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