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 "dm-bio-record.h"
  10
  11 #include <linux/compiler.h>
  12 #include <linux/module.h>
  13 #include <linux/device-mapper.h>
  14 #include <linux/dm-io.h>
  15 #include <linux/vmalloc.h>
  16 #include <linux/sort.h>
  17 #include <linux/rbtree.h>
  18 #include <linux/delay.h>
  19 #include <linux/random.h>
  20 #include <crypto/hash.h>
  21 #include <crypto/skcipher.h>
  22 #include <linux/async_tx.h>
  23 #include <linux/dm-bufio.h>
  24
  25 #define DM_MSG_PREFIX "integrity"
  26
  27 #define DEFAULT_INTERLEAVE_SECTORS      32768
  28 #define DEFAULT_JOURNAL_SIZE_FACTOR     7
  29 #define DEFAULT_BUFFER_SECTORS          128
  30 #define DEFAULT_JOURNAL_WATERMARK       50
  31 #define DEFAULT_SYNC_MSEC               10000
  32 #define DEFAULT_MAX_JOURNAL_SECTORS     131072
  33 #define MIN_LOG2_INTERLEAVE_SECTORS     3
  34 #define MAX_LOG2_INTERLEAVE_SECTORS     31
  35 #define METADATA_WORKQUEUE_MAX_ACTIVE   16
  36 #define RECALC_SECTORS                  8192
  37 #define RECALC_WRITE_SUPER              16
  38
  39 /*
  40  * Warning - DEBUG_PRINT prints security-sensitive data to the log,
  41  * so it should not be enabled in the official kernel
  42  */
  43 //#define DEBUG_PRINT
  44 //#define INTERNAL_VERIFY
  45
  46 /*
  47  * On disk structures
  48  */
  49
  50 #define SB_MAGIC                        "integrt"
  51 #define SB_VERSION_1                    1
  52 #define SB_VERSION_2                    2
  53 #define SB_SECTORS                      8
  54 #define MAX_SECTORS_PER_BLOCK           8
  55
  56 struct superblock {
  57         __u8 magic[8];
  58         __u8 version;
  59         __u8 log2_interleave_sectors;
  60         __u16 integrity_tag_size;
  61         __u32 journal_sections;
  62         __u64 provided_data_sectors;    /* userspace uses this value */
  63         __u32 flags;
  64         __u8 log2_sectors_per_block;
  65         __u8 pad[3];
  66         __u64 recalc_sector;
  67 };
  68
  69 #define SB_FLAG_HAVE_JOURNAL_MAC        0x1
  70 #define SB_FLAG_RECALCULATING           0x2
  71
  72 #define JOURNAL_ENTRY_ROUNDUP           8
  73
  74 typedef __u64 commit_id_t;
  75 #define JOURNAL_MAC_PER_SECTOR          8
  76
  77 struct journal_entry {
  78         union {
  79                 struct {
  80                         __u32 sector_lo;
  81                         __u32 sector_hi;
  82                 } s;
  83                 __u64 sector;
  84         } u;
  85         commit_id_t last_bytes[0];
  86         /* __u8 tag[0]; */
  87 };
  88
  89 #define journal_entry_tag(ic, je)               ((__u8 *)&(je)->last_bytes[(ic)->sectors_per_block])
  90
  91 #if BITS_PER_LONG == 64
  92 #define journal_entry_set_sector(je, x)         do { smp_wmb(); WRITE_ONCE((je)->u.sector, cpu_to_le64(x)); } while (0)
  93 #define journal_entry_get_sector(je)            le64_to_cpu((je)->u.sector)
  94 #elif defined(CONFIG_LBDAF)
  95 #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)
  96 #define journal_entry_get_sector(je)            le64_to_cpu((je)->u.sector)
  97 #else
  98 #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)
  99 #define journal_entry_get_sector(je)            le32_to_cpu((je)->u.s.sector_lo)
 100 #endif
 101 #define journal_entry_is_unused(je)             ((je)->u.s.sector_hi == cpu_to_le32(-1))
 102 #define journal_entry_set_unused(je)            do { ((je)->u.s.sector_hi = cpu_to_le32(-1)); } while (0)
 103 #define journal_entry_is_inprogress(je)         ((je)->u.s.sector_hi == cpu_to_le32(-2))
 104 #define journal_entry_set_inprogress(je)        do { ((je)->u.s.sector_hi = cpu_to_le32(-2)); } while (0)
 105
 106 #define JOURNAL_BLOCK_SECTORS           8
 107 #define JOURNAL_SECTOR_DATA             ((1 << SECTOR_SHIFT) - sizeof(commit_id_t))
 108 #define JOURNAL_MAC_SIZE                (JOURNAL_MAC_PER_SECTOR * JOURNAL_BLOCK_SECTORS)
 109
 110 struct journal_sector {
 111         __u8 entries[JOURNAL_SECTOR_DATA - JOURNAL_MAC_PER_SECTOR];
 112         __u8 mac[JOURNAL_MAC_PER_SECTOR];
 113         commit_id_t commit_id;
 114 };
 115
 116 #define MAX_TAG_SIZE                    (JOURNAL_SECTOR_DATA - JOURNAL_MAC_PER_SECTOR - offsetof(struct journal_entry, last_bytes[MAX_SECTORS_PER_BLOCK]))
 117
 118 #define METADATA_PADDING_SECTORS        8
 119
 120 #define N_COMMIT_IDS                    4
 121
 122 static unsigned char prev_commit_seq(unsigned char seq)
 123 {
 124         return (seq + N_COMMIT_IDS - 1) % N_COMMIT_IDS;
 125 }
 126
 127 static unsigned char next_commit_seq(unsigned char seq)
 128 {
 129         return (seq + 1) % N_COMMIT_IDS;
 130 }
 131
 132 /*
 133  * In-memory structures
 134  */
 135
 136 struct journal_node {
 137         struct rb_node node;
 138         sector_t sector;
 139 };
 140
 141 struct alg_spec {
 142         char *alg_string;
 143         char *key_string;
 144         __u8 *key;
 145         unsigned key_size;
 146 };
 147
 148 struct dm_integrity_c {
 149         struct dm_dev *dev;
 150         struct dm_dev *meta_dev;
 151         unsigned tag_size;
 152         __s8 log2_tag_size;
 153         sector_t start;
 154         mempool_t journal_io_mempool;
 155         struct dm_io_client *io;
 156         struct dm_bufio_client *bufio;
 157         struct workqueue_struct *metadata_wq;
 158         struct superblock *sb;
 159         unsigned journal_pages;
 160         struct page_list *journal;
 161         struct page_list *journal_io;
 162         struct page_list *journal_xor;
 163
 164         struct crypto_skcipher *journal_crypt;
 165         struct scatterlist **journal_scatterlist;
 166         struct scatterlist **journal_io_scatterlist;
 167         struct skcipher_request **sk_requests;
 168
 169         struct crypto_shash *journal_mac;
 170
 171         struct journal_node *journal_tree;
 172         struct rb_root journal_tree_root;
 173
 174         sector_t provided_data_sectors;
 175
 176         unsigned short journal_entry_size;
 177         unsigned char journal_entries_per_sector;
 178         unsigned char journal_section_entries;
 179         unsigned short journal_section_sectors;
 180         unsigned journal_sections;
 181         unsigned journal_entries;
 182         sector_t data_device_sectors;
 183         sector_t meta_device_sectors;
 184         unsigned initial_sectors;
 185         unsigned metadata_run;
 186         __s8 log2_metadata_run;
 187         __u8 log2_buffer_sectors;
 188         __u8 sectors_per_block;
 189
 190         unsigned char mode;
 191
 192         int failed;
 193
 194         struct crypto_shash *internal_hash;
 195
 196         struct dm_target *ti;
 197
 198         /* these variables are locked with endio_wait.lock */
 199         struct rb_root in_progress;
 200         struct list_head wait_list;
 201         wait_queue_head_t endio_wait;
 202         struct workqueue_struct *wait_wq;
 203         struct workqueue_struct *offload_wq;
 204
 205         unsigned char commit_seq;
 206         commit_id_t commit_ids[N_COMMIT_IDS];
 207
 208         unsigned committed_section;
 209         unsigned n_committed_sections;
 210
 211         unsigned uncommitted_section;
 212         unsigned n_uncommitted_sections;
 213
 214         unsigned free_section;
 215         unsigned char free_section_entry;
 216         unsigned free_sectors;
 217
 218         unsigned free_sectors_threshold;
 219
 220         struct workqueue_struct *commit_wq;
 221         struct work_struct commit_work;
 222
 223         struct workqueue_struct *writer_wq;
 224         struct work_struct writer_work;
 225
 226         struct workqueue_struct *recalc_wq;
 227         struct work_struct recalc_work;
 228         u8 *recalc_buffer;
 229         u8 *recalc_tags;
 230
 231         struct bio_list flush_bio_list;
 232
 233         unsigned long autocommit_jiffies;
 234         struct timer_list autocommit_timer;
 235         unsigned autocommit_msec;
 236
 237         wait_queue_head_t copy_to_journal_wait;
 238
 239         struct completion crypto_backoff;
 240
 241         bool journal_uptodate;
 242         bool just_formatted;
 243         bool legacy_recalculate;
 244
 245         struct alg_spec internal_hash_alg;
 246         struct alg_spec journal_crypt_alg;
 247         struct alg_spec journal_mac_alg;
 248
 249         atomic64_t number_of_mismatches;
 250 };
 251
 252 struct dm_integrity_range {
 253         sector_t logical_sector;
 254         unsigned n_sectors;
 255         bool waiting;
 256         union {
 257                 struct rb_node node;
 258                 struct {
 259                         struct task_struct *task;
 260                         struct list_head wait_entry;
 261                 };
 262         };
 263 };
 264
 265 struct dm_integrity_io {
 266         struct work_struct work;
 267
 268         struct dm_integrity_c *ic;
 269         bool write;
 270         bool fua;
 271
 272         struct dm_integrity_range range;
 273
 274         sector_t metadata_block;
 275         unsigned metadata_offset;
 276
 277         atomic_t in_flight;
 278         blk_status_t bi_status;
 279
 280         struct completion *completion;
 281
 282         struct dm_bio_details bio_details;
 283 };
 284
 285 struct journal_completion {
 286         struct dm_integrity_c *ic;
 287         atomic_t in_flight;
 288         struct completion comp;
 289 };
 290
 291 struct journal_io {
 292         struct dm_integrity_range range;
 293         struct journal_completion *comp;
 294 };
 295
 296 static struct kmem_cache *journal_io_cache;
 297
 298 #define JOURNAL_IO_MEMPOOL      32
 299
 300 #ifdef DEBUG_PRINT
 301 #define DEBUG_print(x, ...)     printk(KERN_DEBUG x, ##__VA_ARGS__)
 302 static void __DEBUG_bytes(__u8 *bytes, size_t len, const char *msg, ...)
 303 {
 304         va_list args;
 305         va_start(args, msg);
 306         vprintk(msg, args);
 307         va_end(args);
 308         if (len)
 309                 pr_cont(":");
 310         while (len) {
 311                 pr_cont(" %02x", *bytes);
 312                 bytes++;
 313                 len--;
 314         }
 315         pr_cont("\n");
 316 }
 317 #define DEBUG_bytes(bytes, len, msg, ...)       __DEBUG_bytes(bytes, len, KERN_DEBUG msg, ##__VA_ARGS__)
 318 #else
 319 #define DEBUG_print(x, ...)                     do { } while (0)
 320 #define DEBUG_bytes(bytes, len, msg, ...)       do { } while (0)
 321 #endif
 322
 323 /*
 324  * DM Integrity profile, protection is performed layer above (dm-crypt)
 325  */
 326 static const struct blk_integrity_profile dm_integrity_profile = {
 327         .name                   = "DM-DIF-EXT-TAG",
 328         .generate_fn            = NULL,
 329         .verify_fn              = NULL,
 330 };
 331
 332 static void dm_integrity_map_continue(struct dm_integrity_io *dio, bool from_map);
 333 static void integrity_bio_wait(struct work_struct *w);
 334 static void dm_integrity_dtr(struct dm_target *ti);
 335
 336 static void dm_integrity_io_error(struct dm_integrity_c *ic, const char *msg, int err)
 337 {
 338         if (err == -EILSEQ)
 339                 atomic64_inc(&ic->number_of_mismatches);
 340         if (!cmpxchg(&ic->failed, 0, err))
 341                 DMERR("Error on %s: %d", msg, err);
 342 }
 343
 344 static int dm_integrity_failed(struct dm_integrity_c *ic)
 345 {
 346         return READ_ONCE(ic->failed);
 347 }
 348
 349 static bool dm_integrity_disable_recalculate(struct dm_integrity_c *ic)
 350 {
 351         if ((ic->internal_hash_alg.key || ic->journal_mac_alg.key) &&
 352             !ic->legacy_recalculate)
 353                 return true;
 354         return false;
 355 }
 356
 357 static commit_id_t dm_integrity_commit_id(struct dm_integrity_c *ic, unsigned i,
 358                                           unsigned j, unsigned char seq)
 359 {
 360         /*
 361          * Xor the number with section and sector, so that if a piece of
 362          * journal is written at wrong place, it is detected.
 363          */
 364         return ic->commit_ids[seq] ^ cpu_to_le64(((__u64)i << 32) ^ j);
 365 }
 366
 367 static void get_area_and_offset(struct dm_integrity_c *ic, sector_t data_sector,
 368                                 sector_t *area, sector_t *offset)
 369 {
 370         if (!ic->meta_dev) {
 371                 __u8 log2_interleave_sectors = ic->sb->log2_interleave_sectors;
 372                 *area = data_sector >> log2_interleave_sectors;
 373                 *offset = (unsigned)data_sector & ((1U << log2_interleave_sectors) - 1);
 374         } else {
 375                 *area = 0;
 376                 *offset = data_sector;
 377         }
 378 }
 379
 380 #define sector_to_block(ic, n)                                          \
 381 do {                                                                    \
 382         BUG_ON((n) & (unsigned)((ic)->sectors_per_block - 1));          \
 383         (n) >>= (ic)->sb->log2_sectors_per_block;                       \
 384 } while (0)
 385
 386 static __u64 get_metadata_sector_and_offset(struct dm_integrity_c *ic, sector_t area,
 387                                             sector_t offset, unsigned *metadata_offset)
 388 {
 389         __u64 ms;
 390         unsigned mo;
 391
 392         ms = area << ic->sb->log2_interleave_sectors;
 393         if (likely(ic->log2_metadata_run >= 0))
 394                 ms += area << ic->log2_metadata_run;
 395         else
 396                 ms += area * ic->metadata_run;
 397         ms >>= ic->log2_buffer_sectors;
 398
 399         sector_to_block(ic, offset);
 400
 401         if (likely(ic->log2_tag_size >= 0)) {
 402                 ms += offset >> (SECTOR_SHIFT + ic->log2_buffer_sectors - ic->log2_tag_size);
 403                 mo = (offset << ic->log2_tag_size) & ((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - 1);
 404         } else {
 405                 ms += (__u64)offset * ic->tag_size >> (SECTOR_SHIFT + ic->log2_buffer_sectors);
 406                 mo = (offset * ic->tag_size) & ((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - 1);
 407         }
 408         *metadata_offset = mo;
 409         return ms;
 410 }
 411
 412 static sector_t get_data_sector(struct dm_integrity_c *ic, sector_t area, sector_t offset)
 413 {
 414         sector_t result;
 415
 416         if (ic->meta_dev)
 417                 return offset;
 418
 419         result = area << ic->sb->log2_interleave_sectors;
 420         if (likely(ic->log2_metadata_run >= 0))
 421                 result += (area + 1) << ic->log2_metadata_run;
 422         else
 423                 result += (area + 1) * ic->metadata_run;
 424
 425         result += (sector_t)ic->initial_sectors + offset;
 426         result += ic->start;
 427
 428         return result;
 429 }
 430
 431 static void wraparound_section(struct dm_integrity_c *ic, unsigned *sec_ptr)
 432 {
 433         if (unlikely(*sec_ptr >= ic->journal_sections))
 434                 *sec_ptr -= ic->journal_sections;
 435 }
 436
 437 static void sb_set_version(struct dm_integrity_c *ic)
 438 {
 439         if (ic->meta_dev || ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
 440                 ic->sb->version = SB_VERSION_2;
 441         else
 442                 ic->sb->version = SB_VERSION_1;
 443 }
 444
 445 static int sync_rw_sb(struct dm_integrity_c *ic, int op, int op_flags)
 446 {
 447         struct dm_io_request io_req;
 448         struct dm_io_region io_loc;
 449
 450         io_req.bi_op = op;
 451         io_req.bi_op_flags = op_flags;
 452         io_req.mem.type = DM_IO_KMEM;
 453         io_req.mem.ptr.addr = ic->sb;
 454         io_req.notify.fn = NULL;
 455         io_req.client = ic->io;
 456         io_loc.bdev = ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev;
 457         io_loc.sector = ic->start;
 458         io_loc.count = SB_SECTORS;
 459
 460         return dm_io(&io_req, 1, &io_loc, NULL);
 461 }
 462
 463 static void access_journal_check(struct dm_integrity_c *ic, unsigned section, unsigned offset,
 464                                  bool e, const char *function)
 465 {
 466 #if defined(CONFIG_DM_DEBUG) || defined(INTERNAL_VERIFY)
 467         unsigned limit = e ? ic->journal_section_entries : ic->journal_section_sectors;
 468
 469         if (unlikely(section >= ic->journal_sections) ||
 470             unlikely(offset >= limit)) {
 471                 printk(KERN_CRIT "%s: invalid access at (%u,%u), limit (%u,%u)\n",
 472                         function, section, offset, ic->journal_sections, limit);
 473                 BUG();
 474         }
 475 #endif
 476 }
 477
 478 static void page_list_location(struct dm_integrity_c *ic, unsigned section, unsigned offset,
 479                                unsigned *pl_index, unsigned *pl_offset)
 480 {
 481         unsigned sector;
 482
 483         access_journal_check(ic, section, offset, false, "page_list_location");
 484
 485         sector = section * ic->journal_section_sectors + offset;
 486
 487         *pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
 488         *pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
 489 }
 490
 491 static struct journal_sector *access_page_list(struct dm_integrity_c *ic, struct page_list *pl,
 492                                                unsigned section, unsigned offset, unsigned *n_sectors)
 493 {
 494         unsigned pl_index, pl_offset;
 495         char *va;
 496
 497         page_list_location(ic, section, offset, &pl_index, &pl_offset);
 498
 499         if (n_sectors)
 500                 *n_sectors = (PAGE_SIZE - pl_offset) >> SECTOR_SHIFT;
 501
 502         va = lowmem_page_address(pl[pl_index].page);
 503
 504         return (struct journal_sector *)(va + pl_offset);
 505 }
 506
 507 static struct journal_sector *access_journal(struct dm_integrity_c *ic, unsigned section, unsigned offset)
 508 {
 509         return access_page_list(ic, ic->journal, section, offset, NULL);
 510 }
 511
 512 static struct journal_entry *access_journal_entry(struct dm_integrity_c *ic, unsigned section, unsigned n)
 513 {
 514         unsigned rel_sector, offset;
 515         struct journal_sector *js;
 516
 517         access_journal_check(ic, section, n, true, "access_journal_entry");
 518
 519         rel_sector = n % JOURNAL_BLOCK_SECTORS;
 520         offset = n / JOURNAL_BLOCK_SECTORS;
 521
 522         js = access_journal(ic, section, rel_sector);
 523         return (struct journal_entry *)((char *)js + offset * ic->journal_entry_size);
 524 }
 525
 526 static struct journal_sector *access_journal_data(struct dm_integrity_c *ic, unsigned section, unsigned n)
 527 {
 528         n <<= ic->sb->log2_sectors_per_block;
 529
 530         n += JOURNAL_BLOCK_SECTORS;
 531
 532         access_journal_check(ic, section, n, false, "access_journal_data");
 533
 534         return access_journal(ic, section, n);
 535 }
 536
 537 static void section_mac(struct dm_integrity_c *ic, unsigned section, __u8 result[JOURNAL_MAC_SIZE])
 538 {
 539         SHASH_DESC_ON_STACK(desc, ic->journal_mac);
 540         int r;
 541         unsigned j, size;
 542
 543         desc->tfm = ic->journal_mac;
 544         desc->flags = 0;
 545
 546         r = crypto_shash_init(desc);
 547         if (unlikely(r)) {
 548                 dm_integrity_io_error(ic, "crypto_shash_init", r);
 549                 goto err;
 550         }
 551
 552         for (j = 0; j < ic->journal_section_entries; j++) {
 553                 struct journal_entry *je = access_journal_entry(ic, section, j);
 554                 r = crypto_shash_update(desc, (__u8 *)&je->u.sector, sizeof je->u.sector);
 555                 if (unlikely(r)) {
 556                         dm_integrity_io_error(ic, "crypto_shash_update", r);
 557                         goto err;
 558                 }
 559         }
 560
 561         size = crypto_shash_digestsize(ic->journal_mac);
 562
 563         if (likely(size <= JOURNAL_MAC_SIZE)) {
 564                 r = crypto_shash_final(desc, result);
 565                 if (unlikely(r)) {
 566                         dm_integrity_io_error(ic, "crypto_shash_final", r);
 567                         goto err;
 568                 }
 569                 memset(result + size, 0, JOURNAL_MAC_SIZE - size);
 570         } else {
 571                 __u8 digest[size];
 572                 r = crypto_shash_final(desc, digest);
 573                 if (unlikely(r)) {
 574                         dm_integrity_io_error(ic, "crypto_shash_final", r);
 575                         goto err;
 576                 }
 577                 memcpy(result, digest, JOURNAL_MAC_SIZE);
 578         }
 579
 580         return;
 581 err:
 582         memset(result, 0, JOURNAL_MAC_SIZE);
 583 }
 584
 585 static void rw_section_mac(struct dm_integrity_c *ic, unsigned section, bool wr)
 586 {
 587         __u8 result[JOURNAL_MAC_SIZE];
 588         unsigned j;
 589
 590         if (!ic->journal_mac)
 591                 return;
 592
 593         section_mac(ic, section, result);
 594
 595         for (j = 0; j < JOURNAL_BLOCK_SECTORS; j++) {
 596                 struct journal_sector *js = access_journal(ic, section, j);
 597
 598                 if (likely(wr))
 599                         memcpy(&js->mac, result + (j * JOURNAL_MAC_PER_SECTOR), JOURNAL_MAC_PER_SECTOR);
 600                 else {
 601                         if (memcmp(&js->mac, result + (j * JOURNAL_MAC_PER_SECTOR), JOURNAL_MAC_PER_SECTOR))
 602                                 dm_integrity_io_error(ic, "journal mac", -EILSEQ);
 603                 }
 604         }
 605 }
 606
 607 static void complete_journal_op(void *context)
 608 {
 609         struct journal_completion *comp = context;
 610         BUG_ON(!atomic_read(&comp->in_flight));
 611         if (likely(atomic_dec_and_test(&comp->in_flight)))
 612                 complete(&comp->comp);
 613 }
 614
 615 static void xor_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
 616                         unsigned n_sections, struct journal_completion *comp)
 617 {
 618         struct async_submit_ctl submit;
 619         size_t n_bytes = (size_t)(n_sections * ic->journal_section_sectors) << SECTOR_SHIFT;
 620         unsigned pl_index, pl_offset, section_index;
 621         struct page_list *source_pl, *target_pl;
 622
 623         if (likely(encrypt)) {
 624                 source_pl = ic->journal;
 625                 target_pl = ic->journal_io;
 626         } else {
 627                 source_pl = ic->journal_io;
 628                 target_pl = ic->journal;
 629         }
 630
 631         page_list_location(ic, section, 0, &pl_index, &pl_offset);
 632
 633         atomic_add(roundup(pl_offset + n_bytes, PAGE_SIZE) >> PAGE_SHIFT, &comp->in_flight);
 634
 635         init_async_submit(&submit, ASYNC_TX_XOR_ZERO_DST, NULL, complete_journal_op, comp, NULL);
 636
 637         section_index = pl_index;
 638
 639         do {
 640                 size_t this_step;
 641                 struct page *src_pages[2];
 642                 struct page *dst_page;
 643
 644                 while (unlikely(pl_index == section_index)) {
 645                         unsigned dummy;
 646                         if (likely(encrypt))
 647                                 rw_section_mac(ic, section, true);
 648                         section++;
 649                         n_sections--;
 650                         if (!n_sections)
 651                                 break;
 652                         page_list_location(ic, section, 0, &section_index, &dummy);
 653                 }
 654
 655                 this_step = min(n_bytes, (size_t)PAGE_SIZE - pl_offset);
 656                 dst_page = target_pl[pl_index].page;
 657                 src_pages[0] = source_pl[pl_index].page;
 658                 src_pages[1] = ic->journal_xor[pl_index].page;
 659
 660                 async_xor(dst_page, src_pages, pl_offset, 2, this_step, &submit);
 661
 662                 pl_index++;
 663                 pl_offset = 0;
 664                 n_bytes -= this_step;
 665         } while (n_bytes);
 666
 667         BUG_ON(n_sections);
 668
 669         async_tx_issue_pending_all();
 670 }
 671
 672 static void complete_journal_encrypt(struct crypto_async_request *req, int err)
 673 {
 674         struct journal_completion *comp = req->data;
 675         if (unlikely(err)) {
 676                 if (likely(err == -EINPROGRESS)) {
 677                         complete(&comp->ic->crypto_backoff);
 678                         return;
 679                 }
 680                 dm_integrity_io_error(comp->ic, "asynchronous encrypt", err);
 681         }
 682         complete_journal_op(comp);
 683 }
 684
 685 static bool do_crypt(bool encrypt, struct skcipher_request *req, struct journal_completion *comp)
 686 {
 687         int r;
 688         skcipher_request_set_callback(req, CRYPTO_TFM_REQ_MAY_BACKLOG,
 689                                       complete_journal_encrypt, comp);
 690         if (likely(encrypt))
 691                 r = crypto_skcipher_encrypt(req);
 692         else
 693                 r = crypto_skcipher_decrypt(req);
 694         if (likely(!r))
 695                 return false;
 696         if (likely(r == -EINPROGRESS))
 697                 return true;
 698         if (likely(r == -EBUSY)) {
 699                 wait_for_completion(&comp->ic->crypto_backoff);
 700                 reinit_completion(&comp->ic->crypto_backoff);
 701                 return true;
 702         }
 703         dm_integrity_io_error(comp->ic, "encrypt", r);
 704         return false;
 705 }
 706
 707 static void crypt_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
 708                           unsigned n_sections, struct journal_completion *comp)
 709 {
 710         struct scatterlist **source_sg;
 711         struct scatterlist **target_sg;
 712
 713         atomic_add(2, &comp->in_flight);
 714
 715         if (likely(encrypt)) {
 716                 source_sg = ic->journal_scatterlist;
 717                 target_sg = ic->journal_io_scatterlist;
 718         } else {
 719                 source_sg = ic->journal_io_scatterlist;
 720                 target_sg = ic->journal_scatterlist;
 721         }
 722
 723         do {
 724                 struct skcipher_request *req;
 725                 unsigned ivsize;
 726                 char *iv;
 727
 728                 if (likely(encrypt))
 729                         rw_section_mac(ic, section, true);
 730
 731                 req = ic->sk_requests[section];
 732                 ivsize = crypto_skcipher_ivsize(ic->journal_crypt);
 733                 iv = req->iv;
 734
 735                 memcpy(iv, iv + ivsize, ivsize);
 736
 737                 req->src = source_sg[section];
 738                 req->dst = target_sg[section];
 739
 740                 if (unlikely(do_crypt(encrypt, req, comp)))
 741                         atomic_inc(&comp->in_flight);
 742
 743                 section++;
 744                 n_sections--;
 745         } while (n_sections);
 746
 747         atomic_dec(&comp->in_flight);
 748         complete_journal_op(comp);
 749 }
 750
 751 static void encrypt_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
 752                             unsigned n_sections, struct journal_completion *comp)
 753 {
 754         if (ic->journal_xor)
 755                 return xor_journal(ic, encrypt, section, n_sections, comp);
 756         else
 757                 return crypt_journal(ic, encrypt, section, n_sections, comp);
 758 }
 759
 760 static void complete_journal_io(unsigned long error, void *context)
 761 {
 762         struct journal_completion *comp = context;
 763         if (unlikely(error != 0))
 764                 dm_integrity_io_error(comp->ic, "writing journal", -EIO);
 765         complete_journal_op(comp);
 766 }
 767
 768 static void rw_journal(struct dm_integrity_c *ic, int op, int op_flags, unsigned section,
 769                        unsigned n_sections, struct journal_completion *comp)
 770 {
 771         struct dm_io_request io_req;
 772         struct dm_io_region io_loc;
 773         unsigned sector, n_sectors, pl_index, pl_offset;
 774         int r;
 775
 776         if (unlikely(dm_integrity_failed(ic))) {
 777                 if (comp)
 778                         complete_journal_io(-1UL, comp);
 779                 return;
 780         }
 781
 782         sector = section * ic->journal_section_sectors;
 783         n_sectors = n_sections * ic->journal_section_sectors;
 784
 785         pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
 786         pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
 787
 788         io_req.bi_op = op;
 789         io_req.bi_op_flags = op_flags;
 790         io_req.mem.type = DM_IO_PAGE_LIST;
 791         if (ic->journal_io)
 792                 io_req.mem.ptr.pl = &ic->journal_io[pl_index];
 793         else
 794                 io_req.mem.ptr.pl = &ic->journal[pl_index];
 795         io_req.mem.offset = pl_offset;
 796         if (likely(comp != NULL)) {
 797                 io_req.notify.fn = complete_journal_io;
 798                 io_req.notify.context = comp;
 799         } else {
 800                 io_req.notify.fn = NULL;
 801         }
 802         io_req.client = ic->io;
 803         io_loc.bdev = ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev;
 804         io_loc.sector = ic->start + SB_SECTORS + sector;
 805         io_loc.count = n_sectors;
 806
 807         r = dm_io(&io_req, 1, &io_loc, NULL);
 808         if (unlikely(r)) {
 809                 dm_integrity_io_error(ic, op == REQ_OP_READ ? "reading journal" : "writing journal", r);
 810                 if (comp) {
 811                         WARN_ONCE(1, "asynchronous dm_io failed: %d", r);
 812                         complete_journal_io(-1UL, comp);
 813                 }
 814         }
 815 }
 816
 817 static void write_journal(struct dm_integrity_c *ic, unsigned commit_start, unsigned commit_sections)
 818 {
 819         struct journal_completion io_comp;
 820         struct journal_completion crypt_comp_1;
 821         struct journal_completion crypt_comp_2;
 822         unsigned i;
 823
 824         io_comp.ic = ic;
 825         init_completion(&io_comp.comp);
 826
 827         if (commit_start + commit_sections <= ic->journal_sections) {
 828                 io_comp.in_flight = (atomic_t)ATOMIC_INIT(1);
 829                 if (ic->journal_io) {
 830                         crypt_comp_1.ic = ic;
 831                         init_completion(&crypt_comp_1.comp);
 832                         crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
 833                         encrypt_journal(ic, true, commit_start, commit_sections, &crypt_comp_1);
 834                         wait_for_completion_io(&crypt_comp_1.comp);
 835                 } else {
 836                         for (i = 0; i < commit_sections; i++)
 837                                 rw_section_mac(ic, commit_start + i, true);
 838                 }
 839                 rw_journal(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, commit_start,
 840                            commit_sections, &io_comp);
 841         } else {
 842                 unsigned to_end;
 843                 io_comp.in_flight = (atomic_t)ATOMIC_INIT(2);
 844                 to_end = ic->journal_sections - commit_start;
 845                 if (ic->journal_io) {
 846                         crypt_comp_1.ic = ic;
 847                         init_completion(&crypt_comp_1.comp);
 848                         crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
 849                         encrypt_journal(ic, true, commit_start, to_end, &crypt_comp_1);
 850                         if (try_wait_for_completion(&crypt_comp_1.comp)) {
 851                                 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
 852                                 reinit_completion(&crypt_comp_1.comp);
 853                                 crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
 854                                 encrypt_journal(ic, true, 0, commit_sections - to_end, &crypt_comp_1);
 855                                 wait_for_completion_io(&crypt_comp_1.comp);
 856                         } else {
 857                                 crypt_comp_2.ic = ic;
 858                                 init_completion(&crypt_comp_2.comp);
 859                                 crypt_comp_2.in_flight = (atomic_t)ATOMIC_INIT(0);
 860                                 encrypt_journal(ic, true, 0, commit_sections - to_end, &crypt_comp_2);
 861                                 wait_for_completion_io(&crypt_comp_1.comp);
 862                                 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
 863                                 wait_for_completion_io(&crypt_comp_2.comp);
 864                         }
 865                 } else {
 866                         for (i = 0; i < to_end; i++)
 867                                 rw_section_mac(ic, commit_start + i, true);
 868                         rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
 869                         for (i = 0; i < commit_sections - to_end; i++)
 870                                 rw_section_mac(ic, i, true);
 871                 }
 872                 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, 0, commit_sections - to_end, &io_comp);
 873         }
 874
 875         wait_for_completion_io(&io_comp.comp);
 876 }
 877
 878 static void copy_from_journal(struct dm_integrity_c *ic, unsigned section, unsigned offset,
 879                               unsigned n_sectors, sector_t target, io_notify_fn fn, void *data)
 880 {
 881         struct dm_io_request io_req;
 882         struct dm_io_region io_loc;
 883         int r;
 884         unsigned sector, pl_index, pl_offset;
 885
 886         BUG_ON((target | n_sectors | offset) & (unsigned)(ic->sectors_per_block - 1));
 887
 888         if (unlikely(dm_integrity_failed(ic))) {
 889                 fn(-1UL, data);
 890                 return;
 891         }
 892
 893         sector = section * ic->journal_section_sectors + JOURNAL_BLOCK_SECTORS + offset;
 894
 895         pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
 896         pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
 897
 898         io_req.bi_op = REQ_OP_WRITE;
 899         io_req.bi_op_flags = 0;
 900         io_req.mem.type = DM_IO_PAGE_LIST;
 901         io_req.mem.ptr.pl = &ic->journal[pl_index];
 902         io_req.mem.offset = pl_offset;
 903         io_req.notify.fn = fn;
 904         io_req.notify.context = data;
 905         io_req.client = ic->io;
 906         io_loc.bdev = ic->dev->bdev;
 907         io_loc.sector = target;
 908         io_loc.count = n_sectors;
 909
 910         r = dm_io(&io_req, 1, &io_loc, NULL);
 911         if (unlikely(r)) {
 912                 WARN_ONCE(1, "asynchronous dm_io failed: %d", r);
 913                 fn(-1UL, data);
 914         }
 915 }
 916
 917 static bool ranges_overlap(struct dm_integrity_range *range1, struct dm_integrity_range *range2)
 918 {
 919         return range1->logical_sector < range2->logical_sector + range2->n_sectors &&
 920                range1->logical_sector + range1->n_sectors > range2->logical_sector;
 921 }
 922
 923 static bool add_new_range(struct dm_integrity_c *ic, struct dm_integrity_range *new_range, bool check_waiting)
 924 {
 925         struct rb_node **n = &ic->in_progress.rb_node;
 926         struct rb_node *parent;
 927
 928         BUG_ON((new_range->logical_sector | new_range->n_sectors) & (unsigned)(ic->sectors_per_block - 1));
 929
 930         if (likely(check_waiting)) {
 931                 struct dm_integrity_range *range;
 932                 list_for_each_entry(range, &ic->wait_list, wait_entry) {
 933                         if (unlikely(ranges_overlap(range, new_range)))
 934                                 return false;
 935                 }
 936         }
 937
 938         parent = NULL;
 939
 940         while (*n) {
 941                 struct dm_integrity_range *range = container_of(*n, struct dm_integrity_range, node);
 942
 943                 parent = *n;
 944                 if (new_range->logical_sector + new_range->n_sectors <= range->logical_sector) {
 945                         n = &range->node.rb_left;
 946                 } else if (new_range->logical_sector >= range->logical_sector + range->n_sectors) {
 947                         n = &range->node.rb_right;
 948                 } else {
 949                         return false;
 950                 }
 951         }
 952
 953         rb_link_node(&new_range->node, parent, n);
 954         rb_insert_color(&new_range->node, &ic->in_progress);
 955
 956         return true;
 957 }
 958
 959 static void remove_range_unlocked(struct dm_integrity_c *ic, struct dm_integrity_range *range)
 960 {
 961         rb_erase(&range->node, &ic->in_progress);
 962         while (unlikely(!list_empty(&ic->wait_list))) {
 963                 struct dm_integrity_range *last_range =
 964                         list_first_entry(&ic->wait_list, struct dm_integrity_range, wait_entry);
 965                 struct task_struct *last_range_task;
 966                 last_range_task = last_range->task;
 967                 list_del(&last_range->wait_entry);
 968                 if (!add_new_range(ic, last_range, false)) {
 969                         last_range->task = last_range_task;
 970                         list_add(&last_range->wait_entry, &ic->wait_list);
 971                         break;
 972                 }
 973                 last_range->waiting = false;
 974                 wake_up_process(last_range_task);
 975         }
 976 }
 977
 978 static void remove_range(struct dm_integrity_c *ic, struct dm_integrity_range *range)
 979 {
 980         unsigned long flags;
 981
 982         spin_lock_irqsave(&ic->endio_wait.lock, flags);
 983         remove_range_unlocked(ic, range);
 984         spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
 985 }
 986
 987 static void wait_and_add_new_range(struct dm_integrity_c *ic, struct dm_integrity_range *new_range)
 988 {
 989         new_range->waiting = true;
 990         list_add_tail(&new_range->wait_entry, &ic->wait_list);
 991         new_range->task = current;
 992         do {
 993                 __set_current_state(TASK_UNINTERRUPTIBLE);
 994                 spin_unlock_irq(&ic->endio_wait.lock);
 995                 io_schedule();
 996                 spin_lock_irq(&ic->endio_wait.lock);
 997         } while (unlikely(new_range->waiting));
 998 }
 999
1000 static void init_journal_node(struct journal_node *node)
1001 {
1002         RB_CLEAR_NODE(&node->node);
1003         node->sector = (sector_t)-1;
1004 }
1005
1006 static void add_journal_node(struct dm_integrity_c *ic, struct journal_node *node, sector_t sector)
1007 {
1008         struct rb_node **link;
1009         struct rb_node *parent;
1010
1011         node->sector = sector;
1012         BUG_ON(!RB_EMPTY_NODE(&node->node));
1013
1014         link = &ic->journal_tree_root.rb_node;
1015         parent = NULL;
1016
1017         while (*link) {
1018                 struct journal_node *j;
1019                 parent = *link;
1020                 j = container_of(parent, struct journal_node, node);
1021                 if (sector < j->sector)
1022                         link = &j->node.rb_left;
1023                 else
1024                         link = &j->node.rb_right;
1025         }
1026
1027         rb_link_node(&node->node, parent, link);
1028         rb_insert_color(&node->node, &ic->journal_tree_root);
1029 }
1030
1031 static void remove_journal_node(struct dm_integrity_c *ic, struct journal_node *node)
1032 {
1033         BUG_ON(RB_EMPTY_NODE(&node->node));
1034         rb_erase(&node->node, &ic->journal_tree_root);
1035         init_journal_node(node);
1036 }
1037
1038 #define NOT_FOUND       (-1U)
1039
1040 static unsigned find_journal_node(struct dm_integrity_c *ic, sector_t sector, sector_t *next_sector)
1041 {
1042         struct rb_node *n = ic->journal_tree_root.rb_node;
1043         unsigned found = NOT_FOUND;
1044         *next_sector = (sector_t)-1;
1045         while (n) {
1046                 struct journal_node *j = container_of(n, struct journal_node, node);
1047                 if (sector == j->sector) {
1048                         found = j - ic->journal_tree;
1049                 }
1050                 if (sector < j->sector) {
1051                         *next_sector = j->sector;
1052                         n = j->node.rb_left;
1053                 } else {
1054                         n = j->node.rb_right;
1055                 }
1056         }
1057
1058         return found;
1059 }
1060
1061 static bool test_journal_node(struct dm_integrity_c *ic, unsigned pos, sector_t sector)
1062 {
1063         struct journal_node *node, *next_node;
1064         struct rb_node *next;
1065
1066         if (unlikely(pos >= ic->journal_entries))
1067                 return false;
1068         node = &ic->journal_tree[pos];
1069         if (unlikely(RB_EMPTY_NODE(&node->node)))
1070                 return false;
1071         if (unlikely(node->sector != sector))
1072                 return false;
1073
1074         next = rb_next(&node->node);
1075         if (unlikely(!next))
1076                 return true;
1077
1078         next_node = container_of(next, struct journal_node, node);
1079         return next_node->sector != sector;
1080 }
1081
1082 static bool find_newer_committed_node(struct dm_integrity_c *ic, struct journal_node *node)
1083 {
1084         struct rb_node *next;
1085         struct journal_node *next_node;
1086         unsigned next_section;
1087
1088         BUG_ON(RB_EMPTY_NODE(&node->node));
1089
1090         next = rb_next(&node->node);
1091         if (unlikely(!next))
1092                 return false;
1093
1094         next_node = container_of(next, struct journal_node, node);
1095
1096         if (next_node->sector != node->sector)
1097                 return false;
1098
1099         next_section = (unsigned)(next_node - ic->journal_tree) / ic->journal_section_entries;
1100         if (next_section >= ic->committed_section &&
1101             next_section < ic->committed_section + ic->n_committed_sections)
1102                 return true;
1103         if (next_section + ic->journal_sections < ic->committed_section + ic->n_committed_sections)
1104                 return true;
1105
1106         return false;
1107 }
1108
1109 #define TAG_READ        0
1110 #define TAG_WRITE       1
1111 #define TAG_CMP         2
1112
1113 static int dm_integrity_rw_tag(struct dm_integrity_c *ic, unsigned char *tag, sector_t *metadata_block,
1114                                unsigned *metadata_offset, unsigned total_size, int op)
1115 {
1116         do {
1117                 unsigned char *data, *dp;
1118                 struct dm_buffer *b;
1119                 unsigned to_copy;
1120                 int r;
1121
1122                 r = dm_integrity_failed(ic);
1123                 if (unlikely(r))
1124                         return r;
1125
1126                 data = dm_bufio_read(ic->bufio, *metadata_block, &b);
1127                 if (unlikely(IS_ERR(data)))
1128                         return PTR_ERR(data);
1129
1130                 to_copy = min((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - *metadata_offset, total_size);
1131                 dp = data + *metadata_offset;
1132                 if (op == TAG_READ) {
1133                         memcpy(tag, dp, to_copy);
1134                 } else if (op == TAG_WRITE) {
1135                         memcpy(dp, tag, to_copy);
1136                         dm_bufio_mark_partial_buffer_dirty(b, *metadata_offset, *metadata_offset + to_copy);
1137                 } else  {
1138                         /* e.g.: op == TAG_CMP */
1139                         if (unlikely(memcmp(dp, tag, to_copy))) {
1140                                 unsigned i;
1141
1142                                 for (i = 0; i < to_copy; i++) {
1143                                         if (dp[i] != tag[i])
1144                                                 break;
1145                                         total_size--;
1146                                 }
1147                                 dm_bufio_release(b);
1148                                 return total_size;
1149                         }
1150                 }
1151                 dm_bufio_release(b);
1152
1153                 tag += to_copy;
1154                 *metadata_offset += to_copy;
1155                 if (unlikely(*metadata_offset == 1U << SECTOR_SHIFT << ic->log2_buffer_sectors)) {
1156                         (*metadata_block)++;
1157                         *metadata_offset = 0;
1158                 }
1159                 total_size -= to_copy;
1160         } while (unlikely(total_size));
1161
1162         return 0;
1163 }
1164
1165 struct flush_request {
1166         struct dm_io_request io_req;
1167         struct dm_io_region io_reg;
1168         struct dm_integrity_c *ic;
1169         struct completion comp;
1170 };
1171
1172 static void flush_notify(unsigned long error, void *fr_)
1173 {
1174         struct flush_request *fr = fr_;
1175         if (unlikely(error != 0))
1176                 dm_integrity_io_error(fr->ic, "flusing disk cache", -EIO);
1177         complete(&fr->comp);
1178 }
1179
1180 static void dm_integrity_flush_buffers(struct dm_integrity_c *ic, bool flush_data)
1181 {
1182         int r;
1183
1184         struct flush_request fr;
1185
1186         if (!ic->meta_dev)
1187                 flush_data = false;
1188         if (flush_data) {
1189                 fr.io_req.bi_op = REQ_OP_WRITE,
1190                 fr.io_req.bi_op_flags = REQ_PREFLUSH | REQ_SYNC,
1191                 fr.io_req.mem.type = DM_IO_KMEM,
1192                 fr.io_req.mem.ptr.addr = NULL,
1193                 fr.io_req.notify.fn = flush_notify,
1194                 fr.io_req.notify.context = &fr;
1195                 fr.io_req.client = dm_bufio_get_dm_io_client(ic->bufio),
1196                 fr.io_reg.bdev = ic->dev->bdev,
1197                 fr.io_reg.sector = 0,
1198                 fr.io_reg.count = 0,
1199                 fr.ic = ic;
1200                 init_completion(&fr.comp);
1201                 r = dm_io(&fr.io_req, 1, &fr.io_reg, NULL);
1202                 BUG_ON(r);
1203         }
1204
1205         r = dm_bufio_write_dirty_buffers(ic->bufio);
1206         if (unlikely(r))
1207                 dm_integrity_io_error(ic, "writing tags", r);
1208
1209         if (flush_data)
1210                 wait_for_completion(&fr.comp);
1211 }
1212
1213 static void sleep_on_endio_wait(struct dm_integrity_c *ic)
1214 {
1215         DECLARE_WAITQUEUE(wait, current);
1216         __add_wait_queue(&ic->endio_wait, &wait);
1217         __set_current_state(TASK_UNINTERRUPTIBLE);
1218         spin_unlock_irq(&ic->endio_wait.lock);
1219         io_schedule();
1220         spin_lock_irq(&ic->endio_wait.lock);
1221         __remove_wait_queue(&ic->endio_wait, &wait);
1222 }
1223
1224 static void autocommit_fn(struct timer_list *t)
1225 {
1226         struct dm_integrity_c *ic = from_timer(ic, t, autocommit_timer);
1227
1228         if (likely(!dm_integrity_failed(ic)))
1229                 queue_work(ic->commit_wq, &ic->commit_work);
1230 }
1231
1232 static void schedule_autocommit(struct dm_integrity_c *ic)
1233 {
1234         if (!timer_pending(&ic->autocommit_timer))
1235                 mod_timer(&ic->autocommit_timer, jiffies + ic->autocommit_jiffies);
1236 }
1237
1238 static void submit_flush_bio(struct dm_integrity_c *ic, struct dm_integrity_io *dio)
1239 {
1240         struct bio *bio;
1241         unsigned long flags;
1242
1243         spin_lock_irqsave(&ic->endio_wait.lock, flags);
1244         bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1245         bio_list_add(&ic->flush_bio_list, bio);
1246         spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
1247
1248         queue_work(ic->commit_wq, &ic->commit_work);
1249 }
1250
1251 static void do_endio(struct dm_integrity_c *ic, struct bio *bio)
1252 {
1253         int r = dm_integrity_failed(ic);
1254         if (unlikely(r) && !bio->bi_status)
1255                 bio->bi_status = errno_to_blk_status(r);
1256         bio_endio(bio);
1257 }
1258
1259 static void do_endio_flush(struct dm_integrity_c *ic, struct dm_integrity_io *dio)
1260 {
1261         struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1262
1263         if (unlikely(dio->fua) && likely(!bio->bi_status) && likely(!dm_integrity_failed(ic)))
1264                 submit_flush_bio(ic, dio);
1265         else
1266                 do_endio(ic, bio);
1267 }
1268
1269 static void dec_in_flight(struct dm_integrity_io *dio)
1270 {
1271         if (atomic_dec_and_test(&dio->in_flight)) {
1272                 struct dm_integrity_c *ic = dio->ic;
1273                 struct bio *bio;
1274
1275                 remove_range(ic, &dio->range);
1276
1277                 if (unlikely(dio->write))
1278                         schedule_autocommit(ic);
1279
1280                 bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1281
1282                 if (unlikely(dio->bi_status) && !bio->bi_status)
1283                         bio->bi_status = dio->bi_status;
1284                 if (likely(!bio->bi_status) && unlikely(bio_sectors(bio) != dio->range.n_sectors)) {
1285                         dio->range.logical_sector += dio->range.n_sectors;
1286                         bio_advance(bio, dio->range.n_sectors << SECTOR_SHIFT);
1287                         INIT_WORK(&dio->work, integrity_bio_wait);
1288                         queue_work(ic->offload_wq, &dio->work);
1289                         return;
1290                 }
1291                 do_endio_flush(ic, dio);
1292         }
1293 }
1294
1295 static void integrity_end_io(struct bio *bio)
1296 {
1297         struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
1298
1299         dm_bio_restore(&dio->bio_details, bio);
1300         if (bio->bi_integrity)
1301                 bio->bi_opf |= REQ_INTEGRITY;
1302
1303         if (dio->completion)
1304                 complete(dio->completion);
1305
1306         dec_in_flight(dio);
1307 }
1308
1309 static void integrity_sector_checksum(struct dm_integrity_c *ic, sector_t sector,
1310                                       const char *data, char *result)
1311 {
1312         __u64 sector_le = cpu_to_le64(sector);
1313         SHASH_DESC_ON_STACK(req, ic->internal_hash);
1314         int r;
1315         unsigned digest_size;
1316
1317         req->tfm = ic->internal_hash;
1318         req->flags = 0;
1319
1320         r = crypto_shash_init(req);
1321         if (unlikely(r < 0)) {
1322                 dm_integrity_io_error(ic, "crypto_shash_init", r);
1323                 goto failed;
1324         }
1325
1326         r = crypto_shash_update(req, (const __u8 *)&sector_le, sizeof sector_le);
1327         if (unlikely(r < 0)) {
1328                 dm_integrity_io_error(ic, "crypto_shash_update", r);
1329                 goto failed;
1330         }
1331
1332         r = crypto_shash_update(req, data, ic->sectors_per_block << SECTOR_SHIFT);
1333         if (unlikely(r < 0)) {
1334                 dm_integrity_io_error(ic, "crypto_shash_update", r);
1335                 goto failed;
1336         }
1337
1338         r = crypto_shash_final(req, result);
1339         if (unlikely(r < 0)) {
1340                 dm_integrity_io_error(ic, "crypto_shash_final", r);
1341                 goto failed;
1342         }
1343
1344         digest_size = crypto_shash_digestsize(ic->internal_hash);
1345         if (unlikely(digest_size < ic->tag_size))
1346                 memset(result + digest_size, 0, ic->tag_size - digest_size);
1347
1348         return;
1349
1350 failed:
1351         /* this shouldn't happen anyway, the hash functions have no reason to fail */
1352         get_random_bytes(result, ic->tag_size);
1353 }
1354
1355 static void integrity_metadata(struct work_struct *w)
1356 {
1357         struct dm_integrity_io *dio = container_of(w, struct dm_integrity_io, work);
1358         struct dm_integrity_c *ic = dio->ic;
1359
1360         int r;
1361
1362         if (ic->internal_hash) {
1363                 struct bvec_iter iter;
1364                 struct bio_vec bv;
1365                 unsigned digest_size = crypto_shash_digestsize(ic->internal_hash);
1366                 struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1367                 char *checksums;
1368                 unsigned extra_space = unlikely(digest_size > ic->tag_size) ? digest_size - ic->tag_size : 0;
1369                 char checksums_onstack[ic->tag_size + extra_space];
1370                 unsigned sectors_to_process = dio->range.n_sectors;
1371                 sector_t sector = dio->range.logical_sector;
1372
1373                 if (unlikely(ic->mode == 'R'))
1374                         goto skip_io;
1375
1376                 checksums = kmalloc((PAGE_SIZE >> SECTOR_SHIFT >> ic->sb->log2_sectors_per_block) * ic->tag_size + extra_space,
1377                                     GFP_NOIO | __GFP_NORETRY | __GFP_NOWARN);
1378                 if (!checksums)
1379                         checksums = checksums_onstack;
1380
1381                 __bio_for_each_segment(bv, bio, iter, dio->bio_details.bi_iter) {
1382                         unsigned pos;
1383                         char *mem, *checksums_ptr;
1384
1385 again:
1386                         mem = (char *)kmap_atomic(bv.bv_page) + bv.bv_offset;
1387                         pos = 0;
1388                         checksums_ptr = checksums;
1389                         do {
1390                                 integrity_sector_checksum(ic, sector, mem + pos, checksums_ptr);
1391                                 checksums_ptr += ic->tag_size;
1392                                 sectors_to_process -= ic->sectors_per_block;
1393                                 pos += ic->sectors_per_block << SECTOR_SHIFT;
1394                                 sector += ic->sectors_per_block;
1395                         } while (pos < bv.bv_len && sectors_to_process && checksums != checksums_onstack);
1396                         kunmap_atomic(mem);
1397
1398                         r = dm_integrity_rw_tag(ic, checksums, &dio->metadata_block, &dio->metadata_offset,
1399                                                 checksums_ptr - checksums, !dio->write ? TAG_CMP : TAG_WRITE);
1400                         if (unlikely(r)) {
1401                                 if (r > 0) {
1402                                         DMERR_LIMIT("Checksum failed at sector 0x%llx",
1403                                                     (unsigned long long)(sector - ((r + ic->tag_size - 1) / ic->tag_size)));
1404                                         r = -EILSEQ;
1405                                         atomic64_inc(&ic->number_of_mismatches);
1406                                 }
1407                                 if (likely(checksums != checksums_onstack))
1408                                         kfree(checksums);
1409                                 goto error;
1410                         }
1411
1412                         if (!sectors_to_process)
1413                                 break;
1414
1415                         if (unlikely(pos < bv.bv_len)) {
1416                                 bv.bv_offset += pos;
1417                                 bv.bv_len -= pos;
1418                                 goto again;
1419                         }
1420                 }
1421
1422                 if (likely(checksums != checksums_onstack))
1423                         kfree(checksums);
1424         } else {
1425                 struct bio_integrity_payload *bip = dio->bio_details.bi_integrity;
1426
1427                 if (bip) {
1428                         struct bio_vec biv;
1429                         struct bvec_iter iter;
1430                         unsigned data_to_process = dio->range.n_sectors;
1431                         sector_to_block(ic, data_to_process);
1432                         data_to_process *= ic->tag_size;
1433
1434                         bip_for_each_vec(biv, bip, iter) {
1435                                 unsigned char *tag;
1436                                 unsigned this_len;
1437
1438                                 BUG_ON(PageHighMem(biv.bv_page));
1439                                 tag = lowmem_page_address(biv.bv_page) + biv.bv_offset;
1440                                 this_len = min(biv.bv_len, data_to_process);
1441                                 r = dm_integrity_rw_tag(ic, tag, &dio->metadata_block, &dio->metadata_offset,
1442                                                         this_len, !dio->write ? TAG_READ : TAG_WRITE);
1443                                 if (unlikely(r))
1444                                         goto error;
1445                                 data_to_process -= this_len;
1446                                 if (!data_to_process)
1447                                         break;
1448                         }
1449                 }
1450         }
1451 skip_io:
1452         dec_in_flight(dio);
1453         return;
1454 error:
1455         dio->bi_status = errno_to_blk_status(r);
1456         dec_in_flight(dio);
1457 }
1458
1459 static int dm_integrity_map(struct dm_target *ti, struct bio *bio)
1460 {
1461         struct dm_integrity_c *ic = ti->private;
1462         struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
1463         struct bio_integrity_payload *bip;
1464
1465         sector_t area, offset;
1466
1467         dio->ic = ic;
1468         dio->bi_status = 0;
1469
1470         if (unlikely(bio->bi_opf & REQ_PREFLUSH)) {
1471                 submit_flush_bio(ic, dio);
1472                 return DM_MAPIO_SUBMITTED;
1473         }
1474
1475         dio->range.logical_sector = dm_target_offset(ti, bio->bi_iter.bi_sector);
1476         dio->write = bio_op(bio) == REQ_OP_WRITE;
1477         dio->fua = dio->write && bio->bi_opf & REQ_FUA;
1478         if (unlikely(dio->fua)) {
1479                 /*
1480                  * Don't pass down the FUA flag because we have to flush
1481                  * disk cache anyway.
1482                  */
1483                 bio->bi_opf &= ~REQ_FUA;
1484         }
1485         if (unlikely(dio->range.logical_sector + bio_sectors(bio) > ic->provided_data_sectors)) {
1486                 DMERR("Too big sector number: 0x%llx + 0x%x > 0x%llx",
1487                       (unsigned long long)dio->range.logical_sector, bio_sectors(bio),
1488                       (unsigned long long)ic->provided_data_sectors);
1489                 return DM_MAPIO_KILL;
1490         }
1491         if (unlikely((dio->range.logical_sector | bio_sectors(bio)) & (unsigned)(ic->sectors_per_block - 1))) {
1492                 DMERR("Bio not aligned on %u sectors: 0x%llx, 0x%x",
1493                       ic->sectors_per_block,
1494                       (unsigned long long)dio->range.logical_sector, bio_sectors(bio));
1495                 return DM_MAPIO_KILL;
1496         }
1497
1498         if (ic->sectors_per_block > 1) {
1499                 struct bvec_iter iter;
1500                 struct bio_vec bv;
1501                 bio_for_each_segment(bv, bio, iter) {
1502                         if (unlikely(bv.bv_len & ((ic->sectors_per_block << SECTOR_SHIFT) - 1))) {
1503                                 DMERR("Bio vector (%u,%u) is not aligned on %u-sector boundary",
1504                                         bv.bv_offset, bv.bv_len, ic->sectors_per_block);
1505                                 return DM_MAPIO_KILL;
1506                         }
1507                 }
1508         }
1509
1510         bip = bio_integrity(bio);
1511         if (!ic->internal_hash) {
1512                 if (bip) {
1513                         unsigned wanted_tag_size = bio_sectors(bio) >> ic->sb->log2_sectors_per_block;
1514                         if (ic->log2_tag_size >= 0)
1515                                 wanted_tag_size <<= ic->log2_tag_size;
1516                         else
1517                                 wanted_tag_size *= ic->tag_size;
1518                         if (unlikely(wanted_tag_size != bip->bip_iter.bi_size)) {
1519                                 DMERR("Invalid integrity data size %u, expected %u", bip->bip_iter.bi_size, wanted_tag_size);
1520                                 return DM_MAPIO_KILL;
1521                         }
1522                 }
1523         } else {
1524                 if (unlikely(bip != NULL)) {
1525                         DMERR("Unexpected integrity data when using internal hash");
1526                         return DM_MAPIO_KILL;
1527                 }
1528         }
1529
1530         if (unlikely(ic->mode == 'R') && unlikely(dio->write))
1531                 return DM_MAPIO_KILL;
1532
1533         get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
1534         dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset, &dio->metadata_offset);
1535         bio->bi_iter.bi_sector = get_data_sector(ic, area, offset);
1536
1537         dm_integrity_map_continue(dio, true);
1538         return DM_MAPIO_SUBMITTED;
1539 }
1540
1541 static bool __journal_read_write(struct dm_integrity_io *dio, struct bio *bio,
1542                                  unsigned journal_section, unsigned journal_entry)
1543 {
1544         struct dm_integrity_c *ic = dio->ic;
1545         sector_t logical_sector;
1546         unsigned n_sectors;
1547
1548         logical_sector = dio->range.logical_sector;
1549         n_sectors = dio->range.n_sectors;
1550         do {
1551                 struct bio_vec bv = bio_iovec(bio);
1552                 char *mem;
1553
1554                 if (unlikely(bv.bv_len >> SECTOR_SHIFT > n_sectors))
1555                         bv.bv_len = n_sectors << SECTOR_SHIFT;
1556                 n_sectors -= bv.bv_len >> SECTOR_SHIFT;
1557                 bio_advance_iter(bio, &bio->bi_iter, bv.bv_len);
1558 retry_kmap:
1559                 mem = kmap_atomic(bv.bv_page);
1560                 if (likely(dio->write))
1561                         flush_dcache_page(bv.bv_page);
1562
1563                 do {
1564                         struct journal_entry *je = access_journal_entry(ic, journal_section, journal_entry);
1565
1566                         if (unlikely(!dio->write)) {
1567                                 struct journal_sector *js;
1568                                 char *mem_ptr;
1569                                 unsigned s;
1570
1571                                 if (unlikely(journal_entry_is_inprogress(je))) {
1572                                         flush_dcache_page(bv.bv_page);
1573                                         kunmap_atomic(mem);
1574
1575                                         __io_wait_event(ic->copy_to_journal_wait, !journal_entry_is_inprogress(je));
1576                                         goto retry_kmap;
1577                                 }
1578                                 smp_rmb();
1579                                 BUG_ON(journal_entry_get_sector(je) != logical_sector);
1580                                 js = access_journal_data(ic, journal_section, journal_entry);
1581                                 mem_ptr = mem + bv.bv_offset;
1582                                 s = 0;
1583                                 do {
1584                                         memcpy(mem_ptr, js, JOURNAL_SECTOR_DATA);
1585                                         *(commit_id_t *)(mem_ptr + JOURNAL_SECTOR_DATA) = je->last_bytes[s];
1586                                         js++;
1587                                         mem_ptr += 1 << SECTOR_SHIFT;
1588                                 } while (++s < ic->sectors_per_block);
1589 #ifdef INTERNAL_VERIFY
1590                                 if (ic->internal_hash) {
1591                                         char checksums_onstack[max(crypto_shash_digestsize(ic->internal_hash), ic->tag_size)];
1592
1593                                         integrity_sector_checksum(ic, logical_sector, mem + bv.bv_offset, checksums_onstack);
1594                                         if (unlikely(memcmp(checksums_onstack, journal_entry_tag(ic, je), ic->tag_size))) {
1595                                                 DMERR_LIMIT("Checksum failed when reading from journal, at sector 0x%llx",
1596                                                             (unsigned long long)logical_sector);
1597                                         }
1598                                 }
1599 #endif
1600                         }
1601
1602                         if (!ic->internal_hash) {
1603                                 struct bio_integrity_payload *bip = bio_integrity(bio);
1604                                 unsigned tag_todo = ic->tag_size;
1605                                 char *tag_ptr = journal_entry_tag(ic, je);
1606
1607                                 if (bip) do {
1608                                         struct bio_vec biv = bvec_iter_bvec(bip->bip_vec, bip->bip_iter);
1609                                         unsigned tag_now = min(biv.bv_len, tag_todo);
1610                                         char *tag_addr;
1611                                         BUG_ON(PageHighMem(biv.bv_page));
1612                                         tag_addr = lowmem_page_address(biv.bv_page) + biv.bv_offset;
1613                                         if (likely(dio->write))
1614                                                 memcpy(tag_ptr, tag_addr, tag_now);
1615                                         else
1616                                                 memcpy(tag_addr, tag_ptr, tag_now);
1617                                         bvec_iter_advance(bip->bip_vec, &bip->bip_iter, tag_now);
1618                                         tag_ptr += tag_now;
1619                                         tag_todo -= tag_now;
1620                                 } while (unlikely(tag_todo)); else {
1621                                         if (likely(dio->write))
1622                                                 memset(tag_ptr, 0, tag_todo);
1623                                 }
1624                         }
1625
1626                         if (likely(dio->write)) {
1627                                 struct journal_sector *js;
1628                                 unsigned s;
1629
1630                                 js = access_journal_data(ic, journal_section, journal_entry);
1631                                 memcpy(js, mem + bv.bv_offset, ic->sectors_per_block << SECTOR_SHIFT);
1632
1633                                 s = 0;
1634                                 do {
1635                                         je->last_bytes[s] = js[s].commit_id;
1636                                 } while (++s < ic->sectors_per_block);
1637
1638                                 if (ic->internal_hash) {
1639                                         unsigned digest_size = crypto_shash_digestsize(ic->internal_hash);
1640                                         if (unlikely(digest_size > ic->tag_size)) {
1641                                                 char checksums_onstack[digest_size];
1642                                                 integrity_sector_checksum(ic, logical_sector, (char *)js, checksums_onstack);
1643                                                 memcpy(journal_entry_tag(ic, je), checksums_onstack, ic->tag_size);
1644                                         } else
1645                                                 integrity_sector_checksum(ic, logical_sector, (char *)js, journal_entry_tag(ic, je));
1646                                 }
1647
1648                                 journal_entry_set_sector(je, logical_sector);
1649                         }
1650                         logical_sector += ic->sectors_per_block;
1651
1652                         journal_entry++;
1653                         if (unlikely(journal_entry == ic->journal_section_entries)) {
1654                                 journal_entry = 0;
1655                                 journal_section++;
1656                                 wraparound_section(ic, &journal_section);
1657                         }
1658
1659                         bv.bv_offset += ic->sectors_per_block << SECTOR_SHIFT;
1660                 } while (bv.bv_len -= ic->sectors_per_block << SECTOR_SHIFT);
1661
1662                 if (unlikely(!dio->write))
1663                         flush_dcache_page(bv.bv_page);
1664                 kunmap_atomic(mem);
1665         } while (n_sectors);
1666
1667         if (likely(dio->write)) {
1668                 smp_mb();
1669                 if (unlikely(waitqueue_active(&ic->copy_to_journal_wait)))
1670                         wake_up(&ic->copy_to_journal_wait);
1671                 if (READ_ONCE(ic->free_sectors) <= ic->free_sectors_threshold) {
1672                         queue_work(ic->commit_wq, &ic->commit_work);
1673                 } else {
1674                         schedule_autocommit(ic);
1675                 }
1676         } else {
1677                 remove_range(ic, &dio->range);
1678         }
1679
1680         if (unlikely(bio->bi_iter.bi_size)) {
1681                 sector_t area, offset;
1682
1683                 dio->range.logical_sector = logical_sector;
1684                 get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
1685                 dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset, &dio->metadata_offset);
1686                 return true;
1687         }
1688
1689         return false;
1690 }
1691
1692 static void dm_integrity_map_continue(struct dm_integrity_io *dio, bool from_map)
1693 {
1694         struct dm_integrity_c *ic = dio->ic;
1695         struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1696         unsigned journal_section, journal_entry;
1697         unsigned journal_read_pos;
1698         struct completion read_comp;
1699         bool need_sync_io = ic->internal_hash && !dio->write;
1700
1701         if (need_sync_io && from_map) {
1702                 INIT_WORK(&dio->work, integrity_bio_wait);
1703                 queue_work(ic->offload_wq, &dio->work);
1704                 return;
1705         }
1706
1707 lock_retry:
1708         spin_lock_irq(&ic->endio_wait.lock);
1709 retry:
1710         if (unlikely(dm_integrity_failed(ic))) {
1711                 spin_unlock_irq(&ic->endio_wait.lock);
1712                 do_endio(ic, bio);
1713                 return;
1714         }
1715         dio->range.n_sectors = bio_sectors(bio);
1716         journal_read_pos = NOT_FOUND;
1717         if (likely(ic->mode == 'J')) {
1718                 if (dio->write) {
1719                         unsigned next_entry, i, pos;
1720                         unsigned ws, we, range_sectors;
1721
1722                         dio->range.n_sectors = min(dio->range.n_sectors,
1723                                                    ic->free_sectors << ic->sb->log2_sectors_per_block);
1724                         if (unlikely(!dio->range.n_sectors)) {
1725                                 if (from_map)
1726                                         goto offload_to_thread;
1727                                 sleep_on_endio_wait(ic);
1728                                 goto retry;
1729                         }
1730                         range_sectors = dio->range.n_sectors >> ic->sb->log2_sectors_per_block;
1731                         ic->free_sectors -= range_sectors;
1732                         journal_section = ic->free_section;
1733                         journal_entry = ic->free_section_entry;
1734
1735                         next_entry = ic->free_section_entry + range_sectors;
1736                         ic->free_section_entry = next_entry % ic->journal_section_entries;
1737                         ic->free_section += next_entry / ic->journal_section_entries;
1738                         ic->n_uncommitted_sections += next_entry / ic->journal_section_entries;
1739                         wraparound_section(ic, &ic->free_section);
1740
1741                         pos = journal_section * ic->journal_section_entries + journal_entry;
1742                         ws = journal_section;
1743                         we = journal_entry;
1744                         i = 0;
1745                         do {
1746                                 struct journal_entry *je;
1747
1748                                 add_journal_node(ic, &ic->journal_tree[pos], dio->range.logical_sector + i);
1749                                 pos++;
1750                                 if (unlikely(pos >= ic->journal_entries))
1751                                         pos = 0;
1752
1753                                 je = access_journal_entry(ic, ws, we);
1754                                 BUG_ON(!journal_entry_is_unused(je));
1755                                 journal_entry_set_inprogress(je);
1756                                 we++;
1757                                 if (unlikely(we == ic->journal_section_entries)) {
1758                                         we = 0;
1759                                         ws++;
1760                                         wraparound_section(ic, &ws);
1761                                 }
1762                         } while ((i += ic->sectors_per_block) < dio->range.n_sectors);
1763
1764                         spin_unlock_irq(&ic->endio_wait.lock);
1765                         goto journal_read_write;
1766                 } else {
1767                         sector_t next_sector;
1768                         journal_read_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
1769                         if (likely(journal_read_pos == NOT_FOUND)) {
1770                                 if (unlikely(dio->range.n_sectors > next_sector - dio->range.logical_sector))
1771                                         dio->range.n_sectors = next_sector - dio->range.logical_sector;
1772                         } else {
1773                                 unsigned i;
1774                                 unsigned jp = journal_read_pos + 1;
1775                                 for (i = ic->sectors_per_block; i < dio->range.n_sectors; i += ic->sectors_per_block, jp++) {
1776                                         if (!test_journal_node(ic, jp, dio->range.logical_sector + i))
1777                                                 break;
1778                                 }
1779                                 dio->range.n_sectors = i;
1780                         }
1781                 }
1782         }
1783         if (unlikely(!add_new_range(ic, &dio->range, true))) {
1784                 /*
1785                  * We must not sleep in the request routine because it could
1786                  * stall bios on current->bio_list.
1787                  * So, we offload the bio to a workqueue if we have to sleep.
1788                  */
1789                 if (from_map) {
1790 offload_to_thread:
1791                         spin_unlock_irq(&ic->endio_wait.lock);
1792                         INIT_WORK(&dio->work, integrity_bio_wait);
1793                         queue_work(ic->wait_wq, &dio->work);
1794                         return;
1795                 }
1796                 if (journal_read_pos != NOT_FOUND)
1797                         dio->range.n_sectors = ic->sectors_per_block;
1798                 wait_and_add_new_range(ic, &dio->range);
1799                 /*
1800                  * wait_and_add_new_range drops the spinlock, so the journal
1801                  * may have been changed arbitrarily. We need to recheck.
1802                  * To simplify the code, we restrict I/O size to just one block.
1803                  */
1804                 if (journal_read_pos != NOT_FOUND) {
1805                         sector_t next_sector;
1806                         unsigned new_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
1807                         if (unlikely(new_pos != journal_read_pos)) {
1808                                 remove_range_unlocked(ic, &dio->range);
1809                                 goto retry;
1810                         }
1811                 }
1812         }
1813         spin_unlock_irq(&ic->endio_wait.lock);
1814
1815         if (unlikely(journal_read_pos != NOT_FOUND)) {
1816                 journal_section = journal_read_pos / ic->journal_section_entries;
1817                 journal_entry = journal_read_pos % ic->journal_section_entries;
1818                 goto journal_read_write;
1819         }
1820
1821         dio->in_flight = (atomic_t)ATOMIC_INIT(2);
1822
1823         if (need_sync_io) {
1824                 init_completion(&read_comp);
1825                 dio->completion = &read_comp;
1826         } else
1827                 dio->completion = NULL;
1828
1829         dm_bio_record(&dio->bio_details, bio);
1830         bio_set_dev(bio, ic->dev->bdev);
1831         bio->bi_integrity = NULL;
1832         bio->bi_opf &= ~REQ_INTEGRITY;
1833         bio->bi_end_io = integrity_end_io;
1834         bio->bi_iter.bi_size = dio->range.n_sectors << SECTOR_SHIFT;
1835
1836         generic_make_request(bio);
1837
1838         if (need_sync_io) {
1839                 wait_for_completion_io(&read_comp);
1840                 if (unlikely(ic->recalc_wq != NULL) &&
1841                     ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING) &&
1842                     dio->range.logical_sector + dio->range.n_sectors > le64_to_cpu(ic->sb->recalc_sector))
1843                         goto skip_check;
1844                 if (likely(!bio->bi_status))
1845                         integrity_metadata(&dio->work);
1846                 else
1847 skip_check:
1848                         dec_in_flight(dio);
1849
1850         } else {
1851                 INIT_WORK(&dio->work, integrity_metadata);
1852                 queue_work(ic->metadata_wq, &dio->work);
1853         }
1854
1855         return;
1856
1857 journal_read_write:
1858         if (unlikely(__journal_read_write(dio, bio, journal_section, journal_entry)))
1859                 goto lock_retry;
1860
1861         do_endio_flush(ic, dio);
1862 }
1863
1864
1865 static void integrity_bio_wait(struct work_struct *w)
1866 {
1867         struct dm_integrity_io *dio = container_of(w, struct dm_integrity_io, work);
1868
1869         dm_integrity_map_continue(dio, false);
1870 }
1871
1872 static void pad_uncommitted(struct dm_integrity_c *ic)
1873 {
1874         if (ic->free_section_entry) {
1875                 ic->free_sectors -= ic->journal_section_entries - ic->free_section_entry;
1876                 ic->free_section_entry = 0;
1877                 ic->free_section++;
1878                 wraparound_section(ic, &ic->free_section);
1879                 ic->n_uncommitted_sections++;
1880         }
1881         WARN_ON(ic->journal_sections * ic->journal_section_entries !=
1882                 (ic->n_uncommitted_sections + ic->n_committed_sections) * ic->journal_section_entries + ic->free_sectors);
1883 }
1884
1885 static void integrity_commit(struct work_struct *w)
1886 {
1887         struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, commit_work);
1888         unsigned commit_start, commit_sections;
1889         unsigned i, j, n;
1890         struct bio *flushes;
1891
1892         del_timer(&ic->autocommit_timer);
1893
1894         spin_lock_irq(&ic->endio_wait.lock);
1895         flushes = bio_list_get(&ic->flush_bio_list);
1896         if (unlikely(ic->mode != 'J')) {
1897                 spin_unlock_irq(&ic->endio_wait.lock);
1898                 dm_integrity_flush_buffers(ic, true);
1899                 goto release_flush_bios;
1900         }
1901
1902         pad_uncommitted(ic);
1903         commit_start = ic->uncommitted_section;
1904         commit_sections = ic->n_uncommitted_sections;
1905         spin_unlock_irq(&ic->endio_wait.lock);
1906
1907         if (!commit_sections)
1908                 goto release_flush_bios;
1909
1910         i = commit_start;
1911         for (n = 0; n < commit_sections; n++) {
1912                 for (j = 0; j < ic->journal_section_entries; j++) {
1913                         struct journal_entry *je;
1914                         je = access_journal_entry(ic, i, j);
1915                         io_wait_event(ic->copy_to_journal_wait, !journal_entry_is_inprogress(je));
1916                 }
1917                 for (j = 0; j < ic->journal_section_sectors; j++) {
1918                         struct journal_sector *js;
1919                         js = access_journal(ic, i, j);
1920                         js->commit_id = dm_integrity_commit_id(ic, i, j, ic->commit_seq);
1921                 }
1922                 i++;
1923                 if (unlikely(i >= ic->journal_sections))
1924                         ic->commit_seq = next_commit_seq(ic->commit_seq);
1925                 wraparound_section(ic, &i);
1926         }
1927         smp_rmb();
1928
1929         write_journal(ic, commit_start, commit_sections);
1930
1931         spin_lock_irq(&ic->endio_wait.lock);
1932         ic->uncommitted_section += commit_sections;
1933         wraparound_section(ic, &ic->uncommitted_section);
1934         ic->n_uncommitted_sections -= commit_sections;
1935         ic->n_committed_sections += commit_sections;
1936         spin_unlock_irq(&ic->endio_wait.lock);
1937
1938         if (READ_ONCE(ic->free_sectors) <= ic->free_sectors_threshold)
1939                 queue_work(ic->writer_wq, &ic->writer_work);
1940
1941 release_flush_bios:
1942         while (flushes) {
1943                 struct bio *next = flushes->bi_next;
1944                 flushes->bi_next = NULL;
1945                 do_endio(ic, flushes);
1946                 flushes = next;
1947         }
1948 }
1949
1950 static void complete_copy_from_journal(unsigned long error, void *context)
1951 {
1952         struct journal_io *io = context;
1953         struct journal_completion *comp = io->comp;
1954         struct dm_integrity_c *ic = comp->ic;
1955         remove_range(ic, &io->range);
1956         mempool_free(io, &ic->journal_io_mempool);
1957         if (unlikely(error != 0))
1958                 dm_integrity_io_error(ic, "copying from journal", -EIO);
1959         complete_journal_op(comp);
1960 }
1961
1962 static void restore_last_bytes(struct dm_integrity_c *ic, struct journal_sector *js,
1963                                struct journal_entry *je)
1964 {
1965         unsigned s = 0;
1966         do {
1967                 js->commit_id = je->last_bytes[s];
1968                 js++;
1969         } while (++s < ic->sectors_per_block);
1970 }
1971
1972 static void do_journal_write(struct dm_integrity_c *ic, unsigned write_start,
1973                              unsigned write_sections, bool from_replay)
1974 {
1975         unsigned i, j, n;
1976         struct journal_completion comp;
1977         struct blk_plug plug;
1978
1979         blk_start_plug(&plug);
1980
1981         comp.ic = ic;
1982         comp.in_flight = (atomic_t)ATOMIC_INIT(1);
1983         init_completion(&comp.comp);
1984
1985         i = write_start;
1986         for (n = 0; n < write_sections; n++, i++, wraparound_section(ic, &i)) {
1987 #ifndef INTERNAL_VERIFY
1988                 if (unlikely(from_replay))
1989 #endif
1990                         rw_section_mac(ic, i, false);
1991                 for (j = 0; j < ic->journal_section_entries; j++) {
1992                         struct journal_entry *je = access_journal_entry(ic, i, j);
1993                         sector_t sec, area, offset;
1994                         unsigned k, l, next_loop;
1995                         sector_t metadata_block;
1996                         unsigned metadata_offset;
1997                         struct journal_io *io;
1998
1999                         if (journal_entry_is_unused(je))
2000                                 continue;
2001                         BUG_ON(unlikely(journal_entry_is_inprogress(je)) && !from_replay);
2002                         sec = journal_entry_get_sector(je);
2003                         if (unlikely(from_replay)) {
2004                                 if (unlikely(sec & (unsigned)(ic->sectors_per_block - 1))) {
2005                                         dm_integrity_io_error(ic, "invalid sector in journal", -EIO);
2006                                         sec &= ~(sector_t)(ic->sectors_per_block - 1);
2007                                 }
2008                         }
2009                         get_area_and_offset(ic, sec, &area, &offset);
2010                         restore_last_bytes(ic, access_journal_data(ic, i, j), je);
2011                         for (k = j + 1; k < ic->journal_section_entries; k++) {
2012                                 struct journal_entry *je2 = access_journal_entry(ic, i, k);
2013                                 sector_t sec2, area2, offset2;
2014                                 if (journal_entry_is_unused(je2))
2015                                         break;
2016                                 BUG_ON(unlikely(journal_entry_is_inprogress(je2)) && !from_replay);
2017                                 sec2 = journal_entry_get_sector(je2);
2018                                 get_area_and_offset(ic, sec2, &area2, &offset2);
2019                                 if (area2 != area || offset2 != offset + ((k - j) << ic->sb->log2_sectors_per_block))
2020                                         break;
2021                                 restore_last_bytes(ic, access_journal_data(ic, i, k), je2);
2022                         }
2023                         next_loop = k - 1;
2024
2025                         io = mempool_alloc(&ic->journal_io_mempool, GFP_NOIO);
2026                         io->comp = &comp;
2027                         io->range.logical_sector = sec;
2028                         io->range.n_sectors = (k - j) << ic->sb->log2_sectors_per_block;
2029
2030                         spin_lock_irq(&ic->endio_wait.lock);
2031                         if (unlikely(!add_new_range(ic, &io->range, true)))
2032                                 wait_and_add_new_range(ic, &io->range);
2033
2034                         if (likely(!from_replay)) {
2035                                 struct journal_node *section_node = &ic->journal_tree[i * ic->journal_section_entries];
2036
2037                                 /* don't write if there is newer committed sector */
2038                                 while (j < k && find_newer_committed_node(ic, &section_node[j])) {
2039                                         struct journal_entry *je2 = access_journal_entry(ic, i, j);
2040
2041                                         journal_entry_set_unused(je2);
2042                                         remove_journal_node(ic, &section_node[j]);
2043                                         j++;
2044                                         sec += ic->sectors_per_block;
2045                                         offset += ic->sectors_per_block;
2046                                 }
2047                                 while (j < k && find_newer_committed_node(ic, &section_node[k - 1])) {
2048                                         struct journal_entry *je2 = access_journal_entry(ic, i, k - 1);
2049
2050                                         journal_entry_set_unused(je2);
2051                                         remove_journal_node(ic, &section_node[k - 1]);
2052                                         k--;
2053                                 }
2054                                 if (j == k) {
2055                                         remove_range_unlocked(ic, &io->range);
2056                                         spin_unlock_irq(&ic->endio_wait.lock);
2057                                         mempool_free(io, &ic->journal_io_mempool);
2058                                         goto skip_io;
2059                                 }
2060                                 for (l = j; l < k; l++) {
2061                                         remove_journal_node(ic, &section_node[l]);
2062                                 }
2063                         }
2064                         spin_unlock_irq(&ic->endio_wait.lock);
2065
2066                         metadata_block = get_metadata_sector_and_offset(ic, area, offset, &metadata_offset);
2067                         for (l = j; l < k; l++) {
2068                                 int r;
2069                                 struct journal_entry *je2 = access_journal_entry(ic, i, l);
2070
2071                                 if (
2072 #ifndef INTERNAL_VERIFY
2073                                     unlikely(from_replay) &&
2074 #endif
2075                                     ic->internal_hash) {
2076                                         char test_tag[max(crypto_shash_digestsize(ic->internal_hash), ic->tag_size)];
2077
2078                                         integrity_sector_checksum(ic, sec + ((l - j) << ic->sb->log2_sectors_per_block),
2079                                                                   (char *)access_journal_data(ic, i, l), test_tag);
2080                                         if (unlikely(memcmp(test_tag, journal_entry_tag(ic, je2), ic->tag_size)))
2081                                                 dm_integrity_io_error(ic, "tag mismatch when replaying journal", -EILSEQ);
2082                                 }
2083
2084                                 journal_entry_set_unused(je2);
2085                                 r = dm_integrity_rw_tag(ic, journal_entry_tag(ic, je2), &metadata_block, &metadata_offset,
2086                                                         ic->tag_size, TAG_WRITE);
2087                                 if (unlikely(r)) {
2088                                         dm_integrity_io_error(ic, "reading tags", r);
2089                                 }
2090                         }
2091
2092                         atomic_inc(&comp.in_flight);
2093                         copy_from_journal(ic, i, j << ic->sb->log2_sectors_per_block,
2094                                           (k - j) << ic->sb->log2_sectors_per_block,
2095                                           get_data_sector(ic, area, offset),
2096                                           complete_copy_from_journal, io);
2097 skip_io:
2098                         j = next_loop;
2099                 }
2100         }
2101
2102         dm_bufio_write_dirty_buffers_async(ic->bufio);
2103
2104         blk_finish_plug(&plug);
2105
2106         complete_journal_op(&comp);
2107         wait_for_completion_io(&comp.comp);
2108
2109         dm_integrity_flush_buffers(ic, true);
2110 }
2111
2112 static void integrity_writer(struct work_struct *w)
2113 {
2114         struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, writer_work);
2115         unsigned write_start, write_sections;
2116
2117         unsigned prev_free_sectors;
2118
2119         /* the following test is not needed, but it tests the replay code */
2120         if (unlikely(dm_post_suspending(ic->ti)) && !ic->meta_dev)
2121                 return;
2122
2123         spin_lock_irq(&ic->endio_wait.lock);
2124         write_start = ic->committed_section;
2125         write_sections = ic->n_committed_sections;
2126         spin_unlock_irq(&ic->endio_wait.lock);
2127
2128         if (!write_sections)
2129                 return;
2130
2131         do_journal_write(ic, write_start, write_sections, false);
2132
2133         spin_lock_irq(&ic->endio_wait.lock);
2134
2135         ic->committed_section += write_sections;
2136         wraparound_section(ic, &ic->committed_section);
2137         ic->n_committed_sections -= write_sections;
2138
2139         prev_free_sectors = ic->free_sectors;
2140         ic->free_sectors += write_sections * ic->journal_section_entries;
2141         if (unlikely(!prev_free_sectors))
2142                 wake_up_locked(&ic->endio_wait);
2143
2144         spin_unlock_irq(&ic->endio_wait.lock);
2145 }
2146
2147 static void recalc_write_super(struct dm_integrity_c *ic)
2148 {
2149         int r;
2150
2151         dm_integrity_flush_buffers(ic, false);
2152         if (dm_integrity_failed(ic))
2153                 return;
2154
2155         sb_set_version(ic);
2156         r = sync_rw_sb(ic, REQ_OP_WRITE, 0);
2157         if (unlikely(r))
2158                 dm_integrity_io_error(ic, "writing superblock", r);
2159 }
2160
2161 static void integrity_recalc(struct work_struct *w)
2162 {
2163         struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, recalc_work);
2164         struct dm_integrity_range range;
2165         struct dm_io_request io_req;
2166         struct dm_io_region io_loc;
2167         sector_t area, offset;
2168         sector_t metadata_block;
2169         unsigned metadata_offset;
2170         __u8 *t;
2171         unsigned i;
2172         int r;
2173         unsigned super_counter = 0;
2174
2175         spin_lock_irq(&ic->endio_wait.lock);
2176
2177 next_chunk:
2178
2179         if (unlikely(dm_post_suspending(ic->ti)))
2180                 goto unlock_ret;
2181
2182         range.logical_sector = le64_to_cpu(ic->sb->recalc_sector);
2183         if (unlikely(range.logical_sector >= ic->provided_data_sectors))
2184                 goto unlock_ret;
2185
2186         get_area_and_offset(ic, range.logical_sector, &area, &offset);
2187         range.n_sectors = min((sector_t)RECALC_SECTORS, ic->provided_data_sectors - range.logical_sector);
2188         if (!ic->meta_dev)
2189                 range.n_sectors = min(range.n_sectors, (1U << ic->sb->log2_interleave_sectors) - (unsigned)offset);
2190
2191         if (unlikely(!add_new_range(ic, &range, true)))
2192                 wait_and_add_new_range(ic, &range);
2193
2194         spin_unlock_irq(&ic->endio_wait.lock);
2195
2196         if (unlikely(++super_counter == RECALC_WRITE_SUPER)) {
2197                 recalc_write_super(ic);
2198                 super_counter = 0;
2199         }
2200
2201         if (unlikely(dm_integrity_failed(ic)))
2202                 goto err;
2203
2204         io_req.bi_op = REQ_OP_READ;
2205         io_req.bi_op_flags = 0;
2206         io_req.mem.type = DM_IO_VMA;
2207         io_req.mem.ptr.addr = ic->recalc_buffer;
2208         io_req.notify.fn = NULL;
2209         io_req.client = ic->io;
2210         io_loc.bdev = ic->dev->bdev;
2211         io_loc.sector = get_data_sector(ic, area, offset);
2212         io_loc.count = range.n_sectors;
2213
2214         r = dm_io(&io_req, 1, &io_loc, NULL);
2215         if (unlikely(r)) {
2216                 dm_integrity_io_error(ic, "reading data", r);
2217                 goto err;
2218         }
2219
2220         t = ic->recalc_tags;
2221         for (i = 0; i < range.n_sectors; i += ic->sectors_per_block) {
2222                 integrity_sector_checksum(ic, range.logical_sector + i, ic->recalc_buffer + (i << SECTOR_SHIFT), t);
2223                 t += ic->tag_size;
2224         }
2225
2226         metadata_block = get_metadata_sector_and_offset(ic, area, offset, &metadata_offset);
2227
2228         r = dm_integrity_rw_tag(ic, ic->recalc_tags, &metadata_block, &metadata_offset, t - ic->recalc_tags, TAG_WRITE);
2229         if (unlikely(r)) {
2230                 dm_integrity_io_error(ic, "writing tags", r);
2231                 goto err;
2232         }
2233
2234         spin_lock_irq(&ic->endio_wait.lock);
2235         remove_range_unlocked(ic, &range);
2236         ic->sb->recalc_sector = cpu_to_le64(range.logical_sector + range.n_sectors);
2237         goto next_chunk;
2238
2239 err:
2240         remove_range(ic, &range);
2241         return;
2242
2243 unlock_ret:
2244         spin_unlock_irq(&ic->endio_wait.lock);
2245
2246         recalc_write_super(ic);
2247 }
2248
2249 static void init_journal(struct dm_integrity_c *ic, unsigned start_section,
2250                          unsigned n_sections, unsigned char commit_seq)
2251 {
2252         unsigned i, j, n;
2253
2254         if (!n_sections)
2255                 return;
2256
2257         for (n = 0; n < n_sections; n++) {
2258                 i = start_section + n;
2259                 wraparound_section(ic, &i);
2260                 for (j = 0; j < ic->journal_section_sectors; j++) {
2261                         struct journal_sector *js = access_journal(ic, i, j);
2262                         memset(&js->entries, 0, JOURNAL_SECTOR_DATA);
2263                         js->commit_id = dm_integrity_commit_id(ic, i, j, commit_seq);
2264                 }
2265                 for (j = 0; j < ic->journal_section_entries; j++) {
2266                         struct journal_entry *je = access_journal_entry(ic, i, j);
2267                         journal_entry_set_unused(je);
2268                 }
2269         }
2270
2271         write_journal(ic, start_section, n_sections);
2272 }
2273
2274 static int find_commit_seq(struct dm_integrity_c *ic, unsigned i, unsigned j, commit_id_t id)
2275 {
2276         unsigned char k;
2277         for (k = 0; k < N_COMMIT_IDS; k++) {
2278                 if (dm_integrity_commit_id(ic, i, j, k) == id)
2279                         return k;
2280         }
2281         dm_integrity_io_error(ic, "journal commit id", -EIO);
2282         return -EIO;
2283 }
2284
2285 static void replay_journal(struct dm_integrity_c *ic)
2286 {
2287         unsigned i, j;
2288         bool used_commit_ids[N_COMMIT_IDS];
2289         unsigned max_commit_id_sections[N_COMMIT_IDS];
2290         unsigned write_start, write_sections;
2291         unsigned continue_section;
2292         bool journal_empty;
2293         unsigned char unused, last_used, want_commit_seq;
2294
2295         if (ic->mode == 'R')
2296                 return;
2297
2298         if (ic->journal_uptodate)
2299                 return;
2300
2301         last_used = 0;
2302         write_start = 0;
2303
2304         if (!ic->just_formatted) {
2305                 DEBUG_print("reading journal\n");
2306                 rw_journal(ic, REQ_OP_READ, 0, 0, ic->journal_sections, NULL);
2307                 if (ic->journal_io)
2308                         DEBUG_bytes(lowmem_page_address(ic->journal_io[0].page), 64, "read journal");
2309                 if (ic->journal_io) {
2310                         struct journal_completion crypt_comp;
2311                         crypt_comp.ic = ic;
2312                         init_completion(&crypt_comp.comp);
2313                         crypt_comp.in_flight = (atomic_t)ATOMIC_INIT(0);
2314                         encrypt_journal(ic, false, 0, ic->journal_sections, &crypt_comp);
2315                         wait_for_completion(&crypt_comp.comp);
2316                 }
2317                 DEBUG_bytes(lowmem_page_address(ic->journal[0].page), 64, "decrypted journal");
2318         }
2319
2320         if (dm_integrity_failed(ic))
2321                 goto clear_journal;
2322
2323         journal_empty = true;
2324         memset(used_commit_ids, 0, sizeof used_commit_ids);
2325         memset(max_commit_id_sections, 0, sizeof max_commit_id_sections);
2326         for (i = 0; i < ic->journal_sections; i++) {
2327                 for (j = 0; j < ic->journal_section_sectors; j++) {
2328                         int k;
2329                         struct journal_sector *js = access_journal(ic, i, j);
2330                         k = find_commit_seq(ic, i, j, js->commit_id);
2331                         if (k < 0)
2332                                 goto clear_journal;
2333                         used_commit_ids[k] = true;
2334                         max_commit_id_sections[k] = i;
2335                 }
2336                 if (journal_empty) {
2337                         for (j = 0; j < ic->journal_section_entries; j++) {
2338                                 struct journal_entry *je = access_journal_entry(ic, i, j);
2339                                 if (!journal_entry_is_unused(je)) {
2340                                         journal_empty = false;
2341                                         break;
2342                                 }
2343                         }
2344                 }
2345         }
2346
2347         if (!used_commit_ids[N_COMMIT_IDS - 1]) {
2348                 unused = N_COMMIT_IDS - 1;
2349                 while (unused && !used_commit_ids[unused - 1])
2350                         unused--;
2351         } else {
2352                 for (unused = 0; unused < N_COMMIT_IDS; unused++)
2353                         if (!used_commit_ids[unused])
2354                                 break;
2355                 if (unused == N_COMMIT_IDS) {
2356                         dm_integrity_io_error(ic, "journal commit ids", -EIO);
2357                         goto clear_journal;
2358                 }
2359         }
2360         DEBUG_print("first unused commit seq %d [%d,%d,%d,%d]\n",
2361                     unused, used_commit_ids[0], used_commit_ids[1],
2362                     used_commit_ids[2], used_commit_ids[3]);
2363
2364         last_used = prev_commit_seq(unused);
2365         want_commit_seq = prev_commit_seq(last_used);
2366
2367         if (!used_commit_ids[want_commit_seq] && used_commit_ids[prev_commit_seq(want_commit_seq)])
2368                 journal_empty = true;
2369
2370         write_start = max_commit_id_sections[last_used] + 1;
2371         if (unlikely(write_start >= ic->journal_sections))
2372                 want_commit_seq = next_commit_seq(want_commit_seq);
2373         wraparound_section(ic, &write_start);
2374
2375         i = write_start;
2376         for (write_sections = 0; write_sections < ic->journal_sections; write_sections++) {
2377                 for (j = 0; j < ic->journal_section_sectors; j++) {
2378                         struct journal_sector *js = access_journal(ic, i, j);
2379
2380                         if (js->commit_id != dm_integrity_commit_id(ic, i, j, want_commit_seq)) {
2381                                 /*
2382                                  * This could be caused by crash during writing.
2383                                  * We won't replay the inconsistent part of the
2384                                  * journal.
2385                                  */
2386                                 DEBUG_print("commit id mismatch at position (%u, %u): %d != %d\n",
2387                                             i, j, find_commit_seq(ic, i, j, js->commit_id), want_commit_seq);
2388                                 goto brk;
2389                         }
2390                 }
2391                 i++;
2392                 if (unlikely(i >= ic->journal_sections))
2393                         want_commit_seq = next_commit_seq(want_commit_seq);
2394                 wraparound_section(ic, &i);
2395         }
2396 brk:
2397
2398         if (!journal_empty) {
2399                 DEBUG_print("replaying %u sections, starting at %u, commit seq %d\n",
2400                             write_sections, write_start, want_commit_seq);
2401                 do_journal_write(ic, write_start, write_sections, true);
2402         }
2403
2404         if (write_sections == ic->journal_sections && (ic->mode == 'J' || journal_empty)) {
2405                 continue_section = write_start;
2406                 ic->commit_seq = want_commit_seq;
2407                 DEBUG_print("continuing from section %u, commit seq %d\n", write_start, ic->commit_seq);
2408         } else {
2409                 unsigned s;
2410                 unsigned char erase_seq;
2411 clear_journal:
2412                 DEBUG_print("clearing journal\n");
2413
2414                 erase_seq = prev_commit_seq(prev_commit_seq(last_used));
2415                 s = write_start;
2416                 init_journal(ic, s, 1, erase_seq);
2417                 s++;
2418                 wraparound_section(ic, &s);
2419                 if (ic->journal_sections >= 2) {
2420                         init_journal(ic, s, ic->journal_sections - 2, erase_seq);
2421                         s += ic->journal_sections - 2;
2422                         wraparound_section(ic, &s);
2423                         init_journal(ic, s, 1, erase_seq);
2424                 }
2425
2426                 continue_section = 0;
2427                 ic->commit_seq = next_commit_seq(erase_seq);
2428         }
2429
2430         ic->committed_section = continue_section;
2431         ic->n_committed_sections = 0;
2432
2433         ic->uncommitted_section = continue_section;
2434         ic->n_uncommitted_sections = 0;
2435
2436         ic->free_section = continue_section;
2437         ic->free_section_entry = 0;
2438         ic->free_sectors = ic->journal_entries;
2439
2440         ic->journal_tree_root = RB_ROOT;
2441         for (i = 0; i < ic->journal_entries; i++)
2442                 init_journal_node(&ic->journal_tree[i]);
2443 }
2444
2445 static void dm_integrity_postsuspend(struct dm_target *ti)
2446 {
2447         struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
2448
2449         del_timer_sync(&ic->autocommit_timer);
2450
2451         if (ic->recalc_wq)
2452                 drain_workqueue(ic->recalc_wq);
2453
2454         queue_work(ic->commit_wq, &ic->commit_work);
2455         drain_workqueue(ic->commit_wq);
2456
2457         if (ic->mode == 'J') {
2458                 if (ic->meta_dev)
2459                         queue_work(ic->writer_wq, &ic->writer_work);
2460                 drain_workqueue(ic->writer_wq);
2461                 dm_integrity_flush_buffers(ic, true);
2462         }
2463
2464         BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
2465
2466         ic->journal_uptodate = true;
2467 }
2468
2469 static void dm_integrity_resume(struct dm_target *ti)
2470 {
2471         struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
2472
2473         replay_journal(ic);
2474
2475         if (ic->recalc_wq && ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
2476                 __u64 recalc_pos = le64_to_cpu(ic->sb->recalc_sector);
2477                 if (recalc_pos < ic->provided_data_sectors) {
2478                         queue_work(ic->recalc_wq, &ic->recalc_work);
2479                 } else if (recalc_pos > ic->provided_data_sectors) {
2480                         ic->sb->recalc_sector = cpu_to_le64(ic->provided_data_sectors);
2481                         recalc_write_super(ic);
2482                 }
2483         }
2484 }
2485
2486 static void dm_integrity_status(struct dm_target *ti, status_type_t type,
2487                                 unsigned status_flags, char *result, unsigned maxlen)
2488 {
2489         struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
2490         unsigned arg_count;
2491         size_t sz = 0;
2492
2493         switch (type) {
2494         case STATUSTYPE_INFO:
2495                 DMEMIT("%llu %llu",
2496                         (unsigned long long)atomic64_read(&ic->number_of_mismatches),
2497                         (unsigned long long)ic->provided_data_sectors);
2498                 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
2499                         DMEMIT(" %llu", (unsigned long long)le64_to_cpu(ic->sb->recalc_sector));
2500                 else
2501                         DMEMIT(" -");
2502                 break;
2503
2504         case STATUSTYPE_TABLE: {
2505                 __u64 watermark_percentage = (__u64)(ic->journal_entries - ic->free_sectors_threshold) * 100;
2506                 watermark_percentage += ic->journal_entries / 2;
2507                 do_div(watermark_percentage, ic->journal_entries);
2508                 arg_count = 5;
2509                 arg_count += !!ic->meta_dev;
2510                 arg_count += ic->sectors_per_block != 1;
2511                 arg_count += !!(ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING));
2512                 arg_count += !!ic->internal_hash_alg.alg_string;
2513                 arg_count += !!ic->journal_crypt_alg.alg_string;
2514                 arg_count += !!ic->journal_mac_alg.alg_string;
2515                 arg_count += ic->legacy_recalculate;
2516                 DMEMIT("%s %llu %u %c %u", ic->dev->name, (unsigned long long)ic->start,
2517                        ic->tag_size, ic->mode, arg_count);
2518                 if (ic->meta_dev)
2519                         DMEMIT(" meta_device:%s", ic->meta_dev->name);
2520                 if (ic->sectors_per_block != 1)
2521                         DMEMIT(" block_size:%u", ic->sectors_per_block << SECTOR_SHIFT);
2522                 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
2523                         DMEMIT(" recalculate");
2524                 DMEMIT(" journal_sectors:%u", ic->initial_sectors - SB_SECTORS);
2525                 DMEMIT(" interleave_sectors:%u", 1U << ic->sb->log2_interleave_sectors);
2526                 DMEMIT(" buffer_sectors:%u", 1U << ic->log2_buffer_sectors);
2527                 DMEMIT(" journal_watermark:%u", (unsigned)watermark_percentage);
2528                 DMEMIT(" commit_time:%u", ic->autocommit_msec);
2529                 if (ic->legacy_recalculate)
2530                         DMEMIT(" legacy_recalculate");
2531
2532 #define EMIT_ALG(a, n)                                                  \
2533                 do {                                                    \
2534                         if (ic->a.alg_string) {                         \
2535                                 DMEMIT(" %s:%s", n, ic->a.alg_string);  \
2536                                 if (ic->a.key_string)                   \
2537                                         DMEMIT(":%s", ic->a.key_string);\
2538                         }                                               \
2539                 } while (0)
2540                 EMIT_ALG(internal_hash_alg, "internal_hash");
2541                 EMIT_ALG(journal_crypt_alg, "journal_crypt");
2542                 EMIT_ALG(journal_mac_alg, "journal_mac");
2543                 break;
2544         }
2545         }
2546 }
2547
2548 static int dm_integrity_iterate_devices(struct dm_target *ti,
2549                                         iterate_devices_callout_fn fn, void *data)
2550 {
2551         struct dm_integrity_c *ic = ti->private;
2552
2553         if (!ic->meta_dev)
2554                 return fn(ti, ic->dev, ic->start + ic->initial_sectors + ic->metadata_run, ti->len, data);
2555         else
2556                 return fn(ti, ic->dev, 0, ti->len, data);
2557 }
2558
2559 static void dm_integrity_io_hints(struct dm_target *ti, struct queue_limits *limits)
2560 {
2561         struct dm_integrity_c *ic = ti->private;
2562
2563         if (ic->sectors_per_block > 1) {
2564                 limits->logical_block_size = ic->sectors_per_block << SECTOR_SHIFT;
2565                 limits->physical_block_size = ic->sectors_per_block << SECTOR_SHIFT;
2566                 blk_limits_io_min(limits, ic->sectors_per_block << SECTOR_SHIFT);
2567         }
2568 }
2569
2570 static void calculate_journal_section_size(struct dm_integrity_c *ic)
2571 {
2572         unsigned sector_space = JOURNAL_SECTOR_DATA;
2573
2574         ic->journal_sections = le32_to_cpu(ic->sb->journal_sections);
2575         ic->journal_entry_size = roundup(offsetof(struct journal_entry, last_bytes[ic->sectors_per_block]) + ic->tag_size,
2576                                          JOURNAL_ENTRY_ROUNDUP);
2577
2578         if (ic->sb->flags & cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC))
2579                 sector_space -= JOURNAL_MAC_PER_SECTOR;
2580         ic->journal_entries_per_sector = sector_space / ic->journal_entry_size;
2581         ic->journal_section_entries = ic->journal_entries_per_sector * JOURNAL_BLOCK_SECTORS;
2582         ic->journal_section_sectors = (ic->journal_section_entries << ic->sb->log2_sectors_per_block) + JOURNAL_BLOCK_SECTORS;
2583         ic->journal_entries = ic->journal_section_entries * ic->journal_sections;
2584 }
2585
2586 static int calculate_device_limits(struct dm_integrity_c *ic)
2587 {
2588         __u64 initial_sectors;
2589
2590         calculate_journal_section_size(ic);
2591         initial_sectors = SB_SECTORS + (__u64)ic->journal_section_sectors * ic->journal_sections;
2592         if (initial_sectors + METADATA_PADDING_SECTORS >= ic->meta_device_sectors || initial_sectors > UINT_MAX)
2593                 return -EINVAL;
2594         ic->initial_sectors = initial_sectors;
2595
2596         if (!ic->meta_dev) {
2597                 sector_t last_sector, last_area, last_offset;
2598
2599                 ic->metadata_run = roundup((__u64)ic->tag_size << (ic->sb->log2_interleave_sectors - ic->sb->log2_sectors_per_block),
2600                                            (__u64)(1 << SECTOR_SHIFT << METADATA_PADDING_SECTORS)) >> SECTOR_SHIFT;
2601                 if (!(ic->metadata_run & (ic->metadata_run - 1)))
2602                         ic->log2_metadata_run = __ffs(ic->metadata_run);
2603                 else
2604                         ic->log2_metadata_run = -1;
2605
2606                 get_area_and_offset(ic, ic->provided_data_sectors - 1, &last_area, &last_offset);
2607                 last_sector = get_data_sector(ic, last_area, last_offset);
2608                 if (last_sector < ic->start || last_sector >= ic->meta_device_sectors)
2609                         return -EINVAL;
2610         } else {
2611                 __u64 meta_size = (ic->provided_data_sectors >> ic->sb->log2_sectors_per_block) * ic->tag_size;
2612                 meta_size = (meta_size + ((1U << (ic->log2_buffer_sectors + SECTOR_SHIFT)) - 1))
2613                                 >> (ic->log2_buffer_sectors + SECTOR_SHIFT);
2614                 meta_size <<= ic->log2_buffer_sectors;
2615                 if (ic->initial_sectors + meta_size < ic->initial_sectors ||
2616                     ic->initial_sectors + meta_size > ic->meta_device_sectors)
2617                         return -EINVAL;
2618                 ic->metadata_run = 1;
2619                 ic->log2_metadata_run = 0;
2620         }
2621
2622         return 0;
2623 }
2624
2625 static int initialize_superblock(struct dm_integrity_c *ic, unsigned journal_sectors, unsigned interleave_sectors)
2626 {
2627         unsigned journal_sections;
2628         int test_bit;
2629
2630         memset(ic->sb, 0, SB_SECTORS << SECTOR_SHIFT);
2631         memcpy(ic->sb->magic, SB_MAGIC, 8);
2632         ic->sb->integrity_tag_size = cpu_to_le16(ic->tag_size);
2633         ic->sb->log2_sectors_per_block = __ffs(ic->sectors_per_block);
2634         if (ic->journal_mac_alg.alg_string)
2635                 ic->sb->flags |= cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC);
2636
2637         calculate_journal_section_size(ic);
2638         journal_sections = journal_sectors / ic->journal_section_sectors;
2639         if (!journal_sections)
2640                 journal_sections = 1;
2641
2642         if (!ic->meta_dev) {
2643                 ic->sb->journal_sections = cpu_to_le32(journal_sections);
2644                 if (!interleave_sectors)
2645                         interleave_sectors = DEFAULT_INTERLEAVE_SECTORS;
2646                 ic->sb->log2_interleave_sectors = __fls(interleave_sectors);
2647                 ic->sb->log2_interleave_sectors = max((__u8)MIN_LOG2_INTERLEAVE_SECTORS, ic->sb->log2_interleave_sectors);
2648                 ic->sb->log2_interleave_sectors = min((__u8)MAX_LOG2_INTERLEAVE_SECTORS, ic->sb->log2_interleave_sectors);
2649
2650                 ic->provided_data_sectors = 0;
2651                 for (test_bit = fls64(ic->meta_device_sectors) - 1; test_bit >= 3; test_bit--) {
2652                         __u64 prev_data_sectors = ic->provided_data_sectors;
2653
2654                         ic->provided_data_sectors |= (sector_t)1 << test_bit;
2655                         if (calculate_device_limits(ic))
2656                                 ic->provided_data_sectors = prev_data_sectors;
2657                 }
2658                 if (!ic->provided_data_sectors)
2659                         return -EINVAL;
2660         } else {
2661                 ic->sb->log2_interleave_sectors = 0;
2662                 ic->provided_data_sectors = ic->data_device_sectors;
2663                 ic->provided_data_sectors &= ~(sector_t)(ic->sectors_per_block - 1);
2664
2665 try_smaller_buffer:
2666                 ic->sb->journal_sections = cpu_to_le32(0);
2667                 for (test_bit = fls(journal_sections) - 1; test_bit >= 0; test_bit--) {
2668                         __u32 prev_journal_sections = le32_to_cpu(ic->sb->journal_sections);
2669                         __u32 test_journal_sections = prev_journal_sections | (1U << test_bit);
2670                         if (test_journal_sections > journal_sections)
2671                                 continue;
2672                         ic->sb->journal_sections = cpu_to_le32(test_journal_sections);
2673                         if (calculate_device_limits(ic))
2674                                 ic->sb->journal_sections = cpu_to_le32(prev_journal_sections);
2675
2676                 }
2677                 if (!le32_to_cpu(ic->sb->journal_sections)) {
2678                         if (ic->log2_buffer_sectors > 3) {
2679                                 ic->log2_buffer_sectors--;
2680                                 goto try_smaller_buffer;
2681                         }
2682                         return -EINVAL;
2683                 }
2684         }
2685
2686         ic->sb->provided_data_sectors = cpu_to_le64(ic->provided_data_sectors);
2687
2688         sb_set_version(ic);
2689
2690         return 0;
2691 }
2692
2693 static void dm_integrity_set(struct dm_target *ti, struct dm_integrity_c *ic)
2694 {
2695         struct gendisk *disk = dm_disk(dm_table_get_md(ti->table));
2696         struct blk_integrity bi;
2697
2698         memset(&bi, 0, sizeof(bi));
2699         bi.profile = &dm_integrity_profile;
2700         bi.tuple_size = ic->tag_size;
2701         bi.tag_size = bi.tuple_size;
2702         bi.interval_exp = ic->sb->log2_sectors_per_block + SECTOR_SHIFT;
2703
2704         blk_integrity_register(disk, &bi);
2705         blk_queue_max_integrity_segments(disk->queue, UINT_MAX);
2706 }
2707
2708 static void dm_integrity_free_page_list(struct dm_integrity_c *ic, struct page_list *pl)
2709 {
2710         unsigned i;
2711
2712         if (!pl)
2713                 return;
2714         for (i = 0; i < ic->journal_pages; i++)
2715                 if (pl[i].page)
2716                         __free_page(pl[i].page);
2717         kvfree(pl);
2718 }
2719
2720 static struct page_list *dm_integrity_alloc_page_list(struct dm_integrity_c *ic)
2721 {
2722         size_t page_list_desc_size = ic->journal_pages * sizeof(struct page_list);
2723         struct page_list *pl;
2724         unsigned i;
2725
2726         pl = kvmalloc(page_list_desc_size, GFP_KERNEL | __GFP_ZERO);
2727         if (!pl)
2728                 return NULL;
2729
2730         for (i = 0; i < ic->journal_pages; i++) {
2731                 pl[i].page = alloc_page(GFP_KERNEL);
2732                 if (!pl[i].page) {
2733                         dm_integrity_free_page_list(ic, pl);
2734                         return NULL;
2735                 }
2736                 if (i)
2737                         pl[i - 1].next = &pl[i];
2738         }
2739
2740         return pl;
2741 }
2742
2743 static void dm_integrity_free_journal_scatterlist(struct dm_integrity_c *ic, struct scatterlist **sl)
2744 {
2745         unsigned i;
2746         for (i = 0; i < ic->journal_sections; i++)
2747                 kvfree(sl[i]);
2748         kvfree(sl);
2749 }
2750
2751 static struct scatterlist **dm_integrity_alloc_journal_scatterlist(struct dm_integrity_c *ic, struct page_list *pl)
2752 {
2753         struct scatterlist **sl;
2754         unsigned i;
2755
2756         sl = kvmalloc_array(ic->journal_sections,
2757                             sizeof(struct scatterlist *),
2758                             GFP_KERNEL | __GFP_ZERO);
2759         if (!sl)
2760                 return NULL;
2761
2762         for (i = 0; i < ic->journal_sections; i++) {
2763                 struct scatterlist *s;
2764                 unsigned start_index, start_offset;
2765                 unsigned end_index, end_offset;
2766                 unsigned n_pages;
2767                 unsigned idx;
2768
2769                 page_list_location(ic, i, 0, &start_index, &start_offset);
2770                 page_list_location(ic, i, ic->journal_section_sectors - 1, &end_index, &end_offset);
2771
2772                 n_pages = (end_index - start_index + 1);
2773
2774                 s = kvmalloc_array(n_pages, sizeof(struct scatterlist),
2775                                    GFP_KERNEL);
2776                 if (!s) {
2777                         dm_integrity_free_journal_scatterlist(ic, sl);
2778                         return NULL;
2779                 }
2780
2781                 sg_init_table(s, n_pages);
2782                 for (idx = start_index; idx <= end_index; idx++) {
2783                         char *va = lowmem_page_address(pl[idx].page);
2784                         unsigned start = 0, end = PAGE_SIZE;
2785                         if (idx == start_index)
2786                                 start = start_offset;
2787                         if (idx == end_index)
2788                                 end = end_offset + (1 << SECTOR_SHIFT);
2789                         sg_set_buf(&s[idx - start_index], va + start, end - start);
2790                 }
2791
2792                 sl[i] = s;
2793         }
2794
2795         return sl;
2796 }
2797
2798 static void free_alg(struct alg_spec *a)
2799 {
2800         kzfree(a->alg_string);
2801         kzfree(a->key);
2802         memset(a, 0, sizeof *a);
2803 }
2804
2805 static int get_alg_and_key(const char *arg, struct alg_spec *a, char **error, char *error_inval)
2806 {
2807         char *k;
2808
2809         free_alg(a);
2810
2811         a->alg_string = kstrdup(strchr(arg, ':') + 1, GFP_KERNEL);
2812         if (!a->alg_string)
2813                 goto nomem;
2814
2815         k = strchr(a->alg_string, ':');
2816         if (k) {
2817                 *k = 0;
2818                 a->key_string = k + 1;
2819                 if (strlen(a->key_string) & 1)
2820                         goto inval;
2821
2822                 a->key_size = strlen(a->key_string) / 2;
2823                 a->key = kmalloc(a->key_size, GFP_KERNEL);
2824                 if (!a->key)
2825                         goto nomem;
2826                 if (hex2bin(a->key, a->key_string, a->key_size))
2827                         goto inval;
2828         }
2829
2830         return 0;
2831 inval:
2832         *error = error_inval;
2833         return -EINVAL;
2834 nomem:
2835         *error = "Out of memory for an argument";
2836         return -ENOMEM;
2837 }
2838
2839 static int get_mac(struct crypto_shash **hash, struct alg_spec *a, char **error,
2840                    char *error_alg, char *error_key)
2841 {
2842         int r;
2843
2844         if (a->alg_string) {
2845                 *hash = crypto_alloc_shash(a->alg_string, 0, CRYPTO_ALG_ASYNC);
2846                 if (IS_ERR(*hash)) {
2847                         *error = error_alg;
2848                         r = PTR_ERR(*hash);
2849                         *hash = NULL;
2850                         return r;
2851                 }
2852
2853                 if (a->key) {
2854                         r = crypto_shash_setkey(*hash, a->key, a->key_size);
2855                         if (r) {
2856                                 *error = error_key;
2857                                 return r;
2858                         }
2859                 } else if (crypto_shash_get_flags(*hash) & CRYPTO_TFM_NEED_KEY) {
2860                         *error = error_key;
2861                         return -ENOKEY;
2862                 }
2863         }
2864
2865         return 0;
2866 }
2867
2868 static int create_journal(struct dm_integrity_c *ic, char **error)
2869 {
2870         int r = 0;
2871         unsigned i;
2872         __u64 journal_pages, journal_desc_size, journal_tree_size;
2873         unsigned char *crypt_data = NULL, *crypt_iv = NULL;
2874         struct skcipher_request *req = NULL;
2875
2876         ic->commit_ids[0] = cpu_to_le64(0x1111111111111111ULL);
2877         ic->commit_ids[1] = cpu_to_le64(0x2222222222222222ULL);
2878         ic->commit_ids[2] = cpu_to_le64(0x3333333333333333ULL);
2879         ic->commit_ids[3] = cpu_to_le64(0x4444444444444444ULL);
2880
2881         journal_pages = roundup((__u64)ic->journal_sections * ic->journal_section_sectors,
2882                                 PAGE_SIZE >> SECTOR_SHIFT) >> (PAGE_SHIFT - SECTOR_SHIFT);
2883         journal_desc_size = journal_pages * sizeof(struct page_list);
2884         if (journal_pages >= totalram_pages - totalhigh_pages || journal_desc_size > ULONG_MAX) {
2885                 *error = "Journal doesn't fit into memory";
2886                 r = -ENOMEM;
2887                 goto bad;
2888         }
2889         ic->journal_pages = journal_pages;
2890
2891         ic->journal = dm_integrity_alloc_page_list(ic);
2892         if (!ic->journal) {
2893                 *error = "Could not allocate memory for journal";
2894                 r = -ENOMEM;
2895                 goto bad;
2896         }
2897         if (ic->journal_crypt_alg.alg_string) {
2898                 unsigned ivsize, blocksize;
2899                 struct journal_completion comp;
2900
2901                 comp.ic = ic;
2902                 ic->journal_crypt = crypto_alloc_skcipher(ic->journal_crypt_alg.alg_string, 0, 0);
2903                 if (IS_ERR(ic->journal_crypt)) {
2904                         *error = "Invalid journal cipher";
2905                         r = PTR_ERR(ic->journal_crypt);
2906                         ic->journal_crypt = NULL;
2907                         goto bad;
2908                 }
2909                 ivsize = crypto_skcipher_ivsize(ic->journal_crypt);
2910                 blocksize = crypto_skcipher_blocksize(ic->journal_crypt);
2911
2912                 if (ic->journal_crypt_alg.key) {
2913                         r = crypto_skcipher_setkey(ic->journal_crypt, ic->journal_crypt_alg.key,
2914                                                    ic->journal_crypt_alg.key_size);
2915                         if (r) {
2916                                 *error = "Error setting encryption key";
2917                                 goto bad;
2918                         }
2919                 }
2920                 DEBUG_print("cipher %s, block size %u iv size %u\n",
2921                             ic->journal_crypt_alg.alg_string, blocksize, ivsize);
2922
2923                 ic->journal_io = dm_integrity_alloc_page_list(ic);
2924                 if (!ic->journal_io) {
2925                         *error = "Could not allocate memory for journal io";
2926                         r = -ENOMEM;
2927                         goto bad;
2928                 }
2929
2930                 if (blocksize == 1) {
2931                         struct scatterlist *sg;
2932
2933                         req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
2934                         if (!req) {
2935                                 *error = "Could not allocate crypt request";
2936                                 r = -ENOMEM;
2937                                 goto bad;
2938                         }
2939
2940                         crypt_iv = kmalloc(ivsize, GFP_KERNEL);
2941                         if (!crypt_iv) {
2942                                 *error = "Could not allocate iv";
2943                                 r = -ENOMEM;
2944                                 goto bad;
2945                         }
2946
2947                         ic->journal_xor = dm_integrity_alloc_page_list(ic);
2948                         if (!ic->journal_xor) {
2949                                 *error = "Could not allocate memory for journal xor";
2950                                 r = -ENOMEM;
2951                                 goto bad;
2952                         }
2953
2954                         sg = kvmalloc_array(ic->journal_pages + 1,
2955                                             sizeof(struct scatterlist),
2956                                             GFP_KERNEL);
2957                         if (!sg) {
2958                                 *error = "Unable to allocate sg list";
2959                                 r = -ENOMEM;
2960                                 goto bad;
2961                         }
2962                         sg_init_table(sg, ic->journal_pages + 1);
2963                         for (i = 0; i < ic->journal_pages; i++) {
2964                                 char *va = lowmem_page_address(ic->journal_xor[i].page);
2965                                 clear_page(va);
2966                                 sg_set_buf(&sg[i], va, PAGE_SIZE);
2967                         }
2968                         sg_set_buf(&sg[i], &ic->commit_ids, sizeof ic->commit_ids);
2969                         memset(crypt_iv, 0x00, ivsize);
2970
2971                         skcipher_request_set_crypt(req, sg, sg, PAGE_SIZE * ic->journal_pages + sizeof ic->commit_ids, crypt_iv);
2972                         init_completion(&comp.comp);
2973                         comp.in_flight = (atomic_t)ATOMIC_INIT(1);
2974                         if (do_crypt(true, req, &comp))
2975                                 wait_for_completion(&comp.comp);
2976                         kvfree(sg);
2977                         r = dm_integrity_failed(ic);
2978                         if (r) {
2979                                 *error = "Unable to encrypt journal";
2980                                 goto bad;
2981                         }
2982                         DEBUG_bytes(lowmem_page_address(ic->journal_xor[0].page), 64, "xor data");
2983
2984                         crypto_free_skcipher(ic->journal_crypt);
2985                         ic->journal_crypt = NULL;
2986                 } else {
2987                         unsigned crypt_len = roundup(ivsize, blocksize);
2988
2989                         req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
2990                         if (!req) {
2991                                 *error = "Could not allocate crypt request";
2992                                 r = -ENOMEM;
2993                                 goto bad;
2994                         }
2995
2996                         crypt_iv = kmalloc(ivsize, GFP_KERNEL);
2997                         if (!crypt_iv) {
2998                                 *error = "Could not allocate iv";
2999                                 r = -ENOMEM;
3000                                 goto bad;
3001                         }
3002
3003                         crypt_data = kmalloc(crypt_len, GFP_KERNEL);
3004                         if (!crypt_data) {
3005                                 *error = "Unable to allocate crypt data";
3006                                 r = -ENOMEM;
3007                                 goto bad;
3008                         }
3009
3010                         ic->journal_scatterlist = dm_integrity_alloc_journal_scatterlist(ic, ic->journal);
3011                         if (!ic->journal_scatterlist) {
3012                                 *error = "Unable to allocate sg list";
3013                                 r = -ENOMEM;
3014                                 goto bad;
3015                         }
3016                         ic->journal_io_scatterlist = dm_integrity_alloc_journal_scatterlist(ic, ic->journal_io);
3017                         if (!ic->journal_io_scatterlist) {
3018                                 *error = "Unable to allocate sg list";
3019                                 r = -ENOMEM;
3020                                 goto bad;
3021                         }
3022                         ic->sk_requests = kvmalloc_array(ic->journal_sections,
3023                                                          sizeof(struct skcipher_request *),
3024                                                          GFP_KERNEL | __GFP_ZERO);
3025                         if (!ic->sk_requests) {
3026                                 *error = "Unable to allocate sk requests";
3027                                 r = -ENOMEM;
3028                                 goto bad;
3029                         }
3030                         for (i = 0; i < ic->journal_sections; i++) {
3031                                 struct scatterlist sg;
3032                                 struct skcipher_request *section_req;
3033                                 __u32 section_le = cpu_to_le32(i);
3034
3035                                 memset(crypt_iv, 0x00, ivsize);
3036                                 memset(crypt_data, 0x00, crypt_len);
3037                                 memcpy(crypt_data, &section_le, min((size_t)crypt_len, sizeof(section_le)));
3038
3039                                 sg_init_one(&sg, crypt_data, crypt_len);
3040                                 skcipher_request_set_crypt(req, &sg, &sg, crypt_len, crypt_iv);
3041                                 init_completion(&comp.comp);
3042                                 comp.in_flight = (atomic_t)ATOMIC_INIT(1);
3043                                 if (do_crypt(true, req, &comp))
3044                                         wait_for_completion(&comp.comp);
3045
3046                                 r = dm_integrity_failed(ic);
3047                                 if (r) {
3048                                         *error = "Unable to generate iv";
3049                                         goto bad;
3050                                 }
3051
3052                                 section_req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
3053                                 if (!section_req) {
3054                                         *error = "Unable to allocate crypt request";
3055                                         r = -ENOMEM;
3056                                         goto bad;
3057                                 }
3058                                 section_req->iv = kmalloc_array(ivsize, 2,
3059                                                                 GFP_KERNEL);
3060                                 if (!section_req->iv) {
3061                                         skcipher_request_free(section_req);
3062                                         *error = "Unable to allocate iv";
3063                                         r = -ENOMEM;
3064                                         goto bad;
3065                                 }
3066                                 memcpy(section_req->iv + ivsize, crypt_data, ivsize);
3067                                 section_req->cryptlen = (size_t)ic->journal_section_sectors << SECTOR_SHIFT;
3068                                 ic->sk_requests[i] = section_req;
3069                                 DEBUG_bytes(crypt_data, ivsize, "iv(%u)", i);
3070                         }
3071                 }
3072         }
3073
3074         for (i = 0; i < N_COMMIT_IDS; i++) {
3075                 unsigned j;
3076 retest_commit_id:
3077                 for (j = 0; j < i; j++) {
3078                         if (ic->commit_ids[j] == ic->commit_ids[i]) {
3079                                 ic->commit_ids[i] = cpu_to_le64(le64_to_cpu(ic->commit_ids[i]) + 1);
3080                                 goto retest_commit_id;
3081                         }
3082                 }
3083                 DEBUG_print("commit id %u: %016llx\n", i, ic->commit_ids[i]);
3084         }
3085
3086         journal_tree_size = (__u64)ic->journal_entries * sizeof(struct journal_node);
3087         if (journal_tree_size > ULONG_MAX) {
3088                 *error = "Journal doesn't fit into memory";
3089                 r = -ENOMEM;
3090                 goto bad;
3091         }
3092         ic->journal_tree = kvmalloc(journal_tree_size, GFP_KERNEL);
3093         if (!ic->journal_tree) {
3094                 *error = "Could not allocate memory for journal tree";
3095                 r = -ENOMEM;
3096         }
3097 bad:
3098         kfree(crypt_data);
3099         kfree(crypt_iv);
3100         skcipher_request_free(req);
3101
3102         return r;
3103 }
3104
3105 /*
3106  * Construct a integrity mapping
3107  *
3108  * Arguments:
3109  *      device
3110  *      offset from the start of the device
3111  *      tag size
3112  *      D - direct writes, J - journal writes, R - recovery mode
3113  *      number of optional arguments
3114  *      optional arguments:
3115  *              journal_sectors
3116  *              interleave_sectors
3117  *              buffer_sectors
3118  *              journal_watermark
3119  *              commit_time
3120  *              internal_hash
3121  *              journal_crypt
3122  *              journal_mac
3123  *              block_size
3124  */
3125 static int dm_integrity_ctr(struct dm_target *ti, unsigned argc, char **argv)
3126 {
3127         struct dm_integrity_c *ic;
3128         char dummy;
3129         int r;
3130         unsigned extra_args;
3131         struct dm_arg_set as;
3132         static const struct dm_arg _args[] = {
3133                 {0, 12, "Invalid number of feature args"},
3134         };
3135         unsigned journal_sectors, interleave_sectors, buffer_sectors, journal_watermark, sync_msec;
3136         bool recalculate;
3137         bool should_write_sb;
3138         __u64 threshold;
3139         unsigned long long start;
3140
3141 #define DIRECT_ARGUMENTS        4
3142
3143         if (argc <= DIRECT_ARGUMENTS) {
3144                 ti->error = "Invalid argument count";
3145                 return -EINVAL;
3146         }
3147
3148         ic = kzalloc(sizeof(struct dm_integrity_c), GFP_KERNEL);
3149         if (!ic) {
3150                 ti->error = "Cannot allocate integrity context";
3151                 return -ENOMEM;
3152         }
3153         ti->private = ic;
3154         ti->per_io_data_size = sizeof(struct dm_integrity_io);
3155         ic->ti = ti;
3156
3157         ic->in_progress = RB_ROOT;
3158         INIT_LIST_HEAD(&ic->wait_list);
3159         init_waitqueue_head(&ic->endio_wait);
3160         bio_list_init(&ic->flush_bio_list);
3161         init_waitqueue_head(&ic->copy_to_journal_wait);
3162         init_completion(&ic->crypto_backoff);
3163         atomic64_set(&ic->number_of_mismatches, 0);
3164
3165         r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &ic->dev);
3166         if (r) {
3167                 ti->error = "Device lookup failed";
3168                 goto bad;
3169         }
3170
3171         if (sscanf(argv[1], "%llu%c", &start, &dummy) != 1 || start != (sector_t)start) {
3172                 ti->error = "Invalid starting offset";
3173                 r = -EINVAL;
3174                 goto bad;
3175         }
3176         ic->start = start;
3177
3178         if (strcmp(argv[2], "-")) {
3179                 if (sscanf(argv[2], "%u%c", &ic->tag_size, &dummy) != 1 || !ic->tag_size) {
3180                         ti->error = "Invalid tag size";
3181                         r = -EINVAL;
3182                         goto bad;
3183                 }
3184         }
3185
3186         if (!strcmp(argv[3], "J") || !strcmp(argv[3], "D") || !strcmp(argv[3], "R"))
3187                 ic->mode = argv[3][0];
3188         else {
3189                 ti->error = "Invalid mode (expecting J, D, R)";
3190                 r = -EINVAL;
3191                 goto bad;
3192         }
3193
3194         journal_sectors = 0;
3195         interleave_sectors = DEFAULT_INTERLEAVE_SECTORS;
3196         buffer_sectors = DEFAULT_BUFFER_SECTORS;
3197         journal_watermark = DEFAULT_JOURNAL_WATERMARK;
3198         sync_msec = DEFAULT_SYNC_MSEC;
3199         recalculate = false;
3200         ic->sectors_per_block = 1;
3201
3202         as.argc = argc - DIRECT_ARGUMENTS;
3203         as.argv = argv + DIRECT_ARGUMENTS;
3204         r = dm_read_arg_group(_args, &as, &extra_args, &ti->error);
3205         if (r)
3206                 goto bad;
3207
3208         while (extra_args--) {
3209                 const char *opt_string;
3210                 unsigned val;
3211                 opt_string = dm_shift_arg(&as);
3212                 if (!opt_string) {
3213                         r = -EINVAL;
3214                         ti->error = "Not enough feature arguments";
3215                         goto bad;
3216                 }
3217                 if (sscanf(opt_string, "journal_sectors:%u%c", &val, &dummy) == 1)
3218                         journal_sectors = val ? val : 1;
3219                 else if (sscanf(opt_string, "interleave_sectors:%u%c", &val, &dummy) == 1)
3220                         interleave_sectors = val;
3221                 else if (sscanf(opt_string, "buffer_sectors:%u%c", &val, &dummy) == 1)
3222                         buffer_sectors = val;
3223                 else if (sscanf(opt_string, "journal_watermark:%u%c", &val, &dummy) == 1 && val <= 100)
3224                         journal_watermark = val;
3225                 else if (sscanf(opt_string, "commit_time:%u%c", &val, &dummy) == 1)
3226                         sync_msec = val;
3227                 else if (!strncmp(opt_string, "meta_device:", strlen("meta_device:"))) {
3228                         if (ic->meta_dev) {
3229                                 dm_put_device(ti, ic->meta_dev);
3230                                 ic->meta_dev = NULL;
3231                         }
3232                         r = dm_get_device(ti, strchr(opt_string, ':') + 1, dm_table_get_mode(ti->table), &ic->meta_dev);
3233                         if (r) {
3234                                 ti->error = "Device lookup failed";
3235                                 goto bad;
3236                         }
3237                 } else if (sscanf(opt_string, "block_size:%u%c", &val, &dummy) == 1) {
3238                         if (val < 1 << SECTOR_SHIFT ||
3239                             val > MAX_SECTORS_PER_BLOCK << SECTOR_SHIFT ||
3240                             (val & (val -1))) {
3241                                 r = -EINVAL;
3242                                 ti->error = "Invalid block_size argument";
3243                                 goto bad;
3244                         }
3245                         ic->sectors_per_block = val >> SECTOR_SHIFT;
3246                 } else if (!strncmp(opt_string, "internal_hash:", strlen("internal_hash:"))) {
3247                         r = get_alg_and_key(opt_string, &ic->internal_hash_alg, &ti->error,
3248                                             "Invalid internal_hash argument");
3249                         if (r)
3250                                 goto bad;
3251                 } else if (!strncmp(opt_string, "journal_crypt:", strlen("journal_crypt:"))) {
3252                         r = get_alg_and_key(opt_string, &ic->journal_crypt_alg, &ti->error,
3253                                             "Invalid journal_crypt argument");
3254                         if (r)
3255                                 goto bad;
3256                 } else if (!strncmp(opt_string, "journal_mac:", strlen("journal_mac:"))) {
3257                         r = get_alg_and_key(opt_string, &ic->journal_mac_alg,  &ti->error,
3258                                             "Invalid journal_mac argument");
3259                         if (r)
3260                                 goto bad;
3261                 } else if (!strcmp(opt_string, "recalculate")) {
3262                         recalculate = true;
3263                 } else if (!strcmp(opt_string, "legacy_recalculate")) {
3264                         ic->legacy_recalculate = true;
3265                 } else {
3266                         r = -EINVAL;
3267                         ti->error = "Invalid argument";
3268                         goto bad;
3269                 }
3270         }
3271
3272         ic->data_device_sectors = i_size_read(ic->dev->bdev->bd_inode) >> SECTOR_SHIFT;
3273         if (!ic->meta_dev)
3274                 ic->meta_device_sectors = ic->data_device_sectors;
3275         else
3276                 ic->meta_device_sectors = i_size_read(ic->meta_dev->bdev->bd_inode) >> SECTOR_SHIFT;
3277
3278         if (!journal_sectors) {
3279                 journal_sectors = min((sector_t)DEFAULT_MAX_JOURNAL_SECTORS,
3280                         ic->data_device_sectors >> DEFAULT_JOURNAL_SIZE_FACTOR);
3281         }
3282
3283         if (!buffer_sectors)
3284                 buffer_sectors = 1;
3285         ic->log2_buffer_sectors = min((int)__fls(buffer_sectors), 31 - SECTOR_SHIFT);
3286
3287         r = get_mac(&ic->internal_hash, &ic->internal_hash_alg, &ti->error,
3288                     "Invalid internal hash", "Error setting internal hash key");
3289         if (r)
3290                 goto bad;
3291
3292         r = get_mac(&ic->journal_mac, &ic->journal_mac_alg, &ti->error,
3293                     "Invalid journal mac", "Error setting journal mac key");
3294         if (r)
3295                 goto bad;
3296
3297         if (!ic->tag_size) {
3298                 if (!ic->internal_hash) {
3299                         ti->error = "Unknown tag size";
3300                         r = -EINVAL;
3301                         goto bad;
3302                 }
3303                 ic->tag_size = crypto_shash_digestsize(ic->internal_hash);
3304         }
3305         if (ic->tag_size > MAX_TAG_SIZE) {
3306                 ti->error = "Too big tag size";
3307                 r = -EINVAL;
3308                 goto bad;
3309         }
3310         if (!(ic->tag_size & (ic->tag_size - 1)))
3311                 ic->log2_tag_size = __ffs(ic->tag_size);
3312         else
3313                 ic->log2_tag_size = -1;
3314
3315         ic->autocommit_jiffies = msecs_to_jiffies(sync_msec);
3316         ic->autocommit_msec = sync_msec;
3317         timer_setup(&ic->autocommit_timer, autocommit_fn, 0);
3318
3319         ic->io = dm_io_client_create();
3320         if (IS_ERR(ic->io)) {
3321                 r = PTR_ERR(ic->io);
3322                 ic->io = NULL;
3323                 ti->error = "Cannot allocate dm io";
3324                 goto bad;
3325         }
3326
3327         r = mempool_init_slab_pool(&ic->journal_io_mempool, JOURNAL_IO_MEMPOOL, journal_io_cache);
3328         if (r) {
3329                 ti->error = "Cannot allocate mempool";
3330                 goto bad;
3331         }
3332
3333         ic->metadata_wq = alloc_workqueue("dm-integrity-metadata",
3334                                           WQ_MEM_RECLAIM, METADATA_WORKQUEUE_MAX_ACTIVE);
3335         if (!ic->metadata_wq) {
3336                 ti->error = "Cannot allocate workqueue";
3337                 r = -ENOMEM;
3338                 goto bad;
3339         }
3340
3341         /*
3342          * If this workqueue were percpu, it would cause bio reordering
3343          * and reduced performance.
3344          */
3345         ic->wait_wq = alloc_workqueue("dm-integrity-wait", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3346         if (!ic->wait_wq) {
3347                 ti->error = "Cannot allocate workqueue";
3348                 r = -ENOMEM;
3349                 goto bad;
3350         }
3351
3352         ic->offload_wq = alloc_workqueue("dm-integrity-offload", WQ_MEM_RECLAIM,
3353                                           METADATA_WORKQUEUE_MAX_ACTIVE);
3354         if (!ic->offload_wq) {
3355                 ti->error = "Cannot allocate workqueue";
3356                 r = -ENOMEM;
3357                 goto bad;
3358         }
3359
3360         ic->commit_wq = alloc_workqueue("dm-integrity-commit", WQ_MEM_RECLAIM, 1);
3361         if (!ic->commit_wq) {
3362                 ti->error = "Cannot allocate workqueue";
3363                 r = -ENOMEM;
3364                 goto bad;
3365         }
3366         INIT_WORK(&ic->commit_work, integrity_commit);
3367
3368         if (ic->mode == 'J') {
3369                 ic->writer_wq = alloc_workqueue("dm-integrity-writer", WQ_MEM_RECLAIM, 1);
3370                 if (!ic->writer_wq) {
3371                         ti->error = "Cannot allocate workqueue";
3372                         r = -ENOMEM;
3373                         goto bad;
3374                 }
3375                 INIT_WORK(&ic->writer_work, integrity_writer);
3376         }
3377
3378         ic->sb = alloc_pages_exact(SB_SECTORS << SECTOR_SHIFT, GFP_KERNEL);
3379         if (!ic->sb) {
3380                 r = -ENOMEM;
3381                 ti->error = "Cannot allocate superblock area";
3382                 goto bad;
3383         }
3384
3385         r = sync_rw_sb(ic, REQ_OP_READ, 0);
3386         if (r) {
3387                 ti->error = "Error reading superblock";
3388                 goto bad;
3389         }
3390         should_write_sb = false;
3391         if (memcmp(ic->sb->magic, SB_MAGIC, 8)) {
3392                 if (ic->mode != 'R') {
3393                         if (memchr_inv(ic->sb, 0, SB_SECTORS << SECTOR_SHIFT)) {
3394                                 r = -EINVAL;
3395                                 ti->error = "The device is not initialized";
3396                                 goto bad;
3397                         }
3398                 }
3399
3400                 r = initialize_superblock(ic, journal_sectors, interleave_sectors);
3401                 if (r) {
3402                         ti->error = "Could not initialize superblock";
3403                         goto bad;
3404                 }
3405                 if (ic->mode != 'R')
3406                         should_write_sb = true;
3407         }
3408
3409         if (!ic->sb->version || ic->sb->version > SB_VERSION_2) {
3410                 r = -EINVAL;
3411                 ti->error = "Unknown version";
3412                 goto bad;
3413         }
3414         if (le16_to_cpu(ic->sb->integrity_tag_size) != ic->tag_size) {
3415                 r = -EINVAL;
3416                 ti->error = "Tag size doesn't match the information in superblock";
3417                 goto bad;
3418         }
3419         if (ic->sb->log2_sectors_per_block != __ffs(ic->sectors_per_block)) {
3420                 r = -EINVAL;
3421                 ti->error = "Block size doesn't match the information in superblock";
3422                 goto bad;
3423         }
3424         if (!le32_to_cpu(ic->sb->journal_sections)) {
3425                 r = -EINVAL;
3426                 ti->error = "Corrupted superblock, journal_sections is 0";
3427                 goto bad;
3428         }
3429         /* make sure that ti->max_io_len doesn't overflow */
3430         if (!ic->meta_dev) {
3431                 if (ic->sb->log2_interleave_sectors < MIN_LOG2_INTERLEAVE_SECTORS ||
3432                     ic->sb->log2_interleave_sectors > MAX_LOG2_INTERLEAVE_SECTORS) {
3433                         r = -EINVAL;
3434                         ti->error = "Invalid interleave_sectors in the superblock";
3435                         goto bad;
3436                 }
3437         } else {
3438                 if (ic->sb->log2_interleave_sectors) {
3439                         r = -EINVAL;
3440                         ti->error = "Invalid interleave_sectors in the superblock";
3441                         goto bad;
3442                 }
3443         }
3444         ic->provided_data_sectors = le64_to_cpu(ic->sb->provided_data_sectors);
3445         if (ic->provided_data_sectors != le64_to_cpu(ic->sb->provided_data_sectors)) {
3446                 /* test for overflow */
3447                 r = -EINVAL;
3448                 ti->error = "The superblock has 64-bit device size, but the kernel was compiled with 32-bit sectors";
3449                 goto bad;
3450         }
3451         if (!!(ic->sb->flags & cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC)) != !!ic->journal_mac_alg.alg_string) {
3452                 r = -EINVAL;
3453                 ti->error = "Journal mac mismatch";
3454                 goto bad;
3455         }
3456
3457 try_smaller_buffer:
3458         r = calculate_device_limits(ic);
3459         if (r) {
3460                 if (ic->meta_dev) {
3461                         if (ic->log2_buffer_sectors > 3) {
3462                                 ic->log2_buffer_sectors--;
3463                                 goto try_smaller_buffer;
3464                         }
3465                 }
3466                 ti->error = "The device is too small";
3467                 goto bad;
3468         }
3469         if (!ic->meta_dev)
3470                 ic->log2_buffer_sectors = min(ic->log2_buffer_sectors, (__u8)__ffs(ic->metadata_run));
3471
3472         if (ti->len > ic->provided_data_sectors) {
3473                 r = -EINVAL;
3474                 ti->error = "Not enough provided sectors for requested mapping size";
3475                 goto bad;
3476         }
3477
3478
3479         threshold = (__u64)ic->journal_entries * (100 - journal_watermark);
3480         threshold += 50;
3481         do_div(threshold, 100);
3482         ic->free_sectors_threshold = threshold;
3483
3484         DEBUG_print("initialized:\n");
3485         DEBUG_print("   integrity_tag_size %u\n", le16_to_cpu(ic->sb->integrity_tag_size));
3486         DEBUG_print("   journal_entry_size %u\n", ic->journal_entry_size);
3487         DEBUG_print("   journal_entries_per_sector %u\n", ic->journal_entries_per_sector);
3488         DEBUG_print("   journal_section_entries %u\n", ic->journal_section_entries);
3489         DEBUG_print("   journal_section_sectors %u\n", ic->journal_section_sectors);
3490         DEBUG_print("   journal_sections %u\n", (unsigned)le32_to_cpu(ic->sb->journal_sections));
3491         DEBUG_print("   journal_entries %u\n", ic->journal_entries);
3492         DEBUG_print("   log2_interleave_sectors %d\n", ic->sb->log2_interleave_sectors);
3493         DEBUG_print("   data_device_sectors 0x%llx\n", (unsigned long long)ic->data_device_sectors);
3494         DEBUG_print("   initial_sectors 0x%x\n", ic->initial_sectors);
3495         DEBUG_print("   metadata_run 0x%x\n", ic->metadata_run);
3496         DEBUG_print("   log2_metadata_run %d\n", ic->log2_metadata_run);
3497         DEBUG_print("   provided_data_sectors 0x%llx (%llu)\n", (unsigned long long)ic->provided_data_sectors,
3498                     (unsigned long long)ic->provided_data_sectors);
3499         DEBUG_print("   log2_buffer_sectors %u\n", ic->log2_buffer_sectors);
3500
3501         if (recalculate && !(ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))) {
3502                 ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
3503                 ic->sb->recalc_sector = cpu_to_le64(0);
3504         }
3505
3506         if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
3507                 size_t recalc_tags_size;
3508                 if (!ic->internal_hash) {
3509                         r = -EINVAL;
3510                         ti->error = "Recalculate is only valid with internal hash";
3511                         goto bad;
3512                 }
3513                 ic->recalc_wq = alloc_workqueue("dm-intergrity-recalc", WQ_MEM_RECLAIM, 1);
3514                 if (!ic->recalc_wq ) {
3515                         ti->error = "Cannot allocate workqueue";
3516                         r = -ENOMEM;
3517                         goto bad;
3518                 }
3519                 INIT_WORK(&ic->recalc_work, integrity_recalc);
3520                 ic->recalc_buffer = vmalloc(RECALC_SECTORS << SECTOR_SHIFT);
3521                 if (!ic->recalc_buffer) {
3522                         ti->error = "Cannot allocate buffer for recalculating";
3523                         r = -ENOMEM;
3524                         goto bad;
3525                 }
3526                 recalc_tags_size = (RECALC_SECTORS >> ic->sb->log2_sectors_per_block) * ic->tag_size;
3527                 if (crypto_shash_digestsize(ic->internal_hash) > ic->tag_size)
3528                         recalc_tags_size += crypto_shash_digestsize(ic->internal_hash) - ic->tag_size;
3529                 ic->recalc_tags = kvmalloc(recalc_tags_size, GFP_KERNEL);
3530                 if (!ic->recalc_tags) {
3531                         ti->error = "Cannot allocate tags for recalculating";
3532                         r = -ENOMEM;
3533                         goto bad;
3534                 }
3535         } else {
3536                 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
3537                         ti->error = "Recalculate can only be specified with internal_hash";
3538                         r = -EINVAL;
3539                         goto bad;
3540                 }
3541         }
3542
3543         if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING) &&
3544             le64_to_cpu(ic->sb->recalc_sector) < ic->provided_data_sectors &&
3545             dm_integrity_disable_recalculate(ic)) {
3546                 ti->error = "Recalculating with HMAC is disabled for security reasons - if you really need it, use the argument \"legacy_recalculate\"";
3547                 r = -EOPNOTSUPP;
3548                 goto bad;
3549         }
3550
3551         ic->bufio = dm_bufio_client_create(ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev,
3552                         1U << (SECTOR_SHIFT + ic->log2_buffer_sectors), 1, 0, NULL, NULL);
3553         if (IS_ERR(ic->bufio)) {
3554                 r = PTR_ERR(ic->bufio);
3555                 ti->error = "Cannot initialize dm-bufio";
3556                 ic->bufio = NULL;
3557                 goto bad;
3558         }
3559         dm_bufio_set_sector_offset(ic->bufio, ic->start + ic->initial_sectors);
3560
3561         if (ic->mode != 'R') {
3562                 r = create_journal(ic, &ti->error);
3563                 if (r)
3564                         goto bad;
3565         }
3566
3567         if (should_write_sb) {
3568                 int r;
3569
3570                 init_journal(ic, 0, ic->journal_sections, 0);
3571                 r = dm_integrity_failed(ic);
3572                 if (unlikely(r)) {
3573                         ti->error = "Error initializing journal";
3574                         goto bad;
3575                 }
3576                 r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
3577                 if (r) {
3578                         ti->error = "Error initializing superblock";
3579                         goto bad;
3580                 }
3581                 ic->just_formatted = true;
3582         }
3583
3584         if (!ic->meta_dev) {
3585                 r = dm_set_target_max_io_len(ti, 1U << ic->sb->log2_interleave_sectors);
3586                 if (r)
3587                         goto bad;
3588         }
3589
3590         if (!ic->internal_hash)
3591                 dm_integrity_set(ti, ic);
3592
3593         ti->num_flush_bios = 1;
3594         ti->flush_supported = true;
3595
3596         return 0;
3597 bad:
3598         dm_integrity_dtr(ti);
3599         return r;
3600 }
3601
3602 static void dm_integrity_dtr(struct dm_target *ti)
3603 {
3604         struct dm_integrity_c *ic = ti->private;
3605
3606         BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
3607         BUG_ON(!list_empty(&ic->wait_list));
3608
3609         if (ic->metadata_wq)
3610                 destroy_workqueue(ic->metadata_wq);
3611         if (ic->wait_wq)
3612                 destroy_workqueue(ic->wait_wq);
3613         if (ic->offload_wq)
3614                 destroy_workqueue(ic->offload_wq);
3615         if (ic->commit_wq)
3616                 destroy_workqueue(ic->commit_wq);
3617         if (ic->writer_wq)
3618                 destroy_workqueue(ic->writer_wq);
3619         if (ic->recalc_wq)
3620                 destroy_workqueue(ic->recalc_wq);
3621         if (ic->recalc_buffer)
3622                 vfree(ic->recalc_buffer);
3623         if (ic->recalc_tags)
3624                 kvfree(ic->recalc_tags);
3625         if (ic->bufio)
3626                 dm_bufio_client_destroy(ic->bufio);
3627         mempool_exit(&ic->journal_io_mempool);
3628         if (ic->io)
3629                 dm_io_client_destroy(ic->io);
3630         if (ic->dev)
3631                 dm_put_device(ti, ic->dev);
3632         if (ic->meta_dev)
3633                 dm_put_device(ti, ic->meta_dev);
3634         dm_integrity_free_page_list(ic, ic->journal);
3635         dm_integrity_free_page_list(ic, ic->journal_io);
3636         dm_integrity_free_page_list(ic, ic->journal_xor);
3637         if (ic->journal_scatterlist)
3638                 dm_integrity_free_journal_scatterlist(ic, ic->journal_scatterlist);
3639         if (ic->journal_io_scatterlist)
3640                 dm_integrity_free_journal_scatterlist(ic, ic->journal_io_scatterlist);
3641         if (ic->sk_requests) {
3642                 unsigned i;
3643
3644                 for (i = 0; i < ic->journal_sections; i++) {
3645                         struct skcipher_request *req = ic->sk_requests[i];
3646                         if (req) {
3647                                 kzfree(req->iv);
3648                                 skcipher_request_free(req);
3649                         }
3650                 }
3651                 kvfree(ic->sk_requests);
3652         }
3653         kvfree(ic->journal_tree);
3654         if (ic->sb)
3655                 free_pages_exact(ic->sb, SB_SECTORS << SECTOR_SHIFT);
3656
3657         if (ic->internal_hash)
3658                 crypto_free_shash(ic->internal_hash);
3659         free_alg(&ic->internal_hash_alg);
3660
3661         if (ic->journal_crypt)
3662                 crypto_free_skcipher(ic->journal_crypt);
3663         free_alg(&ic->journal_crypt_alg);
3664
3665         if (ic->journal_mac)
3666                 crypto_free_shash(ic->journal_mac);
3667         free_alg(&ic->journal_mac_alg);
3668
3669         kfree(ic);
3670 }
3671
3672 static struct target_type integrity_target = {
3673         .name                   = "integrity",
3674         .version                = {1, 2, 0},
3675         .module                 = THIS_MODULE,
3676         .features               = DM_TARGET_SINGLETON | DM_TARGET_INTEGRITY,
3677         .ctr                    = dm_integrity_ctr,
3678         .dtr                    = dm_integrity_dtr,
3679         .map                    = dm_integrity_map,
3680         .postsuspend            = dm_integrity_postsuspend,
3681         .resume                 = dm_integrity_resume,
3682         .status                 = dm_integrity_status,
3683         .iterate_devices        = dm_integrity_iterate_devices,
3684         .io_hints               = dm_integrity_io_hints,
3685 };
3686
3687 int __init dm_integrity_init(void)
3688 {
3689         int r;
3690
3691         journal_io_cache = kmem_cache_create("integrity_journal_io",
3692                                              sizeof(struct journal_io), 0, 0, NULL);
3693         if (!journal_io_cache) {
3694                 DMERR("can't allocate journal io cache");
3695                 return -ENOMEM;
3696         }
3697
3698         r = dm_register_target(&integrity_target);
3699
3700         if (r < 0)
3701                 DMERR("register failed %d", r);
3702
3703         return r;
3704 }
3705
3706 void dm_integrity_exit(void)
3707 {
3708         dm_unregister_target(&integrity_target);
3709         kmem_cache_destroy(journal_io_cache);
3710 }
3711
3712 module_init(dm_integrity_init);
3713 module_exit(dm_integrity_exit);
3714
3715 MODULE_AUTHOR("Milan Broz");
3716 MODULE_AUTHOR("Mikulas Patocka");
3717 MODULE_DESCRIPTION(DM_NAME " target for integrity tags extension");
3718 MODULE_LICENSE("GPL");