 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
 848                                 reinit_completion(&crypt_comp_1.comp);
 849                                 crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
 850                                 encrypt_journal(ic, true, 0, commit_sections - to_end, &crypt_comp_1);
 851                                 wait_for_completion_io(&crypt_comp_1.comp);
 852                         } else {
 853                                 crypt_comp_2.ic = ic;
 854                                 init_completion(&crypt_comp_2.comp);
 855                                 crypt_comp_2.in_flight = (atomic_t)ATOMIC_INIT(0);
 856                                 encrypt_journal(ic, true, 0, commit_sections - to_end, &crypt_comp_2);
 857                                 wait_for_completion_io(&crypt_comp_1.comp);
 858                                 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
 859                                 wait_for_completion_io(&crypt_comp_2.comp);
 860                         }
 861                 } else {
 862                         for (i = 0; i < to_end; i++)
 863                                 rw_section_mac(ic, commit_start + i, true);
 864                         rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
 865                         for (i = 0; i < commit_sections - to_end; i++)
 866                                 rw_section_mac(ic, i, true);
 867                 }
 868                 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, 0, commit_sections - to_end, &io_comp);
 869         }
 870
 871         wait_for_completion_io(&io_comp.comp);
 872 }
 873
 874 static void copy_from_journal(struct dm_integrity_c *ic, unsigned section, unsigned offset,
 875                               unsigned n_sectors, sector_t target, io_notify_fn fn, void *data)
 876 {
 877         struct dm_io_request io_req;
 878         struct dm_io_region io_loc;
 879         int r;
 880         unsigned sector, pl_index, pl_offset;
 881
 882         BUG_ON((target | n_sectors | offset) & (unsigned)(ic->sectors_per_block - 1));
 883
 884         if (unlikely(dm_integrity_failed(ic))) {
 885                 fn(-1UL, data);
 886                 return;
 887         }
 888
 889         sector = section * ic->journal_section_sectors + JOURNAL_BLOCK_SECTORS + offset;
 890
 891         pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
 892         pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
 893
 894         io_req.bi_op = REQ_OP_WRITE;
 895         io_req.bi_op_flags = 0;
 896         io_req.mem.type = DM_IO_PAGE_LIST;
 897         io_req.mem.ptr.pl = &ic->journal[pl_index];
 898         io_req.mem.offset = pl_offset;
 899         io_req.notify.fn = fn;
 900         io_req.notify.context = data;
 901         io_req.client = ic->io;
 902         io_loc.bdev = ic->dev->bdev;
 903         io_loc.sector = target;
 904         io_loc.count = n_sectors;
 905
 906         r = dm_io(&io_req, 1, &io_loc, NULL);
 907         if (unlikely(r)) {
 908                 WARN_ONCE(1, "asynchronous dm_io failed: %d", r);
 909                 fn(-1UL, data);
 910         }
 911 }
 912
 913 static bool ranges_overlap(struct dm_integrity_range *range1, struct dm_integrity_range *range2)
 914 {
 915         return range1->logical_sector < range2->logical_sector + range2->n_sectors &&
 916                range2->logical_sector + range2->n_sectors > range2->logical_sector;
 917 }
 918
 919 static bool add_new_range(struct dm_integrity_c *ic, struct dm_integrity_range *new_range, bool check_waiting)
 920 {
 921         struct rb_node **n = &ic->in_progress.rb_node;
 922         struct rb_node *parent;
 923
 924         BUG_ON((new_range->logical_sector | new_range->n_sectors) & (unsigned)(ic->sectors_per_block - 1));
 925
 926         if (likely(check_waiting)) {
 927                 struct dm_integrity_range *range;
 928                 list_for_each_entry(range, &ic->wait_list, wait_entry) {
 929                         if (unlikely(ranges_overlap(range, new_range)))
 930                                 return false;
 931                 }
 932         }
 933
 934         parent = NULL;
 935
 936         while (*n) {
 937                 struct dm_integrity_range *range = container_of(*n, struct dm_integrity_range, node);
 938
 939                 parent = *n;
 940                 if (new_range->logical_sector + new_range->n_sectors <= range->logical_sector) {
 941                         n = &range->node.rb_left;
 942                 } else if (new_range->logical_sector >= range->logical_sector + range->n_sectors) {
 943                         n = &range->node.rb_right;
 944                 } else {
 945                         return false;
 946                 }
 947         }
 948
 949         rb_link_node(&new_range->node, parent, n);
 950         rb_insert_color(&new_range->node, &ic->in_progress);
 951
 952         return true;
 953 }
 954
 955 static void remove_range_unlocked(struct dm_integrity_c *ic, struct dm_integrity_range *range)
 956 {
 957         rb_erase(&range->node, &ic->in_progress);
 958         while (unlikely(!list_empty(&ic->wait_list))) {
 959                 struct dm_integrity_range *last_range =
 960                         list_first_entry(&ic->wait_list, struct dm_integrity_range, wait_entry);
 961                 struct task_struct *last_range_task;
 962                 if (!ranges_overlap(range, last_range))
 963                         break;
 964                 last_range_task = last_range->task;
 965                 list_del(&last_range->wait_entry);
 966                 if (!add_new_range(ic, last_range, false)) {
 967                         last_range->task = last_range_task;
 968                         list_add(&last_range->wait_entry, &ic->wait_list);
 969                         break;
 970                 }
 971                 last_range->waiting = false;
 972                 wake_up_process(last_range_task);
 973         }
 974 }
 975
 976 static void remove_range(struct dm_integrity_c *ic, struct dm_integrity_range *range)
 977 {
 978         unsigned long flags;
 979
 980         spin_lock_irqsave(&ic->endio_wait.lock, flags);
 981         remove_range_unlocked(ic, range);
 982         spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
 983 }
 984
 985 static void wait_and_add_new_range(struct dm_integrity_c *ic, struct dm_integrity_range *new_range)
 986 {
 987         new_range->waiting = true;
 988         list_add_tail(&new_range->wait_entry, &ic->wait_list);
 989         new_range->task = current;
 990         do {
 991                 __set_current_state(TASK_UNINTERRUPTIBLE);
 992                 spin_unlock_irq(&ic->endio_wait.lock);
 993                 io_schedule();
 994                 spin_lock_irq(&ic->endio_wait.lock);
 995         } while (unlikely(new_range->waiting));
 996 }
 997
 998 static void init_journal_node(struct journal_node *node)
 999 {
1000         RB_CLEAR_NODE(&node->node);
1001         node->sector = (sector_t)-1;
1002 }
1003
1004 static void add_journal_node(struct dm_integrity_c *ic, struct journal_node *node, sector_t sector)
1005 {
1006         struct rb_node **link;
1007         struct rb_node *parent;
1008
1009         node->sector = sector;
1010         BUG_ON(!RB_EMPTY_NODE(&node->node));
1011
1012         link = &ic->journal_tree_root.rb_node;
1013         parent = NULL;
1014
1015         while (*link) {
1016                 struct journal_node *j;
1017                 parent = *link;
1018                 j = container_of(parent, struct journal_node, node);
1019                 if (sector < j->sector)
1020                         link = &j->node.rb_left;
1021                 else
1022                         link = &j->node.rb_right;
1023         }
1024
1025         rb_link_node(&node->node, parent, link);
1026         rb_insert_color(&node->node, &ic->journal_tree_root);
1027 }
1028
1029 static void remove_journal_node(struct dm_integrity_c *ic, struct journal_node *node)
1030 {
1031         BUG_ON(RB_EMPTY_NODE(&node->node));
1032         rb_erase(&node->node, &ic->journal_tree_root);
1033         init_journal_node(node);
1034 }
1035
1036 #define NOT_FOUND       (-1U)
1037
1038 static unsigned find_journal_node(struct dm_integrity_c *ic, sector_t sector, sector_t *next_sector)
1039 {
1040         struct rb_node *n = ic->journal_tree_root.rb_node;
1041         unsigned found = NOT_FOUND;
1042         *next_sector = (sector_t)-1;
1043         while (n) {
1044                 struct journal_node *j = container_of(n, struct journal_node, node);
1045                 if (sector == j->sector) {
1046                         found = j - ic->journal_tree;
1047                 }
1048                 if (sector < j->sector) {
1049                         *next_sector = j->sector;
1050                         n = j->node.rb_left;
1051                 } else {
1052                         n = j->node.rb_right;
1053                 }
1054         }
1055
1056         return found;
1057 }
1058
1059 static bool test_journal_node(struct dm_integrity_c *ic, unsigned pos, sector_t sector)
1060 {
1061         struct journal_node *node, *next_node;
1062         struct rb_node *next;
1063
1064         if (unlikely(pos >= ic->journal_entries))
1065                 return false;
1066         node = &ic->journal_tree[pos];
1067         if (unlikely(RB_EMPTY_NODE(&node->node)))
1068                 return false;
1069         if (unlikely(node->sector != sector))
1070                 return false;
1071
1072         next = rb_next(&node->node);
1073         if (unlikely(!next))
1074                 return true;
1075
1076         next_node = container_of(next, struct journal_node, node);
1077         return next_node->sector != sector;
1078 }
1079
1080 static bool find_newer_committed_node(struct dm_integrity_c *ic, struct journal_node *node)
1081 {
1082         struct rb_node *next;
1083         struct journal_node *next_node;
1084         unsigned next_section;
1085
1086         BUG_ON(RB_EMPTY_NODE(&node->node));
1087
1088         next = rb_next(&node->node);
1089         if (unlikely(!next))
1090                 return false;
1091
1092         next_node = container_of(next, struct journal_node, node);
1093
1094         if (next_node->sector != node->sector)
1095                 return false;
1096
1097         next_section = (unsigned)(next_node - ic->journal_tree) / ic->journal_section_entries;
1098         if (next_section >= ic->committed_section &&
1099             next_section < ic->committed_section + ic->n_committed_sections)
1100                 return true;
1101         if (next_section + ic->journal_sections < ic->committed_section + ic->n_committed_sections)
1102                 return true;
1103
1104         return false;
1105 }
1106
1107 #define TAG_READ        0
1108 #define TAG_WRITE       1
1109 #define TAG_CMP         2
1110
1111 static int dm_integrity_rw_tag(struct dm_integrity_c *ic, unsigned char *tag, sector_t *metadata_block,
1112                                unsigned *metadata_offset, unsigned total_size, int op)
1113 {
1114         do {
1115                 unsigned char *data, *dp;
1116                 struct dm_buffer *b;
1117                 unsigned to_copy;
1118                 int r;
1119
1120                 r = dm_integrity_failed(ic);
1121                 if (unlikely(r))
1122                         return r;
1123
1124                 data = dm_bufio_read(ic->bufio, *metadata_block, &b);
1125                 if (unlikely(IS_ERR(data)))
1126                         return PTR_ERR(data);
1127
1128                 to_copy = min((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - *metadata_offset, total_size);
1129                 dp = data + *metadata_offset;
1130                 if (op == TAG_READ) {
1131                         memcpy(tag, dp, to_copy);
1132                 } else if (op == TAG_WRITE) {
1133                         memcpy(dp, tag, to_copy);
1134                         dm_bufio_mark_partial_buffer_dirty(b, *metadata_offset, *metadata_offset + to_copy);
1135                 } else  {
1136                         /* e.g.: op == TAG_CMP */
1137                         if (unlikely(memcmp(dp, tag, to_copy))) {
1138                                 unsigned i;
1139
1140                                 for (i = 0; i < to_copy; i++) {
1141                                         if (dp[i] != tag[i])
1142                                                 break;
1143                                         total_size--;
1144                                 }
1145                                 dm_bufio_release(b);
1146                                 return total_size;
1147                         }
1148                 }
1149                 dm_bufio_release(b);
1150
1151                 tag += to_copy;
1152                 *metadata_offset += to_copy;
1153                 if (unlikely(*metadata_offset == 1U << SECTOR_SHIFT << ic->log2_buffer_sectors)) {
1154                         (*metadata_block)++;
1155                         *metadata_offset = 0;
1156                 }
1157                 total_size -= to_copy;
1158         } while (unlikely(total_size));
1159
1160         return 0;
1161 }
1162
1163 static void dm_integrity_flush_buffers(struct dm_integrity_c *ic)
1164 {
1165         int r;
1166         r = dm_bufio_write_dirty_buffers(ic->bufio);
1167         if (unlikely(r))
1168                 dm_integrity_io_error(ic, "writing tags", r);
1169 }
1170
1171 static void sleep_on_endio_wait(struct dm_integrity_c *ic)
1172 {
1173         DECLARE_WAITQUEUE(wait, current);
1174         __add_wait_queue(&ic->endio_wait, &wait);
1175         __set_current_state(TASK_UNINTERRUPTIBLE);
1176         spin_unlock_irq(&ic->endio_wait.lock);
1177         io_schedule();
1178         spin_lock_irq(&ic->endio_wait.lock);
1179         __remove_wait_queue(&ic->endio_wait, &wait);
1180 }
1181
1182 static void autocommit_fn(struct timer_list *t)
1183 {
1184         struct dm_integrity_c *ic = from_timer(ic, t, autocommit_timer);
1185
1186         if (likely(!dm_integrity_failed(ic)))
1187                 queue_work(ic->commit_wq, &ic->commit_work);
1188 }
1189
1190 static void schedule_autocommit(struct dm_integrity_c *ic)
1191 {
1192         if (!timer_pending(&ic->autocommit_timer))
1193                 mod_timer(&ic->autocommit_timer, jiffies + ic->autocommit_jiffies);
1194 }
1195
1196 static void submit_flush_bio(struct dm_integrity_c *ic, struct dm_integrity_io *dio)
1197 {
1198         struct bio *bio;
1199         unsigned long flags;
1200
1201         spin_lock_irqsave(&ic->endio_wait.lock, flags);
1202         bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1203         bio_list_add(&ic->flush_bio_list, bio);
1204         spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
1205
1206         queue_work(ic->commit_wq, &ic->commit_work);
1207 }
1208
1209 static void do_endio(struct dm_integrity_c *ic, struct bio *bio)
1210 {
1211         int r = dm_integrity_failed(ic);
1212         if (unlikely(r) && !bio->bi_status)
1213                 bio->bi_status = errno_to_blk_status(r);
1214         bio_endio(bio);
1215 }
1216
1217 static void do_endio_flush(struct dm_integrity_c *ic, struct dm_integrity_io *dio)
1218 {
1219         struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1220
1221         if (unlikely(dio->fua) && likely(!bio->bi_status) && likely(!dm_integrity_failed(ic)))
1222                 submit_flush_bio(ic, dio);
1223         else
1224                 do_endio(ic, bio);
1225 }
1226
1227 static void dec_in_flight(struct dm_integrity_io *dio)
1228 {
1229         if (atomic_dec_and_test(&dio->in_flight)) {
1230                 struct dm_integrity_c *ic = dio->ic;
1231                 struct bio *bio;
1232
1233                 remove_range(ic, &dio->range);
1234
1235                 if (unlikely(dio->write))
1236                         schedule_autocommit(ic);
1237
1238                 bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1239
1240                 if (unlikely(dio->bi_status) && !bio->bi_status)
1241                         bio->bi_status = dio->bi_status;
1242                 if (likely(!bio->bi_status) && unlikely(bio_sectors(bio) != dio->range.n_sectors)) {
1243                         dio->range.logical_sector += dio->range.n_sectors;
1244                         bio_advance(bio, dio->range.n_sectors << SECTOR_SHIFT);
1245                         INIT_WORK(&dio->work, integrity_bio_wait);
1246                         queue_work(ic->wait_wq, &dio->work);
1247                         return;
1248                 }
1249                 do_endio_flush(ic, dio);
1250         }
1251 }
1252
1253 static void integrity_end_io(struct bio *bio)
1254 {
1255         struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
1256
1257         bio->bi_iter = dio->orig_bi_iter;
1258         bio->bi_disk = dio->orig_bi_disk;
1259         bio->bi_partno = dio->orig_bi_partno;
1260         if (dio->orig_bi_integrity) {
1261                 bio->bi_integrity = dio->orig_bi_integrity;
1262                 bio->bi_opf |= REQ_INTEGRITY;
1263         }
1264         bio->bi_end_io = dio->orig_bi_end_io;
1265
1266         if (dio->completion)
1267                 complete(dio->completion);
1268
1269         dec_in_flight(dio);
1270 }
1271
1272 static void integrity_sector_checksum(struct dm_integrity_c *ic, sector_t sector,
1273                                       const char *data, char *result)
1274 {
1275         __u64 sector_le = cpu_to_le64(sector);
1276         SHASH_DESC_ON_STACK(req, ic->internal_hash);
1277         int r;
1278         unsigned digest_size;
1279
1280         req->tfm = ic->internal_hash;
1281         req->flags = 0;
1282
1283         r = crypto_shash_init(req);
1284         if (unlikely(r < 0)) {
1285                 dm_integrity_io_error(ic, "crypto_shash_init", r);
1286                 goto failed;
1287         }
1288
1289         r = crypto_shash_update(req, (const __u8 *)&sector_le, sizeof sector_le);
1290         if (unlikely(r < 0)) {
1291                 dm_integrity_io_error(ic, "crypto_shash_update", r);
1292                 goto failed;
1293         }
1294
1295         r = crypto_shash_update(req, data, ic->sectors_per_block << SECTOR_SHIFT);
1296         if (unlikely(r < 0)) {
1297                 dm_integrity_io_error(ic, "crypto_shash_update", r);
1298                 goto failed;
1299         }
1300
1301         r = crypto_shash_final(req, result);
1302         if (unlikely(r < 0)) {
1303                 dm_integrity_io_error(ic, "crypto_shash_final", r);
1304                 goto failed;
1305         }
1306
1307         digest_size = crypto_shash_digestsize(ic->internal_hash);
1308         if (unlikely(digest_size < ic->tag_size))
1309                 memset(result + digest_size, 0, ic->tag_size - digest_size);
1310
1311         return;
1312
1313 failed:
1314         /* this shouldn't happen anyway, the hash functions have no reason to fail */
1315         get_random_bytes(result, ic->tag_size);
1316 }
1317
1318 static void integrity_metadata(struct work_struct *w)
1319 {
1320         struct dm_integrity_io *dio = container_of(w, struct dm_integrity_io, work);
1321         struct dm_integrity_c *ic = dio->ic;
1322
1323         int r;
1324
1325         if (ic->internal_hash) {
1326                 struct bvec_iter iter;
1327                 struct bio_vec bv;
1328                 unsigned digest_size = crypto_shash_digestsize(ic->internal_hash);
1329                 struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1330                 char *checksums;
1331                 unsigned extra_space = unlikely(digest_size > ic->tag_size) ? digest_size - ic->tag_size : 0;
1332                 char checksums_onstack[HASH_MAX_DIGESTSIZE];
1333                 unsigned sectors_to_process = dio->range.n_sectors;
1334                 sector_t sector = dio->range.logical_sector;
1335
1336                 if (unlikely(ic->mode == 'R'))
1337                         goto skip_io;
1338
1339                 checksums = kmalloc((PAGE_SIZE >> SECTOR_SHIFT >> ic->sb->log2_sectors_per_block) * ic->tag_size + extra_space,
1340                                     GFP_NOIO | __GFP_NORETRY | __GFP_NOWARN);
1341                 if (!checksums) {
1342                         checksums = checksums_onstack;
1343                         if (WARN_ON(extra_space &&
1344                                     digest_size > sizeof(checksums_onstack))) {
1345                                 r = -EINVAL;
1346                                 goto error;
1347                         }
1348                 }
1349
1350                 __bio_for_each_segment(bv, bio, iter, dio->orig_bi_iter) {
1351                         unsigned pos;
1352                         char *mem, *checksums_ptr;
1353
1354 again:
1355                         mem = (char *)kmap_atomic(bv.bv_page) + bv.bv_offset;
1356                         pos = 0;
1357                         checksums_ptr = checksums;
1358                         do {
1359                                 integrity_sector_checksum(ic, sector, mem + pos, checksums_ptr);
1360                                 checksums_ptr += ic->tag_size;
1361                                 sectors_to_process -= ic->sectors_per_block;
1362                                 pos += ic->sectors_per_block << SECTOR_SHIFT;
1363                                 sector += ic->sectors_per_block;
1364                         } while (pos < bv.bv_len && sectors_to_process && checksums != checksums_onstack);
1365                         kunmap_atomic(mem);
1366
1367                         r = dm_integrity_rw_tag(ic, checksums, &dio->metadata_block, &dio->metadata_offset,
1368                                                 checksums_ptr - checksums, !dio->write ? TAG_CMP : TAG_WRITE);
1369                         if (unlikely(r)) {
1370                                 if (r > 0) {
1371                                         DMERR("Checksum failed at sector 0x%llx",
1372                                               (unsigned long long)(sector - ((r + ic->tag_size - 1) / ic->tag_size)));
1373                                         r = -EILSEQ;
1374                                         atomic64_inc(&ic->number_of_mismatches);
1375                                 }
1376                                 if (likely(checksums != checksums_onstack))
1377                                         kfree(checksums);
1378                                 goto error;
1379                         }
1380
1381                         if (!sectors_to_process)
1382                                 break;
1383
1384                         if (unlikely(pos < bv.bv_len)) {
1385                                 bv.bv_offset += pos;
1386                                 bv.bv_len -= pos;
1387                                 goto again;
1388                         }
1389                 }
1390
1391                 if (likely(checksums != checksums_onstack))
1392                         kfree(checksums);
1393         } else {
1394                 struct bio_integrity_payload *bip = dio->orig_bi_integrity;
1395
1396                 if (bip) {
1397                         struct bio_vec biv;
1398                         struct bvec_iter iter;
1399                         unsigned data_to_process = dio->range.n_sectors;
1400                         sector_to_block(ic, data_to_process);
1401                         data_to_process *= ic->tag_size;
1402
1403                         bip_for_each_vec(biv, bip, iter) {
1404                                 unsigned char *tag;
1405                                 unsigned this_len;
1406
1407                                 BUG_ON(PageHighMem(biv.bv_page));
1408                                 tag = lowmem_page_address(biv.bv_page) + biv.bv_offset;
1409                                 this_len = min(biv.bv_len, data_to_process);
1410                                 r = dm_integrity_rw_tag(ic, tag, &dio->metadata_block, &dio->metadata_offset,
1411                                                         this_len, !dio->write ? TAG_READ : TAG_WRITE);
1412                                 if (unlikely(r))
1413                                         goto error;
1414                                 data_to_process -= this_len;
1415                                 if (!data_to_process)
1416                                         break;
1417                         }
1418                 }
1419         }
1420 skip_io:
1421         dec_in_flight(dio);
1422         return;
1423 error:
1424         dio->bi_status = errno_to_blk_status(r);
1425         dec_in_flight(dio);
1426 }
1427
1428 static int dm_integrity_map(struct dm_target *ti, struct bio *bio)
1429 {
1430         struct dm_integrity_c *ic = ti->private;
1431         struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
1432         struct bio_integrity_payload *bip;
1433
1434         sector_t area, offset;
1435
1436         dio->ic = ic;
1437         dio->bi_status = 0;
1438
1439         if (unlikely(bio->bi_opf & REQ_PREFLUSH)) {
1440                 submit_flush_bio(ic, dio);
1441                 return DM_MAPIO_SUBMITTED;
1442         }
1443
1444         dio->range.logical_sector = dm_target_offset(ti, bio->bi_iter.bi_sector);
1445         dio->write = bio_op(bio) == REQ_OP_WRITE;
1446         dio->fua = dio->write && bio->bi_opf & REQ_FUA;
1447         if (unlikely(dio->fua)) {
1448                 /*
1449                  * Don't pass down the FUA flag because we have to flush
1450                  * disk cache anyway.
1451                  */
1452                 bio->bi_opf &= ~REQ_FUA;
1453         }
1454         if (unlikely(dio->range.logical_sector + bio_sectors(bio) > ic->provided_data_sectors)) {
1455                 DMERR("Too big sector number: 0x%llx + 0x%x > 0x%llx",
1456                       (unsigned long long)dio->range.logical_sector, bio_sectors(bio),
1457                       (unsigned long long)ic->provided_data_sectors);
1458                 return DM_MAPIO_KILL;
1459         }
1460         if (unlikely((dio->range.logical_sector | bio_sectors(bio)) & (unsigned)(ic->sectors_per_block - 1))) {
1461                 DMERR("Bio not aligned on %u sectors: 0x%llx, 0x%x",
1462                       ic->sectors_per_block,
1463                       (unsigned long long)dio->range.logical_sector, bio_sectors(bio));
1464                 return DM_MAPIO_KILL;
1465         }
1466
1467         if (ic->sectors_per_block > 1) {
1468                 struct bvec_iter iter;
1469                 struct bio_vec bv;
1470                 bio_for_each_segment(bv, bio, iter) {
1471                         if (unlikely(bv.bv_len & ((ic->sectors_per_block << SECTOR_SHIFT) - 1))) {
1472                                 DMERR("Bio vector (%u,%u) is not aligned on %u-sector boundary",
1473                                         bv.bv_offset, bv.bv_len, ic->sectors_per_block);
1474                                 return DM_MAPIO_KILL;
1475                         }
1476                 }
1477         }
1478
1479         bip = bio_integrity(bio);
1480         if (!ic->internal_hash) {
1481                 if (bip) {
1482                         unsigned wanted_tag_size = bio_sectors(bio) >> ic->sb->log2_sectors_per_block;
1483                         if (ic->log2_tag_size >= 0)
1484                                 wanted_tag_size <<= ic->log2_tag_size;
1485                         else
1486                                 wanted_tag_size *= ic->tag_size;
1487                         if (unlikely(wanted_tag_size != bip->bip_iter.bi_size)) {
1488                                 DMERR("Invalid integrity data size %u, expected %u", bip->bip_iter.bi_size, wanted_tag_size);
1489                                 return DM_MAPIO_KILL;
1490                         }
1491                 }
1492         } else {
1493                 if (unlikely(bip != NULL)) {
1494                         DMERR("Unexpected integrity data when using internal hash");
1495                         return DM_MAPIO_KILL;
1496                 }
1497         }
1498
1499         if (unlikely(ic->mode == 'R') && unlikely(dio->write))
1500                 return DM_MAPIO_KILL;
1501
1502         get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
1503         dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset, &dio->metadata_offset);
1504         bio->bi_iter.bi_sector = get_data_sector(ic, area, offset);
1505
1506         dm_integrity_map_continue(dio, true);
1507         return DM_MAPIO_SUBMITTED;
1508 }
1509
1510 static bool __journal_read_write(struct dm_integrity_io *dio, struct bio *bio,
1511                                  unsigned journal_section, unsigned journal_entry)
1512 {
1513         struct dm_integrity_c *ic = dio->ic;
1514         sector_t logical_sector;
1515         unsigned n_sectors;
1516
1517         logical_sector = dio->range.logical_sector;
1518         n_sectors = dio->range.n_sectors;
1519         do {
1520                 struct bio_vec bv = bio_iovec(bio);
1521                 char *mem;
1522
1523                 if (unlikely(bv.bv_len >> SECTOR_SHIFT > n_sectors))
1524                         bv.bv_len = n_sectors << SECTOR_SHIFT;
1525                 n_sectors -= bv.bv_len >> SECTOR_SHIFT;
1526                 bio_advance_iter(bio, &bio->bi_iter, bv.bv_len);
1527 retry_kmap:
1528                 mem = kmap_atomic(bv.bv_page);
1529                 if (likely(dio->write))
1530                         flush_dcache_page(bv.bv_page);
1531
1532                 do {
1533                         struct journal_entry *je = access_journal_entry(ic, journal_section, journal_entry);
1534
1535                         if (unlikely(!dio->write)) {
1536                                 struct journal_sector *js;
1537                                 char *mem_ptr;
1538                                 unsigned s;
1539
1540                                 if (unlikely(journal_entry_is_inprogress(je))) {
1541                                         flush_dcache_page(bv.bv_page);
1542                                         kunmap_atomic(mem);
1543
1544                                         __io_wait_event(ic->copy_to_journal_wait, !journal_entry_is_inprogress(je));
1545                                         goto retry_kmap;
1546                                 }
1547                                 smp_rmb();
1548                                 BUG_ON(journal_entry_get_sector(je) != logical_sector);
1549                                 js = access_journal_data(ic, journal_section, journal_entry);
1550                                 mem_ptr = mem + bv.bv_offset;
1551                                 s = 0;
1552                                 do {
1553                                         memcpy(mem_ptr, js, JOURNAL_SECTOR_DATA);
1554                                         *(commit_id_t *)(mem_ptr + JOURNAL_SECTOR_DATA) = je->last_bytes[s];
1555                                         js++;
1556                                         mem_ptr += 1 << SECTOR_SHIFT;
1557                                 } while (++s < ic->sectors_per_block);
1558 #ifdef INTERNAL_VERIFY
1559                                 if (ic->internal_hash) {
1560                                         char checksums_onstack[max(HASH_MAX_DIGESTSIZE, MAX_TAG_SIZE)];
1561
1562                                         integrity_sector_checksum(ic, logical_sector, mem + bv.bv_offset, checksums_onstack);
1563                                         if (unlikely(memcmp(checksums_onstack, journal_entry_tag(ic, je), ic->tag_size))) {
1564                                                 DMERR("Checksum failed when reading from journal, at sector 0x%llx",
1565                                                       (unsigned long long)logical_sector);
1566                                         }
1567                                 }
1568 #endif
1569                         }
1570
1571                         if (!ic->internal_hash) {
1572                                 struct bio_integrity_payload *bip = bio_integrity(bio);
1573                                 unsigned tag_todo = ic->tag_size;
1574                                 char *tag_ptr = journal_entry_tag(ic, je);
1575
1576                                 if (bip) do {
1577                                         struct bio_vec biv = bvec_iter_bvec(bip->bip_vec, bip->bip_iter);
1578                                         unsigned tag_now = min(biv.bv_len, tag_todo);
1579                                         char *tag_addr;
1580                                         BUG_ON(PageHighMem(biv.bv_page));
1581                                         tag_addr = lowmem_page_address(biv.bv_page) + biv.bv_offset;
1582                                         if (likely(dio->write))
1583                                                 memcpy(tag_ptr, tag_addr, tag_now);
1584                                         else
1585                                                 memcpy(tag_addr, tag_ptr, tag_now);
1586                                         bvec_iter_advance(bip->bip_vec, &bip->bip_iter, tag_now);
1587                                         tag_ptr += tag_now;
1588                                         tag_todo -= tag_now;
1589                                 } while (unlikely(tag_todo)); else {
1590                                         if (likely(dio->write))
1591                                                 memset(tag_ptr, 0, tag_todo);
1592                                 }
1593                         }
1594
1595                         if (likely(dio->write)) {
1596                                 struct journal_sector *js;
1597                                 unsigned s;
1598
1599                                 js = access_journal_data(ic, journal_section, journal_entry);
1600                                 memcpy(js, mem + bv.bv_offset, ic->sectors_per_block << SECTOR_SHIFT);
1601
1602                                 s = 0;
1603                                 do {
1604                                         je->last_bytes[s] = js[s].commit_id;
1605                                 } while (++s < ic->sectors_per_block);
1606
1607                                 if (ic->internal_hash) {
1608                                         unsigned digest_size = crypto_shash_digestsize(ic->internal_hash);
1609                                         if (unlikely(digest_size > ic->tag_size)) {
1610                                                 char checksums_onstack[HASH_MAX_DIGESTSIZE];
1611                                                 integrity_sector_checksum(ic, logical_sector, (char *)js, checksums_onstack);
1612                                                 memcpy(journal_entry_tag(ic, je), checksums_onstack, ic->tag_size);
1613                                         } else
1614                                                 integrity_sector_checksum(ic, logical_sector, (char *)js, journal_entry_tag(ic, je));
1615                                 }
1616
1617                                 journal_entry_set_sector(je, logical_sector);
1618                         }
1619                         logical_sector += ic->sectors_per_block;
1620
1621                         journal_entry++;
1622                         if (unlikely(journal_entry == ic->journal_section_entries)) {
1623                                 journal_entry = 0;
1624                                 journal_section++;
1625                                 wraparound_section(ic, &journal_section);
1626                         }
1627
1628                         bv.bv_offset += ic->sectors_per_block << SECTOR_SHIFT;
1629                 } while (bv.bv_len -= ic->sectors_per_block << SECTOR_SHIFT);
1630
1631                 if (unlikely(!dio->write))
1632                         flush_dcache_page(bv.bv_page);
1633                 kunmap_atomic(mem);
1634         } while (n_sectors);
1635
1636         if (likely(dio->write)) {
1637                 smp_mb();
1638                 if (unlikely(waitqueue_active(&ic->copy_to_journal_wait)))
1639                         wake_up(&ic->copy_to_journal_wait);
1640                 if (READ_ONCE(ic->free_sectors) <= ic->free_sectors_threshold) {
1641                         queue_work(ic->commit_wq, &ic->commit_work);
1642                 } else {
1643                         schedule_autocommit(ic);
1644                 }
1645         } else {
1646                 remove_range(ic, &dio->range);
1647         }
1648
1649         if (unlikely(bio->bi_iter.bi_size)) {
1650                 sector_t area, offset;
1651
1652                 dio->range.logical_sector = logical_sector;
1653                 get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
1654                 dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset, &dio->metadata_offset);
1655                 return true;
1656         }
1657
1658         return false;
1659 }
1660
1661 static void dm_integrity_map_continue(struct dm_integrity_io *dio, bool from_map)
1662 {
1663         struct dm_integrity_c *ic = dio->ic;
1664         struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1665         unsigned journal_section, journal_entry;
1666         unsigned journal_read_pos;
1667         struct completion read_comp;
1668         bool need_sync_io = ic->internal_hash && !dio->write;
1669
1670         if (need_sync_io && from_map) {
1671                 INIT_WORK(&dio->work, integrity_bio_wait);
1672                 queue_work(ic->metadata_wq, &dio->work);
1673                 return;
1674         }
1675
1676 lock_retry:
1677         spin_lock_irq(&ic->endio_wait.lock);
1678 retry:
1679         if (unlikely(dm_integrity_failed(ic))) {
1680                 spin_unlock_irq(&ic->endio_wait.lock);
1681                 do_endio(ic, bio);
1682                 return;
1683         }
1684         dio->range.n_sectors = bio_sectors(bio);
1685         journal_read_pos = NOT_FOUND;
1686         if (likely(ic->mode == 'J')) {
1687                 if (dio->write) {
1688                         unsigned next_entry, i, pos;
1689                         unsigned ws, we, range_sectors;
1690
1691                         dio->range.n_sectors = min(dio->range.n_sectors,
1692                                                    ic->free_sectors << ic->sb->log2_sectors_per_block);
1693                         if (unlikely(!dio->range.n_sectors)) {
1694                                 if (from_map)
1695                                         goto offload_to_thread;
1696                                 sleep_on_endio_wait(ic);
1697                                 goto retry;
1698                         }
1699                         range_sectors = dio->range.n_sectors >> ic->sb->log2_sectors_per_block;
1700                         ic->free_sectors -= range_sectors;
1701                         journal_section = ic->free_section;
1702                         journal_entry = ic->free_section_entry;
1703
1704                         next_entry = ic->free_section_entry + range_sectors;
1705                         ic->free_section_entry = next_entry % ic->journal_section_entries;
1706                         ic->free_section += next_entry / ic->journal_section_entries;
1707                         ic->n_uncommitted_sections += next_entry / ic->journal_section_entries;
1708                         wraparound_section(ic, &ic->free_section);
1709
1710                         pos = journal_section * ic->journal_section_entries + journal_entry;
1711                         ws = journal_section;
1712                         we = journal_entry;
1713                         i = 0;
1714                         do {
1715                                 struct journal_entry *je;
1716
1717                                 add_journal_node(ic, &ic->journal_tree[pos], dio->range.logical_sector + i);
1718                                 pos++;
1719                                 if (unlikely(pos >= ic->journal_entries))
1720                                         pos = 0;
1721
1722                                 je = access_journal_entry(ic, ws, we);
1723                                 BUG_ON(!journal_entry_is_unused(je));
1724                                 journal_entry_set_inprogress(je);
1725                                 we++;
1726                                 if (unlikely(we == ic->journal_section_entries)) {
1727                                         we = 0;
1728                                         ws++;
1729                                         wraparound_section(ic, &ws);
1730                                 }
1731                         } while ((i += ic->sectors_per_block) < dio->range.n_sectors);
1732
1733                         spin_unlock_irq(&ic->endio_wait.lock);
1734                         goto journal_read_write;
1735                 } else {
1736                         sector_t next_sector;
1737                         journal_read_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
1738                         if (likely(journal_read_pos == NOT_FOUND)) {
1739                                 if (unlikely(dio->range.n_sectors > next_sector - dio->range.logical_sector))
1740                                         dio->range.n_sectors = next_sector - dio->range.logical_sector;
1741                         } else {
1742                                 unsigned i;
1743                                 unsigned jp = journal_read_pos + 1;
1744                                 for (i = ic->sectors_per_block; i < dio->range.n_sectors; i += ic->sectors_per_block, jp++) {
1745                                         if (!test_journal_node(ic, jp, dio->range.logical_sector + i))
1746                                                 break;
1747                                 }
1748                                 dio->range.n_sectors = i;
1749                         }
1750                 }
1751         }
1752         if (unlikely(!add_new_range(ic, &dio->range, true))) {
1753                 /*
1754                  * We must not sleep in the request routine because it could
1755                  * stall bios on current->bio_list.
1756                  * So, we offload the bio to a workqueue if we have to sleep.
1757                  */
1758                 if (from_map) {
1759 offload_to_thread:
1760                         spin_unlock_irq(&ic->endio_wait.lock);
1761                         INIT_WORK(&dio->work, integrity_bio_wait);
1762                         queue_work(ic->wait_wq, &dio->work);
1763                         return;
1764                 }
1765                 wait_and_add_new_range(ic, &dio->range);
1766         }
1767         spin_unlock_irq(&ic->endio_wait.lock);
1768
1769         if (unlikely(journal_read_pos != NOT_FOUND)) {
1770                 journal_section = journal_read_pos / ic->journal_section_entries;
1771                 journal_entry = journal_read_pos % ic->journal_section_entries;
1772                 goto journal_read_write;
1773         }
1774
1775         dio->in_flight = (atomic_t)ATOMIC_INIT(2);
1776
1777         if (need_sync_io) {
1778                 init_completion(&read_comp);
1779                 dio->completion = &read_comp;
1780         } else
1781                 dio->completion = NULL;
1782
1783         dio->orig_bi_iter = bio->bi_iter;
1784
1785         dio->orig_bi_disk = bio->bi_disk;
1786         dio->orig_bi_partno = bio->bi_partno;
1787         bio_set_dev(bio, ic->dev->bdev);
1788
1789         dio->orig_bi_integrity = bio_integrity(bio);
1790         bio->bi_integrity = NULL;
1791         bio->bi_opf &= ~REQ_INTEGRITY;
1792
1793         dio->orig_bi_end_io = bio->bi_end_io;
1794         bio->bi_end_io = integrity_end_io;
1795
1796         bio->bi_iter.bi_size = dio->range.n_sectors << SECTOR_SHIFT;
1797         generic_make_request(bio);
1798
1799         if (need_sync_io) {
1800                 wait_for_completion_io(&read_comp);
1801                 if (unlikely(ic->recalc_wq != NULL) &&
1802                     ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING) &&
1803                     dio->range.logical_sector + dio->range.n_sectors > le64_to_cpu(ic->sb->recalc_sector))
1804                         goto skip_check;
1805                 if (likely(!bio->bi_status))
1806                         integrity_metadata(&dio->work);
1807                 else
1808 skip_check:
1809                         dec_in_flight(dio);
1810
1811         } else {
1812                 INIT_WORK(&dio->work, integrity_metadata);
1813                 queue_work(ic->metadata_wq, &dio->work);
1814         }
1815
1816         return;
1817
1818 journal_read_write:
1819         if (unlikely(__journal_read_write(dio, bio, journal_section, journal_entry)))
1820                 goto lock_retry;
1821
1822         do_endio_flush(ic, dio);
1823 }
1824
1825
1826 static void integrity_bio_wait(struct work_struct *w)
1827 {
1828         struct dm_integrity_io *dio = container_of(w, struct dm_integrity_io, work);
1829
1830         dm_integrity_map_continue(dio, false);
1831 }
1832
1833 static void pad_uncommitted(struct dm_integrity_c *ic)
1834 {
1835         if (ic->free_section_entry) {
1836                 ic->free_sectors -= ic->journal_section_entries - ic->free_section_entry;
1837                 ic->free_section_entry = 0;
1838                 ic->free_section++;
1839                 wraparound_section(ic, &ic->free_section);
1840                 ic->n_uncommitted_sections++;
1841         }
1842         WARN_ON(ic->journal_sections * ic->journal_section_entries !=
1843                 (ic->n_uncommitted_sections + ic->n_committed_sections) * ic->journal_section_entries + ic->free_sectors);
1844 }
1845
1846 static void integrity_commit(struct work_struct *w)
1847 {
1848         struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, commit_work);
1849         unsigned commit_start, commit_sections;
1850         unsigned i, j, n;
1851         struct bio *flushes;
1852
1853         del_timer(&ic->autocommit_timer);
1854
1855         spin_lock_irq(&ic->endio_wait.lock);
1856         flushes = bio_list_get(&ic->flush_bio_list);
1857         if (unlikely(ic->mode != 'J')) {
1858                 spin_unlock_irq(&ic->endio_wait.lock);
1859                 dm_integrity_flush_buffers(ic);
1860                 goto release_flush_bios;
1861         }
1862
1863         pad_uncommitted(ic);
1864         commit_start = ic->uncommitted_section;
1865         commit_sections = ic->n_uncommitted_sections;
1866         spin_unlock_irq(&ic->endio_wait.lock);
1867
1868         if (!commit_sections)
1869                 goto release_flush_bios;
1870
1871         i = commit_start;
1872         for (n = 0; n < commit_sections; n++) {
1873                 for (j = 0; j < ic->journal_section_entries; j++) {
1874                         struct journal_entry *je;
1875                         je = access_journal_entry(ic, i, j);
1876                         io_wait_event(ic->copy_to_journal_wait, !journal_entry_is_inprogress(je));
1877                 }
1878                 for (j = 0; j < ic->journal_section_sectors; j++) {
1879                         struct journal_sector *js;
1880                         js = access_journal(ic, i, j);
1881                         js->commit_id = dm_integrity_commit_id(ic, i, j, ic->commit_seq);
1882                 }
1883                 i++;
1884                 if (unlikely(i >= ic->journal_sections))
1885                         ic->commit_seq = next_commit_seq(ic->commit_seq);
1886                 wraparound_section(ic, &i);
1887         }
1888         smp_rmb();
1889
1890         write_journal(ic, commit_start, commit_sections);
1891
1892         spin_lock_irq(&ic->endio_wait.lock);
1893         ic->uncommitted_section += commit_sections;
1894         wraparound_section(ic, &ic->uncommitted_section);
1895         ic->n_uncommitted_sections -= commit_sections;
1896         ic->n_committed_sections += commit_sections;
1897         spin_unlock_irq(&ic->endio_wait.lock);
1898
1899         if (READ_ONCE(ic->free_sectors) <= ic->free_sectors_threshold)
1900                 queue_work(ic->writer_wq, &ic->writer_work);
1901
1902 release_flush_bios:
1903         while (flushes) {
1904                 struct bio *next = flushes->bi_next;
1905                 flushes->bi_next = NULL;
1906                 do_endio(ic, flushes);
1907                 flushes = next;
1908         }
1909 }
1910
1911 static void complete_copy_from_journal(unsigned long error, void *context)
1912 {
1913         struct journal_io *io = context;
1914         struct journal_completion *comp = io->comp;
1915         struct dm_integrity_c *ic = comp->ic;
1916         remove_range(ic, &io->range);
1917         mempool_free(io, &ic->journal_io_mempool);
1918         if (unlikely(error != 0))
1919                 dm_integrity_io_error(ic, "copying from journal", -EIO);
1920         complete_journal_op(comp);
1921 }
1922
1923 static void restore_last_bytes(struct dm_integrity_c *ic, struct journal_sector *js,
1924                                struct journal_entry *je)
1925 {
1926         unsigned s = 0;
1927         do {
1928                 js->commit_id = je->last_bytes[s];
1929                 js++;
1930         } while (++s < ic->sectors_per_block);
1931 }
1932
1933 static void do_journal_write(struct dm_integrity_c *ic, unsigned write_start,
1934                              unsigned write_sections, bool from_replay)
1935 {
1936         unsigned i, j, n;
1937         struct journal_completion comp;
1938         struct blk_plug plug;
1939
1940         blk_start_plug(&plug);
1941
1942         comp.ic = ic;
1943         comp.in_flight = (atomic_t)ATOMIC_INIT(1);
1944         init_completion(&comp.comp);
1945
1946         i = write_start;
1947         for (n = 0; n < write_sections; n++, i++, wraparound_section(ic, &i)) {
1948 #ifndef INTERNAL_VERIFY
1949                 if (unlikely(from_replay))
1950 #endif
1951                         rw_section_mac(ic, i, false);
1952                 for (j = 0; j < ic->journal_section_entries; j++) {
1953                         struct journal_entry *je = access_journal_entry(ic, i, j);
1954                         sector_t sec, area, offset;
1955                         unsigned k, l, next_loop;
1956                         sector_t metadata_block;
1957                         unsigned metadata_offset;
1958                         struct journal_io *io;
1959
1960                         if (journal_entry_is_unused(je))
1961                                 continue;
1962                         BUG_ON(unlikely(journal_entry_is_inprogress(je)) && !from_replay);
1963                         sec = journal_entry_get_sector(je);
1964                         if (unlikely(from_replay)) {
1965                                 if (unlikely(sec & (unsigned)(ic->sectors_per_block - 1))) {
1966                                         dm_integrity_io_error(ic, "invalid sector in journal", -EIO);
1967                                         sec &= ~(sector_t)(ic->sectors_per_block - 1);
1968                                 }
1969                         }
1970                         get_area_and_offset(ic, sec, &area, &offset);
1971                         restore_last_bytes(ic, access_journal_data(ic, i, j), je);
1972                         for (k = j + 1; k < ic->journal_section_entries; k++) {
1973                                 struct journal_entry *je2 = access_journal_entry(ic, i, k);
1974                                 sector_t sec2, area2, offset2;
1975                                 if (journal_entry_is_unused(je2))
1976                                         break;
1977                                 BUG_ON(unlikely(journal_entry_is_inprogress(je2)) && !from_replay);
1978                                 sec2 = journal_entry_get_sector(je2);
1979                                 get_area_and_offset(ic, sec2, &area2, &offset2);
1980                                 if (area2 != area || offset2 != offset + ((k - j) << ic->sb->log2_sectors_per_block))
1981                                         break;
1982                                 restore_last_bytes(ic, access_journal_data(ic, i, k), je2);
1983                         }
1984                         next_loop = k - 1;
1985
1986                         io = mempool_alloc(&ic->journal_io_mempool, GFP_NOIO);
1987                         io->comp = &comp;
1988                         io->range.logical_sector = sec;
1989                         io->range.n_sectors = (k - j) << ic->sb->log2_sectors_per_block;
1990
1991                         spin_lock_irq(&ic->endio_wait.lock);
1992                         if (unlikely(!add_new_range(ic, &io->range, true)))
1993                                 wait_and_add_new_range(ic, &io->range);
1994
1995                         if (likely(!from_replay)) {
1996                                 struct journal_node *section_node = &ic->journal_tree[i * ic->journal_section_entries];
1997
1998                                 /* don't write if there is newer committed sector */
1999                                 while (j < k && find_newer_committed_node(ic, &section_node[j])) {
2000                                         struct journal_entry *je2 = access_journal_entry(ic, i, j);
2001
2002                                         journal_entry_set_unused(je2);
2003                                         remove_journal_node(ic, &section_node[j]);
2004                                         j++;
2005                                         sec += ic->sectors_per_block;
2006                                         offset += ic->sectors_per_block;
2007                                 }
2008                                 while (j < k && find_newer_committed_node(ic, &section_node[k - 1])) {
2009                                         struct journal_entry *je2 = access_journal_entry(ic, i, k - 1);
2010
2011                                         journal_entry_set_unused(je2);
2012                                         remove_journal_node(ic, &section_node[k - 1]);
2013                                         k--;
2014                                 }
2015                                 if (j == k) {
2016                                         remove_range_unlocked(ic, &io->range);
2017                                         spin_unlock_irq(&ic->endio_wait.lock);
2018                                         mempool_free(io, &ic->journal_io_mempool);
2019                                         goto skip_io;
2020                                 }
2021                                 for (l = j; l < k; l++) {
2022                                         remove_journal_node(ic, &section_node[l]);
2023                                 }
2024                         }
2025                         spin_unlock_irq(&ic->endio_wait.lock);
2026
2027                         metadata_block = get_metadata_sector_and_offset(ic, area, offset, &metadata_offset);
2028                         for (l = j; l < k; l++) {
2029                                 int r;
2030                                 struct journal_entry *je2 = access_journal_entry(ic, i, l);
2031
2032                                 if (
2033 #ifndef INTERNAL_VERIFY
2034                                     unlikely(from_replay) &&
2035 #endif
2036                                     ic->internal_hash) {
2037                                         char test_tag[max_t(size_t, HASH_MAX_DIGESTSIZE, MAX_TAG_SIZE)];
2038
2039                                         integrity_sector_checksum(ic, sec + ((l - j) << ic->sb->log2_sectors_per_block),
2040                                                                   (char *)access_journal_data(ic, i, l), test_tag);
2041                                         if (unlikely(memcmp(test_tag, journal_entry_tag(ic, je2), ic->tag_size)))
2042                                                 dm_integrity_io_error(ic, "tag mismatch when replaying journal", -EILSEQ);
2043                                 }
2044
2045                                 journal_entry_set_unused(je2);
2046                                 r = dm_integrity_rw_tag(ic, journal_entry_tag(ic, je2), &metadata_block, &metadata_offset,
2047                                                         ic->tag_size, TAG_WRITE);
2048                                 if (unlikely(r)) {
2049                                         dm_integrity_io_error(ic, "reading tags", r);
2050                                 }
2051                         }
2052
2053                         atomic_inc(&comp.in_flight);
2054                         copy_from_journal(ic, i, j << ic->sb->log2_sectors_per_block,
2055                                           (k - j) << ic->sb->log2_sectors_per_block,
2056                                           get_data_sector(ic, area, offset),
2057                                           complete_copy_from_journal, io);
2058 skip_io:
2059                         j = next_loop;
2060                 }
2061         }
2062
2063         dm_bufio_write_dirty_buffers_async(ic->bufio);
2064
2065         blk_finish_plug(&plug);
2066
2067         complete_journal_op(&comp);
2068         wait_for_completion_io(&comp.comp);
2069
2070         dm_integrity_flush_buffers(ic);
2071 }
2072
2073 static void integrity_writer(struct work_struct *w)
2074 {
2075         struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, writer_work);
2076         unsigned write_start, write_sections;
2077
2078         unsigned prev_free_sectors;
2079
2080         /* the following test is not needed, but it tests the replay code */
2081         if (READ_ONCE(ic->suspending) && !ic->meta_dev)
2082                 return;
2083
2084         spin_lock_irq(&ic->endio_wait.lock);
2085         write_start = ic->committed_section;
2086         write_sections = ic->n_committed_sections;
2087         spin_unlock_irq(&ic->endio_wait.lock);
2088
2089         if (!write_sections)
2090                 return;
2091
2092         do_journal_write(ic, write_start, write_sections, false);
2093
2094         spin_lock_irq(&ic->endio_wait.lock);
2095
2096         ic->committed_section += write_sections;
2097         wraparound_section(ic, &ic->committed_section);
2098         ic->n_committed_sections -= write_sections;
2099
2100         prev_free_sectors = ic->free_sectors;
2101         ic->free_sectors += write_sections * ic->journal_section_entries;
2102         if (unlikely(!prev_free_sectors))
2103                 wake_up_locked(&ic->endio_wait);
2104
2105         spin_unlock_irq(&ic->endio_wait.lock);
2106 }
2107
2108 static void recalc_write_super(struct dm_integrity_c *ic)
2109 {
2110         int r;
2111
2112         dm_integrity_flush_buffers(ic);
2113         if (dm_integrity_failed(ic))
2114                 return;
2115
2116         sb_set_version(ic);
2117         r = sync_rw_sb(ic, REQ_OP_WRITE, 0);
2118         if (unlikely(r))
2119                 dm_integrity_io_error(ic, "writing superblock", r);
2120 }
2121
2122 static void integrity_recalc(struct work_struct *w)
2123 {
2124         struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, recalc_work);
2125         struct dm_integrity_range range;
2126         struct dm_io_request io_req;
2127         struct dm_io_region io_loc;
2128         sector_t area, offset;
2129         sector_t metadata_block;
2130         unsigned metadata_offset;
2131         __u8 *t;
2132         unsigned i;
2133         int r;
2134         unsigned super_counter = 0;
2135
2136         spin_lock_irq(&ic->endio_wait.lock);
2137
2138 next_chunk:
2139
2140         if (unlikely(READ_ONCE(ic->suspending)))
2141                 goto unlock_ret;
2142
2143         range.logical_sector = le64_to_cpu(ic->sb->recalc_sector);
2144         if (unlikely(range.logical_sector >= ic->provided_data_sectors))
2145                 goto unlock_ret;
2146
2147         get_area_and_offset(ic, range.logical_sector, &area, &offset);
2148         range.n_sectors = min((sector_t)RECALC_SECTORS, ic->provided_data_sectors - range.logical_sector);
2149         if (!ic->meta_dev)
2150                 range.n_sectors = min(range.n_sectors, (1U << ic->sb->log2_interleave_sectors) - (unsigned)offset);
2151
2152         if (unlikely(!add_new_range(ic, &range, true)))
2153                 wait_and_add_new_range(ic, &range);
2154
2155         spin_unlock_irq(&ic->endio_wait.lock);
2156
2157         if (unlikely(++super_counter == RECALC_WRITE_SUPER)) {
2158                 recalc_write_super(ic);
2159                 super_counter = 0;
2160         }
2161
2162         if (unlikely(dm_integrity_failed(ic)))
2163                 goto err;
2164
2165         io_req.bi_op = REQ_OP_READ;
2166         io_req.bi_op_flags = 0;
2167         io_req.mem.type = DM_IO_VMA;
2168         io_req.mem.ptr.addr = ic->recalc_buffer;
2169         io_req.notify.fn = NULL;
2170         io_req.client = ic->io;
2171         io_loc.bdev = ic->dev->bdev;
2172         io_loc.sector = get_data_sector(ic, area, offset);
2173         io_loc.count = range.n_sectors;
2174
2175         r = dm_io(&io_req, 1, &io_loc, NULL);
2176         if (unlikely(r)) {
2177                 dm_integrity_io_error(ic, "reading data", r);
2178                 goto err;
2179         }
2180
2181         t = ic->recalc_tags;
2182         for (i = 0; i < range.n_sectors; i += ic->sectors_per_block) {
2183                 integrity_sector_checksum(ic, range.logical_sector + i, ic->recalc_buffer + (i << SECTOR_SHIFT), t);
2184                 t += ic->tag_size;
2185         }
2186
2187         metadata_block = get_metadata_sector_and_offset(ic, area, offset, &metadata_offset);
2188
2189         r = dm_integrity_rw_tag(ic, ic->recalc_tags, &metadata_block, &metadata_offset, t - ic->recalc_tags, TAG_WRITE);
2190         if (unlikely(r)) {
2191                 dm_integrity_io_error(ic, "writing tags", r);
2192                 goto err;
2193         }
2194
2195         spin_lock_irq(&ic->endio_wait.lock);
2196         remove_range_unlocked(ic, &range);
2197         ic->sb->recalc_sector = cpu_to_le64(range.logical_sector + range.n_sectors);
2198         goto next_chunk;
2199
2200 err:
2201         remove_range(ic, &range);
2202         return;
2203
2204 unlock_ret:
2205         spin_unlock_irq(&ic->endio_wait.lock);
2206
2207         recalc_write_super(ic);
2208 }
2209
2210 static void init_journal(struct dm_integrity_c *ic, unsigned start_section,
2211                          unsigned n_sections, unsigned char commit_seq)
2212 {
2213         unsigned i, j, n;
2214
2215         if (!n_sections)
2216                 return;
2217
2218         for (n = 0; n < n_sections; n++) {
2219                 i = start_section + n;
2220                 wraparound_section(ic, &i);
2221                 for (j = 0; j < ic->journal_section_sectors; j++) {
2222                         struct journal_sector *js = access_journal(ic, i, j);
2223                         memset(&js->entries, 0, JOURNAL_SECTOR_DATA);
2224                         js->commit_id = dm_integrity_commit_id(ic, i, j, commit_seq);
2225                 }
2226                 for (j = 0; j < ic->journal_section_entries; j++) {
2227                         struct journal_entry *je = access_journal_entry(ic, i, j);
2228                         journal_entry_set_unused(je);
2229                 }
2230         }
2231
2232         write_journal(ic, start_section, n_sections);
2233 }
2234
2235 static int find_commit_seq(struct dm_integrity_c *ic, unsigned i, unsigned j, commit_id_t id)
2236 {
2237         unsigned char k;
2238         for (k = 0; k < N_COMMIT_IDS; k++) {
2239                 if (dm_integrity_commit_id(ic, i, j, k) == id)
2240                         return k;
2241         }
2242         dm_integrity_io_error(ic, "journal commit id", -EIO);
2243         return -EIO;
2244 }
2245
2246 static void replay_journal(struct dm_integrity_c *ic)
2247 {
2248         unsigned i, j;
2249         bool used_commit_ids[N_COMMIT_IDS];
2250         unsigned max_commit_id_sections[N_COMMIT_IDS];
2251         unsigned write_start, write_sections;
2252         unsigned continue_section;
2253         bool journal_empty;
2254         unsigned char unused, last_used, want_commit_seq;
2255
2256         if (ic->mode == 'R')
2257                 return;
2258
2259         if (ic->journal_uptodate)
2260                 return;
2261
2262         last_used = 0;
2263         write_start = 0;
2264
2265         if (!ic->just_formatted) {
2266                 DEBUG_print("reading journal\n");
2267                 rw_journal(ic, REQ_OP_READ, 0, 0, ic->journal_sections, NULL);
2268                 if (ic->journal_io)
2269                         DEBUG_bytes(lowmem_page_address(ic->journal_io[0].page), 64, "read journal");
2270                 if (ic->journal_io) {
2271                         struct journal_completion crypt_comp;
2272                         crypt_comp.ic = ic;
2273                         init_completion(&crypt_comp.comp);
2274                         crypt_comp.in_flight = (atomic_t)ATOMIC_INIT(0);
2275                         encrypt_journal(ic, false, 0, ic->journal_sections, &crypt_comp);
2276                         wait_for_completion(&crypt_comp.comp);
2277                 }
2278                 DEBUG_bytes(lowmem_page_address(ic->journal[0].page), 64, "decrypted journal");
2279         }
2280
2281         if (dm_integrity_failed(ic))
2282                 goto clear_journal;
2283
2284         journal_empty = true;
2285         memset(used_commit_ids, 0, sizeof used_commit_ids);
2286         memset(max_commit_id_sections, 0, sizeof max_commit_id_sections);
2287         for (i = 0; i < ic->journal_sections; i++) {
2288                 for (j = 0; j < ic->journal_section_sectors; j++) {
2289                         int k;
2290                         struct journal_sector *js = access_journal(ic, i, j);
2291                         k = find_commit_seq(ic, i, j, js->commit_id);
2292                         if (k < 0)
2293                                 goto clear_journal;
2294                         used_commit_ids[k] = true;
2295                         max_commit_id_sections[k] = i;
2296                 }
2297                 if (journal_empty) {
2298                         for (j = 0; j < ic->journal_section_entries; j++) {
2299                                 struct journal_entry *je = access_journal_entry(ic, i, j);
2300                                 if (!journal_entry_is_unused(je)) {
2301                                         journal_empty = false;
2302                                         break;
2303                                 }
2304                         }
2305                 }
2306         }
2307
2308         if (!used_commit_ids[N_COMMIT_IDS - 1]) {
2309                 unused = N_COMMIT_IDS - 1;
2310                 while (unused && !used_commit_ids[unused - 1])
2311                         unused--;
2312         } else {
2313                 for (unused = 0; unused < N_COMMIT_IDS; unused++)
2314                         if (!used_commit_ids[unused])
2315                                 break;
2316                 if (unused == N_COMMIT_IDS) {
2317                         dm_integrity_io_error(ic, "journal commit ids", -EIO);
2318                         goto clear_journal;
2319                 }
2320         }
2321         DEBUG_print("first unused commit seq %d [%d,%d,%d,%d]\n",
2322                     unused, used_commit_ids[0], used_commit_ids[1],
2323                     used_commit_ids[2], used_commit_ids[3]);
2324
2325         last_used = prev_commit_seq(unused);
2326         want_commit_seq = prev_commit_seq(last_used);
2327
2328         if (!used_commit_ids[want_commit_seq] && used_commit_ids[prev_commit_seq(want_commit_seq)])
2329                 journal_empty = true;
2330
2331         write_start = max_commit_id_sections[last_used] + 1;
2332         if (unlikely(write_start >= ic->journal_sections))
2333                 want_commit_seq = next_commit_seq(want_commit_seq);
2334         wraparound_section(ic, &write_start);
2335
2336         i = write_start;
2337         for (write_sections = 0; write_sections < ic->journal_sections; write_sections++) {
2338                 for (j = 0; j < ic->journal_section_sectors; j++) {
2339                         struct journal_sector *js = access_journal(ic, i, j);
2340
2341                         if (js->commit_id != dm_integrity_commit_id(ic, i, j, want_commit_seq)) {
2342                                 /*
2343                                  * This could be caused by crash during writing.
2344                                  * We won't replay the inconsistent part of the
2345                                  * journal.
2346                                  */
2347                                 DEBUG_print("commit id mismatch at position (%u, %u): %d != %d\n",
2348                                             i, j, find_commit_seq(ic, i, j, js->commit_id), want_commit_seq);
2349                                 goto brk;
2350                         }
2351                 }
2352                 i++;
2353                 if (unlikely(i >= ic->journal_sections))
2354                         want_commit_seq = next_commit_seq(want_commit_seq);
2355                 wraparound_section(ic, &i);
2356         }
2357 brk:
2358
2359         if (!journal_empty) {
2360                 DEBUG_print("replaying %u sections, starting at %u, commit seq %d\n",
2361                             write_sections, write_start, want_commit_seq);
2362                 do_journal_write(ic, write_start, write_sections, true);
2363         }
2364
2365         if (write_sections == ic->journal_sections && (ic->mode == 'J' || journal_empty)) {
2366                 continue_section = write_start;
2367                 ic->commit_seq = want_commit_seq;
2368                 DEBUG_print("continuing from section %u, commit seq %d\n", write_start, ic->commit_seq);
2369         } else {
2370                 unsigned s;
2371                 unsigned char erase_seq;
2372 clear_journal:
2373                 DEBUG_print("clearing journal\n");
2374
2375                 erase_seq = prev_commit_seq(prev_commit_seq(last_used));
2376                 s = write_start;
2377                 init_journal(ic, s, 1, erase_seq);
2378                 s++;
2379                 wraparound_section(ic, &s);
2380                 if (ic->journal_sections >= 2) {
2381                         init_journal(ic, s, ic->journal_sections - 2, erase_seq);
2382                         s += ic->journal_sections - 2;
2383                         wraparound_section(ic, &s);
2384                         init_journal(ic, s, 1, erase_seq);
2385                 }
2386
2387                 continue_section = 0;
2388                 ic->commit_seq = next_commit_seq(erase_seq);
2389         }
2390
2391         ic->committed_section = continue_section;
2392         ic->n_committed_sections = 0;
2393
2394         ic->uncommitted_section = continue_section;
2395         ic->n_uncommitted_sections = 0;
2396
2397         ic->free_section = continue_section;
2398         ic->free_section_entry = 0;
2399         ic->free_sectors = ic->journal_entries;
2400
2401         ic->journal_tree_root = RB_ROOT;
2402         for (i = 0; i < ic->journal_entries; i++)
2403                 init_journal_node(&ic->journal_tree[i]);
2404 }
2405
2406 static void dm_integrity_postsuspend(struct dm_target *ti)
2407 {
2408         struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
2409
2410         del_timer_sync(&ic->autocommit_timer);
2411
2412         WRITE_ONCE(ic->suspending, 1);
2413
2414         if (ic->recalc_wq)
2415                 drain_workqueue(ic->recalc_wq);
2416
2417         queue_work(ic->commit_wq, &ic->commit_work);
2418         drain_workqueue(ic->commit_wq);
2419
2420         if (ic->mode == 'J') {
2421                 if (ic->meta_dev)
2422                         queue_work(ic->writer_wq, &ic->writer_work);
2423                 drain_workqueue(ic->writer_wq);
2424                 dm_integrity_flush_buffers(ic);
2425         }
2426
2427         WRITE_ONCE(ic->suspending, 0);
2428
2429         BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
2430
2431         ic->journal_uptodate = true;
2432 }
2433
2434 static void dm_integrity_resume(struct dm_target *ti)
2435 {
2436         struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
2437
2438         replay_journal(ic);
2439
2440         if (ic->recalc_wq && ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
2441                 __u64 recalc_pos = le64_to_cpu(ic->sb->recalc_sector);
2442                 if (recalc_pos < ic->provided_data_sectors) {
2443                         queue_work(ic->recalc_wq, &ic->recalc_work);
2444                 } else if (recalc_pos > ic->provided_data_sectors) {
2445                         ic->sb->recalc_sector = cpu_to_le64(ic->provided_data_sectors);
2446                         recalc_write_super(ic);
2447                 }
2448         }
2449 }
2450
2451 static void dm_integrity_status(struct dm_target *ti, status_type_t type,
2452                                 unsigned status_flags, char *result, unsigned maxlen)
2453 {
2454         struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
2455         unsigned arg_count;
2456         size_t sz = 0;
2457
2458         switch (type) {
2459         case STATUSTYPE_INFO:
2460                 DMEMIT("%llu %llu",
2461                         (unsigned long long)atomic64_read(&ic->number_of_mismatches),
2462                         (unsigned long long)ic->provided_data_sectors);
2463                 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
2464                         DMEMIT(" %llu", (unsigned long long)le64_to_cpu(ic->sb->recalc_sector));
2465                 else
2466                         DMEMIT(" -");
2467                 break;
2468
2469         case STATUSTYPE_TABLE: {
2470                 __u64 watermark_percentage = (__u64)(ic->journal_entries - ic->free_sectors_threshold) * 100;
2471                 watermark_percentage += ic->journal_entries / 2;
2472                 do_div(watermark_percentage, ic->journal_entries);
2473                 arg_count = 5;
2474                 arg_count += !!ic->meta_dev;
2475                 arg_count += ic->sectors_per_block != 1;
2476                 arg_count += !!(ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING));
2477                 arg_count += !!ic->internal_hash_alg.alg_string;
2478                 arg_count += !!ic->journal_crypt_alg.alg_string;
2479                 arg_count += !!ic->journal_mac_alg.alg_string;
2480                 DMEMIT("%s %llu %u %c %u", ic->dev->name, (unsigned long long)ic->start,
2481                        ic->tag_size, ic->mode, arg_count);
2482                 if (ic->meta_dev)
2483                         DMEMIT(" meta_device:%s", ic->meta_dev->name);
2484                 if (ic->sectors_per_block != 1)
2485                         DMEMIT(" block_size:%u", ic->sectors_per_block << SECTOR_SHIFT);
2486                 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
2487                         DMEMIT(" recalculate");
2488                 DMEMIT(" journal_sectors:%u", ic->initial_sectors - SB_SECTORS);
2489                 DMEMIT(" interleave_sectors:%u", 1U << ic->sb->log2_interleave_sectors);
2490                 DMEMIT(" buffer_sectors:%u", 1U << ic->log2_buffer_sectors);
2491                 DMEMIT(" journal_watermark:%u", (unsigned)watermark_percentage);
2492                 DMEMIT(" commit_time:%u", ic->autocommit_msec);
2493
2494 #define EMIT_ALG(a, n)                                                  \
2495                 do {                                                    \
2496                         if (ic->a.alg_string) {                         \
2497                                 DMEMIT(" %s:%s", n, ic->a.alg_string);  \
2498                                 if (ic->a.key_string)                   \
2499                                         DMEMIT(":%s", ic->a.key_string);\
2500                         }                                               \
2501                 } while (0)
2502                 EMIT_ALG(internal_hash_alg, "internal_hash");
2503                 EMIT_ALG(journal_crypt_alg, "journal_crypt");
2504                 EMIT_ALG(journal_mac_alg, "journal_mac");
2505                 break;
2506         }
2507         }
2508 }
2509
2510 static int dm_integrity_iterate_devices(struct dm_target *ti,
2511                                         iterate_devices_callout_fn fn, void *data)
2512 {
2513         struct dm_integrity_c *ic = ti->private;
2514
2515         if (!ic->meta_dev)
2516                 return fn(ti, ic->dev, ic->start + ic->initial_sectors + ic->metadata_run, ti->len, data);
2517         else
2518                 return fn(ti, ic->dev, 0, ti->len, data);
2519 }
2520
2521 static void dm_integrity_io_hints(struct dm_target *ti, struct queue_limits *limits)
2522 {
2523         struct dm_integrity_c *ic = ti->private;
2524
2525         if (ic->sectors_per_block > 1) {
2526                 limits->logical_block_size = ic->sectors_per_block << SECTOR_SHIFT;
2527                 limits->physical_block_size = ic->sectors_per_block << SECTOR_SHIFT;
2528                 blk_limits_io_min(limits, ic->sectors_per_block << SECTOR_SHIFT);
2529         }
2530 }
2531
2532 static void calculate_journal_section_size(struct dm_integrity_c *ic)
2533 {
2534         unsigned sector_space = JOURNAL_SECTOR_DATA;
2535
2536         ic->journal_sections = le32_to_cpu(ic->sb->journal_sections);
2537         ic->journal_entry_size = roundup(offsetof(struct journal_entry, last_bytes[ic->sectors_per_block]) + ic->tag_size,
2538                                          JOURNAL_ENTRY_ROUNDUP);
2539
2540         if (ic->sb->flags & cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC))
2541                 sector_space -= JOURNAL_MAC_PER_SECTOR;
2542         ic->journal_entries_per_sector = sector_space / ic->journal_entry_size;
2543         ic->journal_section_entries = ic->journal_entries_per_sector * JOURNAL_BLOCK_SECTORS;
2544         ic->journal_section_sectors = (ic->journal_section_entries << ic->sb->log2_sectors_per_block) + JOURNAL_BLOCK_SECTORS;
2545         ic->journal_entries = ic->journal_section_entries * ic->journal_sections;
2546 }
2547
2548 static int calculate_device_limits(struct dm_integrity_c *ic)
2549 {
2550         __u64 initial_sectors;
2551
2552         calculate_journal_section_size(ic);
2553         initial_sectors = SB_SECTORS + (__u64)ic->journal_section_sectors * ic->journal_sections;
2554         if (initial_sectors + METADATA_PADDING_SECTORS >= ic->meta_device_sectors || initial_sectors > UINT_MAX)
2555                 return -EINVAL;
2556         ic->initial_sectors = initial_sectors;
2557
2558         if (!ic->meta_dev) {
2559                 sector_t last_sector, last_area, last_offset;
2560
2561                 ic->metadata_run = roundup((__u64)ic->tag_size << (ic->sb->log2_interleave_sectors - ic->sb->log2_sectors_per_block),
2562                                            (__u64)(1 << SECTOR_SHIFT << METADATA_PADDING_SECTORS)) >> SECTOR_SHIFT;
2563                 if (!(ic->metadata_run & (ic->metadata_run - 1)))
2564                         ic->log2_metadata_run = __ffs(ic->metadata_run);
2565                 else
2566                         ic->log2_metadata_run = -1;
2567
2568                 get_area_and_offset(ic, ic->provided_data_sectors - 1, &last_area, &last_offset);
2569                 last_sector = get_data_sector(ic, last_area, last_offset);
2570                 if (last_sector < ic->start || last_sector >= ic->meta_device_sectors)
2571                         return -EINVAL;
2572         } else {
2573                 __u64 meta_size = ic->provided_data_sectors * ic->tag_size;
2574                 meta_size = (meta_size + ((1U << (ic->log2_buffer_sectors + SECTOR_SHIFT)) - 1))
2575                                 >> (ic->log2_buffer_sectors + SECTOR_SHIFT);
2576                 meta_size <<= ic->log2_buffer_sectors;
2577                 if (ic->initial_sectors + meta_size < ic->initial_sectors ||
2578                     ic->initial_sectors + meta_size > ic->meta_device_sectors)
2579                         return -EINVAL;
2580                 ic->metadata_run = 1;
2581                 ic->log2_metadata_run = 0;
2582         }
2583
2584         return 0;
2585 }
2586
2587 static int initialize_superblock(struct dm_integrity_c *ic, unsigned journal_sectors, unsigned interleave_sectors)
2588 {
2589         unsigned journal_sections;
2590         int test_bit;
2591
2592         memset(ic->sb, 0, SB_SECTORS << SECTOR_SHIFT);
2593         memcpy(ic->sb->magic, SB_MAGIC, 8);
2594         ic->sb->integrity_tag_size = cpu_to_le16(ic->tag_size);
2595         ic->sb->log2_sectors_per_block = __ffs(ic->sectors_per_block);
2596         if (ic->journal_mac_alg.alg_string)
2597                 ic->sb->flags |= cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC);
2598
2599         calculate_journal_section_size(ic);
2600         journal_sections = journal_sectors / ic->journal_section_sectors;
2601         if (!journal_sections)
2602                 journal_sections = 1;
2603
2604         if (!ic->meta_dev) {
2605                 ic->sb->journal_sections = cpu_to_le32(journal_sections);
2606                 if (!interleave_sectors)
2607                         interleave_sectors = DEFAULT_INTERLEAVE_SECTORS;
2608                 ic->sb->log2_interleave_sectors = __fls(interleave_sectors);
2609                 ic->sb->log2_interleave_sectors = max((__u8)MIN_LOG2_INTERLEAVE_SECTORS, ic->sb->log2_interleave_sectors);
2610                 ic->sb->log2_interleave_sectors = min((__u8)MAX_LOG2_INTERLEAVE_SECTORS, ic->sb->log2_interleave_sectors);
2611
2612                 ic->provided_data_sectors = 0;
2613                 for (test_bit = fls64(ic->meta_device_sectors) - 1; test_bit >= 3; test_bit--) {
2614                         __u64 prev_data_sectors = ic->provided_data_sectors;
2615
2616                         ic->provided_data_sectors |= (sector_t)1 << test_bit;
2617                         if (calculate_device_limits(ic))
2618                                 ic->provided_data_sectors = prev_data_sectors;
2619                 }
2620                 if (!ic->provided_data_sectors)
2621                         return -EINVAL;
2622         } else {
2623                 ic->sb->log2_interleave_sectors = 0;
2624                 ic->provided_data_sectors = ic->data_device_sectors;
2625                 ic->provided_data_sectors &= ~(sector_t)(ic->sectors_per_block - 1);
2626
2627 try_smaller_buffer:
2628                 ic->sb->journal_sections = cpu_to_le32(0);
2629                 for (test_bit = fls(journal_sections) - 1; test_bit >= 0; test_bit--) {
2630                         __u32 prev_journal_sections = le32_to_cpu(ic->sb->journal_sections);
2631                         __u32 test_journal_sections = prev_journal_sections | (1U << test_bit);
2632                         if (test_journal_sections > journal_sections)
2633                                 continue;
2634                         ic->sb->journal_sections = cpu_to_le32(test_journal_sections);
2635                         if (calculate_device_limits(ic))
2636                                 ic->sb->journal_sections = cpu_to_le32(prev_journal_sections);
2637
2638                 }
2639                 if (!le32_to_cpu(ic->sb->journal_sections)) {
2640                         if (ic->log2_buffer_sectors > 3) {
2641                                 ic->log2_buffer_sectors--;
2642                                 goto try_smaller_buffer;
2643                         }
2644                         return -EINVAL;
2645                 }
2646         }
2647
2648         ic->sb->provided_data_sectors = cpu_to_le64(ic->provided_data_sectors);
2649
2650         sb_set_version(ic);
2651
2652         return 0;
2653 }
2654
2655 static void dm_integrity_set(struct dm_target *ti, struct dm_integrity_c *ic)
2656 {
2657         struct gendisk *disk = dm_disk(dm_table_get_md(ti->table));
2658         struct blk_integrity bi;
2659
2660         memset(&bi, 0, sizeof(bi));
2661         bi.profile = &dm_integrity_profile;
2662         bi.tuple_size = ic->tag_size;
2663         bi.tag_size = bi.tuple_size;
2664         bi.interval_exp = ic->sb->log2_sectors_per_block + SECTOR_SHIFT;
2665
2666         blk_integrity_register(disk, &bi);
2667         blk_queue_max_integrity_segments(disk->queue, UINT_MAX);
2668 }
2669
2670 static void dm_integrity_free_page_list(struct dm_integrity_c *ic, struct page_list *pl)
2671 {
2672         unsigned i;
2673
2674         if (!pl)
2675                 return;
2676         for (i = 0; i < ic->journal_pages; i++)
2677                 if (pl[i].page)
2678                         __free_page(pl[i].page);
2679         kvfree(pl);
2680 }
2681
2682 static struct page_list *dm_integrity_alloc_page_list(struct dm_integrity_c *ic)
2683 {
2684         size_t page_list_desc_size = ic->journal_pages * sizeof(struct page_list);
2685         struct page_list *pl;
2686         unsigned i;
2687
2688         pl = kvmalloc(page_list_desc_size, GFP_KERNEL | __GFP_ZERO);
2689         if (!pl)
2690                 return NULL;
2691
2692         for (i = 0; i < ic->journal_pages; i++) {
2693                 pl[i].page = alloc_page(GFP_KERNEL);
2694                 if (!pl[i].page) {
2695                         dm_integrity_free_page_list(ic, pl);
2696                         return NULL;
2697                 }
2698                 if (i)
2699                         pl[i - 1].next = &pl[i];
2700         }
2701
2702         return pl;
2703 }
2704
2705 static void dm_integrity_free_journal_scatterlist(struct dm_integrity_c *ic, struct scatterlist **sl)
2706 {
2707         unsigned i;
2708         for (i = 0; i < ic->journal_sections; i++)
2709                 kvfree(sl[i]);
2710         kvfree(sl);
2711 }
2712
2713 static struct scatterlist **dm_integrity_alloc_journal_scatterlist(struct dm_integrity_c *ic, struct page_list *pl)
2714 {
2715         struct scatterlist **sl;
2716         unsigned i;
2717
2718         sl = kvmalloc_array(ic->journal_sections,
2719                             sizeof(struct scatterlist *),
2720                             GFP_KERNEL | __GFP_ZERO);
2721         if (!sl)
2722                 return NULL;
2723
2724         for (i = 0; i < ic->journal_sections; i++) {
2725                 struct scatterlist *s;
2726                 unsigned start_index, start_offset;
2727                 unsigned end_index, end_offset;
2728                 unsigned n_pages;
2729                 unsigned idx;
2730
2731                 page_list_location(ic, i, 0, &start_index, &start_offset);
2732                 page_list_location(ic, i, ic->journal_section_sectors - 1, &end_index, &end_offset);
2733
2734                 n_pages = (end_index - start_index + 1);
2735
2736                 s = kvmalloc_array(n_pages, sizeof(struct scatterlist),
2737                                    GFP_KERNEL);
2738                 if (!s) {
2739                         dm_integrity_free_journal_scatterlist(ic, sl);
2740                         return NULL;
2741                 }
2742
2743                 sg_init_table(s, n_pages);
2744                 for (idx = start_index; idx <= end_index; idx++) {
2745                         char *va = lowmem_page_address(pl[idx].page);
2746                         unsigned start = 0, end = PAGE_SIZE;
2747                         if (idx == start_index)
2748                                 start = start_offset;
2749                         if (idx == end_index)
2750                                 end = end_offset + (1 << SECTOR_SHIFT);
2751                         sg_set_buf(&s[idx - start_index], va + start, end - start);
2752                 }
2753
2754                 sl[i] = s;
2755         }
2756
2757         return sl;
2758 }
2759
2760 static void free_alg(struct alg_spec *a)
2761 {
2762         kzfree(a->alg_string);
2763         kzfree(a->key);
2764         memset(a, 0, sizeof *a);
2765 }
2766
2767 static int get_alg_and_key(const char *arg, struct alg_spec *a, char **error, char *error_inval)
2768 {
2769         char *k;
2770
2771         free_alg(a);
2772
2773         a->alg_string = kstrdup(strchr(arg, ':') + 1, GFP_KERNEL);
2774         if (!a->alg_string)
2775                 goto nomem;
2776
2777         k = strchr(a->alg_string, ':');
2778         if (k) {
2779                 *k = 0;
2780                 a->key_string = k + 1;
2781                 if (strlen(a->key_string) & 1)
2782                         goto inval;
2783
2784                 a->key_size = strlen(a->key_string) / 2;
2785                 a->key = kmalloc(a->key_size, GFP_KERNEL);
2786                 if (!a->key)
2787                         goto nomem;
2788                 if (hex2bin(a->key, a->key_string, a->key_size))
2789                         goto inval;
2790         }
2791
2792         return 0;
2793 inval:
2794         *error = error_inval;
2795         return -EINVAL;
2796 nomem:
2797         *error = "Out of memory for an argument";
2798         return -ENOMEM;
2799 }
2800
2801 static int get_mac(struct crypto_shash **hash, struct alg_spec *a, char **error,
2802                    char *error_alg, char *error_key)
2803 {
2804         int r;
2805
2806         if (a->alg_string) {
2807                 *hash = crypto_alloc_shash(a->alg_string, 0, 0);
2808                 if (IS_ERR(*hash)) {
2809                         *error = error_alg;
2810                         r = PTR_ERR(*hash);
2811                         *hash = NULL;
2812                         return r;
2813                 }
2814
2815                 if (a->key) {
2816                         r = crypto_shash_setkey(*hash, a->key, a->key_size);
2817                         if (r) {
2818                                 *error = error_key;
2819                                 return r;
2820                         }
2821                 } else if (crypto_shash_get_flags(*hash) & CRYPTO_TFM_NEED_KEY) {
2822                         *error = error_key;
2823                         return -ENOKEY;
2824                 }
2825         }
2826
2827         return 0;
2828 }
2829
2830 static int create_journal(struct dm_integrity_c *ic, char **error)
2831 {
2832         int r = 0;
2833         unsigned i;
2834         __u64 journal_pages, journal_desc_size, journal_tree_size;
2835         unsigned char *crypt_data = NULL, *crypt_iv = NULL;
2836         struct skcipher_request *req = NULL;
2837
2838         ic->commit_ids[0] = cpu_to_le64(0x1111111111111111ULL);
2839         ic->commit_ids[1] = cpu_to_le64(0x2222222222222222ULL);
2840         ic->commit_ids[2] = cpu_to_le64(0x3333333333333333ULL);
2841         ic->commit_ids[3] = cpu_to_le64(0x4444444444444444ULL);
2842
2843         journal_pages = roundup((__u64)ic->journal_sections * ic->journal_section_sectors,
2844                                 PAGE_SIZE >> SECTOR_SHIFT) >> (PAGE_SHIFT - SECTOR_SHIFT);
2845         journal_desc_size = journal_pages * sizeof(struct page_list);
2846         if (journal_pages >= totalram_pages() - totalhigh_pages() || journal_desc_size > ULONG_MAX) {
2847                 *error = "Journal doesn't fit into memory";
2848                 r = -ENOMEM;
2849                 goto bad;
2850         }
2851         ic->journal_pages = journal_pages;
2852
2853         ic->journal = dm_integrity_alloc_page_list(ic);
2854         if (!ic->journal) {
2855                 *error = "Could not allocate memory for journal";
2856                 r = -ENOMEM;
2857                 goto bad;
2858         }
2859         if (ic->journal_crypt_alg.alg_string) {
2860                 unsigned ivsize, blocksize;
2861                 struct journal_completion comp;
2862
2863                 comp.ic = ic;
2864                 ic->journal_crypt = crypto_alloc_skcipher(ic->journal_crypt_alg.alg_string, 0, 0);
2865                 if (IS_ERR(ic->journal_crypt)) {
2866                         *error = "Invalid journal cipher";
2867                         r = PTR_ERR(ic->journal_crypt);
2868                         ic->journal_crypt = NULL;
2869                         goto bad;
2870                 }
2871                 ivsize = crypto_skcipher_ivsize(ic->journal_crypt);
2872                 blocksize = crypto_skcipher_blocksize(ic->journal_crypt);
2873
2874                 if (ic->journal_crypt_alg.key) {
2875                         r = crypto_skcipher_setkey(ic->journal_crypt, ic->journal_crypt_alg.key,
2876                                                    ic->journal_crypt_alg.key_size);
2877                         if (r) {
2878                                 *error = "Error setting encryption key";
2879                                 goto bad;
2880                         }
2881                 }
2882                 DEBUG_print("cipher %s, block size %u iv size %u\n",
2883                             ic->journal_crypt_alg.alg_string, blocksize, ivsize);
2884
2885                 ic->journal_io = dm_integrity_alloc_page_list(ic);
2886                 if (!ic->journal_io) {
2887                         *error = "Could not allocate memory for journal io";
2888                         r = -ENOMEM;
2889                         goto bad;
2890                 }
2891
2892                 if (blocksize == 1) {
2893                         struct scatterlist *sg;
2894
2895                         req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
2896                         if (!req) {
2897                                 *error = "Could not allocate crypt request";
2898                                 r = -ENOMEM;
2899                                 goto bad;
2900                         }
2901
2902                         crypt_iv = kmalloc(ivsize, GFP_KERNEL);
2903                         if (!crypt_iv) {
2904                                 *error = "Could not allocate iv";
2905                                 r = -ENOMEM;
2906                                 goto bad;
2907                         }
2908
2909                         ic->journal_xor = dm_integrity_alloc_page_list(ic);
2910                         if (!ic->journal_xor) {
2911                                 *error = "Could not allocate memory for journal xor";
2912                                 r = -ENOMEM;
2913                                 goto bad;
2914                         }
2915
2916                         sg = kvmalloc_array(ic->journal_pages + 1,
2917                                             sizeof(struct scatterlist),
2918                                             GFP_KERNEL);
2919                         if (!sg) {
2920                                 *error = "Unable to allocate sg list";
2921                                 r = -ENOMEM;
2922                                 goto bad;
2923                         }
2924                         sg_init_table(sg, ic->journal_pages + 1);
2925                         for (i = 0; i < ic->journal_pages; i++) {
2926                                 char *va = lowmem_page_address(ic->journal_xor[i].page);
2927                                 clear_page(va);
2928                                 sg_set_buf(&sg[i], va, PAGE_SIZE);
2929                         }
2930                         sg_set_buf(&sg[i], &ic->commit_ids, sizeof ic->commit_ids);
2931                         memset(crypt_iv, 0x00, ivsize);
2932
2933                         skcipher_request_set_crypt(req, sg, sg, PAGE_SIZE * ic->journal_pages + sizeof ic->commit_ids, crypt_iv);
2934                         init_completion(&comp.comp);
2935                         comp.in_flight = (atomic_t)ATOMIC_INIT(1);
2936                         if (do_crypt(true, req, &comp))
2937                                 wait_for_completion(&comp.comp);
2938                         kvfree(sg);
2939                         r = dm_integrity_failed(ic);
2940                         if (r) {
2941                                 *error = "Unable to encrypt journal";
2942                                 goto bad;
2943                         }
2944                         DEBUG_bytes(lowmem_page_address(ic->journal_xor[0].page), 64, "xor data");
2945
2946                         crypto_free_skcipher(ic->journal_crypt);
2947                         ic->journal_crypt = NULL;
2948                 } else {
2949                         unsigned crypt_len = roundup(ivsize, blocksize);
2950
2951                         req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
2952                         if (!req) {
2953                                 *error = "Could not allocate crypt request";
2954                                 r = -ENOMEM;
2955                                 goto bad;
2956                         }
2957
2958                         crypt_iv = kmalloc(ivsize, GFP_KERNEL);
2959                         if (!crypt_iv) {
2960                                 *error = "Could not allocate iv";
2961                                 r = -ENOMEM;
2962                                 goto bad;
2963                         }
2964
2965                         crypt_data = kmalloc(crypt_len, GFP_KERNEL);
2966                         if (!crypt_data) {
2967                                 *error = "Unable to allocate crypt data";
2968                                 r = -ENOMEM;
2969                                 goto bad;
2970                         }
2971
2972                         ic->journal_scatterlist = dm_integrity_alloc_journal_scatterlist(ic, ic->journal);
2973                         if (!ic->journal_scatterlist) {
2974                                 *error = "Unable to allocate sg list";
2975                                 r = -ENOMEM;
2976                                 goto bad;
2977                         }
2978                         ic->journal_io_scatterlist = dm_integrity_alloc_journal_scatterlist(ic, ic->journal_io);
2979                         if (!ic->journal_io_scatterlist) {
2980                                 *error = "Unable to allocate sg list";
2981                                 r = -ENOMEM;
2982                                 goto bad;
2983                         }
2984                         ic->sk_requests = kvmalloc_array(ic->journal_sections,
2985                                                          sizeof(struct skcipher_request *),
2986                                                          GFP_KERNEL | __GFP_ZERO);
2987                         if (!ic->sk_requests) {
2988                                 *error = "Unable to allocate sk requests";
2989                                 r = -ENOMEM;
2990                                 goto bad;
2991                         }
2992                         for (i = 0; i < ic->journal_sections; i++) {
2993                                 struct scatterlist sg;
2994                                 struct skcipher_request *section_req;
2995                                 __u32 section_le = cpu_to_le32(i);
2996
2997                                 memset(crypt_iv, 0x00, ivsize);
2998                                 memset(crypt_data, 0x00, crypt_len);
2999                                 memcpy(crypt_data, &section_le, min((size_t)crypt_len, sizeof(section_le)));
3000
3001                                 sg_init_one(&sg, crypt_data, crypt_len);
3002                                 skcipher_request_set_crypt(req, &sg, &sg, crypt_len, crypt_iv);
3003                                 init_completion(&comp.comp);
3004                                 comp.in_flight = (atomic_t)ATOMIC_INIT(1);
3005                                 if (do_crypt(true, req, &comp))
3006                                         wait_for_completion(&comp.comp);
3007
3008                                 r = dm_integrity_failed(ic);
3009                                 if (r) {
3010                                         *error = "Unable to generate iv";
3011                                         goto bad;
3012                                 }
3013
3014                                 section_req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
3015                                 if (!section_req) {
3016                                         *error = "Unable to allocate crypt request";
3017                                         r = -ENOMEM;
3018                                         goto bad;
3019                                 }
3020                                 section_req->iv = kmalloc_array(ivsize, 2,
3021                                                                 GFP_KERNEL);
3022                                 if (!section_req->iv) {
3023                                         skcipher_request_free(section_req);
3024                                         *error = "Unable to allocate iv";
3025                                         r = -ENOMEM;
3026                                         goto bad;
3027                                 }
3028                                 memcpy(section_req->iv + ivsize, crypt_data, ivsize);
3029                                 section_req->cryptlen = (size_t)ic->journal_section_sectors << SECTOR_SHIFT;
3030                                 ic->sk_requests[i] = section_req;
3031                                 DEBUG_bytes(crypt_data, ivsize, "iv(%u)", i);
3032                         }
3033                 }
3034         }
3035
3036         for (i = 0; i < N_COMMIT_IDS; i++) {
3037                 unsigned j;
3038 retest_commit_id:
3039                 for (j = 0; j < i; j++) {
3040                         if (ic->commit_ids[j] == ic->commit_ids[i]) {
3041                                 ic->commit_ids[i] = cpu_to_le64(le64_to_cpu(ic->commit_ids[i]) + 1);
3042                                 goto retest_commit_id;
3043                         }
3044                 }
3045                 DEBUG_print("commit id %u: %016llx\n", i, ic->commit_ids[i]);
3046         }
3047
3048         journal_tree_size = (__u64)ic->journal_entries * sizeof(struct journal_node);
3049         if (journal_tree_size > ULONG_MAX) {
3050                 *error = "Journal doesn't fit into memory";
3051                 r = -ENOMEM;
3052                 goto bad;
3053         }
3054         ic->journal_tree = kvmalloc(journal_tree_size, GFP_KERNEL);
3055         if (!ic->journal_tree) {
3056                 *error = "Could not allocate memory for journal tree";
3057                 r = -ENOMEM;
3058         }
3059 bad:
3060         kfree(crypt_data);
3061         kfree(crypt_iv);
3062         skcipher_request_free(req);
3063
3064         return r;
3065 }
3066
3067 /*
3068  * Construct a integrity mapping
3069  *
3070  * Arguments:
3071  *      device
3072  *      offset from the start of the device
3073  *      tag size
3074  *      D - direct writes, J - journal writes, R - recovery mode
3075  *      number of optional arguments
3076  *      optional arguments:
3077  *              journal_sectors
3078  *              interleave_sectors
3079  *              buffer_sectors
3080  *              journal_watermark
3081  *              commit_time
3082  *              internal_hash
3083  *              journal_crypt
3084  *              journal_mac
3085  *              block_size
3086  */
3087 static int dm_integrity_ctr(struct dm_target *ti, unsigned argc, char **argv)
3088 {
3089         struct dm_integrity_c *ic;
3090         char dummy;
3091         int r;
3092         unsigned extra_args;
3093         struct dm_arg_set as;
3094         static const struct dm_arg _args[] = {
3095                 {0, 9, "Invalid number of feature args"},
3096         };
3097         unsigned journal_sectors, interleave_sectors, buffer_sectors, journal_watermark, sync_msec;
3098         bool recalculate;
3099         bool should_write_sb;
3100         __u64 threshold;
3101         unsigned long long start;
3102
3103 #define DIRECT_ARGUMENTS        4
3104
3105         if (argc <= DIRECT_ARGUMENTS) {
3106                 ti->error = "Invalid argument count";
3107                 return -EINVAL;
3108         }
3109
3110         ic = kzalloc(sizeof(struct dm_integrity_c), GFP_KERNEL);
3111         if (!ic) {
3112                 ti->error = "Cannot allocate integrity context";
3113                 return -ENOMEM;
3114         }
3115         ti->private = ic;
3116         ti->per_io_data_size = sizeof(struct dm_integrity_io);
3117
3118         ic->in_progress = RB_ROOT;
3119         INIT_LIST_HEAD(&ic->wait_list);
3120         init_waitqueue_head(&ic->endio_wait);
3121         bio_list_init(&ic->flush_bio_list);
3122         init_waitqueue_head(&ic->copy_to_journal_wait);
3123         init_completion(&ic->crypto_backoff);
3124         atomic64_set(&ic->number_of_mismatches, 0);
3125
3126         r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &ic->dev);
3127         if (r) {
3128                 ti->error = "Device lookup failed";
3129                 goto bad;
3130         }
3131
3132         if (sscanf(argv[1], "%llu%c", &start, &dummy) != 1 || start != (sector_t)start) {
3133                 ti->error = "Invalid starting offset";
3134                 r = -EINVAL;
3135                 goto bad;
3136         }
3137         ic->start = start;
3138
3139         if (strcmp(argv[2], "-")) {
3140                 if (sscanf(argv[2], "%u%c", &ic->tag_size, &dummy) != 1 || !ic->tag_size) {
3141                         ti->error = "Invalid tag size";
3142                         r = -EINVAL;
3143                         goto bad;
3144                 }
3145         }
3146
3147         if (!strcmp(argv[3], "J") || !strcmp(argv[3], "D") || !strcmp(argv[3], "R"))
3148                 ic->mode = argv[3][0];
3149         else {
3150                 ti->error = "Invalid mode (expecting J, D, R)";
3151                 r = -EINVAL;
3152                 goto bad;
3153         }
3154
3155         journal_sectors = 0;
3156         interleave_sectors = DEFAULT_INTERLEAVE_SECTORS;
3157         buffer_sectors = DEFAULT_BUFFER_SECTORS;
3158         journal_watermark = DEFAULT_JOURNAL_WATERMARK;
3159         sync_msec = DEFAULT_SYNC_MSEC;
3160         recalculate = false;
3161         ic->sectors_per_block = 1;
3162
3163         as.argc = argc - DIRECT_ARGUMENTS;
3164         as.argv = argv + DIRECT_ARGUMENTS;
3165         r = dm_read_arg_group(_args, &as, &extra_args, &ti->error);
3166         if (r)
3167                 goto bad;
3168
3169         while (extra_args--) {
3170                 const char *opt_string;
3171                 unsigned val;
3172                 opt_string = dm_shift_arg(&as);
3173                 if (!opt_string) {
3174                         r = -EINVAL;
3175                         ti->error = "Not enough feature arguments";
3176                         goto bad;
3177                 }
3178                 if (sscanf(opt_string, "journal_sectors:%u%c", &val, &dummy) == 1)
3179                         journal_sectors = val ? val : 1;
3180                 else if (sscanf(opt_string, "interleave_sectors:%u%c", &val, &dummy) == 1)
3181                         interleave_sectors = val;
3182                 else if (sscanf(opt_string, "buffer_sectors:%u%c", &val, &dummy) == 1)
3183                         buffer_sectors = val;
3184                 else if (sscanf(opt_string, "journal_watermark:%u%c", &val, &dummy) == 1 && val <= 100)
3185                         journal_watermark = val;
3186                 else if (sscanf(opt_string, "commit_time:%u%c", &val, &dummy) == 1)
3187                         sync_msec = val;
3188                 else if (!memcmp(opt_string, "meta_device:", strlen("meta_device:"))) {
3189                         if (ic->meta_dev) {
3190                                 dm_put_device(ti, ic->meta_dev);
3191                                 ic->meta_dev = NULL;
3192                         }
3193                         r = dm_get_device(ti, strchr(opt_string, ':') + 1, dm_table_get_mode(ti->table), &ic->meta_dev);
3194                         if (r) {
3195                                 ti->error = "Device lookup failed";
3196                                 goto bad;
3197                         }
3198                 } else if (sscanf(opt_string, "block_size:%u%c", &val, &dummy) == 1) {
3199                         if (val < 1 << SECTOR_SHIFT ||
3200                             val > MAX_SECTORS_PER_BLOCK << SECTOR_SHIFT ||
3201                             (val & (val -1))) {
3202                                 r = -EINVAL;
3203                                 ti->error = "Invalid block_size argument";
3204                                 goto bad;
3205                         }
3206                         ic->sectors_per_block = val >> SECTOR_SHIFT;
3207                 } else if (!memcmp(opt_string, "internal_hash:", strlen("internal_hash:"))) {
3208                         r = get_alg_and_key(opt_string, &ic->internal_hash_alg, &ti->error,
3209                                             "Invalid internal_hash argument");
3210                         if (r)
3211                                 goto bad;
3212                 } else if (!memcmp(opt_string, "journal_crypt:", strlen("journal_crypt:"))) {
3213                         r = get_alg_and_key(opt_string, &ic->journal_crypt_alg, &ti->error,
3214                                             "Invalid journal_crypt argument");
3215                         if (r)
3216                                 goto bad;
3217                 } else if (!memcmp(opt_string, "journal_mac:", strlen("journal_mac:"))) {
3218                         r = get_alg_and_key(opt_string, &ic->journal_mac_alg,  &ti->error,
3219                                             "Invalid journal_mac argument");
3220                         if (r)
3221                                 goto bad;
3222                 } else if (!strcmp(opt_string, "recalculate")) {
3223                         recalculate = true;
3224                 } else {
3225                         r = -EINVAL;
3226                         ti->error = "Invalid argument";
3227                         goto bad;
3228                 }
3229         }
3230
3231         ic->data_device_sectors = i_size_read(ic->dev->bdev->bd_inode) >> SECTOR_SHIFT;
3232         if (!ic->meta_dev)
3233                 ic->meta_device_sectors = ic->data_device_sectors;
3234         else
3235                 ic->meta_device_sectors = i_size_read(ic->meta_dev->bdev->bd_inode) >> SECTOR_SHIFT;
3236
3237         if (!journal_sectors) {
3238                 journal_sectors = min((sector_t)DEFAULT_MAX_JOURNAL_SECTORS,
3239                         ic->data_device_sectors >> DEFAULT_JOURNAL_SIZE_FACTOR);
3240         }
3241
3242         if (!buffer_sectors)
3243                 buffer_sectors = 1;
3244         ic->log2_buffer_sectors = min((int)__fls(buffer_sectors), 31 - SECTOR_SHIFT);
3245
3246         r = get_mac(&ic->internal_hash, &ic->internal_hash_alg, &ti->error,
3247                     "Invalid internal hash", "Error setting internal hash key");
3248         if (r)
3249                 goto bad;
3250
3251         r = get_mac(&ic->journal_mac, &ic->journal_mac_alg, &ti->error,
3252                     "Invalid journal mac", "Error setting journal mac key");
3253         if (r)
3254                 goto bad;
3255
3256         if (!ic->tag_size) {
3257                 if (!ic->internal_hash) {
3258                         ti->error = "Unknown tag size";
3259                         r = -EINVAL;
3260                         goto bad;
3261                 }
3262                 ic->tag_size = crypto_shash_digestsize(ic->internal_hash);
3263         }
3264         if (ic->tag_size > MAX_TAG_SIZE) {
3265                 ti->error = "Too big tag size";
3266                 r = -EINVAL;
3267                 goto bad;
3268         }
3269         if (!(ic->tag_size & (ic->tag_size - 1)))
3270                 ic->log2_tag_size = __ffs(ic->tag_size);
3271         else
3272                 ic->log2_tag_size = -1;
3273
3274         ic->autocommit_jiffies = msecs_to_jiffies(sync_msec);
3275         ic->autocommit_msec = sync_msec;
3276         timer_setup(&ic->autocommit_timer, autocommit_fn, 0);
3277
3278         ic->io = dm_io_client_create();
3279         if (IS_ERR(ic->io)) {
3280                 r = PTR_ERR(ic->io);
3281                 ic->io = NULL;
3282                 ti->error = "Cannot allocate dm io";
3283                 goto bad;
3284         }
3285
3286         r = mempool_init_slab_pool(&ic->journal_io_mempool, JOURNAL_IO_MEMPOOL, journal_io_cache);
3287         if (r) {
3288                 ti->error = "Cannot allocate mempool";
3289                 goto bad;
3290         }
3291
3292         ic->metadata_wq = alloc_workqueue("dm-integrity-metadata",
3293                                           WQ_MEM_RECLAIM, METADATA_WORKQUEUE_MAX_ACTIVE);
3294         if (!ic->metadata_wq) {
3295                 ti->error = "Cannot allocate workqueue";
3296                 r = -ENOMEM;
3297                 goto bad;
3298         }
3299
3300         /*
3301          * If this workqueue were percpu, it would cause bio reordering
3302          * and reduced performance.
3303          */
3304         ic->wait_wq = alloc_workqueue("dm-integrity-wait", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3305         if (!ic->wait_wq) {
3306                 ti->error = "Cannot allocate workqueue";
3307                 r = -ENOMEM;
3308                 goto bad;
3309         }
3310
3311         ic->commit_wq = alloc_workqueue("dm-integrity-commit", WQ_MEM_RECLAIM, 1);
3312         if (!ic->commit_wq) {
3313                 ti->error = "Cannot allocate workqueue";
3314                 r = -ENOMEM;
3315                 goto bad;
3316         }
3317         INIT_WORK(&ic->commit_work, integrity_commit);
3318
3319         if (ic->mode == 'J') {
3320                 ic->writer_wq = alloc_workqueue("dm-integrity-writer", WQ_MEM_RECLAIM, 1);
3321                 if (!ic->writer_wq) {
3322                         ti->error = "Cannot allocate workqueue";
3323                         r = -ENOMEM;
3324                         goto bad;
3325                 }
3326                 INIT_WORK(&ic->writer_work, integrity_writer);
3327         }
3328
3329         ic->sb = alloc_pages_exact(SB_SECTORS << SECTOR_SHIFT, GFP_KERNEL);
3330         if (!ic->sb) {
3331                 r = -ENOMEM;
3332                 ti->error = "Cannot allocate superblock area";
3333                 goto bad;
3334         }
3335
3336         r = sync_rw_sb(ic, REQ_OP_READ, 0);
3337         if (r) {
3338                 ti->error = "Error reading superblock";
3339                 goto bad;
3340         }
3341         should_write_sb = false;
3342         if (memcmp(ic->sb->magic, SB_MAGIC, 8)) {
3343                 if (ic->mode != 'R') {
3344                         if (memchr_inv(ic->sb, 0, SB_SECTORS << SECTOR_SHIFT)) {
3345                                 r = -EINVAL;
3346                                 ti->error = "The device is not initialized";
3347                                 goto bad;
3348                         }
3349                 }
3350
3351                 r = initialize_superblock(ic, journal_sectors, interleave_sectors);
3352                 if (r) {
3353                         ti->error = "Could not initialize superblock";
3354                         goto bad;
3355                 }
3356                 if (ic->mode != 'R')
3357                         should_write_sb = true;
3358         }
3359
3360         if (!ic->sb->version || ic->sb->version > SB_VERSION_2) {
3361                 r = -EINVAL;
3362                 ti->error = "Unknown version";
3363                 goto bad;
3364         }
3365         if (le16_to_cpu(ic->sb->integrity_tag_size) != ic->tag_size) {
3366                 r = -EINVAL;
3367                 ti->error = "Tag size doesn't match the information in superblock";
3368                 goto bad;
3369         }
3370         if (ic->sb->log2_sectors_per_block != __ffs(ic->sectors_per_block)) {
3371                 r = -EINVAL;
3372                 ti->error = "Block size doesn't match the information in superblock";
3373                 goto bad;
3374         }
3375         if (!le32_to_cpu(ic->sb->journal_sections)) {
3376                 r = -EINVAL;
3377                 ti->error = "Corrupted superblock, journal_sections is 0";
3378                 goto bad;
3379         }
3380         /* make sure that ti->max_io_len doesn't overflow */
3381         if (!ic->meta_dev) {
3382                 if (ic->sb->log2_interleave_sectors < MIN_LOG2_INTERLEAVE_SECTORS ||
3383                     ic->sb->log2_interleave_sectors > MAX_LOG2_INTERLEAVE_SECTORS) {
3384                         r = -EINVAL;
3385                         ti->error = "Invalid interleave_sectors in the superblock";
3386                         goto bad;
3387                 }
3388         } else {
3389                 if (ic->sb->log2_interleave_sectors) {
3390                         r = -EINVAL;
3391                         ti->error = "Invalid interleave_sectors in the superblock";
3392                         goto bad;
3393                 }
3394         }
3395         ic->provided_data_sectors = le64_to_cpu(ic->sb->provided_data_sectors);
3396         if (ic->provided_data_sectors != le64_to_cpu(ic->sb->provided_data_sectors)) {
3397                 /* test for overflow */
3398                 r = -EINVAL;
3399                 ti->error = "The superblock has 64-bit device size, but the kernel was compiled with 32-bit sectors";
3400                 goto bad;
3401         }
3402         if (!!(ic->sb->flags & cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC)) != !!ic->journal_mac_alg.alg_string) {
3403                 r = -EINVAL;
3404                 ti->error = "Journal mac mismatch";
3405                 goto bad;
3406         }
3407
3408 try_smaller_buffer:
3409         r = calculate_device_limits(ic);
3410         if (r) {
3411                 if (ic->meta_dev) {
3412                         if (ic->log2_buffer_sectors > 3) {
3413                                 ic->log2_buffer_sectors--;
3414                                 goto try_smaller_buffer;
3415                         }
3416                 }
3417                 ti->error = "The device is too small";
3418                 goto bad;
3419         }
3420         if (!ic->meta_dev)
3421                 ic->log2_buffer_sectors = min(ic->log2_buffer_sectors, (__u8)__ffs(ic->metadata_run));
3422
3423         if (ti->len > ic->provided_data_sectors) {
3424                 r = -EINVAL;
3425                 ti->error = "Not enough provided sectors for requested mapping size";
3426                 goto bad;
3427         }
3428
3429
3430         threshold = (__u64)ic->journal_entries * (100 - journal_watermark);
3431         threshold += 50;
3432         do_div(threshold, 100);
3433         ic->free_sectors_threshold = threshold;
3434
3435         DEBUG_print("initialized:\n");
3436         DEBUG_print("   integrity_tag_size %u\n", le16_to_cpu(ic->sb->integrity_tag_size));
3437         DEBUG_print("   journal_entry_size %u\n", ic->journal_entry_size);
3438         DEBUG_print("   journal_entries_per_sector %u\n", ic->journal_entries_per_sector);
3439         DEBUG_print("   journal_section_entries %u\n", ic->journal_section_entries);
3440         DEBUG_print("   journal_section_sectors %u\n", ic->journal_section_sectors);
3441         DEBUG_print("   journal_sections %u\n", (unsigned)le32_to_cpu(ic->sb->journal_sections));
3442         DEBUG_print("   journal_entries %u\n", ic->journal_entries);
3443         DEBUG_print("   log2_interleave_sectors %d\n", ic->sb->log2_interleave_sectors);
3444         DEBUG_print("   device_sectors 0x%llx\n", (unsigned long long)ic->device_sectors);
3445         DEBUG_print("   initial_sectors 0x%x\n", ic->initial_sectors);
3446         DEBUG_print("   metadata_run 0x%x\n", ic->metadata_run);
3447         DEBUG_print("   log2_metadata_run %d\n", ic->log2_metadata_run);
3448         DEBUG_print("   provided_data_sectors 0x%llx (%llu)\n", (unsigned long long)ic->provided_data_sectors,
3449                     (unsigned long long)ic->provided_data_sectors);
3450         DEBUG_print("   log2_buffer_sectors %u\n", ic->log2_buffer_sectors);
3451
3452         if (recalculate && !(ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))) {
3453                 ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
3454                 ic->sb->recalc_sector = cpu_to_le64(0);
3455         }
3456
3457         if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
3458                 if (!ic->internal_hash) {
3459                         r = -EINVAL;
3460                         ti->error = "Recalculate is only valid with internal hash";
3461                         goto bad;
3462                 }
3463                 ic->recalc_wq = alloc_workqueue("dm-integrity-recalc", WQ_MEM_RECLAIM, 1);
3464                 if (!ic->recalc_wq ) {
3465                         ti->error = "Cannot allocate workqueue";
3466                         r = -ENOMEM;
3467                         goto bad;
3468                 }
3469                 INIT_WORK(&ic->recalc_work, integrity_recalc);
3470                 ic->recalc_buffer = vmalloc(RECALC_SECTORS << SECTOR_SHIFT);
3471                 if (!ic->recalc_buffer) {
3472                         ti->error = "Cannot allocate buffer for recalculating";
3473                         r = -ENOMEM;
3474                         goto bad;
3475                 }
3476                 ic->recalc_tags = kvmalloc_array(RECALC_SECTORS >> ic->sb->log2_sectors_per_block,
3477                                                  ic->tag_size, GFP_KERNEL);
3478                 if (!ic->recalc_tags) {
3479                         ti->error = "Cannot allocate tags for recalculating";
3480                         r = -ENOMEM;
3481                         goto bad;
3482                 }
3483         }
3484
3485         ic->bufio = dm_bufio_client_create(ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev,
3486                         1U << (SECTOR_SHIFT + ic->log2_buffer_sectors), 1, 0, NULL, NULL);
3487         if (IS_ERR(ic->bufio)) {
3488                 r = PTR_ERR(ic->bufio);
3489                 ti->error = "Cannot initialize dm-bufio";
3490                 ic->bufio = NULL;
3491                 goto bad;
3492         }
3493         dm_bufio_set_sector_offset(ic->bufio, ic->start + ic->initial_sectors);
3494
3495         if (ic->mode != 'R') {
3496                 r = create_journal(ic, &ti->error);
3497                 if (r)
3498                         goto bad;
3499         }
3500
3501         if (should_write_sb) {
3502                 int r;
3503
3504                 init_journal(ic, 0, ic->journal_sections, 0);
3505                 r = dm_integrity_failed(ic);
3506                 if (unlikely(r)) {
3507                         ti->error = "Error initializing journal";
3508                         goto bad;
3509                 }
3510                 r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
3511                 if (r) {
3512                         ti->error = "Error initializing superblock";
3513                         goto bad;
3514                 }
3515                 ic->just_formatted = true;
3516         }
3517
3518         if (!ic->meta_dev) {
3519                 r = dm_set_target_max_io_len(ti, 1U << ic->sb->log2_interleave_sectors);
3520                 if (r)
3521                         goto bad;
3522         }
3523
3524         if (!ic->internal_hash)
3525                 dm_integrity_set(ti, ic);
3526
3527         ti->num_flush_bios = 1;
3528         ti->flush_supported = true;
3529
3530         return 0;
3531 bad:
3532         dm_integrity_dtr(ti);
3533         return r;
3534 }
3535
3536 static void dm_integrity_dtr(struct dm_target *ti)
3537 {
3538         struct dm_integrity_c *ic = ti->private;
3539
3540         BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
3541         BUG_ON(!list_empty(&ic->wait_list));
3542
3543         if (ic->metadata_wq)
3544                 destroy_workqueue(ic->metadata_wq);
3545         if (ic->wait_wq)
3546                 destroy_workqueue(ic->wait_wq);
3547         if (ic->commit_wq)
3548                 destroy_workqueue(ic->commit_wq);
3549         if (ic->writer_wq)
3550                 destroy_workqueue(ic->writer_wq);
3551         if (ic->recalc_wq)
3552                 destroy_workqueue(ic->recalc_wq);
3553         if (ic->recalc_buffer)
3554                 vfree(ic->recalc_buffer);
3555         if (ic->recalc_tags)
3556                 kvfree(ic->recalc_tags);
3557         if (ic->bufio)
3558                 dm_bufio_client_destroy(ic->bufio);
3559         mempool_exit(&ic->journal_io_mempool);
3560         if (ic->io)
3561                 dm_io_client_destroy(ic->io);
3562         if (ic->dev)
3563                 dm_put_device(ti, ic->dev);
3564         if (ic->meta_dev)
3565                 dm_put_device(ti, ic->meta_dev);
3566         dm_integrity_free_page_list(ic, ic->journal);
3567         dm_integrity_free_page_list(ic, ic->journal_io);
3568         dm_integrity_free_page_list(ic, ic->journal_xor);
3569         if (ic->journal_scatterlist)
3570                 dm_integrity_free_journal_scatterlist(ic, ic->journal_scatterlist);
3571         if (ic->journal_io_scatterlist)
3572                 dm_integrity_free_journal_scatterlist(ic, ic->journal_io_scatterlist);
3573         if (ic->sk_requests) {
3574                 unsigned i;
3575
3576                 for (i = 0; i < ic->journal_sections; i++) {
3577                         struct skcipher_request *req = ic->sk_requests[i];
3578                         if (req) {
3579                                 kzfree(req->iv);
3580                                 skcipher_request_free(req);
3581                         }
3582                 }
3583                 kvfree(ic->sk_requests);
3584         }
3585         kvfree(ic->journal_tree);
3586         if (ic->sb)
3587                 free_pages_exact(ic->sb, SB_SECTORS << SECTOR_SHIFT);
3588
3589         if (ic->internal_hash)
3590                 crypto_free_shash(ic->internal_hash);
3591         free_alg(&ic->internal_hash_alg);
3592
3593         if (ic->journal_crypt)
3594                 crypto_free_skcipher(ic->journal_crypt);
3595         free_alg(&ic->journal_crypt_alg);
3596
3597         if (ic->journal_mac)
3598                 crypto_free_shash(ic->journal_mac);
3599         free_alg(&ic->journal_mac_alg);
3600
3601         kfree(ic);
3602 }
3603
3604 static struct target_type integrity_target = {
3605         .name                   = "integrity",
3606         .version                = {1, 2, 0},
3607         .module                 = THIS_MODULE,
3608         .features               = DM_TARGET_SINGLETON | DM_TARGET_INTEGRITY,
3609         .ctr                    = dm_integrity_ctr,
3610         .dtr                    = dm_integrity_dtr,
3611         .map                    = dm_integrity_map,
3612         .postsuspend            = dm_integrity_postsuspend,
3613         .resume                 = dm_integrity_resume,
3614         .status                 = dm_integrity_status,
3615         .iterate_devices        = dm_integrity_iterate_devices,
3616         .io_hints               = dm_integrity_io_hints,
3617 };
3618
3619 int __init dm_integrity_init(void)
3620 {
3621         int r;
3622
3623         journal_io_cache = kmem_cache_create("integrity_journal_io",
3624                                              sizeof(struct journal_io), 0, 0, NULL);
3625         if (!journal_io_cache) {
3626                 DMERR("can't allocate journal io cache");
3627                 return -ENOMEM;
3628         }
3629
3630         r = dm_register_target(&integrity_target);
3631
3632         if (r < 0)
3633                 DMERR("register failed %d", r);
3634
3635         return r;
3636 }
3637
3638 void dm_integrity_exit(void)
3639 {
3640         dm_unregister_target(&integrity_target);
3641         kmem_cache_destroy(journal_io_cache);
3642 }
3643
3644 module_init(dm_integrity_init);
3645 module_exit(dm_integrity_exit);
3646
3647 MODULE_AUTHOR("Milan Broz");
3648 MODULE_AUTHOR("Mikulas Patocka");
3649 MODULE_DESCRIPTION(DM_NAME " target for integrity tags extension");
3650 MODULE_LICENSE("GPL");