drivers/md/dm-integrity.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright (C) 2016-2017 Red Hat, Inc. All rights reserved.
   4  * Copyright (C) 2016-2017 Milan Broz
   5  * Copyright (C) 2016-2017 Mikulas Patocka
   6  *
   7  * This file is released under the GPL.
   8  */
   9
  10 #include "dm-bio-record.h"
  11
  12 #include <linux/compiler.h>
  13 #include <linux/module.h>
  14 #include <linux/device-mapper.h>
  15 #include <linux/dm-io.h>
  16 #include <linux/vmalloc.h>
  17 #include <linux/sort.h>
  18 #include <linux/rbtree.h>
  19 #include <linux/delay.h>
  20 #include <linux/random.h>
  21 #include <linux/reboot.h>
  22 #include <crypto/hash.h>
  23 #include <crypto/skcipher.h>
  24 #include <linux/async_tx.h>
  25 #include <linux/dm-bufio.h>
  26
  27 #include "dm-audit.h"
  28
  29 #define DM_MSG_PREFIX "integrity"
  30
  31 #define DEFAULT_INTERLEAVE_SECTORS      32768
  32 #define DEFAULT_JOURNAL_SIZE_FACTOR     7
  33 #define DEFAULT_SECTORS_PER_BITMAP_BIT  32768
  34 #define DEFAULT_BUFFER_SECTORS          128
  35 #define DEFAULT_JOURNAL_WATERMARK       50
  36 #define DEFAULT_SYNC_MSEC               10000
  37 #define DEFAULT_MAX_JOURNAL_SECTORS     (IS_ENABLED(CONFIG_64BIT) ? 131072 : 8192)
  38 #define MIN_LOG2_INTERLEAVE_SECTORS     3
  39 #define MAX_LOG2_INTERLEAVE_SECTORS     31
  40 #define METADATA_WORKQUEUE_MAX_ACTIVE   16
  41 #define RECALC_SECTORS                  (IS_ENABLED(CONFIG_64BIT) ? 32768 : 2048)
  42 #define RECALC_WRITE_SUPER              16
  43 #define BITMAP_BLOCK_SIZE               4096    /* don't change it */
  44 #define BITMAP_FLUSH_INTERVAL           (10 * HZ)
  45 #define DISCARD_FILLER                  0xf6
  46 #define SALT_SIZE                       16
  47
  48 /*
  49  * Warning - DEBUG_PRINT prints security-sensitive data to the log,
  50  * so it should not be enabled in the official kernel
  51  */
  52 //#define DEBUG_PRINT
  53 //#define INTERNAL_VERIFY
  54
  55 /*
  56  * On disk structures
  57  */
  58
  59 #define SB_MAGIC                        "integrt"
  60 #define SB_VERSION_1                    1
  61 #define SB_VERSION_2                    2
  62 #define SB_VERSION_3                    3
  63 #define SB_VERSION_4                    4
  64 #define SB_VERSION_5                    5
  65 #define SB_SECTORS                      8
  66 #define MAX_SECTORS_PER_BLOCK           8
  67
  68 struct superblock {
  69         __u8 magic[8];
  70         __u8 version;
  71         __u8 log2_interleave_sectors;
  72         __le16 integrity_tag_size;
  73         __le32 journal_sections;
  74         __le64 provided_data_sectors;   /* userspace uses this value */
  75         __le32 flags;
  76         __u8 log2_sectors_per_block;
  77         __u8 log2_blocks_per_bitmap_bit;
  78         __u8 pad[2];
  79         __le64 recalc_sector;
  80         __u8 pad2[8];
  81         __u8 salt[SALT_SIZE];
  82 };
  83
  84 #define SB_FLAG_HAVE_JOURNAL_MAC        0x1
  85 #define SB_FLAG_RECALCULATING           0x2
  86 #define SB_FLAG_DIRTY_BITMAP            0x4
  87 #define SB_FLAG_FIXED_PADDING           0x8
  88 #define SB_FLAG_FIXED_HMAC              0x10
  89
  90 #define JOURNAL_ENTRY_ROUNDUP           8
  91
  92 typedef __le64 commit_id_t;
  93 #define JOURNAL_MAC_PER_SECTOR          8
  94
  95 struct journal_entry {
  96         union {
  97                 struct {
  98                         __le32 sector_lo;
  99                         __le32 sector_hi;
 100                 } s;
 101                 __le64 sector;
 102         } u;
 103         commit_id_t last_bytes[];
 104         /* __u8 tag[0]; */
 105 };
 106
 107 #define journal_entry_tag(ic, je)               ((__u8 *)&(je)->last_bytes[(ic)->sectors_per_block])
 108
 109 #if BITS_PER_LONG == 64
 110 #define journal_entry_set_sector(je, x)         do { smp_wmb(); WRITE_ONCE((je)->u.sector, cpu_to_le64(x)); } while (0)
 111 #else
 112 #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)
 113 #endif
 114 #define journal_entry_get_sector(je)            le64_to_cpu((je)->u.sector)
 115 #define journal_entry_is_unused(je)             ((je)->u.s.sector_hi == cpu_to_le32(-1))
 116 #define journal_entry_set_unused(je)            ((je)->u.s.sector_hi = cpu_to_le32(-1))
 117 #define journal_entry_is_inprogress(je)         ((je)->u.s.sector_hi == cpu_to_le32(-2))
 118 #define journal_entry_set_inprogress(je)        ((je)->u.s.sector_hi = cpu_to_le32(-2))
 119
 120 #define JOURNAL_BLOCK_SECTORS           8
 121 #define JOURNAL_SECTOR_DATA             ((1 << SECTOR_SHIFT) - sizeof(commit_id_t))
 122 #define JOURNAL_MAC_SIZE                (JOURNAL_MAC_PER_SECTOR * JOURNAL_BLOCK_SECTORS)
 123
 124 struct journal_sector {
 125         struct_group(sectors,
 126                 __u8 entries[JOURNAL_SECTOR_DATA - JOURNAL_MAC_PER_SECTOR];
 127                 __u8 mac[JOURNAL_MAC_PER_SECTOR];
 128         );
 129         commit_id_t commit_id;
 130 };
 131
 132 #define MAX_TAG_SIZE                    (JOURNAL_SECTOR_DATA - JOURNAL_MAC_PER_SECTOR - offsetof(struct journal_entry, last_bytes[MAX_SECTORS_PER_BLOCK]))
 133
 134 #define METADATA_PADDING_SECTORS        8
 135
 136 #define N_COMMIT_IDS                    4
 137
 138 static unsigned char prev_commit_seq(unsigned char seq)
 139 {
 140         return (seq + N_COMMIT_IDS - 1) % N_COMMIT_IDS;
 141 }
 142
 143 static unsigned char next_commit_seq(unsigned char seq)
 144 {
 145         return (seq + 1) % N_COMMIT_IDS;
 146 }
 147
 148 /*
 149  * In-memory structures
 150  */
 151
 152 struct journal_node {
 153         struct rb_node node;
 154         sector_t sector;
 155 };
 156
 157 struct alg_spec {
 158         char *alg_string;
 159         char *key_string;
 160         __u8 *key;
 161         unsigned int key_size;
 162 };
 163
 164 struct dm_integrity_c {
 165         struct dm_dev *dev;
 166         struct dm_dev *meta_dev;
 167         unsigned int tag_size;
 168         __s8 log2_tag_size;
 169         sector_t start;
 170         mempool_t journal_io_mempool;
 171         struct dm_io_client *io;
 172         struct dm_bufio_client *bufio;
 173         struct workqueue_struct *metadata_wq;
 174         struct superblock *sb;
 175         unsigned int journal_pages;
 176         unsigned int n_bitmap_blocks;
 177
 178         struct page_list *journal;
 179         struct page_list *journal_io;
 180         struct page_list *journal_xor;
 181         struct page_list *recalc_bitmap;
 182         struct page_list *may_write_bitmap;
 183         struct bitmap_block_status *bbs;
 184         unsigned int bitmap_flush_interval;
 185         int synchronous_mode;
 186         struct bio_list synchronous_bios;
 187         struct delayed_work bitmap_flush_work;
 188
 189         struct crypto_skcipher *journal_crypt;
 190         struct scatterlist **journal_scatterlist;
 191         struct scatterlist **journal_io_scatterlist;
 192         struct skcipher_request **sk_requests;
 193
 194         struct crypto_shash *journal_mac;
 195
 196         struct journal_node *journal_tree;
 197         struct rb_root journal_tree_root;
 198
 199         sector_t provided_data_sectors;
 200
 201         unsigned short journal_entry_size;
 202         unsigned char journal_entries_per_sector;
 203         unsigned char journal_section_entries;
 204         unsigned short journal_section_sectors;
 205         unsigned int journal_sections;
 206         unsigned int journal_entries;
 207         sector_t data_device_sectors;
 208         sector_t meta_device_sectors;
 209         unsigned int initial_sectors;
 210         unsigned int metadata_run;
 211         __s8 log2_metadata_run;
 212         __u8 log2_buffer_sectors;
 213         __u8 sectors_per_block;
 214         __u8 log2_blocks_per_bitmap_bit;
 215
 216         unsigned char mode;
 217
 218         int failed;
 219
 220         struct crypto_shash *internal_hash;
 221
 222         struct dm_target *ti;
 223
 224         /* these variables are locked with endio_wait.lock */
 225         struct rb_root in_progress;
 226         struct list_head wait_list;
 227         wait_queue_head_t endio_wait;
 228         struct workqueue_struct *wait_wq;
 229         struct workqueue_struct *offload_wq;
 230
 231         unsigned char commit_seq;
 232         commit_id_t commit_ids[N_COMMIT_IDS];
 233
 234         unsigned int committed_section;
 235         unsigned int n_committed_sections;
 236
 237         unsigned int uncommitted_section;
 238         unsigned int n_uncommitted_sections;
 239
 240         unsigned int free_section;
 241         unsigned char free_section_entry;
 242         unsigned int free_sectors;
 243
 244         unsigned int free_sectors_threshold;
 245
 246         struct workqueue_struct *commit_wq;
 247         struct work_struct commit_work;
 248
 249         struct workqueue_struct *writer_wq;
 250         struct work_struct writer_work;
 251
 252         struct workqueue_struct *recalc_wq;
 253         struct work_struct recalc_work;
 254
 255         struct bio_list flush_bio_list;
 256
 257         unsigned long autocommit_jiffies;
 258         struct timer_list autocommit_timer;
 259         unsigned int autocommit_msec;
 260
 261         wait_queue_head_t copy_to_journal_wait;
 262
 263         struct completion crypto_backoff;
 264
 265         bool wrote_to_journal;
 266         bool journal_uptodate;
 267         bool just_formatted;
 268         bool recalculate_flag;
 269         bool reset_recalculate_flag;
 270         bool discard;
 271         bool fix_padding;
 272         bool fix_hmac;
 273         bool legacy_recalculate;
 274
 275         struct alg_spec internal_hash_alg;
 276         struct alg_spec journal_crypt_alg;
 277         struct alg_spec journal_mac_alg;
 278
 279         atomic64_t number_of_mismatches;
 280
 281         struct notifier_block reboot_notifier;
 282 };
 283
 284 struct dm_integrity_range {
 285         sector_t logical_sector;
 286         sector_t n_sectors;
 287         bool waiting;
 288         union {
 289                 struct rb_node node;
 290                 struct {
 291                         struct task_struct *task;
 292                         struct list_head wait_entry;
 293                 };
 294         };
 295 };
 296
 297 struct dm_integrity_io {
 298         struct work_struct work;
 299
 300         struct dm_integrity_c *ic;
 301         enum req_op op;
 302         bool fua;
 303
 304         struct dm_integrity_range range;
 305
 306         sector_t metadata_block;
 307         unsigned int metadata_offset;
 308
 309         atomic_t in_flight;
 310         blk_status_t bi_status;
 311
 312         struct completion *completion;
 313
 314         struct dm_bio_details bio_details;
 315 };
 316
 317 struct journal_completion {
 318         struct dm_integrity_c *ic;
 319         atomic_t in_flight;
 320         struct completion comp;
 321 };
 322
 323 struct journal_io {
 324         struct dm_integrity_range range;
 325         struct journal_completion *comp;
 326 };
 327
 328 struct bitmap_block_status {
 329         struct work_struct work;
 330         struct dm_integrity_c *ic;
 331         unsigned int idx;
 332         unsigned long *bitmap;
 333         struct bio_list bio_queue;
 334         spinlock_t bio_queue_lock;
 335
 336 };
 337
 338 static struct kmem_cache *journal_io_cache;
 339
 340 #define JOURNAL_IO_MEMPOOL      32
 341
 342 #ifdef DEBUG_PRINT
 343 #define DEBUG_print(x, ...)                     printk(KERN_DEBUG x, ##__VA_ARGS__)
 344 #define DEBUG_bytes(bytes, len, msg, ...)       printk(KERN_DEBUG msg "%s%*ph\n", ##__VA_ARGS__, \
 345                                                        len ? ": " : "", len, bytes)
 346 #else
 347 #define DEBUG_print(x, ...)                     do { } while (0)
 348 #define DEBUG_bytes(bytes, len, msg, ...)       do { } while (0)
 349 #endif
 350
 351 static void dm_integrity_prepare(struct request *rq)
 352 {
 353 }
 354
 355 static void dm_integrity_complete(struct request *rq, unsigned int nr_bytes)
 356 {
 357 }
 358
 359 /*
 360  * DM Integrity profile, protection is performed layer above (dm-crypt)
 361  */
 362 static const struct blk_integrity_profile dm_integrity_profile = {
 363         .name                   = "DM-DIF-EXT-TAG",
 364         .generate_fn            = NULL,
 365         .verify_fn              = NULL,
 366         .prepare_fn             = dm_integrity_prepare,
 367         .complete_fn            = dm_integrity_complete,
 368 };
 369
 370 static void dm_integrity_map_continue(struct dm_integrity_io *dio, bool from_map);
 371 static void integrity_bio_wait(struct work_struct *w);
 372 static void dm_integrity_dtr(struct dm_target *ti);
 373
 374 static void dm_integrity_io_error(struct dm_integrity_c *ic, const char *msg, int err)
 375 {
 376         if (err == -EILSEQ)
 377                 atomic64_inc(&ic->number_of_mismatches);
 378         if (!cmpxchg(&ic->failed, 0, err))
 379                 DMERR("Error on %s: %d", msg, err);
 380 }
 381
 382 static int dm_integrity_failed(struct dm_integrity_c *ic)
 383 {
 384         return READ_ONCE(ic->failed);
 385 }
 386
 387 static bool dm_integrity_disable_recalculate(struct dm_integrity_c *ic)
 388 {
 389         if (ic->legacy_recalculate)
 390                 return false;
 391         if (!(ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) ?
 392             ic->internal_hash_alg.key || ic->journal_mac_alg.key :
 393             ic->internal_hash_alg.key && !ic->journal_mac_alg.key)
 394                 return true;
 395         return false;
 396 }
 397
 398 static commit_id_t dm_integrity_commit_id(struct dm_integrity_c *ic, unsigned int i,
 399                                           unsigned int j, unsigned char seq)
 400 {
 401         /*
 402          * Xor the number with section and sector, so that if a piece of
 403          * journal is written at wrong place, it is detected.
 404          */
 405         return ic->commit_ids[seq] ^ cpu_to_le64(((__u64)i << 32) ^ j);
 406 }
 407
 408 static void get_area_and_offset(struct dm_integrity_c *ic, sector_t data_sector,
 409                                 sector_t *area, sector_t *offset)
 410 {
 411         if (!ic->meta_dev) {
 412                 __u8 log2_interleave_sectors = ic->sb->log2_interleave_sectors;
 413                 *area = data_sector >> log2_interleave_sectors;
 414                 *offset = (unsigned int)data_sector & ((1U << log2_interleave_sectors) - 1);
 415         } else {
 416                 *area = 0;
 417                 *offset = data_sector;
 418         }
 419 }
 420
 421 #define sector_to_block(ic, n)                                          \
 422 do {                                                                    \
 423         BUG_ON((n) & (unsigned int)((ic)->sectors_per_block - 1));              \
 424         (n) >>= (ic)->sb->log2_sectors_per_block;                       \
 425 } while (0)
 426
 427 static __u64 get_metadata_sector_and_offset(struct dm_integrity_c *ic, sector_t area,
 428                                             sector_t offset, unsigned int *metadata_offset)
 429 {
 430         __u64 ms;
 431         unsigned int mo;
 432
 433         ms = area << ic->sb->log2_interleave_sectors;
 434         if (likely(ic->log2_metadata_run >= 0))
 435                 ms += area << ic->log2_metadata_run;
 436         else
 437                 ms += area * ic->metadata_run;
 438         ms >>= ic->log2_buffer_sectors;
 439
 440         sector_to_block(ic, offset);
 441
 442         if (likely(ic->log2_tag_size >= 0)) {
 443                 ms += offset >> (SECTOR_SHIFT + ic->log2_buffer_sectors - ic->log2_tag_size);
 444                 mo = (offset << ic->log2_tag_size) & ((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - 1);
 445         } else {
 446                 ms += (__u64)offset * ic->tag_size >> (SECTOR_SHIFT + ic->log2_buffer_sectors);
 447                 mo = (offset * ic->tag_size) & ((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - 1);
 448         }
 449         *metadata_offset = mo;
 450         return ms;
 451 }
 452
 453 static sector_t get_data_sector(struct dm_integrity_c *ic, sector_t area, sector_t offset)
 454 {
 455         sector_t result;
 456
 457         if (ic->meta_dev)
 458                 return offset;
 459
 460         result = area << ic->sb->log2_interleave_sectors;
 461         if (likely(ic->log2_metadata_run >= 0))
 462                 result += (area + 1) << ic->log2_metadata_run;
 463         else
 464                 result += (area + 1) * ic->metadata_run;
 465
 466         result += (sector_t)ic->initial_sectors + offset;
 467         result += ic->start;
 468
 469         return result;
 470 }
 471
 472 static void wraparound_section(struct dm_integrity_c *ic, unsigned int *sec_ptr)
 473 {
 474         if (unlikely(*sec_ptr >= ic->journal_sections))
 475                 *sec_ptr -= ic->journal_sections;
 476 }
 477
 478 static void sb_set_version(struct dm_integrity_c *ic)
 479 {
 480         if (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC))
 481                 ic->sb->version = SB_VERSION_5;
 482         else if (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING))
 483                 ic->sb->version = SB_VERSION_4;
 484         else if (ic->mode == 'B' || ic->sb->flags & cpu_to_le32(SB_FLAG_DIRTY_BITMAP))
 485                 ic->sb->version = SB_VERSION_3;
 486         else if (ic->meta_dev || ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
 487                 ic->sb->version = SB_VERSION_2;
 488         else
 489                 ic->sb->version = SB_VERSION_1;
 490 }
 491
 492 static int sb_mac(struct dm_integrity_c *ic, bool wr)
 493 {
 494         SHASH_DESC_ON_STACK(desc, ic->journal_mac);
 495         int r;
 496         unsigned int size = crypto_shash_digestsize(ic->journal_mac);
 497
 498         if (sizeof(struct superblock) + size > 1 << SECTOR_SHIFT) {
 499                 dm_integrity_io_error(ic, "digest is too long", -EINVAL);
 500                 return -EINVAL;
 501         }
 502
 503         desc->tfm = ic->journal_mac;
 504
 505         r = crypto_shash_init(desc);
 506         if (unlikely(r < 0)) {
 507                 dm_integrity_io_error(ic, "crypto_shash_init", r);
 508                 return r;
 509         }
 510
 511         r = crypto_shash_update(desc, (__u8 *)ic->sb, (1 << SECTOR_SHIFT) - size);
 512         if (unlikely(r < 0)) {
 513                 dm_integrity_io_error(ic, "crypto_shash_update", r);
 514                 return r;
 515         }
 516
 517         if (likely(wr)) {
 518                 r = crypto_shash_final(desc, (__u8 *)ic->sb + (1 << SECTOR_SHIFT) - size);
 519                 if (unlikely(r < 0)) {
 520                         dm_integrity_io_error(ic, "crypto_shash_final", r);
 521                         return r;
 522                 }
 523         } else {
 524                 __u8 result[HASH_MAX_DIGESTSIZE];
 525
 526                 r = crypto_shash_final(desc, result);
 527                 if (unlikely(r < 0)) {
 528                         dm_integrity_io_error(ic, "crypto_shash_final", r);
 529                         return r;
 530                 }
 531                 if (memcmp((__u8 *)ic->sb + (1 << SECTOR_SHIFT) - size, result, size)) {
 532                         dm_integrity_io_error(ic, "superblock mac", -EILSEQ);
 533                         dm_audit_log_target(DM_MSG_PREFIX, "mac-superblock", ic->ti, 0);
 534                         return -EILSEQ;
 535                 }
 536         }
 537
 538         return 0;
 539 }
 540
 541 static int sync_rw_sb(struct dm_integrity_c *ic, blk_opf_t opf)
 542 {
 543         struct dm_io_request io_req;
 544         struct dm_io_region io_loc;
 545         const enum req_op op = opf & REQ_OP_MASK;
 546         int r;
 547
 548         io_req.bi_opf = opf;
 549         io_req.mem.type = DM_IO_KMEM;
 550         io_req.mem.ptr.addr = ic->sb;
 551         io_req.notify.fn = NULL;
 552         io_req.client = ic->io;
 553         io_loc.bdev = ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev;
 554         io_loc.sector = ic->start;
 555         io_loc.count = SB_SECTORS;
 556
 557         if (op == REQ_OP_WRITE) {
 558                 sb_set_version(ic);
 559                 if (ic->journal_mac && ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) {
 560                         r = sb_mac(ic, true);
 561                         if (unlikely(r))
 562                                 return r;
 563                 }
 564         }
 565
 566         r = dm_io(&io_req, 1, &io_loc, NULL);
 567         if (unlikely(r))
 568                 return r;
 569
 570         if (op == REQ_OP_READ) {
 571                 if (ic->mode != 'R' && ic->journal_mac && ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) {
 572                         r = sb_mac(ic, false);
 573                         if (unlikely(r))
 574                                 return r;
 575                 }
 576         }
 577
 578         return 0;
 579 }
 580
 581 #define BITMAP_OP_TEST_ALL_SET          0
 582 #define BITMAP_OP_TEST_ALL_CLEAR        1
 583 #define BITMAP_OP_SET                   2
 584 #define BITMAP_OP_CLEAR                 3
 585
 586 static bool block_bitmap_op(struct dm_integrity_c *ic, struct page_list *bitmap,
 587                             sector_t sector, sector_t n_sectors, int mode)
 588 {
 589         unsigned long bit, end_bit, this_end_bit, page, end_page;
 590         unsigned long *data;
 591
 592         if (unlikely(((sector | n_sectors) & ((1 << ic->sb->log2_sectors_per_block) - 1)) != 0)) {
 593                 DMCRIT("invalid bitmap access (%llx,%llx,%d,%d,%d)",
 594                         sector,
 595                         n_sectors,
 596                         ic->sb->log2_sectors_per_block,
 597                         ic->log2_blocks_per_bitmap_bit,
 598                         mode);
 599                 BUG();
 600         }
 601
 602         if (unlikely(!n_sectors))
 603                 return true;
 604
 605         bit = sector >> (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
 606         end_bit = (sector + n_sectors - 1) >>
 607                 (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
 608
 609         page = bit / (PAGE_SIZE * 8);
 610         bit %= PAGE_SIZE * 8;
 611
 612         end_page = end_bit / (PAGE_SIZE * 8);
 613         end_bit %= PAGE_SIZE * 8;
 614
 615 repeat:
 616         if (page < end_page)
 617                 this_end_bit = PAGE_SIZE * 8 - 1;
 618         else
 619                 this_end_bit = end_bit;
 620
 621         data = lowmem_page_address(bitmap[page].page);
 622
 623         if (mode == BITMAP_OP_TEST_ALL_SET) {
 624                 while (bit <= this_end_bit) {
 625                         if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
 626                                 do {
 627                                         if (data[bit / BITS_PER_LONG] != -1)
 628                                                 return false;
 629                                         bit += BITS_PER_LONG;
 630                                 } while (this_end_bit >= bit + BITS_PER_LONG - 1);
 631                                 continue;
 632                         }
 633                         if (!test_bit(bit, data))
 634                                 return false;
 635                         bit++;
 636                 }
 637         } else if (mode == BITMAP_OP_TEST_ALL_CLEAR) {
 638                 while (bit <= this_end_bit) {
 639                         if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
 640                                 do {
 641                                         if (data[bit / BITS_PER_LONG] != 0)
 642                                                 return false;
 643                                         bit += BITS_PER_LONG;
 644                                 } while (this_end_bit >= bit + BITS_PER_LONG - 1);
 645                                 continue;
 646                         }
 647                         if (test_bit(bit, data))
 648                                 return false;
 649                         bit++;
 650                 }
 651         } else if (mode == BITMAP_OP_SET) {
 652                 while (bit <= this_end_bit) {
 653                         if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
 654                                 do {
 655                                         data[bit / BITS_PER_LONG] = -1;
 656                                         bit += BITS_PER_LONG;
 657                                 } while (this_end_bit >= bit + BITS_PER_LONG - 1);
 658                                 continue;
 659                         }
 660                         __set_bit(bit, data);
 661                         bit++;
 662                 }
 663         } else if (mode == BITMAP_OP_CLEAR) {
 664                 if (!bit && this_end_bit == PAGE_SIZE * 8 - 1)
 665                         clear_page(data);
 666                 else {
 667                         while (bit <= this_end_bit) {
 668                                 if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
 669                                         do {
 670                                                 data[bit / BITS_PER_LONG] = 0;
 671                                                 bit += BITS_PER_LONG;
 672                                         } while (this_end_bit >= bit + BITS_PER_LONG - 1);
 673                                         continue;
 674                                 }
 675                                 __clear_bit(bit, data);
 676                                 bit++;
 677                         }
 678                 }
 679         } else {
 680                 BUG();
 681         }
 682
 683         if (unlikely(page < end_page)) {
 684                 bit = 0;
 685                 page++;
 686                 goto repeat;
 687         }
 688
 689         return true;
 690 }
 691
 692 static void block_bitmap_copy(struct dm_integrity_c *ic, struct page_list *dst, struct page_list *src)
 693 {
 694         unsigned int n_bitmap_pages = DIV_ROUND_UP(ic->n_bitmap_blocks, PAGE_SIZE / BITMAP_BLOCK_SIZE);
 695         unsigned int i;
 696
 697         for (i = 0; i < n_bitmap_pages; i++) {
 698                 unsigned long *dst_data = lowmem_page_address(dst[i].page);
 699                 unsigned long *src_data = lowmem_page_address(src[i].page);
 700
 701                 copy_page(dst_data, src_data);
 702         }
 703 }
 704
 705 static struct bitmap_block_status *sector_to_bitmap_block(struct dm_integrity_c *ic, sector_t sector)
 706 {
 707         unsigned int bit = sector >> (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
 708         unsigned int bitmap_block = bit / (BITMAP_BLOCK_SIZE * 8);
 709
 710         BUG_ON(bitmap_block >= ic->n_bitmap_blocks);
 711         return &ic->bbs[bitmap_block];
 712 }
 713
 714 static void access_journal_check(struct dm_integrity_c *ic, unsigned int section, unsigned int offset,
 715                                  bool e, const char *function)
 716 {
 717 #if defined(CONFIG_DM_DEBUG) || defined(INTERNAL_VERIFY)
 718         unsigned int limit = e ? ic->journal_section_entries : ic->journal_section_sectors;
 719
 720         if (unlikely(section >= ic->journal_sections) ||
 721             unlikely(offset >= limit)) {
 722                 DMCRIT("%s: invalid access at (%u,%u), limit (%u,%u)",
 723                        function, section, offset, ic->journal_sections, limit);
 724                 BUG();
 725         }
 726 #endif
 727 }
 728
 729 static void page_list_location(struct dm_integrity_c *ic, unsigned int section, unsigned int offset,
 730                                unsigned int *pl_index, unsigned int *pl_offset)
 731 {
 732         unsigned int sector;
 733
 734         access_journal_check(ic, section, offset, false, "page_list_location");
 735
 736         sector = section * ic->journal_section_sectors + offset;
 737
 738         *pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
 739         *pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
 740 }
 741
 742 static struct journal_sector *access_page_list(struct dm_integrity_c *ic, struct page_list *pl,
 743                                                unsigned int section, unsigned int offset, unsigned int *n_sectors)
 744 {
 745         unsigned int pl_index, pl_offset;
 746         char *va;
 747
 748         page_list_location(ic, section, offset, &pl_index, &pl_offset);
 749
 750         if (n_sectors)
 751                 *n_sectors = (PAGE_SIZE - pl_offset) >> SECTOR_SHIFT;
 752
 753         va = lowmem_page_address(pl[pl_index].page);
 754
 755         return (struct journal_sector *)(va + pl_offset);
 756 }
 757
 758 static struct journal_sector *access_journal(struct dm_integrity_c *ic, unsigned int section, unsigned int offset)
 759 {
 760         return access_page_list(ic, ic->journal, section, offset, NULL);
 761 }
 762
 763 static struct journal_entry *access_journal_entry(struct dm_integrity_c *ic, unsigned int section, unsigned int n)
 764 {
 765         unsigned int rel_sector, offset;
 766         struct journal_sector *js;
 767
 768         access_journal_check(ic, section, n, true, "access_journal_entry");
 769
 770         rel_sector = n % JOURNAL_BLOCK_SECTORS;
 771         offset = n / JOURNAL_BLOCK_SECTORS;
 772
 773         js = access_journal(ic, section, rel_sector);
 774         return (struct journal_entry *)((char *)js + offset * ic->journal_entry_size);
 775 }
 776
 777 static struct journal_sector *access_journal_data(struct dm_integrity_c *ic, unsigned int section, unsigned int n)
 778 {
 779         n <<= ic->sb->log2_sectors_per_block;
 780
 781         n += JOURNAL_BLOCK_SECTORS;
 782
 783         access_journal_check(ic, section, n, false, "access_journal_data");
 784
 785         return access_journal(ic, section, n);
 786 }
 787
 788 static void section_mac(struct dm_integrity_c *ic, unsigned int section, __u8 result[JOURNAL_MAC_SIZE])
 789 {
 790         SHASH_DESC_ON_STACK(desc, ic->journal_mac);
 791         int r;
 792         unsigned int j, size;
 793
 794         desc->tfm = ic->journal_mac;
 795
 796         r = crypto_shash_init(desc);
 797         if (unlikely(r < 0)) {
 798                 dm_integrity_io_error(ic, "crypto_shash_init", r);
 799                 goto err;
 800         }
 801
 802         if (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) {
 803                 __le64 section_le;
 804
 805                 r = crypto_shash_update(desc, (__u8 *)&ic->sb->salt, SALT_SIZE);
 806                 if (unlikely(r < 0)) {
 807                         dm_integrity_io_error(ic, "crypto_shash_update", r);
 808                         goto err;
 809                 }
 810
 811                 section_le = cpu_to_le64(section);
 812                 r = crypto_shash_update(desc, (__u8 *)&section_le, sizeof(section_le));
 813                 if (unlikely(r < 0)) {
 814                         dm_integrity_io_error(ic, "crypto_shash_update", r);
 815                         goto err;
 816                 }
 817         }
 818
 819         for (j = 0; j < ic->journal_section_entries; j++) {
 820                 struct journal_entry *je = access_journal_entry(ic, section, j);
 821
 822                 r = crypto_shash_update(desc, (__u8 *)&je->u.sector, sizeof(je->u.sector));
 823                 if (unlikely(r < 0)) {
 824                         dm_integrity_io_error(ic, "crypto_shash_update", r);
 825                         goto err;
 826                 }
 827         }
 828
 829         size = crypto_shash_digestsize(ic->journal_mac);
 830
 831         if (likely(size <= JOURNAL_MAC_SIZE)) {
 832                 r = crypto_shash_final(desc, result);
 833                 if (unlikely(r < 0)) {
 834                         dm_integrity_io_error(ic, "crypto_shash_final", r);
 835                         goto err;
 836                 }
 837                 memset(result + size, 0, JOURNAL_MAC_SIZE - size);
 838         } else {
 839                 __u8 digest[HASH_MAX_DIGESTSIZE];
 840
 841                 if (WARN_ON(size > sizeof(digest))) {
 842                         dm_integrity_io_error(ic, "digest_size", -EINVAL);
 843                         goto err;
 844                 }
 845                 r = crypto_shash_final(desc, digest);
 846                 if (unlikely(r < 0)) {
 847                         dm_integrity_io_error(ic, "crypto_shash_final", r);
 848                         goto err;
 849                 }
 850                 memcpy(result, digest, JOURNAL_MAC_SIZE);
 851         }
 852
 853         return;
 854 err:
 855         memset(result, 0, JOURNAL_MAC_SIZE);
 856 }
 857
 858 static void rw_section_mac(struct dm_integrity_c *ic, unsigned int section, bool wr)
 859 {
 860         __u8 result[JOURNAL_MAC_SIZE];
 861         unsigned int j;
 862
 863         if (!ic->journal_mac)
 864                 return;
 865
 866         section_mac(ic, section, result);
 867
 868         for (j = 0; j < JOURNAL_BLOCK_SECTORS; j++) {
 869                 struct journal_sector *js = access_journal(ic, section, j);
 870
 871                 if (likely(wr))
 872                         memcpy(&js->mac, result + (j * JOURNAL_MAC_PER_SECTOR), JOURNAL_MAC_PER_SECTOR);
 873                 else {
 874                         if (memcmp(&js->mac, result + (j * JOURNAL_MAC_PER_SECTOR), JOURNAL_MAC_PER_SECTOR)) {
 875                                 dm_integrity_io_error(ic, "journal mac", -EILSEQ);
 876                                 dm_audit_log_target(DM_MSG_PREFIX, "mac-journal", ic->ti, 0);
 877                         }
 878                 }
 879         }
 880 }
 881
 882 static void complete_journal_op(void *context)
 883 {
 884         struct journal_completion *comp = context;
 885
 886         BUG_ON(!atomic_read(&comp->in_flight));
 887         if (likely(atomic_dec_and_test(&comp->in_flight)))
 888                 complete(&comp->comp);
 889 }
 890
 891 static void xor_journal(struct dm_integrity_c *ic, bool encrypt, unsigned int section,
 892                         unsigned int n_sections, struct journal_completion *comp)
 893 {
 894         struct async_submit_ctl submit;
 895         size_t n_bytes = (size_t)(n_sections * ic->journal_section_sectors) << SECTOR_SHIFT;
 896         unsigned int pl_index, pl_offset, section_index;
 897         struct page_list *source_pl, *target_pl;
 898
 899         if (likely(encrypt)) {
 900                 source_pl = ic->journal;
 901                 target_pl = ic->journal_io;
 902         } else {
 903                 source_pl = ic->journal_io;
 904                 target_pl = ic->journal;
 905         }
 906
 907         page_list_location(ic, section, 0, &pl_index, &pl_offset);
 908
 909         atomic_add(roundup(pl_offset + n_bytes, PAGE_SIZE) >> PAGE_SHIFT, &comp->in_flight);
 910
 911         init_async_submit(&submit, ASYNC_TX_XOR_ZERO_DST, NULL, complete_journal_op, comp, NULL);
 912
 913         section_index = pl_index;
 914
 915         do {
 916                 size_t this_step;
 917                 struct page *src_pages[2];
 918                 struct page *dst_page;
 919
 920                 while (unlikely(pl_index == section_index)) {
 921                         unsigned int dummy;
 922
 923                         if (likely(encrypt))
 924                                 rw_section_mac(ic, section, true);
 925                         section++;
 926                         n_sections--;
 927                         if (!n_sections)
 928                                 break;
 929                         page_list_location(ic, section, 0, &section_index, &dummy);
 930                 }
 931
 932                 this_step = min(n_bytes, (size_t)PAGE_SIZE - pl_offset);
 933                 dst_page = target_pl[pl_index].page;
 934                 src_pages[0] = source_pl[pl_index].page;
 935                 src_pages[1] = ic->journal_xor[pl_index].page;
 936
 937                 async_xor(dst_page, src_pages, pl_offset, 2, this_step, &submit);
 938
 939                 pl_index++;
 940                 pl_offset = 0;
 941                 n_bytes -= this_step;
 942         } while (n_bytes);
 943
 944         BUG_ON(n_sections);
 945
 946         async_tx_issue_pending_all();
 947 }
 948
 949 static void complete_journal_encrypt(void *data, int err)
 950 {
 951         struct journal_completion *comp = data;
 952
 953         if (unlikely(err)) {
 954                 if (likely(err == -EINPROGRESS)) {
 955                         complete(&comp->ic->crypto_backoff);
 956                         return;
 957                 }
 958                 dm_integrity_io_error(comp->ic, "asynchronous encrypt", err);
 959         }
 960         complete_journal_op(comp);
 961 }
 962
 963 static bool do_crypt(bool encrypt, struct skcipher_request *req, struct journal_completion *comp)
 964 {
 965         int r;
 966
 967         skcipher_request_set_callback(req, CRYPTO_TFM_REQ_MAY_BACKLOG,
 968                                       complete_journal_encrypt, comp);
 969         if (likely(encrypt))
 970                 r = crypto_skcipher_encrypt(req);
 971         else
 972                 r = crypto_skcipher_decrypt(req);
 973         if (likely(!r))
 974                 return false;
 975         if (likely(r == -EINPROGRESS))
 976                 return true;
 977         if (likely(r == -EBUSY)) {
 978                 wait_for_completion(&comp->ic->crypto_backoff);
 979                 reinit_completion(&comp->ic->crypto_backoff);
 980                 return true;
 981         }
 982         dm_integrity_io_error(comp->ic, "encrypt", r);
 983         return false;
 984 }
 985
 986 static void crypt_journal(struct dm_integrity_c *ic, bool encrypt, unsigned int section,
 987                           unsigned int n_sections, struct journal_completion *comp)
 988 {
 989         struct scatterlist **source_sg;
 990         struct scatterlist **target_sg;
 991
 992         atomic_add(2, &comp->in_flight);
 993
 994         if (likely(encrypt)) {
 995                 source_sg = ic->journal_scatterlist;
 996                 target_sg = ic->journal_io_scatterlist;
 997         } else {
 998                 source_sg = ic->journal_io_scatterlist;
 999                 target_sg = ic->journal_scatterlist;
1000         }
1001
1002         do {
1003                 struct skcipher_request *req;
1004                 unsigned int ivsize;
1005                 char *iv;
1006
1007                 if (likely(encrypt))
1008                         rw_section_mac(ic, section, true);
1009
1010                 req = ic->sk_requests[section];
1011                 ivsize = crypto_skcipher_ivsize(ic->journal_crypt);
1012                 iv = req->iv;
1013
1014                 memcpy(iv, iv + ivsize, ivsize);
1015
1016                 req->src = source_sg[section];
1017                 req->dst = target_sg[section];
1018
1019                 if (unlikely(do_crypt(encrypt, req, comp)))
1020                         atomic_inc(&comp->in_flight);
1021
1022                 section++;
1023                 n_sections--;
1024         } while (n_sections);
1025
1026         atomic_dec(&comp->in_flight);
1027         complete_journal_op(comp);
1028 }
1029
1030 static void encrypt_journal(struct dm_integrity_c *ic, bool encrypt, unsigned int section,
1031                             unsigned int n_sections, struct journal_completion *comp)
1032 {
1033         if (ic->journal_xor)
1034                 return xor_journal(ic, encrypt, section, n_sections, comp);
1035         else
1036                 return crypt_journal(ic, encrypt, section, n_sections, comp);
1037 }
1038
1039 static void complete_journal_io(unsigned long error, void *context)
1040 {
1041         struct journal_completion *comp = context;
1042
1043         if (unlikely(error != 0))
1044                 dm_integrity_io_error(comp->ic, "writing journal", -EIO);
1045         complete_journal_op(comp);
1046 }
1047
1048 static void rw_journal_sectors(struct dm_integrity_c *ic, blk_opf_t opf,
1049                                unsigned int sector, unsigned int n_sectors,
1050                                struct journal_completion *comp)
1051 {
1052         struct dm_io_request io_req;
1053         struct dm_io_region io_loc;
1054         unsigned int pl_index, pl_offset;
1055         int r;
1056
1057         if (unlikely(dm_integrity_failed(ic))) {
1058                 if (comp)
1059                         complete_journal_io(-1UL, comp);
1060                 return;
1061         }
1062
1063         pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
1064         pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
1065
1066         io_req.bi_opf = opf;
1067         io_req.mem.type = DM_IO_PAGE_LIST;
1068         if (ic->journal_io)
1069                 io_req.mem.ptr.pl = &ic->journal_io[pl_index];
1070         else
1071                 io_req.mem.ptr.pl = &ic->journal[pl_index];
1072         io_req.mem.offset = pl_offset;
1073         if (likely(comp != NULL)) {
1074                 io_req.notify.fn = complete_journal_io;
1075                 io_req.notify.context = comp;
1076         } else {
1077                 io_req.notify.fn = NULL;
1078         }
1079         io_req.client = ic->io;
1080         io_loc.bdev = ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev;
1081         io_loc.sector = ic->start + SB_SECTORS + sector;
1082         io_loc.count = n_sectors;
1083
1084         r = dm_io(&io_req, 1, &io_loc, NULL);
1085         if (unlikely(r)) {
1086                 dm_integrity_io_error(ic, (opf & REQ_OP_MASK) == REQ_OP_READ ?
1087                                       "reading journal" : "writing journal", r);
1088                 if (comp) {
1089                         WARN_ONCE(1, "asynchronous dm_io failed: %d", r);
1090                         complete_journal_io(-1UL, comp);
1091                 }
1092         }
1093 }
1094
1095 static void rw_journal(struct dm_integrity_c *ic, blk_opf_t opf,
1096                        unsigned int section, unsigned int n_sections,
1097                        struct journal_completion *comp)
1098 {
1099         unsigned int sector, n_sectors;
1100
1101         sector = section * ic->journal_section_sectors;
1102         n_sectors = n_sections * ic->journal_section_sectors;
1103
1104         rw_journal_sectors(ic, opf, sector, n_sectors, comp);
1105 }
1106
1107 static void write_journal(struct dm_integrity_c *ic, unsigned int commit_start, unsigned int commit_sections)
1108 {
1109         struct journal_completion io_comp;
1110         struct journal_completion crypt_comp_1;
1111         struct journal_completion crypt_comp_2;
1112         unsigned int i;
1113
1114         io_comp.ic = ic;
1115         init_completion(&io_comp.comp);
1116
1117         if (commit_start + commit_sections <= ic->journal_sections) {
1118                 io_comp.in_flight = (atomic_t)ATOMIC_INIT(1);
1119                 if (ic->journal_io) {
1120                         crypt_comp_1.ic = ic;
1121                         init_completion(&crypt_comp_1.comp);
1122                         crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
1123                         encrypt_journal(ic, true, commit_start, commit_sections, &crypt_comp_1);
1124                         wait_for_completion_io(&crypt_comp_1.comp);
1125                 } else {
1126                         for (i = 0; i < commit_sections; i++)
1127                                 rw_section_mac(ic, commit_start + i, true);
1128                 }
1129                 rw_journal(ic, REQ_OP_WRITE | REQ_FUA | REQ_SYNC, commit_start,
1130                            commit_sections, &io_comp);
1131         } else {
1132                 unsigned int to_end;
1133
1134                 io_comp.in_flight = (atomic_t)ATOMIC_INIT(2);
1135                 to_end = ic->journal_sections - commit_start;
1136                 if (ic->journal_io) {
1137                         crypt_comp_1.ic = ic;
1138                         init_completion(&crypt_comp_1.comp);
1139                         crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
1140                         encrypt_journal(ic, true, commit_start, to_end, &crypt_comp_1);
1141                         if (try_wait_for_completion(&crypt_comp_1.comp)) {
1142                                 rw_journal(ic, REQ_OP_WRITE | REQ_FUA,
1143                                            commit_start, to_end, &io_comp);
1144                                 reinit_completion(&crypt_comp_1.comp);
1145                                 crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
1146                                 encrypt_journal(ic, true, 0, commit_sections - to_end, &crypt_comp_1);
1147                                 wait_for_completion_io(&crypt_comp_1.comp);
1148                         } else {
1149                                 crypt_comp_2.ic = ic;
1150                                 init_completion(&crypt_comp_2.comp);
1151                                 crypt_comp_2.in_flight = (atomic_t)ATOMIC_INIT(0);
1152                                 encrypt_journal(ic, true, 0, commit_sections - to_end, &crypt_comp_2);
1153                                 wait_for_completion_io(&crypt_comp_1.comp);
1154                                 rw_journal(ic, REQ_OP_WRITE | REQ_FUA, commit_start, to_end, &io_comp);
1155                                 wait_for_completion_io(&crypt_comp_2.comp);
1156                         }
1157                 } else {
1158                         for (i = 0; i < to_end; i++)
1159                                 rw_section_mac(ic, commit_start + i, true);
1160                         rw_journal(ic, REQ_OP_WRITE | REQ_FUA, commit_start, to_end, &io_comp);
1161                         for (i = 0; i < commit_sections - to_end; i++)
1162                                 rw_section_mac(ic, i, true);
1163                 }
1164                 rw_journal(ic, REQ_OP_WRITE | REQ_FUA, 0, commit_sections - to_end, &io_comp);
1165         }
1166
1167         wait_for_completion_io(&io_comp.comp);
1168 }
1169
1170 static void copy_from_journal(struct dm_integrity_c *ic, unsigned int section, unsigned int offset,
1171                               unsigned int n_sectors, sector_t target, io_notify_fn fn, void *data)
1172 {
1173         struct dm_io_request io_req;
1174         struct dm_io_region io_loc;
1175         int r;
1176         unsigned int sector, pl_index, pl_offset;
1177
1178         BUG_ON((target | n_sectors | offset) & (unsigned int)(ic->sectors_per_block - 1));
1179
1180         if (unlikely(dm_integrity_failed(ic))) {
1181                 fn(-1UL, data);
1182                 return;
1183         }
1184
1185         sector = section * ic->journal_section_sectors + JOURNAL_BLOCK_SECTORS + offset;
1186
1187         pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
1188         pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
1189
1190         io_req.bi_opf = REQ_OP_WRITE;
1191         io_req.mem.type = DM_IO_PAGE_LIST;
1192         io_req.mem.ptr.pl = &ic->journal[pl_index];
1193         io_req.mem.offset = pl_offset;
1194         io_req.notify.fn = fn;
1195         io_req.notify.context = data;
1196         io_req.client = ic->io;
1197         io_loc.bdev = ic->dev->bdev;
1198         io_loc.sector = target;
1199         io_loc.count = n_sectors;
1200
1201         r = dm_io(&io_req, 1, &io_loc, NULL);
1202         if (unlikely(r)) {
1203                 WARN_ONCE(1, "asynchronous dm_io failed: %d", r);
1204                 fn(-1UL, data);
1205         }
1206 }
1207
1208 static bool ranges_overlap(struct dm_integrity_range *range1, struct dm_integrity_range *range2)
1209 {
1210         return range1->logical_sector < range2->logical_sector + range2->n_sectors &&
1211                range1->logical_sector + range1->n_sectors > range2->logical_sector;
1212 }
1213
1214 static bool add_new_range(struct dm_integrity_c *ic, struct dm_integrity_range *new_range, bool check_waiting)
1215 {
1216         struct rb_node **n = &ic->in_progress.rb_node;
1217         struct rb_node *parent;
1218
1219         BUG_ON((new_range->logical_sector | new_range->n_sectors) & (unsigned int)(ic->sectors_per_block - 1));
1220
1221         if (likely(check_waiting)) {
1222                 struct dm_integrity_range *range;
1223
1224                 list_for_each_entry(range, &ic->wait_list, wait_entry) {
1225                         if (unlikely(ranges_overlap(range, new_range)))
1226                                 return false;
1227                 }
1228         }
1229
1230         parent = NULL;
1231
1232         while (*n) {
1233                 struct dm_integrity_range *range = container_of(*n, struct dm_integrity_range, node);
1234
1235                 parent = *n;
1236                 if (new_range->logical_sector + new_range->n_sectors <= range->logical_sector)
1237                         n = &range->node.rb_left;
1238                 else if (new_range->logical_sector >= range->logical_sector + range->n_sectors)
1239                         n = &range->node.rb_right;
1240                 else
1241                         return false;
1242         }
1243
1244         rb_link_node(&new_range->node, parent, n);
1245         rb_insert_color(&new_range->node, &ic->in_progress);
1246
1247         return true;
1248 }
1249
1250 static void remove_range_unlocked(struct dm_integrity_c *ic, struct dm_integrity_range *range)
1251 {
1252         rb_erase(&range->node, &ic->in_progress);
1253         while (unlikely(!list_empty(&ic->wait_list))) {
1254                 struct dm_integrity_range *last_range =
1255                         list_first_entry(&ic->wait_list, struct dm_integrity_range, wait_entry);
1256                 struct task_struct *last_range_task;
1257
1258                 last_range_task = last_range->task;
1259                 list_del(&last_range->wait_entry);
1260                 if (!add_new_range(ic, last_range, false)) {
1261                         last_range->task = last_range_task;
1262                         list_add(&last_range->wait_entry, &ic->wait_list);
1263                         break;
1264                 }
1265                 last_range->waiting = false;
1266                 wake_up_process(last_range_task);
1267         }
1268 }
1269
1270 static void remove_range(struct dm_integrity_c *ic, struct dm_integrity_range *range)
1271 {
1272         unsigned long flags;
1273
1274         spin_lock_irqsave(&ic->endio_wait.lock, flags);
1275         remove_range_unlocked(ic, range);
1276         spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
1277 }
1278
1279 static void wait_and_add_new_range(struct dm_integrity_c *ic, struct dm_integrity_range *new_range)
1280 {
1281         new_range->waiting = true;
1282         list_add_tail(&new_range->wait_entry, &ic->wait_list);
1283         new_range->task = current;
1284         do {
1285                 __set_current_state(TASK_UNINTERRUPTIBLE);
1286                 spin_unlock_irq(&ic->endio_wait.lock);
1287                 io_schedule();
1288                 spin_lock_irq(&ic->endio_wait.lock);
1289         } while (unlikely(new_range->waiting));
1290 }
1291
1292 static void add_new_range_and_wait(struct dm_integrity_c *ic, struct dm_integrity_range *new_range)
1293 {
1294         if (unlikely(!add_new_range(ic, new_range, true)))
1295                 wait_and_add_new_range(ic, new_range);
1296 }
1297
1298 static void init_journal_node(struct journal_node *node)
1299 {
1300         RB_CLEAR_NODE(&node->node);
1301         node->sector = (sector_t)-1;
1302 }
1303
1304 static void add_journal_node(struct dm_integrity_c *ic, struct journal_node *node, sector_t sector)
1305 {
1306         struct rb_node **link;
1307         struct rb_node *parent;
1308
1309         node->sector = sector;
1310         BUG_ON(!RB_EMPTY_NODE(&node->node));
1311
1312         link = &ic->journal_tree_root.rb_node;
1313         parent = NULL;
1314
1315         while (*link) {
1316                 struct journal_node *j;
1317
1318                 parent = *link;
1319                 j = container_of(parent, struct journal_node, node);
1320                 if (sector < j->sector)
1321                         link = &j->node.rb_left;
1322                 else
1323                         link = &j->node.rb_right;
1324         }
1325
1326         rb_link_node(&node->node, parent, link);
1327         rb_insert_color(&node->node, &ic->journal_tree_root);
1328 }
1329
1330 static void remove_journal_node(struct dm_integrity_c *ic, struct journal_node *node)
1331 {
1332         BUG_ON(RB_EMPTY_NODE(&node->node));
1333         rb_erase(&node->node, &ic->journal_tree_root);
1334         init_journal_node(node);
1335 }
1336
1337 #define NOT_FOUND       (-1U)
1338
1339 static unsigned int find_journal_node(struct dm_integrity_c *ic, sector_t sector, sector_t *next_sector)
1340 {
1341         struct rb_node *n = ic->journal_tree_root.rb_node;
1342         unsigned int found = NOT_FOUND;
1343
1344         *next_sector = (sector_t)-1;
1345         while (n) {
1346                 struct journal_node *j = container_of(n, struct journal_node, node);
1347
1348                 if (sector == j->sector)
1349                         found = j - ic->journal_tree;
1350
1351                 if (sector < j->sector) {
1352                         *next_sector = j->sector;
1353                         n = j->node.rb_left;
1354                 } else
1355                         n = j->node.rb_right;
1356         }
1357
1358         return found;
1359 }
1360
1361 static bool test_journal_node(struct dm_integrity_c *ic, unsigned int pos, sector_t sector)
1362 {
1363         struct journal_node *node, *next_node;
1364         struct rb_node *next;
1365
1366         if (unlikely(pos >= ic->journal_entries))
1367                 return false;
1368         node = &ic->journal_tree[pos];
1369         if (unlikely(RB_EMPTY_NODE(&node->node)))
1370                 return false;
1371         if (unlikely(node->sector != sector))
1372                 return false;
1373
1374         next = rb_next(&node->node);
1375         if (unlikely(!next))
1376                 return true;
1377
1378         next_node = container_of(next, struct journal_node, node);
1379         return next_node->sector != sector;
1380 }
1381
1382 static bool find_newer_committed_node(struct dm_integrity_c *ic, struct journal_node *node)
1383 {
1384         struct rb_node *next;
1385         struct journal_node *next_node;
1386         unsigned int next_section;
1387
1388         BUG_ON(RB_EMPTY_NODE(&node->node));
1389
1390         next = rb_next(&node->node);
1391         if (unlikely(!next))
1392                 return false;
1393
1394         next_node = container_of(next, struct journal_node, node);
1395
1396         if (next_node->sector != node->sector)
1397                 return false;
1398
1399         next_section = (unsigned int)(next_node - ic->journal_tree) / ic->journal_section_entries;
1400         if (next_section >= ic->committed_section &&
1401             next_section < ic->committed_section + ic->n_committed_sections)
1402                 return true;
1403         if (next_section + ic->journal_sections < ic->committed_section + ic->n_committed_sections)
1404                 return true;
1405
1406         return false;
1407 }
1408
1409 #define TAG_READ        0
1410 #define TAG_WRITE       1
1411 #define TAG_CMP         2
1412
1413 static int dm_integrity_rw_tag(struct dm_integrity_c *ic, unsigned char *tag, sector_t *metadata_block,
1414                                unsigned int *metadata_offset, unsigned int total_size, int op)
1415 {
1416 #define MAY_BE_FILLER           1
1417 #define MAY_BE_HASH             2
1418         unsigned int hash_offset = 0;
1419         unsigned int may_be = MAY_BE_HASH | (ic->discard ? MAY_BE_FILLER : 0);
1420
1421         do {
1422                 unsigned char *data, *dp;
1423                 struct dm_buffer *b;
1424                 unsigned int to_copy;
1425                 int r;
1426
1427                 r = dm_integrity_failed(ic);
1428                 if (unlikely(r))
1429                         return r;
1430
1431                 data = dm_bufio_read(ic->bufio, *metadata_block, &b);
1432                 if (IS_ERR(data))
1433                         return PTR_ERR(data);
1434
1435                 to_copy = min((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - *metadata_offset, total_size);
1436                 dp = data + *metadata_offset;
1437                 if (op == TAG_READ) {
1438                         memcpy(tag, dp, to_copy);
1439                 } else if (op == TAG_WRITE) {
1440                         if (memcmp(dp, tag, to_copy)) {
1441                                 memcpy(dp, tag, to_copy);
1442                                 dm_bufio_mark_partial_buffer_dirty(b, *metadata_offset, *metadata_offset + to_copy);
1443                         }
1444                 } else {
1445                         /* e.g.: op == TAG_CMP */
1446
1447                         if (likely(is_power_of_2(ic->tag_size))) {
1448                                 if (unlikely(memcmp(dp, tag, to_copy)))
1449                                         if (unlikely(!ic->discard) ||
1450                                             unlikely(memchr_inv(dp, DISCARD_FILLER, to_copy) != NULL)) {
1451                                                 goto thorough_test;
1452                                 }
1453                         } else {
1454                                 unsigned int i, ts;
1455 thorough_test:
1456                                 ts = total_size;
1457
1458                                 for (i = 0; i < to_copy; i++, ts--) {
1459                                         if (unlikely(dp[i] != tag[i]))
1460                                                 may_be &= ~MAY_BE_HASH;
1461                                         if (likely(dp[i] != DISCARD_FILLER))
1462                                                 may_be &= ~MAY_BE_FILLER;
1463                                         hash_offset++;
1464                                         if (unlikely(hash_offset == ic->tag_size)) {
1465                                                 if (unlikely(!may_be)) {
1466                                                         dm_bufio_release(b);
1467                                                         return ts;
1468                                                 }
1469                                                 hash_offset = 0;
1470                                                 may_be = MAY_BE_HASH | (ic->discard ? MAY_BE_FILLER : 0);
1471                                         }
1472                                 }
1473                         }
1474                 }
1475                 dm_bufio_release(b);
1476
1477                 tag += to_copy;
1478                 *metadata_offset += to_copy;
1479                 if (unlikely(*metadata_offset == 1U << SECTOR_SHIFT << ic->log2_buffer_sectors)) {
1480                         (*metadata_block)++;
1481                         *metadata_offset = 0;
1482                 }
1483
1484                 if (unlikely(!is_power_of_2(ic->tag_size)))
1485                         hash_offset = (hash_offset + to_copy) % ic->tag_size;
1486
1487                 total_size -= to_copy;
1488         } while (unlikely(total_size));
1489
1490         return 0;
1491 #undef MAY_BE_FILLER
1492 #undef MAY_BE_HASH
1493 }
1494
1495 struct flush_request {
1496         struct dm_io_request io_req;
1497         struct dm_io_region io_reg;
1498         struct dm_integrity_c *ic;
1499         struct completion comp;
1500 };
1501
1502 static void flush_notify(unsigned long error, void *fr_)
1503 {
1504         struct flush_request *fr = fr_;
1505
1506         if (unlikely(error != 0))
1507                 dm_integrity_io_error(fr->ic, "flushing disk cache", -EIO);
1508         complete(&fr->comp);
1509 }
1510
1511 static void dm_integrity_flush_buffers(struct dm_integrity_c *ic, bool flush_data)
1512 {
1513         int r;
1514         struct flush_request fr;
1515
1516         if (!ic->meta_dev)
1517                 flush_data = false;
1518         if (flush_data) {
1519                 fr.io_req.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC,
1520                 fr.io_req.mem.type = DM_IO_KMEM,
1521                 fr.io_req.mem.ptr.addr = NULL,
1522                 fr.io_req.notify.fn = flush_notify,
1523                 fr.io_req.notify.context = &fr;
1524                 fr.io_req.client = dm_bufio_get_dm_io_client(ic->bufio),
1525                 fr.io_reg.bdev = ic->dev->bdev,
1526                 fr.io_reg.sector = 0,
1527                 fr.io_reg.count = 0,
1528                 fr.ic = ic;
1529                 init_completion(&fr.comp);
1530                 r = dm_io(&fr.io_req, 1, &fr.io_reg, NULL);
1531                 BUG_ON(r);
1532         }
1533
1534         r = dm_bufio_write_dirty_buffers(ic->bufio);
1535         if (unlikely(r))
1536                 dm_integrity_io_error(ic, "writing tags", r);
1537
1538         if (flush_data)
1539                 wait_for_completion(&fr.comp);
1540 }
1541
1542 static void sleep_on_endio_wait(struct dm_integrity_c *ic)
1543 {
1544         DECLARE_WAITQUEUE(wait, current);
1545
1546         __add_wait_queue(&ic->endio_wait, &wait);
1547         __set_current_state(TASK_UNINTERRUPTIBLE);
1548         spin_unlock_irq(&ic->endio_wait.lock);
1549         io_schedule();
1550         spin_lock_irq(&ic->endio_wait.lock);
1551         __remove_wait_queue(&ic->endio_wait, &wait);
1552 }
1553
1554 static void autocommit_fn(struct timer_list *t)
1555 {
1556         struct dm_integrity_c *ic = from_timer(ic, t, autocommit_timer);
1557
1558         if (likely(!dm_integrity_failed(ic)))
1559                 queue_work(ic->commit_wq, &ic->commit_work);
1560 }
1561
1562 static void schedule_autocommit(struct dm_integrity_c *ic)
1563 {
1564         if (!timer_pending(&ic->autocommit_timer))
1565                 mod_timer(&ic->autocommit_timer, jiffies + ic->autocommit_jiffies);
1566 }
1567
1568 static void submit_flush_bio(struct dm_integrity_c *ic, struct dm_integrity_io *dio)
1569 {
1570         struct bio *bio;
1571         unsigned long flags;
1572
1573         spin_lock_irqsave(&ic->endio_wait.lock, flags);
1574         bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1575         bio_list_add(&ic->flush_bio_list, bio);
1576         spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
1577
1578         queue_work(ic->commit_wq, &ic->commit_work);
1579 }
1580
1581 static void do_endio(struct dm_integrity_c *ic, struct bio *bio)
1582 {
1583         int r;
1584
1585         r = dm_integrity_failed(ic);
1586         if (unlikely(r) && !bio->bi_status)
1587                 bio->bi_status = errno_to_blk_status(r);
1588         if (unlikely(ic->synchronous_mode) && bio_op(bio) == REQ_OP_WRITE) {
1589                 unsigned long flags;
1590
1591                 spin_lock_irqsave(&ic->endio_wait.lock, flags);
1592                 bio_list_add(&ic->synchronous_bios, bio);
1593                 queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, 0);
1594                 spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
1595                 return;
1596         }
1597         bio_endio(bio);
1598 }
1599
1600 static void do_endio_flush(struct dm_integrity_c *ic, struct dm_integrity_io *dio)
1601 {
1602         struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1603
1604         if (unlikely(dio->fua) && likely(!bio->bi_status) && likely(!dm_integrity_failed(ic)))
1605                 submit_flush_bio(ic, dio);
1606         else
1607                 do_endio(ic, bio);
1608 }
1609
1610 static void dec_in_flight(struct dm_integrity_io *dio)
1611 {
1612         if (atomic_dec_and_test(&dio->in_flight)) {
1613                 struct dm_integrity_c *ic = dio->ic;
1614                 struct bio *bio;
1615
1616                 remove_range(ic, &dio->range);
1617
1618                 if (dio->op == REQ_OP_WRITE || unlikely(dio->op == REQ_OP_DISCARD))
1619                         schedule_autocommit(ic);
1620
1621                 bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1622                 if (unlikely(dio->bi_status) && !bio->bi_status)
1623                         bio->bi_status = dio->bi_status;
1624                 if (likely(!bio->bi_status) && unlikely(bio_sectors(bio) != dio->range.n_sectors)) {
1625                         dio->range.logical_sector += dio->range.n_sectors;
1626                         bio_advance(bio, dio->range.n_sectors << SECTOR_SHIFT);
1627                         INIT_WORK(&dio->work, integrity_bio_wait);
1628                         queue_work(ic->offload_wq, &dio->work);
1629                         return;
1630                 }
1631                 do_endio_flush(ic, dio);
1632         }
1633 }
1634
1635 static void integrity_end_io(struct bio *bio)
1636 {
1637         struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
1638
1639         dm_bio_restore(&dio->bio_details, bio);
1640         if (bio->bi_integrity)
1641                 bio->bi_opf |= REQ_INTEGRITY;
1642
1643         if (dio->completion)
1644                 complete(dio->completion);
1645
1646         dec_in_flight(dio);
1647 }
1648
1649 static void integrity_sector_checksum(struct dm_integrity_c *ic, sector_t sector,
1650                                       const char *data, char *result)
1651 {
1652         __le64 sector_le = cpu_to_le64(sector);
1653         SHASH_DESC_ON_STACK(req, ic->internal_hash);
1654         int r;
1655         unsigned int digest_size;
1656
1657         req->tfm = ic->internal_hash;
1658
1659         r = crypto_shash_init(req);
1660         if (unlikely(r < 0)) {
1661                 dm_integrity_io_error(ic, "crypto_shash_init", r);
1662                 goto failed;
1663         }
1664
1665         if (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) {
1666                 r = crypto_shash_update(req, (__u8 *)&ic->sb->salt, SALT_SIZE);
1667                 if (unlikely(r < 0)) {
1668                         dm_integrity_io_error(ic, "crypto_shash_update", r);
1669                         goto failed;
1670                 }
1671         }
1672
1673         r = crypto_shash_update(req, (const __u8 *)&sector_le, sizeof(sector_le));
1674         if (unlikely(r < 0)) {
1675                 dm_integrity_io_error(ic, "crypto_shash_update", r);
1676                 goto failed;
1677         }
1678
1679         r = crypto_shash_update(req, data, ic->sectors_per_block << SECTOR_SHIFT);
1680         if (unlikely(r < 0)) {
1681                 dm_integrity_io_error(ic, "crypto_shash_update", r);
1682                 goto failed;
1683         }
1684
1685         r = crypto_shash_final(req, result);
1686         if (unlikely(r < 0)) {
1687                 dm_integrity_io_error(ic, "crypto_shash_final", r);
1688                 goto failed;
1689         }
1690
1691         digest_size = crypto_shash_digestsize(ic->internal_hash);
1692         if (unlikely(digest_size < ic->tag_size))
1693                 memset(result + digest_size, 0, ic->tag_size - digest_size);
1694
1695         return;
1696
1697 failed:
1698         /* this shouldn't happen anyway, the hash functions have no reason to fail */
1699         get_random_bytes(result, ic->tag_size);
1700 }
1701
1702 static void integrity_metadata(struct work_struct *w)
1703 {
1704         struct dm_integrity_io *dio = container_of(w, struct dm_integrity_io, work);
1705         struct dm_integrity_c *ic = dio->ic;
1706
1707         int r;
1708
1709         if (ic->internal_hash) {
1710                 struct bvec_iter iter;
1711                 struct bio_vec bv;
1712                 unsigned int digest_size = crypto_shash_digestsize(ic->internal_hash);
1713                 struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1714                 char *checksums;
1715                 unsigned int extra_space = unlikely(digest_size > ic->tag_size) ? digest_size - ic->tag_size : 0;
1716                 char checksums_onstack[max_t(size_t, HASH_MAX_DIGESTSIZE, MAX_TAG_SIZE)];
1717                 sector_t sector;
1718                 unsigned int sectors_to_process;
1719
1720                 if (unlikely(ic->mode == 'R'))
1721                         goto skip_io;
1722
1723                 if (likely(dio->op != REQ_OP_DISCARD))
1724                         checksums = kmalloc((PAGE_SIZE >> SECTOR_SHIFT >> ic->sb->log2_sectors_per_block) * ic->tag_size + extra_space,
1725                                             GFP_NOIO | __GFP_NORETRY | __GFP_NOWARN);
1726                 else
1727                         checksums = kmalloc(PAGE_SIZE, GFP_NOIO | __GFP_NORETRY | __GFP_NOWARN);
1728                 if (!checksums) {
1729                         checksums = checksums_onstack;
1730                         if (WARN_ON(extra_space &&
1731                                     digest_size > sizeof(checksums_onstack))) {
1732                                 r = -EINVAL;
1733                                 goto error;
1734                         }
1735                 }
1736
1737                 if (unlikely(dio->op == REQ_OP_DISCARD)) {
1738                         unsigned int bi_size = dio->bio_details.bi_iter.bi_size;
1739                         unsigned int max_size = likely(checksums != checksums_onstack) ? PAGE_SIZE : HASH_MAX_DIGESTSIZE;
1740                         unsigned int max_blocks = max_size / ic->tag_size;
1741
1742                         memset(checksums, DISCARD_FILLER, max_size);
1743
1744                         while (bi_size) {
1745                                 unsigned int this_step_blocks = bi_size >> (SECTOR_SHIFT + ic->sb->log2_sectors_per_block);
1746
1747                                 this_step_blocks = min(this_step_blocks, max_blocks);
1748                                 r = dm_integrity_rw_tag(ic, checksums, &dio->metadata_block, &dio->metadata_offset,
1749                                                         this_step_blocks * ic->tag_size, TAG_WRITE);
1750                                 if (unlikely(r)) {
1751                                         if (likely(checksums != checksums_onstack))
1752                                                 kfree(checksums);
1753                                         goto error;
1754                                 }
1755
1756                                 bi_size -= this_step_blocks << (SECTOR_SHIFT + ic->sb->log2_sectors_per_block);
1757                         }
1758
1759                         if (likely(checksums != checksums_onstack))
1760                                 kfree(checksums);
1761                         goto skip_io;
1762                 }
1763
1764                 sector = dio->range.logical_sector;
1765                 sectors_to_process = dio->range.n_sectors;
1766
1767                 __bio_for_each_segment(bv, bio, iter, dio->bio_details.bi_iter) {
1768                         struct bio_vec bv_copy = bv;
1769                         unsigned int pos;
1770                         char *mem, *checksums_ptr;
1771
1772 again:
1773                         mem = bvec_kmap_local(&bv_copy);
1774                         pos = 0;
1775                         checksums_ptr = checksums;
1776                         do {
1777                                 integrity_sector_checksum(ic, sector, mem + pos, checksums_ptr);
1778                                 checksums_ptr += ic->tag_size;
1779                                 sectors_to_process -= ic->sectors_per_block;
1780                                 pos += ic->sectors_per_block << SECTOR_SHIFT;
1781                                 sector += ic->sectors_per_block;
1782                         } while (pos < bv_copy.bv_len && sectors_to_process && checksums != checksums_onstack);
1783                         kunmap_local(mem);
1784
1785                         r = dm_integrity_rw_tag(ic, checksums, &dio->metadata_block, &dio->metadata_offset,
1786                                                 checksums_ptr - checksums, dio->op == REQ_OP_READ ? TAG_CMP : TAG_WRITE);
1787                         if (unlikely(r)) {
1788                                 if (r > 0) {
1789                                         sector_t s;
1790
1791                                         s = sector - ((r + ic->tag_size - 1) / ic->tag_size);
1792                                         DMERR_LIMIT("%pg: Checksum failed at sector 0x%llx",
1793                                                     bio->bi_bdev, s);
1794                                         r = -EILSEQ;
1795                                         atomic64_inc(&ic->number_of_mismatches);
1796                                         dm_audit_log_bio(DM_MSG_PREFIX, "integrity-checksum",
1797                                                          bio, s, 0);
1798                                 }
1799                                 if (likely(checksums != checksums_onstack))
1800                                         kfree(checksums);
1801                                 goto error;
1802                         }
1803
1804                         if (!sectors_to_process)
1805                                 break;
1806
1807                         if (unlikely(pos < bv_copy.bv_len)) {
1808                                 bv_copy.bv_offset += pos;
1809                                 bv_copy.bv_len -= pos;
1810                                 goto again;
1811                         }
1812                 }
1813
1814                 if (likely(checksums != checksums_onstack))
1815                         kfree(checksums);
1816         } else {
1817                 struct bio_integrity_payload *bip = dio->bio_details.bi_integrity;
1818
1819                 if (bip) {
1820                         struct bio_vec biv;
1821                         struct bvec_iter iter;
1822                         unsigned int data_to_process = dio->range.n_sectors;
1823
1824                         sector_to_block(ic, data_to_process);
1825                         data_to_process *= ic->tag_size;
1826
1827                         bip_for_each_vec(biv, bip, iter) {
1828                                 unsigned char *tag;
1829                                 unsigned int this_len;
1830
1831                                 BUG_ON(PageHighMem(biv.bv_page));
1832                                 tag = bvec_virt(&biv);
1833                                 this_len = min(biv.bv_len, data_to_process);
1834                                 r = dm_integrity_rw_tag(ic, tag, &dio->metadata_block, &dio->metadata_offset,
1835                                                         this_len, dio->op == REQ_OP_READ ? TAG_READ : TAG_WRITE);
1836                                 if (unlikely(r))
1837                                         goto error;
1838                                 data_to_process -= this_len;
1839                                 if (!data_to_process)
1840                                         break;
1841                         }
1842                 }
1843         }
1844 skip_io:
1845         dec_in_flight(dio);
1846         return;
1847 error:
1848         dio->bi_status = errno_to_blk_status(r);
1849         dec_in_flight(dio);
1850 }
1851
1852 static int dm_integrity_map(struct dm_target *ti, struct bio *bio)
1853 {
1854         struct dm_integrity_c *ic = ti->private;
1855         struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
1856         struct bio_integrity_payload *bip;
1857
1858         sector_t area, offset;
1859
1860         dio->ic = ic;
1861         dio->bi_status = 0;
1862         dio->op = bio_op(bio);
1863
1864         if (unlikely(dio->op == REQ_OP_DISCARD)) {
1865                 if (ti->max_io_len) {
1866                         sector_t sec = dm_target_offset(ti, bio->bi_iter.bi_sector);
1867                         unsigned int log2_max_io_len = __fls(ti->max_io_len);
1868                         sector_t start_boundary = sec >> log2_max_io_len;
1869                         sector_t end_boundary = (sec + bio_sectors(bio) - 1) >> log2_max_io_len;
1870
1871                         if (start_boundary < end_boundary) {
1872                                 sector_t len = ti->max_io_len - (sec & (ti->max_io_len - 1));
1873
1874                                 dm_accept_partial_bio(bio, len);
1875                         }
1876                 }
1877         }
1878
1879         if (unlikely(bio->bi_opf & REQ_PREFLUSH)) {
1880                 submit_flush_bio(ic, dio);
1881                 return DM_MAPIO_SUBMITTED;
1882         }
1883
1884         dio->range.logical_sector = dm_target_offset(ti, bio->bi_iter.bi_sector);
1885         dio->fua = dio->op == REQ_OP_WRITE && bio->bi_opf & REQ_FUA;
1886         if (unlikely(dio->fua)) {
1887                 /*
1888                  * Don't pass down the FUA flag because we have to flush
1889                  * disk cache anyway.
1890                  */
1891                 bio->bi_opf &= ~REQ_FUA;
1892         }
1893         if (unlikely(dio->range.logical_sector + bio_sectors(bio) > ic->provided_data_sectors)) {
1894                 DMERR("Too big sector number: 0x%llx + 0x%x > 0x%llx",
1895                       dio->range.logical_sector, bio_sectors(bio),
1896                       ic->provided_data_sectors);
1897                 return DM_MAPIO_KILL;
1898         }
1899         if (unlikely((dio->range.logical_sector | bio_sectors(bio)) & (unsigned int)(ic->sectors_per_block - 1))) {
1900                 DMERR("Bio not aligned on %u sectors: 0x%llx, 0x%x",
1901                       ic->sectors_per_block,
1902                       dio->range.logical_sector, bio_sectors(bio));
1903                 return DM_MAPIO_KILL;
1904         }
1905
1906         if (ic->sectors_per_block > 1 && likely(dio->op != REQ_OP_DISCARD)) {
1907                 struct bvec_iter iter;
1908                 struct bio_vec bv;
1909
1910                 bio_for_each_segment(bv, bio, iter) {
1911                         if (unlikely(bv.bv_len & ((ic->sectors_per_block << SECTOR_SHIFT) - 1))) {
1912                                 DMERR("Bio vector (%u,%u) is not aligned on %u-sector boundary",
1913                                         bv.bv_offset, bv.bv_len, ic->sectors_per_block);
1914                                 return DM_MAPIO_KILL;
1915                         }
1916                 }
1917         }
1918
1919         bip = bio_integrity(bio);
1920         if (!ic->internal_hash) {
1921                 if (bip) {
1922                         unsigned int wanted_tag_size = bio_sectors(bio) >> ic->sb->log2_sectors_per_block;
1923
1924                         if (ic->log2_tag_size >= 0)
1925                                 wanted_tag_size <<= ic->log2_tag_size;
1926                         else
1927                                 wanted_tag_size *= ic->tag_size;
1928                         if (unlikely(wanted_tag_size != bip->bip_iter.bi_size)) {
1929                                 DMERR("Invalid integrity data size %u, expected %u",
1930                                       bip->bip_iter.bi_size, wanted_tag_size);
1931                                 return DM_MAPIO_KILL;
1932                         }
1933                 }
1934         } else {
1935                 if (unlikely(bip != NULL)) {
1936                         DMERR("Unexpected integrity data when using internal hash");
1937                         return DM_MAPIO_KILL;
1938                 }
1939         }
1940
1941         if (unlikely(ic->mode == 'R') && unlikely(dio->op != REQ_OP_READ))
1942                 return DM_MAPIO_KILL;
1943
1944         get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
1945         dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset, &dio->metadata_offset);
1946         bio->bi_iter.bi_sector = get_data_sector(ic, area, offset);
1947
1948         dm_integrity_map_continue(dio, true);
1949         return DM_MAPIO_SUBMITTED;
1950 }
1951
1952 static bool __journal_read_write(struct dm_integrity_io *dio, struct bio *bio,
1953                                  unsigned int journal_section, unsigned int journal_entry)
1954 {
1955         struct dm_integrity_c *ic = dio->ic;
1956         sector_t logical_sector;
1957         unsigned int n_sectors;
1958
1959         logical_sector = dio->range.logical_sector;
1960         n_sectors = dio->range.n_sectors;
1961         do {
1962                 struct bio_vec bv = bio_iovec(bio);
1963                 char *mem;
1964
1965                 if (unlikely(bv.bv_len >> SECTOR_SHIFT > n_sectors))
1966                         bv.bv_len = n_sectors << SECTOR_SHIFT;
1967                 n_sectors -= bv.bv_len >> SECTOR_SHIFT;
1968                 bio_advance_iter(bio, &bio->bi_iter, bv.bv_len);
1969 retry_kmap:
1970                 mem = kmap_local_page(bv.bv_page);
1971                 if (likely(dio->op == REQ_OP_WRITE))
1972                         flush_dcache_page(bv.bv_page);
1973
1974                 do {
1975                         struct journal_entry *je = access_journal_entry(ic, journal_section, journal_entry);
1976
1977                         if (unlikely(dio->op == REQ_OP_READ)) {
1978                                 struct journal_sector *js;
1979                                 char *mem_ptr;
1980                                 unsigned int s;
1981
1982                                 if (unlikely(journal_entry_is_inprogress(je))) {
1983                                         flush_dcache_page(bv.bv_page);
1984                                         kunmap_local(mem);
1985
1986                                         __io_wait_event(ic->copy_to_journal_wait, !journal_entry_is_inprogress(je));
1987                                         goto retry_kmap;
1988                                 }
1989                                 smp_rmb();
1990                                 BUG_ON(journal_entry_get_sector(je) != logical_sector);
1991                                 js = access_journal_data(ic, journal_section, journal_entry);
1992                                 mem_ptr = mem + bv.bv_offset;
1993                                 s = 0;
1994                                 do {
1995                                         memcpy(mem_ptr, js, JOURNAL_SECTOR_DATA);
1996                                         *(commit_id_t *)(mem_ptr + JOURNAL_SECTOR_DATA) = je->last_bytes[s];
1997                                         js++;
1998                                         mem_ptr += 1 << SECTOR_SHIFT;
1999                                 } while (++s < ic->sectors_per_block);
2000 #ifdef INTERNAL_VERIFY
2001                                 if (ic->internal_hash) {
2002                                         char checksums_onstack[max_t(size_t, HASH_MAX_DIGESTSIZE, MAX_TAG_SIZE)];
2003
2004                                         integrity_sector_checksum(ic, logical_sector, mem + bv.bv_offset, checksums_onstack);
2005                                         if (unlikely(memcmp(checksums_onstack, journal_entry_tag(ic, je), ic->tag_size))) {
2006                                                 DMERR_LIMIT("Checksum failed when reading from journal, at sector 0x%llx",
2007                                                             logical_sector);
2008                                                 dm_audit_log_bio(DM_MSG_PREFIX, "journal-checksum",
2009                                                                  bio, logical_sector, 0);
2010                                         }
2011                                 }
2012 #endif
2013                         }
2014
2015                         if (!ic->internal_hash) {
2016                                 struct bio_integrity_payload *bip = bio_integrity(bio);
2017                                 unsigned int tag_todo = ic->tag_size;
2018                                 char *tag_ptr = journal_entry_tag(ic, je);
2019
2020                                 if (bip) {
2021                                         do {
2022                                                 struct bio_vec biv = bvec_iter_bvec(bip->bip_vec, bip->bip_iter);
2023                                                 unsigned int tag_now = min(biv.bv_len, tag_todo);
2024                                                 char *tag_addr;
2025
2026                                                 BUG_ON(PageHighMem(biv.bv_page));
2027                                                 tag_addr = bvec_virt(&biv);
2028                                                 if (likely(dio->op == REQ_OP_WRITE))
2029                                                         memcpy(tag_ptr, tag_addr, tag_now);
2030                                                 else
2031                                                         memcpy(tag_addr, tag_ptr, tag_now);
2032                                                 bvec_iter_advance(bip->bip_vec, &bip->bip_iter, tag_now);
2033                                                 tag_ptr += tag_now;
2034                                                 tag_todo -= tag_now;
2035                                         } while (unlikely(tag_todo));
2036                                 } else if (likely(dio->op == REQ_OP_WRITE))
2037                                         memset(tag_ptr, 0, tag_todo);
2038                         }
2039
2040                         if (likely(dio->op == REQ_OP_WRITE)) {
2041                                 struct journal_sector *js;
2042                                 unsigned int s;
2043
2044                                 js = access_journal_data(ic, journal_section, journal_entry);
2045                                 memcpy(js, mem + bv.bv_offset, ic->sectors_per_block << SECTOR_SHIFT);
2046
2047                                 s = 0;
2048                                 do {
2049                                         je->last_bytes[s] = js[s].commit_id;
2050                                 } while (++s < ic->sectors_per_block);
2051
2052                                 if (ic->internal_hash) {
2053                                         unsigned int digest_size = crypto_shash_digestsize(ic->internal_hash);
2054
2055                                         if (unlikely(digest_size > ic->tag_size)) {
2056                                                 char checksums_onstack[HASH_MAX_DIGESTSIZE];
2057
2058                                                 integrity_sector_checksum(ic, logical_sector, (char *)js, checksums_onstack);
2059                                                 memcpy(journal_entry_tag(ic, je), checksums_onstack, ic->tag_size);
2060                                         } else
2061                                                 integrity_sector_checksum(ic, logical_sector, (char *)js, journal_entry_tag(ic, je));
2062                                 }
2063
2064                                 journal_entry_set_sector(je, logical_sector);
2065                         }
2066                         logical_sector += ic->sectors_per_block;
2067
2068                         journal_entry++;
2069                         if (unlikely(journal_entry == ic->journal_section_entries)) {
2070                                 journal_entry = 0;
2071                                 journal_section++;
2072                                 wraparound_section(ic, &journal_section);
2073                         }
2074
2075                         bv.bv_offset += ic->sectors_per_block << SECTOR_SHIFT;
2076                 } while (bv.bv_len -= ic->sectors_per_block << SECTOR_SHIFT);
2077
2078                 if (unlikely(dio->op == REQ_OP_READ))
2079                         flush_dcache_page(bv.bv_page);
2080                 kunmap_local(mem);
2081         } while (n_sectors);
2082
2083         if (likely(dio->op == REQ_OP_WRITE)) {
2084                 smp_mb();
2085                 if (unlikely(waitqueue_active(&ic->copy_to_journal_wait)))
2086                         wake_up(&ic->copy_to_journal_wait);
2087                 if (READ_ONCE(ic->free_sectors) <= ic->free_sectors_threshold)
2088                         queue_work(ic->commit_wq, &ic->commit_work);
2089                 else
2090                         schedule_autocommit(ic);
2091         } else
2092                 remove_range(ic, &dio->range);
2093
2094         if (unlikely(bio->bi_iter.bi_size)) {
2095                 sector_t area, offset;
2096
2097                 dio->range.logical_sector = logical_sector;
2098                 get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
2099                 dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset, &dio->metadata_offset);
2100                 return true;
2101         }
2102
2103         return false;
2104 }
2105
2106 static void dm_integrity_map_continue(struct dm_integrity_io *dio, bool from_map)
2107 {
2108         struct dm_integrity_c *ic = dio->ic;
2109         struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
2110         unsigned int journal_section, journal_entry;
2111         unsigned int journal_read_pos;
2112         struct completion read_comp;
2113         bool discard_retried = false;
2114         bool need_sync_io = ic->internal_hash && dio->op == REQ_OP_READ;
2115
2116         if (unlikely(dio->op == REQ_OP_DISCARD) && ic->mode != 'D')
2117                 need_sync_io = true;
2118
2119         if (need_sync_io && from_map) {
2120                 INIT_WORK(&dio->work, integrity_bio_wait);
2121                 queue_work(ic->offload_wq, &dio->work);
2122                 return;
2123         }
2124
2125 lock_retry:
2126         spin_lock_irq(&ic->endio_wait.lock);
2127 retry:
2128         if (unlikely(dm_integrity_failed(ic))) {
2129                 spin_unlock_irq(&ic->endio_wait.lock);
2130                 do_endio(ic, bio);
2131                 return;
2132         }
2133         dio->range.n_sectors = bio_sectors(bio);
2134         journal_read_pos = NOT_FOUND;
2135         if (ic->mode == 'J' && likely(dio->op != REQ_OP_DISCARD)) {
2136                 if (dio->op == REQ_OP_WRITE) {
2137                         unsigned int next_entry, i, pos;
2138                         unsigned int ws, we, range_sectors;
2139
2140                         dio->range.n_sectors = min(dio->range.n_sectors,
2141                                                    (sector_t)ic->free_sectors << ic->sb->log2_sectors_per_block);
2142                         if (unlikely(!dio->range.n_sectors)) {
2143                                 if (from_map)
2144                                         goto offload_to_thread;
2145                                 sleep_on_endio_wait(ic);
2146                                 goto retry;
2147                         }
2148                         range_sectors = dio->range.n_sectors >> ic->sb->log2_sectors_per_block;
2149                         ic->free_sectors -= range_sectors;
2150                         journal_section = ic->free_section;
2151                         journal_entry = ic->free_section_entry;
2152
2153                         next_entry = ic->free_section_entry + range_sectors;
2154                         ic->free_section_entry = next_entry % ic->journal_section_entries;
2155                         ic->free_section += next_entry / ic->journal_section_entries;
2156                         ic->n_uncommitted_sections += next_entry / ic->journal_section_entries;
2157                         wraparound_section(ic, &ic->free_section);
2158
2159                         pos = journal_section * ic->journal_section_entries + journal_entry;
2160                         ws = journal_section;
2161                         we = journal_entry;
2162                         i = 0;
2163                         do {
2164                                 struct journal_entry *je;
2165
2166                                 add_journal_node(ic, &ic->journal_tree[pos], dio->range.logical_sector + i);
2167                                 pos++;
2168                                 if (unlikely(pos >= ic->journal_entries))
2169                                         pos = 0;
2170
2171                                 je = access_journal_entry(ic, ws, we);
2172                                 BUG_ON(!journal_entry_is_unused(je));
2173                                 journal_entry_set_inprogress(je);
2174                                 we++;
2175                                 if (unlikely(we == ic->journal_section_entries)) {
2176                                         we = 0;
2177                                         ws++;
2178                                         wraparound_section(ic, &ws);
2179                                 }
2180                         } while ((i += ic->sectors_per_block) < dio->range.n_sectors);
2181
2182                         spin_unlock_irq(&ic->endio_wait.lock);
2183                         goto journal_read_write;
2184                 } else {
2185                         sector_t next_sector;
2186
2187                         journal_read_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
2188                         if (likely(journal_read_pos == NOT_FOUND)) {
2189                                 if (unlikely(dio->range.n_sectors > next_sector - dio->range.logical_sector))
2190                                         dio->range.n_sectors = next_sector - dio->range.logical_sector;
2191                         } else {
2192                                 unsigned int i;
2193                                 unsigned int jp = journal_read_pos + 1;
2194
2195                                 for (i = ic->sectors_per_block; i < dio->range.n_sectors; i += ic->sectors_per_block, jp++) {
2196                                         if (!test_journal_node(ic, jp, dio->range.logical_sector + i))
2197                                                 break;
2198                                 }
2199                                 dio->range.n_sectors = i;
2200                         }
2201                 }
2202         }
2203         if (unlikely(!add_new_range(ic, &dio->range, true))) {
2204                 /*
2205                  * We must not sleep in the request routine because it could
2206                  * stall bios on current->bio_list.
2207                  * So, we offload the bio to a workqueue if we have to sleep.
2208                  */
2209                 if (from_map) {
2210 offload_to_thread:
2211                         spin_unlock_irq(&ic->endio_wait.lock);
2212                         INIT_WORK(&dio->work, integrity_bio_wait);
2213                         queue_work(ic->wait_wq, &dio->work);
2214                         return;
2215                 }
2216                 if (journal_read_pos != NOT_FOUND)
2217                         dio->range.n_sectors = ic->sectors_per_block;
2218                 wait_and_add_new_range(ic, &dio->range);
2219                 /*
2220                  * wait_and_add_new_range drops the spinlock, so the journal
2221                  * may have been changed arbitrarily. We need to recheck.
2222                  * To simplify the code, we restrict I/O size to just one block.
2223                  */
2224                 if (journal_read_pos != NOT_FOUND) {
2225                         sector_t next_sector;
2226                         unsigned int new_pos;
2227
2228                         new_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
2229                         if (unlikely(new_pos != journal_read_pos)) {
2230                                 remove_range_unlocked(ic, &dio->range);
2231                                 goto retry;
2232                         }
2233                 }
2234         }
2235         if (ic->mode == 'J' && likely(dio->op == REQ_OP_DISCARD) && !discard_retried) {
2236                 sector_t next_sector;
2237                 unsigned int new_pos;
2238
2239                 new_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
2240                 if (unlikely(new_pos != NOT_FOUND) ||
2241                     unlikely(next_sector < dio->range.logical_sector - dio->range.n_sectors)) {
2242                         remove_range_unlocked(ic, &dio->range);
2243                         spin_unlock_irq(&ic->endio_wait.lock);
2244                         queue_work(ic->commit_wq, &ic->commit_work);
2245                         flush_workqueue(ic->commit_wq);
2246                         queue_work(ic->writer_wq, &ic->writer_work);
2247                         flush_workqueue(ic->writer_wq);
2248                         discard_retried = true;
2249                         goto lock_retry;
2250                 }
2251         }
2252         spin_unlock_irq(&ic->endio_wait.lock);
2253
2254         if (unlikely(journal_read_pos != NOT_FOUND)) {
2255                 journal_section = journal_read_pos / ic->journal_section_entries;
2256                 journal_entry = journal_read_pos % ic->journal_section_entries;
2257                 goto journal_read_write;
2258         }
2259
2260         if (ic->mode == 'B' && (dio->op == REQ_OP_WRITE || unlikely(dio->op == REQ_OP_DISCARD))) {
2261                 if (!block_bitmap_op(ic, ic->may_write_bitmap, dio->range.logical_sector,
2262                                      dio->range.n_sectors, BITMAP_OP_TEST_ALL_SET)) {
2263                         struct bitmap_block_status *bbs;
2264
2265                         bbs = sector_to_bitmap_block(ic, dio->range.logical_sector);
2266                         spin_lock(&bbs->bio_queue_lock);
2267                         bio_list_add(&bbs->bio_queue, bio);
2268                         spin_unlock(&bbs->bio_queue_lock);
2269                         queue_work(ic->writer_wq, &bbs->work);
2270                         return;
2271                 }
2272         }
2273
2274         dio->in_flight = (atomic_t)ATOMIC_INIT(2);
2275
2276         if (need_sync_io) {
2277                 init_completion(&read_comp);
2278                 dio->completion = &read_comp;
2279         } else
2280                 dio->completion = NULL;
2281
2282         dm_bio_record(&dio->bio_details, bio);
2283         bio_set_dev(bio, ic->dev->bdev);
2284         bio->bi_integrity = NULL;
2285         bio->bi_opf &= ~REQ_INTEGRITY;
2286         bio->bi_end_io = integrity_end_io;
2287         bio->bi_iter.bi_size = dio->range.n_sectors << SECTOR_SHIFT;
2288
2289         if (unlikely(dio->op == REQ_OP_DISCARD) && likely(ic->mode != 'D')) {
2290                 integrity_metadata(&dio->work);
2291                 dm_integrity_flush_buffers(ic, false);
2292
2293                 dio->in_flight = (atomic_t)ATOMIC_INIT(1);
2294                 dio->completion = NULL;
2295
2296                 submit_bio_noacct(bio);
2297
2298                 return;
2299         }
2300
2301         submit_bio_noacct(bio);
2302
2303         if (need_sync_io) {
2304                 wait_for_completion_io(&read_comp);
2305                 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING) &&
2306                     dio->range.logical_sector + dio->range.n_sectors > le64_to_cpu(ic->sb->recalc_sector))
2307                         goto skip_check;
2308                 if (ic->mode == 'B') {
2309                         if (!block_bitmap_op(ic, ic->recalc_bitmap, dio->range.logical_sector,
2310                                              dio->range.n_sectors, BITMAP_OP_TEST_ALL_CLEAR))
2311                                 goto skip_check;
2312                 }
2313
2314                 if (likely(!bio->bi_status))
2315                         integrity_metadata(&dio->work);
2316                 else
2317 skip_check:
2318                         dec_in_flight(dio);
2319         } else {
2320                 INIT_WORK(&dio->work, integrity_metadata);
2321                 queue_work(ic->metadata_wq, &dio->work);
2322         }
2323
2324         return;
2325
2326 journal_read_write:
2327         if (unlikely(__journal_read_write(dio, bio, journal_section, journal_entry)))
2328                 goto lock_retry;
2329
2330         do_endio_flush(ic, dio);
2331 }
2332
2333
2334 static void integrity_bio_wait(struct work_struct *w)
2335 {
2336         struct dm_integrity_io *dio = container_of(w, struct dm_integrity_io, work);
2337
2338         dm_integrity_map_continue(dio, false);
2339 }
2340
2341 static void pad_uncommitted(struct dm_integrity_c *ic)
2342 {
2343         if (ic->free_section_entry) {
2344                 ic->free_sectors -= ic->journal_section_entries - ic->free_section_entry;
2345                 ic->free_section_entry = 0;
2346                 ic->free_section++;
2347                 wraparound_section(ic, &ic->free_section);
2348                 ic->n_uncommitted_sections++;
2349         }
2350         if (WARN_ON(ic->journal_sections * ic->journal_section_entries !=
2351                     (ic->n_uncommitted_sections + ic->n_committed_sections) *
2352                     ic->journal_section_entries + ic->free_sectors)) {
2353                 DMCRIT("journal_sections %u, journal_section_entries %u, "
2354                        "n_uncommitted_sections %u, n_committed_sections %u, "
2355                        "journal_section_entries %u, free_sectors %u",
2356                        ic->journal_sections, ic->journal_section_entries,
2357                        ic->n_uncommitted_sections, ic->n_committed_sections,
2358                        ic->journal_section_entries, ic->free_sectors);
2359         }
2360 }
2361
2362 static void integrity_commit(struct work_struct *w)
2363 {
2364         struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, commit_work);
2365         unsigned int commit_start, commit_sections;
2366         unsigned int i, j, n;
2367         struct bio *flushes;
2368
2369         del_timer(&ic->autocommit_timer);
2370
2371         spin_lock_irq(&ic->endio_wait.lock);
2372         flushes = bio_list_get(&ic->flush_bio_list);
2373         if (unlikely(ic->mode != 'J')) {
2374                 spin_unlock_irq(&ic->endio_wait.lock);
2375                 dm_integrity_flush_buffers(ic, true);
2376                 goto release_flush_bios;
2377         }
2378
2379         pad_uncommitted(ic);
2380         commit_start = ic->uncommitted_section;
2381         commit_sections = ic->n_uncommitted_sections;
2382         spin_unlock_irq(&ic->endio_wait.lock);
2383
2384         if (!commit_sections)
2385                 goto release_flush_bios;
2386
2387         ic->wrote_to_journal = true;
2388
2389         i = commit_start;
2390         for (n = 0; n < commit_sections; n++) {
2391                 for (j = 0; j < ic->journal_section_entries; j++) {
2392                         struct journal_entry *je;
2393
2394                         je = access_journal_entry(ic, i, j);
2395                         io_wait_event(ic->copy_to_journal_wait, !journal_entry_is_inprogress(je));
2396                 }
2397                 for (j = 0; j < ic->journal_section_sectors; j++) {
2398                         struct journal_sector *js;
2399
2400                         js = access_journal(ic, i, j);
2401                         js->commit_id = dm_integrity_commit_id(ic, i, j, ic->commit_seq);
2402                 }
2403                 i++;
2404                 if (unlikely(i >= ic->journal_sections))
2405                         ic->commit_seq = next_commit_seq(ic->commit_seq);
2406                 wraparound_section(ic, &i);
2407         }
2408         smp_rmb();
2409
2410         write_journal(ic, commit_start, commit_sections);
2411
2412         spin_lock_irq(&ic->endio_wait.lock);
2413         ic->uncommitted_section += commit_sections;
2414         wraparound_section(ic, &ic->uncommitted_section);
2415         ic->n_uncommitted_sections -= commit_sections;
2416         ic->n_committed_sections += commit_sections;
2417         spin_unlock_irq(&ic->endio_wait.lock);
2418
2419         if (READ_ONCE(ic->free_sectors) <= ic->free_sectors_threshold)
2420                 queue_work(ic->writer_wq, &ic->writer_work);
2421
2422 release_flush_bios:
2423         while (flushes) {
2424                 struct bio *next = flushes->bi_next;
2425
2426                 flushes->bi_next = NULL;
2427                 do_endio(ic, flushes);
2428                 flushes = next;
2429         }
2430 }
2431
2432 static void complete_copy_from_journal(unsigned long error, void *context)
2433 {
2434         struct journal_io *io = context;
2435         struct journal_completion *comp = io->comp;
2436         struct dm_integrity_c *ic = comp->ic;
2437
2438         remove_range(ic, &io->range);
2439         mempool_free(io, &ic->journal_io_mempool);
2440         if (unlikely(error != 0))
2441                 dm_integrity_io_error(ic, "copying from journal", -EIO);
2442         complete_journal_op(comp);
2443 }
2444
2445 static void restore_last_bytes(struct dm_integrity_c *ic, struct journal_sector *js,
2446                                struct journal_entry *je)
2447 {
2448         unsigned int s = 0;
2449
2450         do {
2451                 js->commit_id = je->last_bytes[s];
2452                 js++;
2453         } while (++s < ic->sectors_per_block);
2454 }
2455
2456 static void do_journal_write(struct dm_integrity_c *ic, unsigned int write_start,
2457                              unsigned int write_sections, bool from_replay)
2458 {
2459         unsigned int i, j, n;
2460         struct journal_completion comp;
2461         struct blk_plug plug;
2462
2463         blk_start_plug(&plug);
2464
2465         comp.ic = ic;
2466         comp.in_flight = (atomic_t)ATOMIC_INIT(1);
2467         init_completion(&comp.comp);
2468
2469         i = write_start;
2470         for (n = 0; n < write_sections; n++, i++, wraparound_section(ic, &i)) {
2471 #ifndef INTERNAL_VERIFY
2472                 if (unlikely(from_replay))
2473 #endif
2474                         rw_section_mac(ic, i, false);
2475                 for (j = 0; j < ic->journal_section_entries; j++) {
2476                         struct journal_entry *je = access_journal_entry(ic, i, j);
2477                         sector_t sec, area, offset;
2478                         unsigned int k, l, next_loop;
2479                         sector_t metadata_block;
2480                         unsigned int metadata_offset;
2481                         struct journal_io *io;
2482
2483                         if (journal_entry_is_unused(je))
2484                                 continue;
2485                         BUG_ON(unlikely(journal_entry_is_inprogress(je)) && !from_replay);
2486                         sec = journal_entry_get_sector(je);
2487                         if (unlikely(from_replay)) {
2488                                 if (unlikely(sec & (unsigned int)(ic->sectors_per_block - 1))) {
2489                                         dm_integrity_io_error(ic, "invalid sector in journal", -EIO);
2490                                         sec &= ~(sector_t)(ic->sectors_per_block - 1);
2491                                 }
2492                                 if (unlikely(sec >= ic->provided_data_sectors)) {
2493                                         journal_entry_set_unused(je);
2494                                         continue;
2495                                 }
2496                         }
2497                         get_area_and_offset(ic, sec, &area, &offset);
2498                         restore_last_bytes(ic, access_journal_data(ic, i, j), je);
2499                         for (k = j + 1; k < ic->journal_section_entries; k++) {
2500                                 struct journal_entry *je2 = access_journal_entry(ic, i, k);
2501                                 sector_t sec2, area2, offset2;
2502
2503                                 if (journal_entry_is_unused(je2))
2504                                         break;
2505                                 BUG_ON(unlikely(journal_entry_is_inprogress(je2)) && !from_replay);
2506                                 sec2 = journal_entry_get_sector(je2);
2507                                 if (unlikely(sec2 >= ic->provided_data_sectors))
2508                                         break;
2509                                 get_area_and_offset(ic, sec2, &area2, &offset2);
2510                                 if (area2 != area || offset2 != offset + ((k - j) << ic->sb->log2_sectors_per_block))
2511                                         break;
2512                                 restore_last_bytes(ic, access_journal_data(ic, i, k), je2);
2513                         }
2514                         next_loop = k - 1;
2515
2516                         io = mempool_alloc(&ic->journal_io_mempool, GFP_NOIO);
2517                         io->comp = &comp;
2518                         io->range.logical_sector = sec;
2519                         io->range.n_sectors = (k - j) << ic->sb->log2_sectors_per_block;
2520
2521                         spin_lock_irq(&ic->endio_wait.lock);
2522                         add_new_range_and_wait(ic, &io->range);
2523
2524                         if (likely(!from_replay)) {
2525                                 struct journal_node *section_node = &ic->journal_tree[i * ic->journal_section_entries];
2526
2527                                 /* don't write if there is newer committed sector */
2528                                 while (j < k && find_newer_committed_node(ic, &section_node[j])) {
2529                                         struct journal_entry *je2 = access_journal_entry(ic, i, j);
2530
2531                                         journal_entry_set_unused(je2);
2532                                         remove_journal_node(ic, &section_node[j]);
2533                                         j++;
2534                                         sec += ic->sectors_per_block;
2535                                         offset += ic->sectors_per_block;
2536                                 }
2537                                 while (j < k && find_newer_committed_node(ic, &section_node[k - 1])) {
2538                                         struct journal_entry *je2 = access_journal_entry(ic, i, k - 1);
2539
2540                                         journal_entry_set_unused(je2);
2541                                         remove_journal_node(ic, &section_node[k - 1]);
2542                                         k--;
2543                                 }
2544                                 if (j == k) {
2545                                         remove_range_unlocked(ic, &io->range);
2546                                         spin_unlock_irq(&ic->endio_wait.lock);
2547                                         mempool_free(io, &ic->journal_io_mempool);
2548                                         goto skip_io;
2549                                 }
2550                                 for (l = j; l < k; l++)
2551                                         remove_journal_node(ic, &section_node[l]);
2552                         }
2553                         spin_unlock_irq(&ic->endio_wait.lock);
2554
2555                         metadata_block = get_metadata_sector_and_offset(ic, area, offset, &metadata_offset);
2556                         for (l = j; l < k; l++) {
2557                                 int r;
2558                                 struct journal_entry *je2 = access_journal_entry(ic, i, l);
2559
2560                                 if (
2561 #ifndef INTERNAL_VERIFY
2562                                     unlikely(from_replay) &&
2563 #endif
2564                                     ic->internal_hash) {
2565                                         char test_tag[max_t(size_t, HASH_MAX_DIGESTSIZE, MAX_TAG_SIZE)];
2566
2567                                         integrity_sector_checksum(ic, sec + ((l - j) << ic->sb->log2_sectors_per_block),
2568                                                                   (char *)access_journal_data(ic, i, l), test_tag);
2569                                         if (unlikely(memcmp(test_tag, journal_entry_tag(ic, je2), ic->tag_size))) {
2570                                                 dm_integrity_io_error(ic, "tag mismatch when replaying journal", -EILSEQ);
2571                                                 dm_audit_log_target(DM_MSG_PREFIX, "integrity-replay-journal", ic->ti, 0);
2572                                         }
2573                                 }
2574
2575                                 journal_entry_set_unused(je2);
2576                                 r = dm_integrity_rw_tag(ic, journal_entry_tag(ic, je2), &metadata_block, &metadata_offset,
2577                                                         ic->tag_size, TAG_WRITE);
2578                                 if (unlikely(r))
2579                                         dm_integrity_io_error(ic, "reading tags", r);
2580                         }
2581
2582                         atomic_inc(&comp.in_flight);
2583                         copy_from_journal(ic, i, j << ic->sb->log2_sectors_per_block,
2584                                           (k - j) << ic->sb->log2_sectors_per_block,
2585                                           get_data_sector(ic, area, offset),
2586                                           complete_copy_from_journal, io);
2587 skip_io:
2588                         j = next_loop;
2589                 }
2590         }
2591
2592         dm_bufio_write_dirty_buffers_async(ic->bufio);
2593
2594         blk_finish_plug(&plug);
2595
2596         complete_journal_op(&comp);
2597         wait_for_completion_io(&comp.comp);
2598
2599         dm_integrity_flush_buffers(ic, true);
2600 }
2601
2602 static void integrity_writer(struct work_struct *w)
2603 {
2604         struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, writer_work);
2605         unsigned int write_start, write_sections;
2606         unsigned int prev_free_sectors;
2607
2608         spin_lock_irq(&ic->endio_wait.lock);
2609         write_start = ic->committed_section;
2610         write_sections = ic->n_committed_sections;
2611         spin_unlock_irq(&ic->endio_wait.lock);
2612
2613         if (!write_sections)
2614                 return;
2615
2616         do_journal_write(ic, write_start, write_sections, false);
2617
2618         spin_lock_irq(&ic->endio_wait.lock);
2619
2620         ic->committed_section += write_sections;
2621         wraparound_section(ic, &ic->committed_section);
2622         ic->n_committed_sections -= write_sections;
2623
2624         prev_free_sectors = ic->free_sectors;
2625         ic->free_sectors += write_sections * ic->journal_section_entries;
2626         if (unlikely(!prev_free_sectors))
2627                 wake_up_locked(&ic->endio_wait);
2628
2629         spin_unlock_irq(&ic->endio_wait.lock);
2630 }
2631
2632 static void recalc_write_super(struct dm_integrity_c *ic)
2633 {
2634         int r;
2635
2636         dm_integrity_flush_buffers(ic, false);
2637         if (dm_integrity_failed(ic))
2638                 return;
2639
2640         r = sync_rw_sb(ic, REQ_OP_WRITE);
2641         if (unlikely(r))
2642                 dm_integrity_io_error(ic, "writing superblock", r);
2643 }
2644
2645 static void integrity_recalc(struct work_struct *w)
2646 {
2647         struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, recalc_work);
2648         size_t recalc_tags_size;
2649         u8 *recalc_buffer = NULL;
2650         u8 *recalc_tags = NULL;
2651         struct dm_integrity_range range;
2652         struct dm_io_request io_req;
2653         struct dm_io_region io_loc;
2654         sector_t area, offset;
2655         sector_t metadata_block;
2656         unsigned int metadata_offset;
2657         sector_t logical_sector, n_sectors;
2658         __u8 *t;
2659         unsigned int i;
2660         int r;
2661         unsigned int super_counter = 0;
2662         unsigned recalc_sectors = RECALC_SECTORS;
2663
2664 retry:
2665         recalc_buffer = __vmalloc(recalc_sectors << SECTOR_SHIFT, GFP_NOIO);
2666         if (!recalc_buffer) {
2667 oom:
2668                 recalc_sectors >>= 1;
2669                 if (recalc_sectors >= 1U << ic->sb->log2_sectors_per_block)
2670                         goto retry;
2671                 DMCRIT("out of memory for recalculate buffer - recalculation disabled");
2672                 goto free_ret;
2673         }
2674         recalc_tags_size = (recalc_sectors >> ic->sb->log2_sectors_per_block) * ic->tag_size;
2675         if (crypto_shash_digestsize(ic->internal_hash) > ic->tag_size)
2676                 recalc_tags_size += crypto_shash_digestsize(ic->internal_hash) - ic->tag_size;
2677         recalc_tags = kvmalloc(recalc_tags_size, GFP_NOIO);
2678         if (!recalc_tags) {
2679                 vfree(recalc_buffer);
2680                 recalc_buffer = NULL;
2681                 goto oom;
2682         }
2683
2684         DEBUG_print("start recalculation... (position %llx)\n", le64_to_cpu(ic->sb->recalc_sector));
2685
2686         spin_lock_irq(&ic->endio_wait.lock);
2687
2688 next_chunk:
2689
2690         if (unlikely(dm_post_suspending(ic->ti)))
2691                 goto unlock_ret;
2692
2693         range.logical_sector = le64_to_cpu(ic->sb->recalc_sector);
2694         if (unlikely(range.logical_sector >= ic->provided_data_sectors)) {
2695                 if (ic->mode == 'B') {
2696                         block_bitmap_op(ic, ic->recalc_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_CLEAR);
2697                         DEBUG_print("queue_delayed_work: bitmap_flush_work\n");
2698                         queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, 0);
2699                 }
2700                 goto unlock_ret;
2701         }
2702
2703         get_area_and_offset(ic, range.logical_sector, &area, &offset);
2704         range.n_sectors = min((sector_t)recalc_sectors, ic->provided_data_sectors - range.logical_sector);
2705         if (!ic->meta_dev)
2706                 range.n_sectors = min(range.n_sectors, ((sector_t)1U << ic->sb->log2_interleave_sectors) - (unsigned int)offset);
2707
2708         add_new_range_and_wait(ic, &range);
2709         spin_unlock_irq(&ic->endio_wait.lock);
2710         logical_sector = range.logical_sector;
2711         n_sectors = range.n_sectors;
2712
2713         if (ic->mode == 'B') {
2714                 if (block_bitmap_op(ic, ic->recalc_bitmap, logical_sector, n_sectors, BITMAP_OP_TEST_ALL_CLEAR))
2715                         goto advance_and_next;
2716
2717                 while (block_bitmap_op(ic, ic->recalc_bitmap, logical_sector,
2718                                        ic->sectors_per_block, BITMAP_OP_TEST_ALL_CLEAR)) {
2719                         logical_sector += ic->sectors_per_block;
2720                         n_sectors -= ic->sectors_per_block;
2721                         cond_resched();
2722                 }
2723                 while (block_bitmap_op(ic, ic->recalc_bitmap, logical_sector + n_sectors - ic->sectors_per_block,
2724                                        ic->sectors_per_block, BITMAP_OP_TEST_ALL_CLEAR)) {
2725                         n_sectors -= ic->sectors_per_block;
2726                         cond_resched();
2727                 }
2728                 get_area_and_offset(ic, logical_sector, &area, &offset);
2729         }
2730
2731         DEBUG_print("recalculating: %llx, %llx\n", logical_sector, n_sectors);
2732
2733         if (unlikely(++super_counter == RECALC_WRITE_SUPER)) {
2734                 recalc_write_super(ic);
2735                 if (ic->mode == 'B')
2736                         queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, ic->bitmap_flush_interval);
2737
2738                 super_counter = 0;
2739         }
2740
2741         if (unlikely(dm_integrity_failed(ic)))
2742                 goto err;
2743
2744         io_req.bi_opf = REQ_OP_READ;
2745         io_req.mem.type = DM_IO_VMA;
2746         io_req.mem.ptr.addr = recalc_buffer;
2747         io_req.notify.fn = NULL;
2748         io_req.client = ic->io;
2749         io_loc.bdev = ic->dev->bdev;
2750         io_loc.sector = get_data_sector(ic, area, offset);
2751         io_loc.count = n_sectors;
2752
2753         r = dm_io(&io_req, 1, &io_loc, NULL);
2754         if (unlikely(r)) {
2755                 dm_integrity_io_error(ic, "reading data", r);
2756                 goto err;
2757         }
2758
2759         t = recalc_tags;
2760         for (i = 0; i < n_sectors; i += ic->sectors_per_block) {
2761                 integrity_sector_checksum(ic, logical_sector + i, recalc_buffer + (i << SECTOR_SHIFT), t);
2762                 t += ic->tag_size;
2763         }
2764
2765         metadata_block = get_metadata_sector_and_offset(ic, area, offset, &metadata_offset);
2766
2767         r = dm_integrity_rw_tag(ic, recalc_tags, &metadata_block, &metadata_offset, t - recalc_tags, TAG_WRITE);
2768         if (unlikely(r)) {
2769                 dm_integrity_io_error(ic, "writing tags", r);
2770                 goto err;
2771         }
2772
2773         if (ic->mode == 'B') {
2774                 sector_t start, end;
2775
2776                 start = (range.logical_sector >>
2777                          (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit)) <<
2778                         (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
2779                 end = ((range.logical_sector + range.n_sectors) >>
2780                        (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit)) <<
2781                         (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
2782                 block_bitmap_op(ic, ic->recalc_bitmap, start, end - start, BITMAP_OP_CLEAR);
2783         }
2784
2785 advance_and_next:
2786         cond_resched();
2787
2788         spin_lock_irq(&ic->endio_wait.lock);
2789         remove_range_unlocked(ic, &range);
2790         ic->sb->recalc_sector = cpu_to_le64(range.logical_sector + range.n_sectors);
2791         goto next_chunk;
2792
2793 err:
2794         remove_range(ic, &range);
2795         goto free_ret;
2796
2797 unlock_ret:
2798         spin_unlock_irq(&ic->endio_wait.lock);
2799
2800         recalc_write_super(ic);
2801
2802 free_ret:
2803         vfree(recalc_buffer);
2804         kvfree(recalc_tags);
2805 }
2806
2807 static void bitmap_block_work(struct work_struct *w)
2808 {
2809         struct bitmap_block_status *bbs = container_of(w, struct bitmap_block_status, work);
2810         struct dm_integrity_c *ic = bbs->ic;
2811         struct bio *bio;
2812         struct bio_list bio_queue;
2813         struct bio_list waiting;
2814
2815         bio_list_init(&waiting);
2816
2817         spin_lock(&bbs->bio_queue_lock);
2818         bio_queue = bbs->bio_queue;
2819         bio_list_init(&bbs->bio_queue);
2820         spin_unlock(&bbs->bio_queue_lock);
2821
2822         while ((bio = bio_list_pop(&bio_queue))) {
2823                 struct dm_integrity_io *dio;
2824
2825                 dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
2826
2827                 if (block_bitmap_op(ic, ic->may_write_bitmap, dio->range.logical_sector,
2828                                     dio->range.n_sectors, BITMAP_OP_TEST_ALL_SET)) {
2829                         remove_range(ic, &dio->range);
2830                         INIT_WORK(&dio->work, integrity_bio_wait);
2831                         queue_work(ic->offload_wq, &dio->work);
2832                 } else {
2833                         block_bitmap_op(ic, ic->journal, dio->range.logical_sector,
2834                                         dio->range.n_sectors, BITMAP_OP_SET);
2835                         bio_list_add(&waiting, bio);
2836                 }
2837         }
2838
2839         if (bio_list_empty(&waiting))
2840                 return;
2841
2842         rw_journal_sectors(ic, REQ_OP_WRITE | REQ_FUA | REQ_SYNC,
2843                            bbs->idx * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT),
2844                            BITMAP_BLOCK_SIZE >> SECTOR_SHIFT, NULL);
2845
2846         while ((bio = bio_list_pop(&waiting))) {
2847                 struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
2848
2849                 block_bitmap_op(ic, ic->may_write_bitmap, dio->range.logical_sector,
2850                                 dio->range.n_sectors, BITMAP_OP_SET);
2851
2852                 remove_range(ic, &dio->range);
2853                 INIT_WORK(&dio->work, integrity_bio_wait);
2854                 queue_work(ic->offload_wq, &dio->work);
2855         }
2856
2857         queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, ic->bitmap_flush_interval);
2858 }
2859
2860 static void bitmap_flush_work(struct work_struct *work)
2861 {
2862         struct dm_integrity_c *ic = container_of(work, struct dm_integrity_c, bitmap_flush_work.work);
2863         struct dm_integrity_range range;
2864         unsigned long limit;
2865         struct bio *bio;
2866
2867         dm_integrity_flush_buffers(ic, false);
2868
2869         range.logical_sector = 0;
2870         range.n_sectors = ic->provided_data_sectors;
2871
2872         spin_lock_irq(&ic->endio_wait.lock);
2873         add_new_range_and_wait(ic, &range);
2874         spin_unlock_irq(&ic->endio_wait.lock);
2875
2876         dm_integrity_flush_buffers(ic, true);
2877
2878         limit = ic->provided_data_sectors;
2879         if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
2880                 limit = le64_to_cpu(ic->sb->recalc_sector)
2881                         >> (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit)
2882                         << (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
2883         }
2884         /*DEBUG_print("zeroing journal\n");*/
2885         block_bitmap_op(ic, ic->journal, 0, limit, BITMAP_OP_CLEAR);
2886         block_bitmap_op(ic, ic->may_write_bitmap, 0, limit, BITMAP_OP_CLEAR);
2887
2888         rw_journal_sectors(ic, REQ_OP_WRITE | REQ_FUA | REQ_SYNC, 0,
2889                            ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
2890
2891         spin_lock_irq(&ic->endio_wait.lock);
2892         remove_range_unlocked(ic, &range);
2893         while (unlikely((bio = bio_list_pop(&ic->synchronous_bios)) != NULL)) {
2894                 bio_endio(bio);
2895                 spin_unlock_irq(&ic->endio_wait.lock);
2896                 spin_lock_irq(&ic->endio_wait.lock);
2897         }
2898         spin_unlock_irq(&ic->endio_wait.lock);
2899 }
2900
2901
2902 static void init_journal(struct dm_integrity_c *ic, unsigned int start_section,
2903                          unsigned int n_sections, unsigned char commit_seq)
2904 {
2905         unsigned int i, j, n;
2906
2907         if (!n_sections)
2908                 return;
2909
2910         for (n = 0; n < n_sections; n++) {
2911                 i = start_section + n;
2912                 wraparound_section(ic, &i);
2913                 for (j = 0; j < ic->journal_section_sectors; j++) {
2914                         struct journal_sector *js = access_journal(ic, i, j);
2915
2916                         BUILD_BUG_ON(sizeof(js->sectors) != JOURNAL_SECTOR_DATA);
2917                         memset(&js->sectors, 0, sizeof(js->sectors));
2918                         js->commit_id = dm_integrity_commit_id(ic, i, j, commit_seq);
2919                 }
2920                 for (j = 0; j < ic->journal_section_entries; j++) {
2921                         struct journal_entry *je = access_journal_entry(ic, i, j);
2922
2923                         journal_entry_set_unused(je);
2924                 }
2925         }
2926
2927         write_journal(ic, start_section, n_sections);
2928 }
2929
2930 static int find_commit_seq(struct dm_integrity_c *ic, unsigned int i, unsigned int j, commit_id_t id)
2931 {
2932         unsigned char k;
2933
2934         for (k = 0; k < N_COMMIT_IDS; k++) {
2935                 if (dm_integrity_commit_id(ic, i, j, k) == id)
2936                         return k;
2937         }
2938         dm_integrity_io_error(ic, "journal commit id", -EIO);
2939         return -EIO;
2940 }
2941
2942 static void replay_journal(struct dm_integrity_c *ic)
2943 {
2944         unsigned int i, j;
2945         bool used_commit_ids[N_COMMIT_IDS];
2946         unsigned int max_commit_id_sections[N_COMMIT_IDS];
2947         unsigned int write_start, write_sections;
2948         unsigned int continue_section;
2949         bool journal_empty;
2950         unsigned char unused, last_used, want_commit_seq;
2951
2952         if (ic->mode == 'R')
2953                 return;
2954
2955         if (ic->journal_uptodate)
2956                 return;
2957
2958         last_used = 0;
2959         write_start = 0;
2960
2961         if (!ic->just_formatted) {
2962                 DEBUG_print("reading journal\n");
2963                 rw_journal(ic, REQ_OP_READ, 0, ic->journal_sections, NULL);
2964                 if (ic->journal_io)
2965                         DEBUG_bytes(lowmem_page_address(ic->journal_io[0].page), 64, "read journal");
2966                 if (ic->journal_io) {
2967                         struct journal_completion crypt_comp;
2968
2969                         crypt_comp.ic = ic;
2970                         init_completion(&crypt_comp.comp);
2971                         crypt_comp.in_flight = (atomic_t)ATOMIC_INIT(0);
2972                         encrypt_journal(ic, false, 0, ic->journal_sections, &crypt_comp);
2973                         wait_for_completion(&crypt_comp.comp);
2974                 }
2975                 DEBUG_bytes(lowmem_page_address(ic->journal[0].page), 64, "decrypted journal");
2976         }
2977
2978         if (dm_integrity_failed(ic))
2979                 goto clear_journal;
2980
2981         journal_empty = true;
2982         memset(used_commit_ids, 0, sizeof(used_commit_ids));
2983         memset(max_commit_id_sections, 0, sizeof(max_commit_id_sections));
2984         for (i = 0; i < ic->journal_sections; i++) {
2985                 for (j = 0; j < ic->journal_section_sectors; j++) {
2986                         int k;
2987                         struct journal_sector *js = access_journal(ic, i, j);
2988
2989                         k = find_commit_seq(ic, i, j, js->commit_id);
2990                         if (k < 0)
2991                                 goto clear_journal;
2992                         used_commit_ids[k] = true;
2993                         max_commit_id_sections[k] = i;
2994                 }
2995                 if (journal_empty) {
2996                         for (j = 0; j < ic->journal_section_entries; j++) {
2997                                 struct journal_entry *je = access_journal_entry(ic, i, j);
2998
2999                                 if (!journal_entry_is_unused(je)) {
3000                                         journal_empty = false;
3001                                         break;
3002                                 }
3003                         }
3004                 }
3005         }
3006
3007         if (!used_commit_ids[N_COMMIT_IDS - 1]) {
3008                 unused = N_COMMIT_IDS - 1;
3009                 while (unused && !used_commit_ids[unused - 1])
3010                         unused--;
3011         } else {
3012                 for (unused = 0; unused < N_COMMIT_IDS; unused++)
3013                         if (!used_commit_ids[unused])
3014                                 break;
3015                 if (unused == N_COMMIT_IDS) {
3016                         dm_integrity_io_error(ic, "journal commit ids", -EIO);
3017                         goto clear_journal;
3018                 }
3019         }
3020         DEBUG_print("first unused commit seq %d [%d,%d,%d,%d]\n",
3021                     unused, used_commit_ids[0], used_commit_ids[1],
3022                     used_commit_ids[2], used_commit_ids[3]);
3023
3024         last_used = prev_commit_seq(unused);
3025         want_commit_seq = prev_commit_seq(last_used);
3026
3027         if (!used_commit_ids[want_commit_seq] && used_commit_ids[prev_commit_seq(want_commit_seq)])
3028                 journal_empty = true;
3029
3030         write_start = max_commit_id_sections[last_used] + 1;
3031         if (unlikely(write_start >= ic->journal_sections))
3032                 want_commit_seq = next_commit_seq(want_commit_seq);
3033         wraparound_section(ic, &write_start);
3034
3035         i = write_start;
3036         for (write_sections = 0; write_sections < ic->journal_sections; write_sections++) {
3037                 for (j = 0; j < ic->journal_section_sectors; j++) {
3038                         struct journal_sector *js = access_journal(ic, i, j);
3039
3040                         if (js->commit_id != dm_integrity_commit_id(ic, i, j, want_commit_seq)) {
3041                                 /*
3042                                  * This could be caused by crash during writing.
3043                                  * We won't replay the inconsistent part of the
3044                                  * journal.
3045                                  */
3046                                 DEBUG_print("commit id mismatch at position (%u, %u): %d != %d\n",
3047                                             i, j, find_commit_seq(ic, i, j, js->commit_id), want_commit_seq);
3048                                 goto brk;
3049                         }
3050                 }
3051                 i++;
3052                 if (unlikely(i >= ic->journal_sections))
3053                         want_commit_seq = next_commit_seq(want_commit_seq);
3054                 wraparound_section(ic, &i);
3055         }
3056 brk:
3057
3058         if (!journal_empty) {
3059                 DEBUG_print("replaying %u sections, starting at %u, commit seq %d\n",
3060                             write_sections, write_start, want_commit_seq);
3061                 do_journal_write(ic, write_start, write_sections, true);
3062         }
3063
3064         if (write_sections == ic->journal_sections && (ic->mode == 'J' || journal_empty)) {
3065                 continue_section = write_start;
3066                 ic->commit_seq = want_commit_seq;
3067                 DEBUG_print("continuing from section %u, commit seq %d\n", write_start, ic->commit_seq);
3068         } else {
3069                 unsigned int s;
3070                 unsigned char erase_seq;
3071
3072 clear_journal:
3073                 DEBUG_print("clearing journal\n");
3074
3075                 erase_seq = prev_commit_seq(prev_commit_seq(last_used));
3076                 s = write_start;
3077                 init_journal(ic, s, 1, erase_seq);
3078                 s++;
3079                 wraparound_section(ic, &s);
3080                 if (ic->journal_sections >= 2) {
3081                         init_journal(ic, s, ic->journal_sections - 2, erase_seq);
3082                         s += ic->journal_sections - 2;
3083                         wraparound_section(ic, &s);
3084                         init_journal(ic, s, 1, erase_seq);
3085                 }
3086
3087                 continue_section = 0;
3088                 ic->commit_seq = next_commit_seq(erase_seq);
3089         }
3090
3091         ic->committed_section = continue_section;
3092         ic->n_committed_sections = 0;
3093
3094         ic->uncommitted_section = continue_section;
3095         ic->n_uncommitted_sections = 0;
3096
3097         ic->free_section = continue_section;
3098         ic->free_section_entry = 0;
3099         ic->free_sectors = ic->journal_entries;
3100
3101         ic->journal_tree_root = RB_ROOT;
3102         for (i = 0; i < ic->journal_entries; i++)
3103                 init_journal_node(&ic->journal_tree[i]);
3104 }
3105
3106 static void dm_integrity_enter_synchronous_mode(struct dm_integrity_c *ic)
3107 {
3108         DEBUG_print("%s\n", __func__);
3109
3110         if (ic->mode == 'B') {
3111                 ic->bitmap_flush_interval = msecs_to_jiffies(10) + 1;
3112                 ic->synchronous_mode = 1;
3113
3114                 cancel_delayed_work_sync(&ic->bitmap_flush_work);
3115                 queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, 0);
3116                 flush_workqueue(ic->commit_wq);
3117         }
3118 }
3119
3120 static int dm_integrity_reboot(struct notifier_block *n, unsigned long code, void *x)
3121 {
3122         struct dm_integrity_c *ic = container_of(n, struct dm_integrity_c, reboot_notifier);
3123
3124         DEBUG_print("%s\n", __func__);
3125
3126         dm_integrity_enter_synchronous_mode(ic);
3127
3128         return NOTIFY_DONE;
3129 }
3130
3131 static void dm_integrity_postsuspend(struct dm_target *ti)
3132 {
3133         struct dm_integrity_c *ic = ti->private;
3134         int r;
3135
3136         WARN_ON(unregister_reboot_notifier(&ic->reboot_notifier));
3137
3138         del_timer_sync(&ic->autocommit_timer);
3139
3140         if (ic->recalc_wq)
3141                 drain_workqueue(ic->recalc_wq);
3142
3143         if (ic->mode == 'B')
3144                 cancel_delayed_work_sync(&ic->bitmap_flush_work);
3145
3146         queue_work(ic->commit_wq, &ic->commit_work);
3147         drain_workqueue(ic->commit_wq);
3148
3149         if (ic->mode == 'J') {
3150                 queue_work(ic->writer_wq, &ic->writer_work);
3151                 drain_workqueue(ic->writer_wq);
3152                 dm_integrity_flush_buffers(ic, true);
3153                 if (ic->wrote_to_journal) {
3154                         init_journal(ic, ic->free_section,
3155                                      ic->journal_sections - ic->free_section, ic->commit_seq);
3156                         if (ic->free_section) {
3157                                 init_journal(ic, 0, ic->free_section,
3158                                              next_commit_seq(ic->commit_seq));
3159                         }
3160                 }
3161         }
3162
3163         if (ic->mode == 'B') {
3164                 dm_integrity_flush_buffers(ic, true);
3165 #if 1
3166                 /* set to 0 to test bitmap replay code */
3167                 init_journal(ic, 0, ic->journal_sections, 0);
3168                 ic->sb->flags &= ~cpu_to_le32(SB_FLAG_DIRTY_BITMAP);
3169                 r = sync_rw_sb(ic, REQ_OP_WRITE | REQ_FUA);
3170                 if (unlikely(r))
3171                         dm_integrity_io_error(ic, "writing superblock", r);
3172 #endif
3173         }
3174
3175         BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
3176
3177         ic->journal_uptodate = true;
3178 }
3179
3180 static void dm_integrity_resume(struct dm_target *ti)
3181 {
3182         struct dm_integrity_c *ic = ti->private;
3183         __u64 old_provided_data_sectors = le64_to_cpu(ic->sb->provided_data_sectors);
3184         int r;
3185
3186         DEBUG_print("resume\n");
3187
3188         ic->wrote_to_journal = false;
3189
3190         if (ic->provided_data_sectors != old_provided_data_sectors) {
3191                 if (ic->provided_data_sectors > old_provided_data_sectors &&
3192                     ic->mode == 'B' &&
3193                     ic->sb->log2_blocks_per_bitmap_bit == ic->log2_blocks_per_bitmap_bit) {
3194                         rw_journal_sectors(ic, REQ_OP_READ, 0,
3195                                            ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3196                         block_bitmap_op(ic, ic->journal, old_provided_data_sectors,
3197                                         ic->provided_data_sectors - old_provided_data_sectors, BITMAP_OP_SET);
3198                         rw_journal_sectors(ic, REQ_OP_WRITE | REQ_FUA | REQ_SYNC, 0,
3199                                            ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3200                 }
3201
3202                 ic->sb->provided_data_sectors = cpu_to_le64(ic->provided_data_sectors);
3203                 r = sync_rw_sb(ic, REQ_OP_WRITE | REQ_FUA);
3204                 if (unlikely(r))
3205                         dm_integrity_io_error(ic, "writing superblock", r);
3206         }
3207
3208         if (ic->sb->flags & cpu_to_le32(SB_FLAG_DIRTY_BITMAP)) {
3209                 DEBUG_print("resume dirty_bitmap\n");
3210                 rw_journal_sectors(ic, REQ_OP_READ, 0,
3211                                    ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3212                 if (ic->mode == 'B') {
3213                         if (ic->sb->log2_blocks_per_bitmap_bit == ic->log2_blocks_per_bitmap_bit &&
3214                             !ic->reset_recalculate_flag) {
3215                                 block_bitmap_copy(ic, ic->recalc_bitmap, ic->journal);
3216                                 block_bitmap_copy(ic, ic->may_write_bitmap, ic->journal);
3217                                 if (!block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors,
3218                                                      BITMAP_OP_TEST_ALL_CLEAR)) {
3219                                         ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
3220                                         ic->sb->recalc_sector = cpu_to_le64(0);
3221                                 }
3222                         } else {
3223                                 DEBUG_print("non-matching blocks_per_bitmap_bit: %u, %u\n",
3224                                             ic->sb->log2_blocks_per_bitmap_bit, ic->log2_blocks_per_bitmap_bit);
3225                                 ic->sb->log2_blocks_per_bitmap_bit = ic->log2_blocks_per_bitmap_bit;
3226                                 block_bitmap_op(ic, ic->recalc_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_SET);
3227                                 block_bitmap_op(ic, ic->may_write_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_SET);
3228                                 block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors, BITMAP_OP_SET);
3229                                 rw_journal_sectors(ic, REQ_OP_WRITE | REQ_FUA | REQ_SYNC, 0,
3230                                                    ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3231                                 ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
3232                                 ic->sb->recalc_sector = cpu_to_le64(0);
3233                         }
3234                 } else {
3235                         if (!(ic->sb->log2_blocks_per_bitmap_bit == ic->log2_blocks_per_bitmap_bit &&
3236                               block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors, BITMAP_OP_TEST_ALL_CLEAR)) ||
3237                             ic->reset_recalculate_flag) {
3238                                 ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
3239                                 ic->sb->recalc_sector = cpu_to_le64(0);
3240                         }
3241                         init_journal(ic, 0, ic->journal_sections, 0);
3242                         replay_journal(ic);
3243                         ic->sb->flags &= ~cpu_to_le32(SB_FLAG_DIRTY_BITMAP);
3244                 }
3245                 r = sync_rw_sb(ic, REQ_OP_WRITE | REQ_FUA);
3246                 if (unlikely(r))
3247                         dm_integrity_io_error(ic, "writing superblock", r);
3248         } else {
3249                 replay_journal(ic);
3250                 if (ic->reset_recalculate_flag) {
3251                         ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
3252                         ic->sb->recalc_sector = cpu_to_le64(0);
3253                 }
3254                 if (ic->mode == 'B') {
3255                         ic->sb->flags |= cpu_to_le32(SB_FLAG_DIRTY_BITMAP);
3256                         ic->sb->log2_blocks_per_bitmap_bit = ic->log2_blocks_per_bitmap_bit;
3257                         r = sync_rw_sb(ic, REQ_OP_WRITE | REQ_FUA);
3258                         if (unlikely(r))
3259                                 dm_integrity_io_error(ic, "writing superblock", r);
3260
3261                         block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors, BITMAP_OP_CLEAR);
3262                         block_bitmap_op(ic, ic->recalc_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_CLEAR);
3263                         block_bitmap_op(ic, ic->may_write_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_CLEAR);
3264                         if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING) &&
3265                             le64_to_cpu(ic->sb->recalc_sector) < ic->provided_data_sectors) {
3266                                 block_bitmap_op(ic, ic->journal, le64_to_cpu(ic->sb->recalc_sector),
3267                                                 ic->provided_data_sectors - le64_to_cpu(ic->sb->recalc_sector), BITMAP_OP_SET);
3268                                 block_bitmap_op(ic, ic->recalc_bitmap, le64_to_cpu(ic->sb->recalc_sector),
3269                                                 ic->provided_data_sectors - le64_to_cpu(ic->sb->recalc_sector), BITMAP_OP_SET);
3270                                 block_bitmap_op(ic, ic->may_write_bitmap, le64_to_cpu(ic->sb->recalc_sector),
3271                                                 ic->provided_data_sectors - le64_to_cpu(ic->sb->recalc_sector), BITMAP_OP_SET);
3272                         }
3273                         rw_journal_sectors(ic, REQ_OP_WRITE | REQ_FUA | REQ_SYNC, 0,
3274                                            ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3275                 }
3276         }
3277
3278         DEBUG_print("testing recalc: %x\n", ic->sb->flags);
3279         if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
3280                 __u64 recalc_pos = le64_to_cpu(ic->sb->recalc_sector);
3281
3282                 DEBUG_print("recalc pos: %llx / %llx\n", recalc_pos, ic->provided_data_sectors);
3283                 if (recalc_pos < ic->provided_data_sectors) {
3284                         queue_work(ic->recalc_wq, &ic->recalc_work);
3285                 } else if (recalc_pos > ic->provided_data_sectors) {
3286                         ic->sb->recalc_sector = cpu_to_le64(ic->provided_data_sectors);
3287                         recalc_write_super(ic);
3288                 }
3289         }
3290
3291         ic->reboot_notifier.notifier_call = dm_integrity_reboot;
3292         ic->reboot_notifier.next = NULL;
3293         ic->reboot_notifier.priority = INT_MAX - 1;     /* be notified after md and before hardware drivers */
3294         WARN_ON(register_reboot_notifier(&ic->reboot_notifier));
3295
3296 #if 0
3297         /* set to 1 to stress test synchronous mode */
3298         dm_integrity_enter_synchronous_mode(ic);
3299 #endif
3300 }
3301
3302 static void dm_integrity_status(struct dm_target *ti, status_type_t type,
3303                                 unsigned int status_flags, char *result, unsigned int maxlen)
3304 {
3305         struct dm_integrity_c *ic = ti->private;
3306         unsigned int arg_count;
3307         size_t sz = 0;
3308
3309         switch (type) {
3310         case STATUSTYPE_INFO:
3311                 DMEMIT("%llu %llu",
3312                         (unsigned long long)atomic64_read(&ic->number_of_mismatches),
3313                         ic->provided_data_sectors);
3314                 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
3315                         DMEMIT(" %llu", le64_to_cpu(ic->sb->recalc_sector));
3316                 else
3317                         DMEMIT(" -");
3318                 break;
3319
3320         case STATUSTYPE_TABLE: {
3321                 __u64 watermark_percentage = (__u64)(ic->journal_entries - ic->free_sectors_threshold) * 100;
3322
3323                 watermark_percentage += ic->journal_entries / 2;
3324                 do_div(watermark_percentage, ic->journal_entries);
3325                 arg_count = 3;
3326                 arg_count += !!ic->meta_dev;
3327                 arg_count += ic->sectors_per_block != 1;
3328                 arg_count += !!(ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING));
3329                 arg_count += ic->reset_recalculate_flag;
3330                 arg_count += ic->discard;
3331                 arg_count += ic->mode == 'J';
3332                 arg_count += ic->mode == 'J';
3333                 arg_count += ic->mode == 'B';
3334                 arg_count += ic->mode == 'B';
3335                 arg_count += !!ic->internal_hash_alg.alg_string;
3336                 arg_count += !!ic->journal_crypt_alg.alg_string;
3337                 arg_count += !!ic->journal_mac_alg.alg_string;
3338                 arg_count += (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING)) != 0;
3339                 arg_count += (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) != 0;
3340                 arg_count += ic->legacy_recalculate;
3341                 DMEMIT("%s %llu %u %c %u", ic->dev->name, ic->start,
3342                        ic->tag_size, ic->mode, arg_count);
3343                 if (ic->meta_dev)
3344                         DMEMIT(" meta_device:%s", ic->meta_dev->name);
3345                 if (ic->sectors_per_block != 1)
3346                         DMEMIT(" block_size:%u", ic->sectors_per_block << SECTOR_SHIFT);
3347                 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
3348                         DMEMIT(" recalculate");
3349                 if (ic->reset_recalculate_flag)
3350                         DMEMIT(" reset_recalculate");
3351                 if (ic->discard)
3352                         DMEMIT(" allow_discards");
3353                 DMEMIT(" journal_sectors:%u", ic->initial_sectors - SB_SECTORS);
3354                 DMEMIT(" interleave_sectors:%u", 1U << ic->sb->log2_interleave_sectors);
3355                 DMEMIT(" buffer_sectors:%u", 1U << ic->log2_buffer_sectors);
3356                 if (ic->mode == 'J') {
3357                         DMEMIT(" journal_watermark:%u", (unsigned int)watermark_percentage);
3358                         DMEMIT(" commit_time:%u", ic->autocommit_msec);
3359                 }
3360                 if (ic->mode == 'B') {
3361                         DMEMIT(" sectors_per_bit:%llu", (sector_t)ic->sectors_per_block << ic->log2_blocks_per_bitmap_bit);
3362                         DMEMIT(" bitmap_flush_interval:%u", jiffies_to_msecs(ic->bitmap_flush_interval));
3363                 }
3364                 if ((ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING)) != 0)
3365                         DMEMIT(" fix_padding");
3366                 if ((ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) != 0)
3367                         DMEMIT(" fix_hmac");
3368                 if (ic->legacy_recalculate)
3369                         DMEMIT(" legacy_recalculate");
3370
3371 #define EMIT_ALG(a, n)                                                  \
3372                 do {                                                    \
3373                         if (ic->a.alg_string) {                         \
3374                                 DMEMIT(" %s:%s", n, ic->a.alg_string);  \
3375                                 if (ic->a.key_string)                   \
3376                                         DMEMIT(":%s", ic->a.key_string);\
3377                         }                                               \
3378                 } while (0)
3379                 EMIT_ALG(internal_hash_alg, "internal_hash");
3380                 EMIT_ALG(journal_crypt_alg, "journal_crypt");
3381                 EMIT_ALG(journal_mac_alg, "journal_mac");
3382                 break;
3383         }
3384         case STATUSTYPE_IMA:
3385                 DMEMIT_TARGET_NAME_VERSION(ti->type);
3386                 DMEMIT(",dev_name=%s,start=%llu,tag_size=%u,mode=%c",
3387                         ic->dev->name, ic->start, ic->tag_size, ic->mode);
3388
3389                 if (ic->meta_dev)
3390                         DMEMIT(",meta_device=%s", ic->meta_dev->name);
3391                 if (ic->sectors_per_block != 1)
3392                         DMEMIT(",block_size=%u", ic->sectors_per_block << SECTOR_SHIFT);
3393
3394                 DMEMIT(",recalculate=%c", (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) ?
3395                        'y' : 'n');
3396                 DMEMIT(",allow_discards=%c", ic->discard ? 'y' : 'n');
3397                 DMEMIT(",fix_padding=%c",
3398                        ((ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING)) != 0) ? 'y' : 'n');
3399                 DMEMIT(",fix_hmac=%c",
3400                        ((ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) != 0) ? 'y' : 'n');
3401                 DMEMIT(",legacy_recalculate=%c", ic->legacy_recalculate ? 'y' : 'n');
3402
3403                 DMEMIT(",journal_sectors=%u", ic->initial_sectors - SB_SECTORS);
3404                 DMEMIT(",interleave_sectors=%u", 1U << ic->sb->log2_interleave_sectors);
3405                 DMEMIT(",buffer_sectors=%u", 1U << ic->log2_buffer_sectors);
3406                 DMEMIT(";");
3407                 break;
3408         }
3409 }
3410
3411 static int dm_integrity_iterate_devices(struct dm_target *ti,
3412                                         iterate_devices_callout_fn fn, void *data)
3413 {
3414         struct dm_integrity_c *ic = ti->private;
3415
3416         if (!ic->meta_dev)
3417                 return fn(ti, ic->dev, ic->start + ic->initial_sectors + ic->metadata_run, ti->len, data);
3418         else
3419                 return fn(ti, ic->dev, 0, ti->len, data);
3420 }
3421
3422 static void dm_integrity_io_hints(struct dm_target *ti, struct queue_limits *limits)
3423 {
3424         struct dm_integrity_c *ic = ti->private;
3425
3426         if (ic->sectors_per_block > 1) {
3427                 limits->logical_block_size = ic->sectors_per_block << SECTOR_SHIFT;
3428                 limits->physical_block_size = ic->sectors_per_block << SECTOR_SHIFT;
3429                 blk_limits_io_min(limits, ic->sectors_per_block << SECTOR_SHIFT);
3430                 limits->dma_alignment = limits->logical_block_size - 1;
3431         }
3432 }
3433
3434 static void calculate_journal_section_size(struct dm_integrity_c *ic)
3435 {
3436         unsigned int sector_space = JOURNAL_SECTOR_DATA;
3437
3438         ic->journal_sections = le32_to_cpu(ic->sb->journal_sections);
3439         ic->journal_entry_size = roundup(offsetof(struct journal_entry, last_bytes[ic->sectors_per_block]) + ic->tag_size,
3440                                          JOURNAL_ENTRY_ROUNDUP);
3441
3442         if (ic->sb->flags & cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC))
3443                 sector_space -= JOURNAL_MAC_PER_SECTOR;
3444         ic->journal_entries_per_sector = sector_space / ic->journal_entry_size;
3445         ic->journal_section_entries = ic->journal_entries_per_sector * JOURNAL_BLOCK_SECTORS;
3446         ic->journal_section_sectors = (ic->journal_section_entries << ic->sb->log2_sectors_per_block) + JOURNAL_BLOCK_SECTORS;
3447         ic->journal_entries = ic->journal_section_entries * ic->journal_sections;
3448 }
3449
3450 static int calculate_device_limits(struct dm_integrity_c *ic)
3451 {
3452         __u64 initial_sectors;
3453
3454         calculate_journal_section_size(ic);
3455         initial_sectors = SB_SECTORS + (__u64)ic->journal_section_sectors * ic->journal_sections;
3456         if (initial_sectors + METADATA_PADDING_SECTORS >= ic->meta_device_sectors || initial_sectors > UINT_MAX)
3457                 return -EINVAL;
3458         ic->initial_sectors = initial_sectors;
3459
3460         if (!ic->meta_dev) {
3461                 sector_t last_sector, last_area, last_offset;
3462
3463                 /* we have to maintain excessive padding for compatibility with existing volumes */
3464                 __u64 metadata_run_padding =
3465                         ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING) ?
3466                         (__u64)(METADATA_PADDING_SECTORS << SECTOR_SHIFT) :
3467                         (__u64)(1 << SECTOR_SHIFT << METADATA_PADDING_SECTORS);
3468
3469                 ic->metadata_run = round_up((__u64)ic->tag_size << (ic->sb->log2_interleave_sectors - ic->sb->log2_sectors_per_block),
3470                                             metadata_run_padding) >> SECTOR_SHIFT;
3471                 if (!(ic->metadata_run & (ic->metadata_run - 1)))
3472                         ic->log2_metadata_run = __ffs(ic->metadata_run);
3473                 else
3474                         ic->log2_metadata_run = -1;
3475
3476                 get_area_and_offset(ic, ic->provided_data_sectors - 1, &last_area, &last_offset);
3477                 last_sector = get_data_sector(ic, last_area, last_offset);
3478                 if (last_sector < ic->start || last_sector >= ic->meta_device_sectors)
3479                         return -EINVAL;
3480         } else {
3481                 __u64 meta_size = (ic->provided_data_sectors >> ic->sb->log2_sectors_per_block) * ic->tag_size;
3482
3483                 meta_size = (meta_size + ((1U << (ic->log2_buffer_sectors + SECTOR_SHIFT)) - 1))
3484                                 >> (ic->log2_buffer_sectors + SECTOR_SHIFT);
3485                 meta_size <<= ic->log2_buffer_sectors;
3486                 if (ic->initial_sectors + meta_size < ic->initial_sectors ||
3487                     ic->initial_sectors + meta_size > ic->meta_device_sectors)
3488                         return -EINVAL;
3489                 ic->metadata_run = 1;
3490                 ic->log2_metadata_run = 0;
3491         }
3492
3493         return 0;
3494 }
3495
3496 static void get_provided_data_sectors(struct dm_integrity_c *ic)
3497 {
3498         if (!ic->meta_dev) {
3499                 int test_bit;
3500
3501                 ic->provided_data_sectors = 0;
3502                 for (test_bit = fls64(ic->meta_device_sectors) - 1; test_bit >= 3; test_bit--) {
3503                         __u64 prev_data_sectors = ic->provided_data_sectors;
3504
3505                         ic->provided_data_sectors |= (sector_t)1 << test_bit;
3506                         if (calculate_device_limits(ic))
3507                                 ic->provided_data_sectors = prev_data_sectors;
3508                 }
3509         } else {
3510                 ic->provided_data_sectors = ic->data_device_sectors;
3511                 ic->provided_data_sectors &= ~(sector_t)(ic->sectors_per_block - 1);
3512         }
3513 }
3514
3515 static int initialize_superblock(struct dm_integrity_c *ic,
3516                                  unsigned int journal_sectors, unsigned int interleave_sectors)
3517 {
3518         unsigned int journal_sections;
3519         int test_bit;
3520
3521         memset(ic->sb, 0, SB_SECTORS << SECTOR_SHIFT);
3522         memcpy(ic->sb->magic, SB_MAGIC, 8);
3523         ic->sb->integrity_tag_size = cpu_to_le16(ic->tag_size);
3524         ic->sb->log2_sectors_per_block = __ffs(ic->sectors_per_block);
3525         if (ic->journal_mac_alg.alg_string)
3526                 ic->sb->flags |= cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC);
3527
3528         calculate_journal_section_size(ic);
3529         journal_sections = journal_sectors / ic->journal_section_sectors;
3530         if (!journal_sections)
3531                 journal_sections = 1;
3532
3533         if (ic->fix_hmac && (ic->internal_hash_alg.alg_string || ic->journal_mac_alg.alg_string)) {
3534                 ic->sb->flags |= cpu_to_le32(SB_FLAG_FIXED_HMAC);
3535                 get_random_bytes(ic->sb->salt, SALT_SIZE);
3536         }
3537
3538         if (!ic->meta_dev) {
3539                 if (ic->fix_padding)
3540                         ic->sb->flags |= cpu_to_le32(SB_FLAG_FIXED_PADDING);
3541                 ic->sb->journal_sections = cpu_to_le32(journal_sections);
3542                 if (!interleave_sectors)
3543                         interleave_sectors = DEFAULT_INTERLEAVE_SECTORS;
3544                 ic->sb->log2_interleave_sectors = __fls(interleave_sectors);
3545                 ic->sb->log2_interleave_sectors = max_t(__u8, MIN_LOG2_INTERLEAVE_SECTORS, ic->sb->log2_interleave_sectors);
3546                 ic->sb->log2_interleave_sectors = min_t(__u8, MAX_LOG2_INTERLEAVE_SECTORS, ic->sb->log2_interleave_sectors);
3547
3548                 get_provided_data_sectors(ic);
3549                 if (!ic->provided_data_sectors)
3550                         return -EINVAL;
3551         } else {
3552                 ic->sb->log2_interleave_sectors = 0;
3553
3554                 get_provided_data_sectors(ic);
3555                 if (!ic->provided_data_sectors)
3556                         return -EINVAL;
3557
3558 try_smaller_buffer:
3559                 ic->sb->journal_sections = cpu_to_le32(0);
3560                 for (test_bit = fls(journal_sections) - 1; test_bit >= 0; test_bit--) {
3561                         __u32 prev_journal_sections = le32_to_cpu(ic->sb->journal_sections);
3562                         __u32 test_journal_sections = prev_journal_sections | (1U << test_bit);
3563
3564                         if (test_journal_sections > journal_sections)
3565                                 continue;
3566                         ic->sb->journal_sections = cpu_to_le32(test_journal_sections);
3567                         if (calculate_device_limits(ic))
3568                                 ic->sb->journal_sections = cpu_to_le32(prev_journal_sections);
3569
3570                 }
3571                 if (!le32_to_cpu(ic->sb->journal_sections)) {
3572                         if (ic->log2_buffer_sectors > 3) {
3573                                 ic->log2_buffer_sectors--;
3574                                 goto try_smaller_buffer;
3575                         }
3576                         return -EINVAL;
3577                 }
3578         }
3579
3580         ic->sb->provided_data_sectors = cpu_to_le64(ic->provided_data_sectors);
3581
3582         sb_set_version(ic);
3583
3584         return 0;
3585 }
3586
3587 static void dm_integrity_set(struct dm_target *ti, struct dm_integrity_c *ic)
3588 {
3589         struct gendisk *disk = dm_disk(dm_table_get_md(ti->table));
3590         struct blk_integrity bi;
3591
3592         memset(&bi, 0, sizeof(bi));
3593         bi.profile = &dm_integrity_profile;
3594         bi.tuple_size = ic->tag_size;
3595         bi.tag_size = bi.tuple_size;
3596         bi.interval_exp = ic->sb->log2_sectors_per_block + SECTOR_SHIFT;
3597
3598         blk_integrity_register(disk, &bi);
3599         blk_queue_max_integrity_segments(disk->queue, UINT_MAX);
3600 }
3601
3602 static void dm_integrity_free_page_list(struct page_list *pl)
3603 {
3604         unsigned int i;
3605
3606         if (!pl)
3607                 return;
3608         for (i = 0; pl[i].page; i++)
3609                 __free_page(pl[i].page);
3610         kvfree(pl);
3611 }
3612
3613 static struct page_list *dm_integrity_alloc_page_list(unsigned int n_pages)
3614 {
3615         struct page_list *pl;
3616         unsigned int i;
3617
3618         pl = kvmalloc_array(n_pages + 1, sizeof(struct page_list), GFP_KERNEL | __GFP_ZERO);
3619         if (!pl)
3620                 return NULL;
3621
3622         for (i = 0; i < n_pages; i++) {
3623                 pl[i].page = alloc_page(GFP_KERNEL);
3624                 if (!pl[i].page) {
3625                         dm_integrity_free_page_list(pl);
3626                         return NULL;
3627                 }
3628                 if (i)
3629                         pl[i - 1].next = &pl[i];
3630         }
3631         pl[i].page = NULL;
3632         pl[i].next = NULL;
3633
3634         return pl;
3635 }
3636
3637 static void dm_integrity_free_journal_scatterlist(struct dm_integrity_c *ic, struct scatterlist **sl)
3638 {
3639         unsigned int i;
3640
3641         for (i = 0; i < ic->journal_sections; i++)
3642                 kvfree(sl[i]);
3643         kvfree(sl);
3644 }
3645
3646 static struct scatterlist **dm_integrity_alloc_journal_scatterlist(struct dm_integrity_c *ic,
3647                                                                    struct page_list *pl)
3648 {
3649         struct scatterlist **sl;
3650         unsigned int i;
3651
3652         sl = kvmalloc_array(ic->journal_sections,
3653                             sizeof(struct scatterlist *),
3654                             GFP_KERNEL | __GFP_ZERO);
3655         if (!sl)
3656                 return NULL;
3657
3658         for (i = 0; i < ic->journal_sections; i++) {
3659                 struct scatterlist *s;
3660                 unsigned int start_index, start_offset;
3661                 unsigned int end_index, end_offset;
3662                 unsigned int n_pages;
3663                 unsigned int idx;
3664
3665                 page_list_location(ic, i, 0, &start_index, &start_offset);
3666                 page_list_location(ic, i, ic->journal_section_sectors - 1,
3667                                    &end_index, &end_offset);
3668
3669                 n_pages = (end_index - start_index + 1);
3670
3671                 s = kvmalloc_array(n_pages, sizeof(struct scatterlist),
3672                                    GFP_KERNEL);
3673                 if (!s) {
3674                         dm_integrity_free_journal_scatterlist(ic, sl);
3675                         return NULL;
3676                 }
3677
3678                 sg_init_table(s, n_pages);
3679                 for (idx = start_index; idx <= end_index; idx++) {
3680                         char *va = lowmem_page_address(pl[idx].page);
3681                         unsigned int start = 0, end = PAGE_SIZE;
3682
3683                         if (idx == start_index)
3684                                 start = start_offset;
3685                         if (idx == end_index)
3686                                 end = end_offset + (1 << SECTOR_SHIFT);
3687                         sg_set_buf(&s[idx - start_index], va + start, end - start);
3688                 }
3689
3690                 sl[i] = s;
3691         }
3692
3693         return sl;
3694 }
3695
3696 static void free_alg(struct alg_spec *a)
3697 {
3698         kfree_sensitive(a->alg_string);
3699         kfree_sensitive(a->key);
3700         memset(a, 0, sizeof(*a));
3701 }
3702
3703 static int get_alg_and_key(const char *arg, struct alg_spec *a, char **error, char *error_inval)
3704 {
3705         char *k;
3706
3707         free_alg(a);
3708
3709         a->alg_string = kstrdup(strchr(arg, ':') + 1, GFP_KERNEL);
3710         if (!a->alg_string)
3711                 goto nomem;
3712
3713         k = strchr(a->alg_string, ':');
3714         if (k) {
3715                 *k = 0;
3716                 a->key_string = k + 1;
3717                 if (strlen(a->key_string) & 1)
3718                         goto inval;
3719
3720                 a->key_size = strlen(a->key_string) / 2;
3721                 a->key = kmalloc(a->key_size, GFP_KERNEL);
3722                 if (!a->key)
3723                         goto nomem;
3724                 if (hex2bin(a->key, a->key_string, a->key_size))
3725                         goto inval;
3726         }
3727
3728         return 0;
3729 inval:
3730         *error = error_inval;
3731         return -EINVAL;
3732 nomem:
3733         *error = "Out of memory for an argument";
3734         return -ENOMEM;
3735 }
3736
3737 static int get_mac(struct crypto_shash **hash, struct alg_spec *a, char **error,
3738                    char *error_alg, char *error_key)
3739 {
3740         int r;
3741
3742         if (a->alg_string) {
3743                 *hash = crypto_alloc_shash(a->alg_string, 0, CRYPTO_ALG_ALLOCATES_MEMORY);
3744                 if (IS_ERR(*hash)) {
3745                         *error = error_alg;
3746                         r = PTR_ERR(*hash);
3747                         *hash = NULL;
3748                         return r;
3749                 }
3750
3751                 if (a->key) {
3752                         r = crypto_shash_setkey(*hash, a->key, a->key_size);
3753                         if (r) {
3754                                 *error = error_key;
3755                                 return r;
3756                         }
3757                 } else if (crypto_shash_get_flags(*hash) & CRYPTO_TFM_NEED_KEY) {
3758                         *error = error_key;
3759                         return -ENOKEY;
3760                 }
3761         }
3762
3763         return 0;
3764 }
3765
3766 static int create_journal(struct dm_integrity_c *ic, char **error)
3767 {
3768         int r = 0;
3769         unsigned int i;
3770         __u64 journal_pages, journal_desc_size, journal_tree_size;
3771         unsigned char *crypt_data = NULL, *crypt_iv = NULL;
3772         struct skcipher_request *req = NULL;
3773
3774         ic->commit_ids[0] = cpu_to_le64(0x1111111111111111ULL);
3775         ic->commit_ids[1] = cpu_to_le64(0x2222222222222222ULL);
3776         ic->commit_ids[2] = cpu_to_le64(0x3333333333333333ULL);
3777         ic->commit_ids[3] = cpu_to_le64(0x4444444444444444ULL);
3778
3779         journal_pages = roundup((__u64)ic->journal_sections * ic->journal_section_sectors,
3780                                 PAGE_SIZE >> SECTOR_SHIFT) >> (PAGE_SHIFT - SECTOR_SHIFT);
3781         journal_desc_size = journal_pages * sizeof(struct page_list);
3782         if (journal_pages >= totalram_pages() - totalhigh_pages() || journal_desc_size > ULONG_MAX) {
3783                 *error = "Journal doesn't fit into memory";
3784                 r = -ENOMEM;
3785                 goto bad;
3786         }
3787         ic->journal_pages = journal_pages;
3788
3789         ic->journal = dm_integrity_alloc_page_list(ic->journal_pages);
3790         if (!ic->journal) {
3791                 *error = "Could not allocate memory for journal";
3792                 r = -ENOMEM;
3793                 goto bad;
3794         }
3795         if (ic->journal_crypt_alg.alg_string) {
3796                 unsigned int ivsize, blocksize;
3797                 struct journal_completion comp;
3798
3799                 comp.ic = ic;
3800                 ic->journal_crypt = crypto_alloc_skcipher(ic->journal_crypt_alg.alg_string, 0, CRYPTO_ALG_ALLOCATES_MEMORY);
3801                 if (IS_ERR(ic->journal_crypt)) {
3802                         *error = "Invalid journal cipher";
3803                         r = PTR_ERR(ic->journal_crypt);
3804                         ic->journal_crypt = NULL;
3805                         goto bad;
3806                 }
3807                 ivsize = crypto_skcipher_ivsize(ic->journal_crypt);
3808                 blocksize = crypto_skcipher_blocksize(ic->journal_crypt);
3809
3810                 if (ic->journal_crypt_alg.key) {
3811                         r = crypto_skcipher_setkey(ic->journal_crypt, ic->journal_crypt_alg.key,
3812                                                    ic->journal_crypt_alg.key_size);
3813                         if (r) {
3814                                 *error = "Error setting encryption key";
3815                                 goto bad;
3816                         }
3817                 }
3818                 DEBUG_print("cipher %s, block size %u iv size %u\n",
3819                             ic->journal_crypt_alg.alg_string, blocksize, ivsize);
3820
3821                 ic->journal_io = dm_integrity_alloc_page_list(ic->journal_pages);
3822                 if (!ic->journal_io) {
3823                         *error = "Could not allocate memory for journal io";
3824                         r = -ENOMEM;
3825                         goto bad;
3826                 }
3827
3828                 if (blocksize == 1) {
3829                         struct scatterlist *sg;
3830
3831                         req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
3832                         if (!req) {
3833                                 *error = "Could not allocate crypt request";
3834                                 r = -ENOMEM;
3835                                 goto bad;
3836                         }
3837
3838                         crypt_iv = kzalloc(ivsize, GFP_KERNEL);
3839                         if (!crypt_iv) {
3840                                 *error = "Could not allocate iv";
3841                                 r = -ENOMEM;
3842                                 goto bad;
3843                         }
3844
3845                         ic->journal_xor = dm_integrity_alloc_page_list(ic->journal_pages);
3846                         if (!ic->journal_xor) {
3847                                 *error = "Could not allocate memory for journal xor";
3848                                 r = -ENOMEM;
3849                                 goto bad;
3850                         }
3851
3852                         sg = kvmalloc_array(ic->journal_pages + 1,
3853                                             sizeof(struct scatterlist),
3854                                             GFP_KERNEL);
3855                         if (!sg) {
3856                                 *error = "Unable to allocate sg list";
3857                                 r = -ENOMEM;
3858                                 goto bad;
3859                         }
3860                         sg_init_table(sg, ic->journal_pages + 1);
3861                         for (i = 0; i < ic->journal_pages; i++) {
3862                                 char *va = lowmem_page_address(ic->journal_xor[i].page);
3863
3864                                 clear_page(va);
3865                                 sg_set_buf(&sg[i], va, PAGE_SIZE);
3866                         }
3867                         sg_set_buf(&sg[i], &ic->commit_ids, sizeof(ic->commit_ids));
3868
3869                         skcipher_request_set_crypt(req, sg, sg,
3870                                                    PAGE_SIZE * ic->journal_pages + sizeof(ic->commit_ids), crypt_iv);
3871                         init_completion(&comp.comp);
3872                         comp.in_flight = (atomic_t)ATOMIC_INIT(1);
3873                         if (do_crypt(true, req, &comp))
3874                                 wait_for_completion(&comp.comp);
3875                         kvfree(sg);
3876                         r = dm_integrity_failed(ic);
3877                         if (r) {
3878                                 *error = "Unable to encrypt journal";
3879                                 goto bad;
3880                         }
3881                         DEBUG_bytes(lowmem_page_address(ic->journal_xor[0].page), 64, "xor data");
3882
3883                         crypto_free_skcipher(ic->journal_crypt);
3884                         ic->journal_crypt = NULL;
3885                 } else {
3886                         unsigned int crypt_len = roundup(ivsize, blocksize);
3887
3888                         req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
3889                         if (!req) {
3890                                 *error = "Could not allocate crypt request";
3891                                 r = -ENOMEM;
3892                                 goto bad;
3893                         }
3894
3895                         crypt_iv = kmalloc(ivsize, GFP_KERNEL);
3896                         if (!crypt_iv) {
3897                                 *error = "Could not allocate iv";
3898                                 r = -ENOMEM;
3899                                 goto bad;
3900                         }
3901
3902                         crypt_data = kmalloc(crypt_len, GFP_KERNEL);
3903                         if (!crypt_data) {
3904                                 *error = "Unable to allocate crypt data";
3905                                 r = -ENOMEM;
3906                                 goto bad;
3907                         }
3908
3909                         ic->journal_scatterlist = dm_integrity_alloc_journal_scatterlist(ic, ic->journal);
3910                         if (!ic->journal_scatterlist) {
3911                                 *error = "Unable to allocate sg list";
3912                                 r = -ENOMEM;
3913                                 goto bad;
3914                         }
3915                         ic->journal_io_scatterlist = dm_integrity_alloc_journal_scatterlist(ic, ic->journal_io);
3916                         if (!ic->journal_io_scatterlist) {
3917                                 *error = "Unable to allocate sg list";
3918                                 r = -ENOMEM;
3919                                 goto bad;
3920                         }
3921                         ic->sk_requests = kvmalloc_array(ic->journal_sections,
3922                                                          sizeof(struct skcipher_request *),
3923                                                          GFP_KERNEL | __GFP_ZERO);
3924                         if (!ic->sk_requests) {
3925                                 *error = "Unable to allocate sk requests";
3926                                 r = -ENOMEM;
3927                                 goto bad;
3928                         }
3929                         for (i = 0; i < ic->journal_sections; i++) {
3930                                 struct scatterlist sg;
3931                                 struct skcipher_request *section_req;
3932                                 __le32 section_le = cpu_to_le32(i);
3933
3934                                 memset(crypt_iv, 0x00, ivsize);
3935                                 memset(crypt_data, 0x00, crypt_len);
3936                                 memcpy(crypt_data, &section_le, min_t(size_t, crypt_len, sizeof(section_le)));
3937
3938                                 sg_init_one(&sg, crypt_data, crypt_len);
3939                                 skcipher_request_set_crypt(req, &sg, &sg, crypt_len, crypt_iv);
3940                                 init_completion(&comp.comp);
3941                                 comp.in_flight = (atomic_t)ATOMIC_INIT(1);
3942                                 if (do_crypt(true, req, &comp))
3943                                         wait_for_completion(&comp.comp);
3944
3945                                 r = dm_integrity_failed(ic);
3946                                 if (r) {
3947                                         *error = "Unable to generate iv";
3948                                         goto bad;
3949                                 }
3950
3951                                 section_req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
3952                                 if (!section_req) {
3953                                         *error = "Unable to allocate crypt request";
3954                                         r = -ENOMEM;
3955                                         goto bad;
3956                                 }
3957                                 section_req->iv = kmalloc_array(ivsize, 2,
3958                                                                 GFP_KERNEL);
3959                                 if (!section_req->iv) {
3960                                         skcipher_request_free(section_req);
3961                                         *error = "Unable to allocate iv";
3962                                         r = -ENOMEM;
3963                                         goto bad;
3964                                 }
3965                                 memcpy(section_req->iv + ivsize, crypt_data, ivsize);
3966                                 section_req->cryptlen = (size_t)ic->journal_section_sectors << SECTOR_SHIFT;
3967                                 ic->sk_requests[i] = section_req;
3968                                 DEBUG_bytes(crypt_data, ivsize, "iv(%u)", i);
3969                         }
3970                 }
3971         }
3972
3973         for (i = 0; i < N_COMMIT_IDS; i++) {
3974                 unsigned int j;
3975
3976 retest_commit_id:
3977                 for (j = 0; j < i; j++) {
3978                         if (ic->commit_ids[j] == ic->commit_ids[i]) {
3979                                 ic->commit_ids[i] = cpu_to_le64(le64_to_cpu(ic->commit_ids[i]) + 1);
3980                                 goto retest_commit_id;
3981                         }
3982                 }
3983                 DEBUG_print("commit id %u: %016llx\n", i, ic->commit_ids[i]);
3984         }
3985
3986         journal_tree_size = (__u64)ic->journal_entries * sizeof(struct journal_node);
3987         if (journal_tree_size > ULONG_MAX) {
3988                 *error = "Journal doesn't fit into memory";
3989                 r = -ENOMEM;
3990                 goto bad;
3991         }
3992         ic->journal_tree = kvmalloc(journal_tree_size, GFP_KERNEL);
3993         if (!ic->journal_tree) {
3994                 *error = "Could not allocate memory for journal tree";
3995                 r = -ENOMEM;
3996         }
3997 bad:
3998         kfree(crypt_data);
3999         kfree(crypt_iv);
4000         skcipher_request_free(req);
4001
4002         return r;
4003 }
4004
4005 /*
4006  * Construct a integrity mapping
4007  *
4008  * Arguments:
4009  *      device
4010  *      offset from the start of the device
4011  *      tag size
4012  *      D - direct writes, J - journal writes, B - bitmap mode, R - recovery mode
4013  *      number of optional arguments
4014  *      optional arguments:
4015  *              journal_sectors
4016  *              interleave_sectors
4017  *              buffer_sectors
4018  *              journal_watermark
4019  *              commit_time
4020  *              meta_device
4021  *              block_size
4022  *              sectors_per_bit
4023  *              bitmap_flush_interval
4024  *              internal_hash
4025  *              journal_crypt
4026  *              journal_mac
4027  *              recalculate
4028  */
4029 static int dm_integrity_ctr(struct dm_target *ti, unsigned int argc, char **argv)
4030 {
4031         struct dm_integrity_c *ic;
4032         char dummy;
4033         int r;
4034         unsigned int extra_args;
4035         struct dm_arg_set as;
4036         static const struct dm_arg _args[] = {
4037                 {0, 18, "Invalid number of feature args"},
4038         };
4039         unsigned int journal_sectors, interleave_sectors, buffer_sectors, journal_watermark, sync_msec;
4040         bool should_write_sb;
4041         __u64 threshold;
4042         unsigned long long start;
4043         __s8 log2_sectors_per_bitmap_bit = -1;
4044         __s8 log2_blocks_per_bitmap_bit;
4045         __u64 bits_in_journal;
4046         __u64 n_bitmap_bits;
4047
4048 #define DIRECT_ARGUMENTS        4
4049
4050         if (argc <= DIRECT_ARGUMENTS) {
4051                 ti->error = "Invalid argument count";
4052                 return -EINVAL;
4053         }
4054
4055         ic = kzalloc(sizeof(struct dm_integrity_c), GFP_KERNEL);
4056         if (!ic) {
4057                 ti->error = "Cannot allocate integrity context";
4058                 return -ENOMEM;
4059         }
4060         ti->private = ic;
4061         ti->per_io_data_size = sizeof(struct dm_integrity_io);
4062         ic->ti = ti;
4063
4064         ic->in_progress = RB_ROOT;
4065         INIT_LIST_HEAD(&ic->wait_list);
4066         init_waitqueue_head(&ic->endio_wait);
4067         bio_list_init(&ic->flush_bio_list);
4068         init_waitqueue_head(&ic->copy_to_journal_wait);
4069         init_completion(&ic->crypto_backoff);
4070         atomic64_set(&ic->number_of_mismatches, 0);
4071         ic->bitmap_flush_interval = BITMAP_FLUSH_INTERVAL;
4072
4073         r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &ic->dev);
4074         if (r) {
4075                 ti->error = "Device lookup failed";
4076                 goto bad;
4077         }
4078
4079         if (sscanf(argv[1], "%llu%c", &start, &dummy) != 1 || start != (sector_t)start) {
4080                 ti->error = "Invalid starting offset";
4081                 r = -EINVAL;
4082                 goto bad;
4083         }
4084         ic->start = start;
4085
4086         if (strcmp(argv[2], "-")) {
4087                 if (sscanf(argv[2], "%u%c", &ic->tag_size, &dummy) != 1 || !ic->tag_size) {
4088                         ti->error = "Invalid tag size";
4089                         r = -EINVAL;
4090                         goto bad;
4091                 }
4092         }
4093
4094         if (!strcmp(argv[3], "J") || !strcmp(argv[3], "B") ||
4095             !strcmp(argv[3], "D") || !strcmp(argv[3], "R")) {
4096                 ic->mode = argv[3][0];
4097         } else {
4098                 ti->error = "Invalid mode (expecting J, B, D, R)";
4099                 r = -EINVAL;
4100                 goto bad;
4101         }
4102
4103         journal_sectors = 0;
4104         interleave_sectors = DEFAULT_INTERLEAVE_SECTORS;
4105         buffer_sectors = DEFAULT_BUFFER_SECTORS;
4106         journal_watermark = DEFAULT_JOURNAL_WATERMARK;
4107         sync_msec = DEFAULT_SYNC_MSEC;
4108         ic->sectors_per_block = 1;
4109
4110         as.argc = argc - DIRECT_ARGUMENTS;
4111         as.argv = argv + DIRECT_ARGUMENTS;
4112         r = dm_read_arg_group(_args, &as, &extra_args, &ti->error);
4113         if (r)
4114                 goto bad;
4115
4116         while (extra_args--) {
4117                 const char *opt_string;
4118                 unsigned int val;
4119                 unsigned long long llval;
4120
4121                 opt_string = dm_shift_arg(&as);
4122                 if (!opt_string) {
4123                         r = -EINVAL;
4124                         ti->error = "Not enough feature arguments";
4125                         goto bad;
4126                 }
4127                 if (sscanf(opt_string, "journal_sectors:%u%c", &val, &dummy) == 1)
4128                         journal_sectors = val ? val : 1;
4129                 else if (sscanf(opt_string, "interleave_sectors:%u%c", &val, &dummy) == 1)
4130                         interleave_sectors = val;
4131                 else if (sscanf(opt_string, "buffer_sectors:%u%c", &val, &dummy) == 1)
4132                         buffer_sectors = val;
4133                 else if (sscanf(opt_string, "journal_watermark:%u%c", &val, &dummy) == 1 && val <= 100)
4134                         journal_watermark = val;
4135                 else if (sscanf(opt_string, "commit_time:%u%c", &val, &dummy) == 1)
4136                         sync_msec = val;
4137                 else if (!strncmp(opt_string, "meta_device:", strlen("meta_device:"))) {
4138                         if (ic->meta_dev) {
4139                                 dm_put_device(ti, ic->meta_dev);
4140                                 ic->meta_dev = NULL;
4141                         }
4142                         r = dm_get_device(ti, strchr(opt_string, ':') + 1,
4143                                           dm_table_get_mode(ti->table), &ic->meta_dev);
4144                         if (r) {
4145                                 ti->error = "Device lookup failed";
4146                                 goto bad;
4147                         }
4148                 } else if (sscanf(opt_string, "block_size:%u%c", &val, &dummy) == 1) {
4149                         if (val < 1 << SECTOR_SHIFT ||
4150                             val > MAX_SECTORS_PER_BLOCK << SECTOR_SHIFT ||
4151                             (val & (val - 1))) {
4152                                 r = -EINVAL;
4153                                 ti->error = "Invalid block_size argument";
4154                                 goto bad;
4155                         }
4156                         ic->sectors_per_block = val >> SECTOR_SHIFT;
4157                 } else if (sscanf(opt_string, "sectors_per_bit:%llu%c", &llval, &dummy) == 1) {
4158                         log2_sectors_per_bitmap_bit = !llval ? 0 : __ilog2_u64(llval);
4159                 } else if (sscanf(opt_string, "bitmap_flush_interval:%u%c", &val, &dummy) == 1) {
4160                         if (val >= (uint64_t)UINT_MAX * 1000 / HZ) {
4161                                 r = -EINVAL;
4162                                 ti->error = "Invalid bitmap_flush_interval argument";
4163                                 goto bad;
4164                         }
4165                         ic->bitmap_flush_interval = msecs_to_jiffies(val);
4166                 } else if (!strncmp(opt_string, "internal_hash:", strlen("internal_hash:"))) {
4167                         r = get_alg_and_key(opt_string, &ic->internal_hash_alg, &ti->error,
4168                                             "Invalid internal_hash argument");
4169                         if (r)
4170                                 goto bad;
4171                 } else if (!strncmp(opt_string, "journal_crypt:", strlen("journal_crypt:"))) {
4172                         r = get_alg_and_key(opt_string, &ic->journal_crypt_alg, &ti->error,
4173                                             "Invalid journal_crypt argument");
4174                         if (r)
4175                                 goto bad;
4176                 } else if (!strncmp(opt_string, "journal_mac:", strlen("journal_mac:"))) {
4177                         r = get_alg_and_key(opt_string, &ic->journal_mac_alg, &ti->error,
4178                                             "Invalid journal_mac argument");
4179                         if (r)
4180                                 goto bad;
4181                 } else if (!strcmp(opt_string, "recalculate")) {
4182                         ic->recalculate_flag = true;
4183                 } else if (!strcmp(opt_string, "reset_recalculate")) {
4184                         ic->recalculate_flag = true;
4185                         ic->reset_recalculate_flag = true;
4186                 } else if (!strcmp(opt_string, "allow_discards")) {
4187                         ic->discard = true;
4188                 } else if (!strcmp(opt_string, "fix_padding")) {
4189                         ic->fix_padding = true;
4190                 } else if (!strcmp(opt_string, "fix_hmac")) {
4191                         ic->fix_hmac = true;
4192                 } else if (!strcmp(opt_string, "legacy_recalculate")) {
4193                         ic->legacy_recalculate = true;
4194                 } else {
4195                         r = -EINVAL;
4196                         ti->error = "Invalid argument";
4197                         goto bad;
4198                 }
4199         }
4200
4201         ic->data_device_sectors = bdev_nr_sectors(ic->dev->bdev);
4202         if (!ic->meta_dev)
4203                 ic->meta_device_sectors = ic->data_device_sectors;
4204         else
4205                 ic->meta_device_sectors = bdev_nr_sectors(ic->meta_dev->bdev);
4206
4207         if (!journal_sectors) {
4208                 journal_sectors = min((sector_t)DEFAULT_MAX_JOURNAL_SECTORS,
4209                                       ic->data_device_sectors >> DEFAULT_JOURNAL_SIZE_FACTOR);
4210         }
4211
4212         if (!buffer_sectors)
4213                 buffer_sectors = 1;
4214         ic->log2_buffer_sectors = min((int)__fls(buffer_sectors), 31 - SECTOR_SHIFT);
4215
4216         r = get_mac(&ic->internal_hash, &ic->internal_hash_alg, &ti->error,
4217                     "Invalid internal hash", "Error setting internal hash key");
4218         if (r)
4219                 goto bad;
4220
4221         r = get_mac(&ic->journal_mac, &ic->journal_mac_alg, &ti->error,
4222                     "Invalid journal mac", "Error setting journal mac key");
4223         if (r)
4224                 goto bad;
4225
4226         if (!ic->tag_size) {
4227                 if (!ic->internal_hash) {
4228                         ti->error = "Unknown tag size";
4229                         r = -EINVAL;
4230                         goto bad;
4231                 }
4232                 ic->tag_size = crypto_shash_digestsize(ic->internal_hash);
4233         }
4234         if (ic->tag_size > MAX_TAG_SIZE) {
4235                 ti->error = "Too big tag size";
4236                 r = -EINVAL;
4237                 goto bad;
4238         }
4239         if (!(ic->tag_size & (ic->tag_size - 1)))
4240                 ic->log2_tag_size = __ffs(ic->tag_size);
4241         else
4242                 ic->log2_tag_size = -1;
4243
4244         if (ic->mode == 'B' && !ic->internal_hash) {
4245                 r = -EINVAL;
4246                 ti->error = "Bitmap mode can be only used with internal hash";
4247                 goto bad;
4248         }
4249
4250         if (ic->discard && !ic->internal_hash) {
4251                 r = -EINVAL;
4252                 ti->error = "Discard can be only used with internal hash";
4253                 goto bad;
4254         }
4255
4256         ic->autocommit_jiffies = msecs_to_jiffies(sync_msec);
4257         ic->autocommit_msec = sync_msec;
4258         timer_setup(&ic->autocommit_timer, autocommit_fn, 0);
4259
4260         ic->io = dm_io_client_create();
4261         if (IS_ERR(ic->io)) {
4262                 r = PTR_ERR(ic->io);
4263                 ic->io = NULL;
4264                 ti->error = "Cannot allocate dm io";
4265                 goto bad;
4266         }
4267
4268         r = mempool_init_slab_pool(&ic->journal_io_mempool, JOURNAL_IO_MEMPOOL, journal_io_cache);
4269         if (r) {
4270                 ti->error = "Cannot allocate mempool";
4271                 goto bad;
4272         }
4273
4274         ic->metadata_wq = alloc_workqueue("dm-integrity-metadata",
4275                                           WQ_MEM_RECLAIM, METADATA_WORKQUEUE_MAX_ACTIVE);
4276         if (!ic->metadata_wq) {
4277                 ti->error = "Cannot allocate workqueue";
4278                 r = -ENOMEM;
4279                 goto bad;
4280         }
4281
4282         /*
4283          * If this workqueue weren't ordered, it would cause bio reordering
4284          * and reduced performance.
4285          */
4286         ic->wait_wq = alloc_ordered_workqueue("dm-integrity-wait", WQ_MEM_RECLAIM);
4287         if (!ic->wait_wq) {
4288                 ti->error = "Cannot allocate workqueue";
4289                 r = -ENOMEM;
4290                 goto bad;
4291         }
4292
4293         ic->offload_wq = alloc_workqueue("dm-integrity-offload", WQ_MEM_RECLAIM,
4294                                           METADATA_WORKQUEUE_MAX_ACTIVE);
4295         if (!ic->offload_wq) {
4296                 ti->error = "Cannot allocate workqueue";
4297                 r = -ENOMEM;
4298                 goto bad;
4299         }
4300
4301         ic->commit_wq = alloc_workqueue("dm-integrity-commit", WQ_MEM_RECLAIM, 1);
4302         if (!ic->commit_wq) {
4303                 ti->error = "Cannot allocate workqueue";
4304                 r = -ENOMEM;
4305                 goto bad;
4306         }
4307         INIT_WORK(&ic->commit_work, integrity_commit);
4308
4309         if (ic->mode == 'J' || ic->mode == 'B') {
4310                 ic->writer_wq = alloc_workqueue("dm-integrity-writer", WQ_MEM_RECLAIM, 1);
4311                 if (!ic->writer_wq) {
4312                         ti->error = "Cannot allocate workqueue";
4313                         r = -ENOMEM;
4314                         goto bad;
4315                 }
4316                 INIT_WORK(&ic->writer_work, integrity_writer);
4317         }
4318
4319         ic->sb = alloc_pages_exact(SB_SECTORS << SECTOR_SHIFT, GFP_KERNEL);
4320         if (!ic->sb) {
4321                 r = -ENOMEM;
4322                 ti->error = "Cannot allocate superblock area";
4323                 goto bad;
4324         }
4325
4326         r = sync_rw_sb(ic, REQ_OP_READ);
4327         if (r) {
4328                 ti->error = "Error reading superblock";
4329                 goto bad;
4330         }
4331         should_write_sb = false;
4332         if (memcmp(ic->sb->magic, SB_MAGIC, 8)) {
4333                 if (ic->mode != 'R') {
4334                         if (memchr_inv(ic->sb, 0, SB_SECTORS << SECTOR_SHIFT)) {
4335                                 r = -EINVAL;
4336                                 ti->error = "The device is not initialized";
4337                                 goto bad;
4338                         }
4339                 }
4340
4341                 r = initialize_superblock(ic, journal_sectors, interleave_sectors);
4342                 if (r) {
4343                         ti->error = "Could not initialize superblock";
4344                         goto bad;
4345                 }
4346                 if (ic->mode != 'R')
4347                         should_write_sb = true;
4348         }
4349
4350         if (!ic->sb->version || ic->sb->version > SB_VERSION_5) {
4351                 r = -EINVAL;
4352                 ti->error = "Unknown version";
4353                 goto bad;
4354         }
4355         if (le16_to_cpu(ic->sb->integrity_tag_size) != ic->tag_size) {
4356                 r = -EINVAL;
4357                 ti->error = "Tag size doesn't match the information in superblock";
4358                 goto bad;
4359         }
4360         if (ic->sb->log2_sectors_per_block != __ffs(ic->sectors_per_block)) {
4361                 r = -EINVAL;
4362                 ti->error = "Block size doesn't match the information in superblock";
4363                 goto bad;
4364         }
4365         if (!le32_to_cpu(ic->sb->journal_sections)) {
4366                 r = -EINVAL;
4367                 ti->error = "Corrupted superblock, journal_sections is 0";
4368                 goto bad;
4369         }
4370         /* make sure that ti->max_io_len doesn't overflow */
4371         if (!ic->meta_dev) {
4372                 if (ic->sb->log2_interleave_sectors < MIN_LOG2_INTERLEAVE_SECTORS ||
4373                     ic->sb->log2_interleave_sectors > MAX_LOG2_INTERLEAVE_SECTORS) {
4374                         r = -EINVAL;
4375                         ti->error = "Invalid interleave_sectors in the superblock";
4376                         goto bad;
4377                 }
4378         } else {
4379                 if (ic->sb->log2_interleave_sectors) {
4380                         r = -EINVAL;
4381                         ti->error = "Invalid interleave_sectors in the superblock";
4382                         goto bad;
4383                 }
4384         }
4385         if (!!(ic->sb->flags & cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC)) != !!ic->journal_mac_alg.alg_string) {
4386                 r = -EINVAL;
4387                 ti->error = "Journal mac mismatch";
4388                 goto bad;
4389         }
4390
4391         get_provided_data_sectors(ic);
4392         if (!ic->provided_data_sectors) {
4393                 r = -EINVAL;
4394                 ti->error = "The device is too small";
4395                 goto bad;
4396         }
4397
4398 try_smaller_buffer:
4399         r = calculate_device_limits(ic);
4400         if (r) {
4401                 if (ic->meta_dev) {
4402                         if (ic->log2_buffer_sectors > 3) {
4403                                 ic->log2_buffer_sectors--;
4404                                 goto try_smaller_buffer;
4405                         }
4406                 }
4407                 ti->error = "The device is too small";
4408                 goto bad;
4409         }
4410
4411         if (log2_sectors_per_bitmap_bit < 0)
4412                 log2_sectors_per_bitmap_bit = __fls(DEFAULT_SECTORS_PER_BITMAP_BIT);
4413         if (log2_sectors_per_bitmap_bit < ic->sb->log2_sectors_per_block)
4414                 log2_sectors_per_bitmap_bit = ic->sb->log2_sectors_per_block;
4415
4416         bits_in_journal = ((__u64)ic->journal_section_sectors * ic->journal_sections) << (SECTOR_SHIFT + 3);
4417         if (bits_in_journal > UINT_MAX)
4418                 bits_in_journal = UINT_MAX;
4419         while (bits_in_journal < (ic->provided_data_sectors + ((sector_t)1 << log2_sectors_per_bitmap_bit) - 1) >> log2_sectors_per_bitmap_bit)
4420                 log2_sectors_per_bitmap_bit++;
4421
4422         log2_blocks_per_bitmap_bit = log2_sectors_per_bitmap_bit - ic->sb->log2_sectors_per_block;
4423         ic->log2_blocks_per_bitmap_bit = log2_blocks_per_bitmap_bit;
4424         if (should_write_sb)
4425                 ic->sb->log2_blocks_per_bitmap_bit = log2_blocks_per_bitmap_bit;
4426
4427         n_bitmap_bits = ((ic->provided_data_sectors >> ic->sb->log2_sectors_per_block)
4428                                 + (((sector_t)1 << log2_blocks_per_bitmap_bit) - 1)) >> log2_blocks_per_bitmap_bit;
4429         ic->n_bitmap_blocks = DIV_ROUND_UP(n_bitmap_bits, BITMAP_BLOCK_SIZE * 8);
4430
4431         if (!ic->meta_dev)
4432                 ic->log2_buffer_sectors = min(ic->log2_buffer_sectors, (__u8)__ffs(ic->metadata_run));
4433
4434         if (ti->len > ic->provided_data_sectors) {
4435                 r = -EINVAL;
4436                 ti->error = "Not enough provided sectors for requested mapping size";
4437                 goto bad;
4438         }
4439
4440
4441         threshold = (__u64)ic->journal_entries * (100 - journal_watermark);
4442         threshold += 50;
4443         do_div(threshold, 100);
4444         ic->free_sectors_threshold = threshold;
4445
4446         DEBUG_print("initialized:\n");
4447         DEBUG_print("   integrity_tag_size %u\n", le16_to_cpu(ic->sb->integrity_tag_size));
4448         DEBUG_print("   journal_entry_size %u\n", ic->journal_entry_size);
4449         DEBUG_print("   journal_entries_per_sector %u\n", ic->journal_entries_per_sector);
4450         DEBUG_print("   journal_section_entries %u\n", ic->journal_section_entries);
4451         DEBUG_print("   journal_section_sectors %u\n", ic->journal_section_sectors);
4452         DEBUG_print("   journal_sections %u\n", (unsigned int)le32_to_cpu(ic->sb->journal_sections));
4453         DEBUG_print("   journal_entries %u\n", ic->journal_entries);
4454         DEBUG_print("   log2_interleave_sectors %d\n", ic->sb->log2_interleave_sectors);
4455         DEBUG_print("   data_device_sectors 0x%llx\n", bdev_nr_sectors(ic->dev->bdev));
4456         DEBUG_print("   initial_sectors 0x%x\n", ic->initial_sectors);
4457         DEBUG_print("   metadata_run 0x%x\n", ic->metadata_run);
4458         DEBUG_print("   log2_metadata_run %d\n", ic->log2_metadata_run);
4459         DEBUG_print("   provided_data_sectors 0x%llx (%llu)\n", ic->provided_data_sectors, ic->provided_data_sectors);
4460         DEBUG_print("   log2_buffer_sectors %u\n", ic->log2_buffer_sectors);
4461         DEBUG_print("   bits_in_journal %llu\n", bits_in_journal);
4462
4463         if (ic->recalculate_flag && !(ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))) {
4464                 ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
4465                 ic->sb->recalc_sector = cpu_to_le64(0);
4466         }
4467
4468         if (ic->internal_hash) {
4469                 ic->recalc_wq = alloc_workqueue("dm-integrity-recalc", WQ_MEM_RECLAIM, 1);
4470                 if (!ic->recalc_wq) {
4471                         ti->error = "Cannot allocate workqueue";
4472                         r = -ENOMEM;
4473                         goto bad;
4474                 }
4475                 INIT_WORK(&ic->recalc_work, integrity_recalc);
4476         } else {
4477                 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
4478                         ti->error = "Recalculate can only be specified with internal_hash";
4479                         r = -EINVAL;
4480                         goto bad;
4481                 }
4482         }
4483
4484         if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING) &&
4485             le64_to_cpu(ic->sb->recalc_sector) < ic->provided_data_sectors &&
4486             dm_integrity_disable_recalculate(ic)) {
4487                 ti->error = "Recalculating with HMAC is disabled for security reasons - if you really need it, use the argument \"legacy_recalculate\"";
4488                 r = -EOPNOTSUPP;
4489                 goto bad;
4490         }
4491
4492         ic->bufio = dm_bufio_client_create(ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev,
4493                         1U << (SECTOR_SHIFT + ic->log2_buffer_sectors), 1, 0, NULL, NULL, 0);
4494         if (IS_ERR(ic->bufio)) {
4495                 r = PTR_ERR(ic->bufio);
4496                 ti->error = "Cannot initialize dm-bufio";
4497                 ic->bufio = NULL;
4498                 goto bad;
4499         }
4500         dm_bufio_set_sector_offset(ic->bufio, ic->start + ic->initial_sectors);
4501
4502         if (ic->mode != 'R') {
4503                 r = create_journal(ic, &ti->error);
4504                 if (r)
4505                         goto bad;
4506
4507         }
4508
4509         if (ic->mode == 'B') {
4510                 unsigned int i;
4511                 unsigned int n_bitmap_pages = DIV_ROUND_UP(ic->n_bitmap_blocks, PAGE_SIZE / BITMAP_BLOCK_SIZE);
4512
4513                 ic->recalc_bitmap = dm_integrity_alloc_page_list(n_bitmap_pages);
4514                 if (!ic->recalc_bitmap) {
4515                         r = -ENOMEM;
4516                         goto bad;
4517                 }
4518                 ic->may_write_bitmap = dm_integrity_alloc_page_list(n_bitmap_pages);
4519                 if (!ic->may_write_bitmap) {
4520                         r = -ENOMEM;
4521                         goto bad;
4522                 }
4523                 ic->bbs = kvmalloc_array(ic->n_bitmap_blocks, sizeof(struct bitmap_block_status), GFP_KERNEL);
4524                 if (!ic->bbs) {
4525                         r = -ENOMEM;
4526                         goto bad;
4527                 }
4528                 INIT_DELAYED_WORK(&ic->bitmap_flush_work, bitmap_flush_work);
4529                 for (i = 0; i < ic->n_bitmap_blocks; i++) {
4530                         struct bitmap_block_status *bbs = &ic->bbs[i];
4531                         unsigned int sector, pl_index, pl_offset;
4532
4533                         INIT_WORK(&bbs->work, bitmap_block_work);
4534                         bbs->ic = ic;
4535                         bbs->idx = i;
4536                         bio_list_init(&bbs->bio_queue);
4537                         spin_lock_init(&bbs->bio_queue_lock);
4538
4539                         sector = i * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT);
4540                         pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
4541                         pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
4542
4543                         bbs->bitmap = lowmem_page_address(ic->journal[pl_index].page) + pl_offset;
4544                 }
4545         }
4546
4547         if (should_write_sb) {
4548                 init_journal(ic, 0, ic->journal_sections, 0);
4549                 r = dm_integrity_failed(ic);
4550                 if (unlikely(r)) {
4551                         ti->error = "Error initializing journal";
4552                         goto bad;
4553                 }
4554                 r = sync_rw_sb(ic, REQ_OP_WRITE | REQ_FUA);
4555                 if (r) {
4556                         ti->error = "Error initializing superblock";
4557                         goto bad;
4558                 }
4559                 ic->just_formatted = true;
4560         }
4561
4562         if (!ic->meta_dev) {
4563                 r = dm_set_target_max_io_len(ti, 1U << ic->sb->log2_interleave_sectors);
4564                 if (r)
4565                         goto bad;
4566         }
4567         if (ic->mode == 'B') {
4568                 unsigned int max_io_len;
4569
4570                 max_io_len = ((sector_t)ic->sectors_per_block << ic->log2_blocks_per_bitmap_bit) * (BITMAP_BLOCK_SIZE * 8);
4571                 if (!max_io_len)
4572                         max_io_len = 1U << 31;
4573                 DEBUG_print("max_io_len: old %u, new %u\n", ti->max_io_len, max_io_len);
4574                 if (!ti->max_io_len || ti->max_io_len > max_io_len) {
4575                         r = dm_set_target_max_io_len(ti, max_io_len);
4576                         if (r)
4577                                 goto bad;
4578                 }
4579         }
4580
4581         if (!ic->internal_hash)
4582                 dm_integrity_set(ti, ic);
4583
4584         ti->num_flush_bios = 1;
4585         ti->flush_supported = true;
4586         if (ic->discard)
4587                 ti->num_discard_bios = 1;
4588
4589         dm_audit_log_ctr(DM_MSG_PREFIX, ti, 1);
4590         return 0;
4591
4592 bad:
4593         dm_audit_log_ctr(DM_MSG_PREFIX, ti, 0);
4594         dm_integrity_dtr(ti);
4595         return r;
4596 }
4597
4598 static void dm_integrity_dtr(struct dm_target *ti)
4599 {
4600         struct dm_integrity_c *ic = ti->private;
4601
4602         BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
4603         BUG_ON(!list_empty(&ic->wait_list));
4604
4605         if (ic->mode == 'B')
4606                 cancel_delayed_work_sync(&ic->bitmap_flush_work);
4607         if (ic->metadata_wq)
4608                 destroy_workqueue(ic->metadata_wq);
4609         if (ic->wait_wq)
4610                 destroy_workqueue(ic->wait_wq);
4611         if (ic->offload_wq)
4612                 destroy_workqueue(ic->offload_wq);
4613         if (ic->commit_wq)
4614                 destroy_workqueue(ic->commit_wq);
4615         if (ic->writer_wq)
4616                 destroy_workqueue(ic->writer_wq);
4617         if (ic->recalc_wq)
4618                 destroy_workqueue(ic->recalc_wq);
4619         kvfree(ic->bbs);
4620         if (ic->bufio)
4621                 dm_bufio_client_destroy(ic->bufio);
4622         mempool_exit(&ic->journal_io_mempool);
4623         if (ic->io)
4624                 dm_io_client_destroy(ic->io);
4625         if (ic->dev)
4626                 dm_put_device(ti, ic->dev);
4627         if (ic->meta_dev)
4628                 dm_put_device(ti, ic->meta_dev);
4629         dm_integrity_free_page_list(ic->journal);
4630         dm_integrity_free_page_list(ic->journal_io);
4631         dm_integrity_free_page_list(ic->journal_xor);
4632         dm_integrity_free_page_list(ic->recalc_bitmap);
4633         dm_integrity_free_page_list(ic->may_write_bitmap);
4634         if (ic->journal_scatterlist)
4635                 dm_integrity_free_journal_scatterlist(ic, ic->journal_scatterlist);
4636         if (ic->journal_io_scatterlist)
4637                 dm_integrity_free_journal_scatterlist(ic, ic->journal_io_scatterlist);
4638         if (ic->sk_requests) {
4639                 unsigned int i;
4640
4641                 for (i = 0; i < ic->journal_sections; i++) {
4642                         struct skcipher_request *req;
4643
4644                         req = ic->sk_requests[i];
4645                         if (req) {
4646                                 kfree_sensitive(req->iv);
4647                                 skcipher_request_free(req);
4648                         }
4649                 }
4650                 kvfree(ic->sk_requests);
4651         }
4652         kvfree(ic->journal_tree);
4653         if (ic->sb)
4654                 free_pages_exact(ic->sb, SB_SECTORS << SECTOR_SHIFT);
4655
4656         if (ic->internal_hash)
4657                 crypto_free_shash(ic->internal_hash);
4658         free_alg(&ic->internal_hash_alg);
4659
4660         if (ic->journal_crypt)
4661                 crypto_free_skcipher(ic->journal_crypt);
4662         free_alg(&ic->journal_crypt_alg);
4663
4664         if (ic->journal_mac)
4665                 crypto_free_shash(ic->journal_mac);
4666         free_alg(&ic->journal_mac_alg);
4667
4668         kfree(ic);
4669         dm_audit_log_dtr(DM_MSG_PREFIX, ti, 1);
4670 }
4671
4672 static struct target_type integrity_target = {
4673         .name                   = "integrity",
4674         .version                = {1, 10, 0},
4675         .module                 = THIS_MODULE,
4676         .features               = DM_TARGET_SINGLETON | DM_TARGET_INTEGRITY,
4677         .ctr                    = dm_integrity_ctr,
4678         .dtr                    = dm_integrity_dtr,
4679         .map                    = dm_integrity_map,
4680         .postsuspend            = dm_integrity_postsuspend,
4681         .resume                 = dm_integrity_resume,
4682         .status                 = dm_integrity_status,
4683         .iterate_devices        = dm_integrity_iterate_devices,
4684         .io_hints               = dm_integrity_io_hints,
4685 };
4686
4687 static int __init dm_integrity_init(void)
4688 {
4689         int r;
4690
4691         journal_io_cache = kmem_cache_create("integrity_journal_io",
4692                                              sizeof(struct journal_io), 0, 0, NULL);
4693         if (!journal_io_cache) {
4694                 DMERR("can't allocate journal io cache");
4695                 return -ENOMEM;
4696         }
4697
4698         r = dm_register_target(&integrity_target);
4699         if (r < 0) {
4700                 kmem_cache_destroy(journal_io_cache);
4701                 return r;
4702         }
4703
4704         return 0;
4705 }
4706
4707 static void __exit dm_integrity_exit(void)
4708 {
4709         dm_unregister_target(&integrity_target);
4710         kmem_cache_destroy(journal_io_cache);
4711 }
4712
4713 module_init(dm_integrity_init);
4714 module_exit(dm_integrity_exit);
4715
4716 MODULE_AUTHOR("Milan Broz");
4717 MODULE_AUTHOR("Mikulas Patocka");
4718 MODULE_DESCRIPTION(DM_NAME " target for integrity tags extension");
4719 MODULE_LICENSE("GPL");