drivers/md/dm-integrity.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright (C) 2016-2017 Red Hat, Inc. All rights reserved.
   4  * Copyright (C) 2016-2017 Milan Broz
   5  * Copyright (C) 2016-2017 Mikulas Patocka
   6  *
   7  * This file is released under the GPL.
   8  */
   9
  10 #include "dm-bio-record.h"
  11
  12 #include <linux/compiler.h>
  13 #include <linux/module.h>
  14 #include <linux/device-mapper.h>
  15 #include <linux/dm-io.h>
  16 #include <linux/vmalloc.h>
  17 #include <linux/sort.h>
  18 #include <linux/rbtree.h>
  19 #include <linux/delay.h>
  20 #include <linux/random.h>
  21 #include <linux/reboot.h>
  22 #include <crypto/hash.h>
  23 #include <crypto/skcipher.h>
  24 #include <linux/async_tx.h>
  25 #include <linux/dm-bufio.h>
  26
  27 #include "dm-audit.h"
  28
  29 #define DM_MSG_PREFIX "integrity"
  30
  31 #define DEFAULT_INTERLEAVE_SECTORS      32768
  32 #define DEFAULT_JOURNAL_SIZE_FACTOR     7
  33 #define DEFAULT_SECTORS_PER_BITMAP_BIT  32768
  34 #define DEFAULT_BUFFER_SECTORS          128
  35 #define DEFAULT_JOURNAL_WATERMARK       50
  36 #define DEFAULT_SYNC_MSEC               10000
  37 #define DEFAULT_MAX_JOURNAL_SECTORS     (IS_ENABLED(CONFIG_64BIT) ? 131072 : 8192)
  38 #define MIN_LOG2_INTERLEAVE_SECTORS     3
  39 #define MAX_LOG2_INTERLEAVE_SECTORS     31
  40 #define METADATA_WORKQUEUE_MAX_ACTIVE   16
  41 #define RECALC_SECTORS                  (IS_ENABLED(CONFIG_64BIT) ? 32768 : 2048)
  42 #define RECALC_WRITE_SUPER              16
  43 #define BITMAP_BLOCK_SIZE               4096    /* don't change it */
  44 #define BITMAP_FLUSH_INTERVAL           (10 * HZ)
  45 #define DISCARD_FILLER                  0xf6
  46 #define SALT_SIZE                       16
  47
  48 /*
  49  * Warning - DEBUG_PRINT prints security-sensitive data to the log,
  50  * so it should not be enabled in the official kernel
  51  */
  52 //#define DEBUG_PRINT
  53 //#define INTERNAL_VERIFY
  54
  55 /*
  56  * On disk structures
  57  */
  58
  59 #define SB_MAGIC                        "integrt"
  60 #define SB_VERSION_1                    1
  61 #define SB_VERSION_2                    2
  62 #define SB_VERSION_3                    3
  63 #define SB_VERSION_4                    4
  64 #define SB_VERSION_5                    5
  65 #define SB_SECTORS                      8
  66 #define MAX_SECTORS_PER_BLOCK           8
  67
  68 struct superblock {
  69         __u8 magic[8];
  70         __u8 version;
  71         __u8 log2_interleave_sectors;
  72         __le16 integrity_tag_size;
  73         __le32 journal_sections;
  74         __le64 provided_data_sectors;   /* userspace uses this value */
  75         __le32 flags;
  76         __u8 log2_sectors_per_block;
  77         __u8 log2_blocks_per_bitmap_bit;
  78         __u8 pad[2];
  79         __le64 recalc_sector;
  80         __u8 pad2[8];
  81         __u8 salt[SALT_SIZE];
  82 };
  83
  84 #define SB_FLAG_HAVE_JOURNAL_MAC        0x1
  85 #define SB_FLAG_RECALCULATING           0x2
  86 #define SB_FLAG_DIRTY_BITMAP            0x4
  87 #define SB_FLAG_FIXED_PADDING           0x8
  88 #define SB_FLAG_FIXED_HMAC              0x10
  89
  90 #define JOURNAL_ENTRY_ROUNDUP           8
  91
  92 typedef __le64 commit_id_t;
  93 #define JOURNAL_MAC_PER_SECTOR          8
  94
  95 struct journal_entry {
  96         union {
  97                 struct {
  98                         __le32 sector_lo;
  99                         __le32 sector_hi;
 100                 } s;
 101                 __le64 sector;
 102         } u;
 103         commit_id_t last_bytes[];
 104         /* __u8 tag[0]; */
 105 };
 106
 107 #define journal_entry_tag(ic, je)               ((__u8 *)&(je)->last_bytes[(ic)->sectors_per_block])
 108
 109 #if BITS_PER_LONG == 64
 110 #define journal_entry_set_sector(je, x)         do { smp_wmb(); WRITE_ONCE((je)->u.sector, cpu_to_le64(x)); } while (0)
 111 #else
 112 #define journal_entry_set_sector(je, x)         do { (je)->u.s.sector_lo = cpu_to_le32(x); smp_wmb(); WRITE_ONCE((je)->u.s.sector_hi, cpu_to_le32((x) >> 32)); } while (0)
 113 #endif
 114 #define journal_entry_get_sector(je)            le64_to_cpu((je)->u.sector)
 115 #define journal_entry_is_unused(je)             ((je)->u.s.sector_hi == cpu_to_le32(-1))
 116 #define journal_entry_set_unused(je)            ((je)->u.s.sector_hi = cpu_to_le32(-1))
 117 #define journal_entry_is_inprogress(je)         ((je)->u.s.sector_hi == cpu_to_le32(-2))
 118 #define journal_entry_set_inprogress(je)        ((je)->u.s.sector_hi = cpu_to_le32(-2))
 119
 120 #define JOURNAL_BLOCK_SECTORS           8
 121 #define JOURNAL_SECTOR_DATA             ((1 << SECTOR_SHIFT) - sizeof(commit_id_t))
 122 #define JOURNAL_MAC_SIZE                (JOURNAL_MAC_PER_SECTOR * JOURNAL_BLOCK_SECTORS)
 123
 124 struct journal_sector {
 125         struct_group(sectors,
 126                 __u8 entries[JOURNAL_SECTOR_DATA - JOURNAL_MAC_PER_SECTOR];
 127                 __u8 mac[JOURNAL_MAC_PER_SECTOR];
 128         );
 129         commit_id_t commit_id;
 130 };
 131
 132 #define MAX_TAG_SIZE                    (JOURNAL_SECTOR_DATA - JOURNAL_MAC_PER_SECTOR - offsetof(struct journal_entry, last_bytes[MAX_SECTORS_PER_BLOCK]))
 133
 134 #define METADATA_PADDING_SECTORS        8
 135
 136 #define N_COMMIT_IDS                    4
 137
 138 static unsigned char prev_commit_seq(unsigned char seq)
 139 {
 140         return (seq + N_COMMIT_IDS - 1) % N_COMMIT_IDS;
 141 }
 142
 143 static unsigned char next_commit_seq(unsigned char seq)
 144 {
 145         return (seq + 1) % N_COMMIT_IDS;
 146 }
 147
 148 /*
 149  * In-memory structures
 150  */
 151
 152 struct journal_node {
 153         struct rb_node node;
 154         sector_t sector;
 155 };
 156
 157 struct alg_spec {
 158         char *alg_string;
 159         char *key_string;
 160         __u8 *key;
 161         unsigned int key_size;
 162 };
 163
 164 struct dm_integrity_c {
 165         struct dm_dev *dev;
 166         struct dm_dev *meta_dev;
 167         unsigned int tag_size;
 168         __s8 log2_tag_size;
 169         sector_t start;
 170         mempool_t journal_io_mempool;
 171         struct dm_io_client *io;
 172         struct dm_bufio_client *bufio;
 173         struct workqueue_struct *metadata_wq;
 174         struct superblock *sb;
 175         unsigned int journal_pages;
 176         unsigned int n_bitmap_blocks;
 177
 178         struct page_list *journal;
 179         struct page_list *journal_io;
 180         struct page_list *journal_xor;
 181         struct page_list *recalc_bitmap;
 182         struct page_list *may_write_bitmap;
 183         struct bitmap_block_status *bbs;
 184         unsigned int bitmap_flush_interval;
 185         int synchronous_mode;
 186         struct bio_list synchronous_bios;
 187         struct delayed_work bitmap_flush_work;
 188
 189         struct crypto_skcipher *journal_crypt;
 190         struct scatterlist **journal_scatterlist;
 191         struct scatterlist **journal_io_scatterlist;
 192         struct skcipher_request **sk_requests;
 193
 194         struct crypto_shash *journal_mac;
 195
 196         struct journal_node *journal_tree;
 197         struct rb_root journal_tree_root;
 198
 199         sector_t provided_data_sectors;
 200
 201         unsigned short journal_entry_size;
 202         unsigned char journal_entries_per_sector;
 203         unsigned char journal_section_entries;
 204         unsigned short journal_section_sectors;
 205         unsigned int journal_sections;
 206         unsigned int journal_entries;
 207         sector_t data_device_sectors;
 208         sector_t meta_device_sectors;
 209         unsigned int initial_sectors;
 210         unsigned int metadata_run;
 211         __s8 log2_metadata_run;
 212         __u8 log2_buffer_sectors;
 213         __u8 sectors_per_block;
 214         __u8 log2_blocks_per_bitmap_bit;
 215
 216         unsigned char mode;
 217
 218         int failed;
 219
 220         struct crypto_shash *internal_hash;
 221
 222         struct dm_target *ti;
 223
 224         /* these variables are locked with endio_wait.lock */
 225         struct rb_root in_progress;
 226         struct list_head wait_list;
 227         wait_queue_head_t endio_wait;
 228         struct workqueue_struct *wait_wq;
 229         struct workqueue_struct *offload_wq;
 230
 231         unsigned char commit_seq;
 232         commit_id_t commit_ids[N_COMMIT_IDS];
 233
 234         unsigned int committed_section;
 235         unsigned int n_committed_sections;
 236
 237         unsigned int uncommitted_section;
 238         unsigned int n_uncommitted_sections;
 239
 240         unsigned int free_section;
 241         unsigned char free_section_entry;
 242         unsigned int free_sectors;
 243
 244         unsigned int free_sectors_threshold;
 245
 246         struct workqueue_struct *commit_wq;
 247         struct work_struct commit_work;
 248
 249         struct workqueue_struct *writer_wq;
 250         struct work_struct writer_work;
 251
 252         struct workqueue_struct *recalc_wq;
 253         struct work_struct recalc_work;
 254
 255         struct bio_list flush_bio_list;
 256
 257         unsigned long autocommit_jiffies;
 258         struct timer_list autocommit_timer;
 259         unsigned int autocommit_msec;
 260
 261         wait_queue_head_t copy_to_journal_wait;
 262
 263         struct completion crypto_backoff;
 264
 265         bool wrote_to_journal;
 266         bool journal_uptodate;
 267         bool just_formatted;
 268         bool recalculate_flag;
 269         bool reset_recalculate_flag;
 270         bool discard;
 271         bool fix_padding;
 272         bool fix_hmac;
 273         bool legacy_recalculate;
 274
 275         struct alg_spec internal_hash_alg;
 276         struct alg_spec journal_crypt_alg;
 277         struct alg_spec journal_mac_alg;
 278
 279         atomic64_t number_of_mismatches;
 280
 281         mempool_t recheck_pool;
 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 mac_size = crypto_shash_digestsize(ic->journal_mac);
 499         __u8 *sb = (__u8 *)ic->sb;
 500         __u8 *mac = sb + (1 << SECTOR_SHIFT) - mac_size;
 501
 502         if (sizeof(struct superblock) + mac_size > 1 << SECTOR_SHIFT) {
 503                 dm_integrity_io_error(ic, "digest is too long", -EINVAL);
 504                 return -EINVAL;
 505         }
 506
 507         desc->tfm = ic->journal_mac;
 508
 509         if (likely(wr)) {
 510                 r = crypto_shash_digest(desc, sb, mac - sb, mac);
 511                 if (unlikely(r < 0)) {
 512                         dm_integrity_io_error(ic, "crypto_shash_digest", r);
 513                         return r;
 514                 }
 515         } else {
 516                 __u8 actual_mac[HASH_MAX_DIGESTSIZE];
 517
 518                 r = crypto_shash_digest(desc, sb, mac - sb, actual_mac);
 519                 if (unlikely(r < 0)) {
 520                         dm_integrity_io_error(ic, "crypto_shash_digest", r);
 521                         return r;
 522                 }
 523                 if (memcmp(mac, actual_mac, mac_size)) {
 524                         dm_integrity_io_error(ic, "superblock mac", -EILSEQ);
 525                         dm_audit_log_target(DM_MSG_PREFIX, "mac-superblock", ic->ti, 0);
 526                         return -EILSEQ;
 527                 }
 528         }
 529
 530         return 0;
 531 }
 532
 533 static int sync_rw_sb(struct dm_integrity_c *ic, blk_opf_t opf)
 534 {
 535         struct dm_io_request io_req;
 536         struct dm_io_region io_loc;
 537         const enum req_op op = opf & REQ_OP_MASK;
 538         int r;
 539
 540         io_req.bi_opf = opf;
 541         io_req.mem.type = DM_IO_KMEM;
 542         io_req.mem.ptr.addr = ic->sb;
 543         io_req.notify.fn = NULL;
 544         io_req.client = ic->io;
 545         io_loc.bdev = ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev;
 546         io_loc.sector = ic->start;
 547         io_loc.count = SB_SECTORS;
 548
 549         if (op == REQ_OP_WRITE) {
 550                 sb_set_version(ic);
 551                 if (ic->journal_mac && ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) {
 552                         r = sb_mac(ic, true);
 553                         if (unlikely(r))
 554                                 return r;
 555                 }
 556         }
 557
 558         r = dm_io(&io_req, 1, &io_loc, NULL);
 559         if (unlikely(r))
 560                 return r;
 561
 562         if (op == REQ_OP_READ) {
 563                 if (ic->mode != 'R' && ic->journal_mac && ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) {
 564                         r = sb_mac(ic, false);
 565                         if (unlikely(r))
 566                                 return r;
 567                 }
 568         }
 569
 570         return 0;
 571 }
 572
 573 #define BITMAP_OP_TEST_ALL_SET          0
 574 #define BITMAP_OP_TEST_ALL_CLEAR        1
 575 #define BITMAP_OP_SET                   2
 576 #define BITMAP_OP_CLEAR                 3
 577
 578 static bool block_bitmap_op(struct dm_integrity_c *ic, struct page_list *bitmap,
 579                             sector_t sector, sector_t n_sectors, int mode)
 580 {
 581         unsigned long bit, end_bit, this_end_bit, page, end_page;
 582         unsigned long *data;
 583
 584         if (unlikely(((sector | n_sectors) & ((1 << ic->sb->log2_sectors_per_block) - 1)) != 0)) {
 585                 DMCRIT("invalid bitmap access (%llx,%llx,%d,%d,%d)",
 586                         sector,
 587                         n_sectors,
 588                         ic->sb->log2_sectors_per_block,
 589                         ic->log2_blocks_per_bitmap_bit,
 590                         mode);
 591                 BUG();
 592         }
 593
 594         if (unlikely(!n_sectors))
 595                 return true;
 596
 597         bit = sector >> (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
 598         end_bit = (sector + n_sectors - 1) >>
 599                 (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
 600
 601         page = bit / (PAGE_SIZE * 8);
 602         bit %= PAGE_SIZE * 8;
 603
 604         end_page = end_bit / (PAGE_SIZE * 8);
 605         end_bit %= PAGE_SIZE * 8;
 606
 607 repeat:
 608         if (page < end_page)
 609                 this_end_bit = PAGE_SIZE * 8 - 1;
 610         else
 611                 this_end_bit = end_bit;
 612
 613         data = lowmem_page_address(bitmap[page].page);
 614
 615         if (mode == BITMAP_OP_TEST_ALL_SET) {
 616                 while (bit <= this_end_bit) {
 617                         if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
 618                                 do {
 619                                         if (data[bit / BITS_PER_LONG] != -1)
 620                                                 return false;
 621                                         bit += BITS_PER_LONG;
 622                                 } while (this_end_bit >= bit + BITS_PER_LONG - 1);
 623                                 continue;
 624                         }
 625                         if (!test_bit(bit, data))
 626                                 return false;
 627                         bit++;
 628                 }
 629         } else if (mode == BITMAP_OP_TEST_ALL_CLEAR) {
 630                 while (bit <= this_end_bit) {
 631                         if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
 632                                 do {
 633                                         if (data[bit / BITS_PER_LONG] != 0)
 634                                                 return false;
 635                                         bit += BITS_PER_LONG;
 636                                 } while (this_end_bit >= bit + BITS_PER_LONG - 1);
 637                                 continue;
 638                         }
 639                         if (test_bit(bit, data))
 640                                 return false;
 641                         bit++;
 642                 }
 643         } else if (mode == BITMAP_OP_SET) {
 644                 while (bit <= this_end_bit) {
 645                         if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
 646                                 do {
 647                                         data[bit / BITS_PER_LONG] = -1;
 648                                         bit += BITS_PER_LONG;
 649                                 } while (this_end_bit >= bit + BITS_PER_LONG - 1);
 650                                 continue;
 651                         }
 652                         __set_bit(bit, data);
 653                         bit++;
 654                 }
 655         } else if (mode == BITMAP_OP_CLEAR) {
 656                 if (!bit && this_end_bit == PAGE_SIZE * 8 - 1)
 657                         clear_page(data);
 658                 else {
 659                         while (bit <= this_end_bit) {
 660                                 if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
 661                                         do {
 662                                                 data[bit / BITS_PER_LONG] = 0;
 663                                                 bit += BITS_PER_LONG;
 664                                         } while (this_end_bit >= bit + BITS_PER_LONG - 1);
 665                                         continue;
 666                                 }
 667                                 __clear_bit(bit, data);
 668                                 bit++;
 669                         }
 670                 }
 671         } else {
 672                 BUG();
 673         }
 674
 675         if (unlikely(page < end_page)) {
 676                 bit = 0;
 677                 page++;
 678                 goto repeat;
 679         }
 680
 681         return true;
 682 }
 683
 684 static void block_bitmap_copy(struct dm_integrity_c *ic, struct page_list *dst, struct page_list *src)
 685 {
 686         unsigned int n_bitmap_pages = DIV_ROUND_UP(ic->n_bitmap_blocks, PAGE_SIZE / BITMAP_BLOCK_SIZE);
 687         unsigned int i;
 688
 689         for (i = 0; i < n_bitmap_pages; i++) {
 690                 unsigned long *dst_data = lowmem_page_address(dst[i].page);
 691                 unsigned long *src_data = lowmem_page_address(src[i].page);
 692
 693                 copy_page(dst_data, src_data);
 694         }
 695 }
 696
 697 static struct bitmap_block_status *sector_to_bitmap_block(struct dm_integrity_c *ic, sector_t sector)
 698 {
 699         unsigned int bit = sector >> (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
 700         unsigned int bitmap_block = bit / (BITMAP_BLOCK_SIZE * 8);
 701
 702         BUG_ON(bitmap_block >= ic->n_bitmap_blocks);
 703         return &ic->bbs[bitmap_block];
 704 }
 705
 706 static void access_journal_check(struct dm_integrity_c *ic, unsigned int section, unsigned int offset,
 707                                  bool e, const char *function)
 708 {
 709 #if defined(CONFIG_DM_DEBUG) || defined(INTERNAL_VERIFY)
 710         unsigned int limit = e ? ic->journal_section_entries : ic->journal_section_sectors;
 711
 712         if (unlikely(section >= ic->journal_sections) ||
 713             unlikely(offset >= limit)) {
 714                 DMCRIT("%s: invalid access at (%u,%u), limit (%u,%u)",
 715                        function, section, offset, ic->journal_sections, limit);
 716                 BUG();
 717         }
 718 #endif
 719 }
 720
 721 static void page_list_location(struct dm_integrity_c *ic, unsigned int section, unsigned int offset,
 722                                unsigned int *pl_index, unsigned int *pl_offset)
 723 {
 724         unsigned int sector;
 725
 726         access_journal_check(ic, section, offset, false, "page_list_location");
 727
 728         sector = section * ic->journal_section_sectors + offset;
 729
 730         *pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
 731         *pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
 732 }
 733
 734 static struct journal_sector *access_page_list(struct dm_integrity_c *ic, struct page_list *pl,
 735                                                unsigned int section, unsigned int offset, unsigned int *n_sectors)
 736 {
 737         unsigned int pl_index, pl_offset;
 738         char *va;
 739
 740         page_list_location(ic, section, offset, &pl_index, &pl_offset);
 741
 742         if (n_sectors)
 743                 *n_sectors = (PAGE_SIZE - pl_offset) >> SECTOR_SHIFT;
 744
 745         va = lowmem_page_address(pl[pl_index].page);
 746
 747         return (struct journal_sector *)(va + pl_offset);
 748 }
 749
 750 static struct journal_sector *access_journal(struct dm_integrity_c *ic, unsigned int section, unsigned int offset)
 751 {
 752         return access_page_list(ic, ic->journal, section, offset, NULL);
 753 }
 754
 755 static struct journal_entry *access_journal_entry(struct dm_integrity_c *ic, unsigned int section, unsigned int n)
 756 {
 757         unsigned int rel_sector, offset;
 758         struct journal_sector *js;
 759
 760         access_journal_check(ic, section, n, true, "access_journal_entry");
 761
 762         rel_sector = n % JOURNAL_BLOCK_SECTORS;
 763         offset = n / JOURNAL_BLOCK_SECTORS;
 764
 765         js = access_journal(ic, section, rel_sector);
 766         return (struct journal_entry *)((char *)js + offset * ic->journal_entry_size);
 767 }
 768
 769 static struct journal_sector *access_journal_data(struct dm_integrity_c *ic, unsigned int section, unsigned int n)
 770 {
 771         n <<= ic->sb->log2_sectors_per_block;
 772
 773         n += JOURNAL_BLOCK_SECTORS;
 774
 775         access_journal_check(ic, section, n, false, "access_journal_data");
 776
 777         return access_journal(ic, section, n);
 778 }
 779
 780 static void section_mac(struct dm_integrity_c *ic, unsigned int section, __u8 result[JOURNAL_MAC_SIZE])
 781 {
 782         SHASH_DESC_ON_STACK(desc, ic->journal_mac);
 783         int r;
 784         unsigned int j, size;
 785
 786         desc->tfm = ic->journal_mac;
 787
 788         r = crypto_shash_init(desc);
 789         if (unlikely(r < 0)) {
 790                 dm_integrity_io_error(ic, "crypto_shash_init", r);
 791                 goto err;
 792         }
 793
 794         if (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) {
 795                 __le64 section_le;
 796
 797                 r = crypto_shash_update(desc, (__u8 *)&ic->sb->salt, SALT_SIZE);
 798                 if (unlikely(r < 0)) {
 799                         dm_integrity_io_error(ic, "crypto_shash_update", r);
 800                         goto err;
 801                 }
 802
 803                 section_le = cpu_to_le64(section);
 804                 r = crypto_shash_update(desc, (__u8 *)&section_le, sizeof(section_le));
 805                 if (unlikely(r < 0)) {
 806                         dm_integrity_io_error(ic, "crypto_shash_update", r);
 807                         goto err;
 808                 }
 809         }
 810
 811         for (j = 0; j < ic->journal_section_entries; j++) {
 812                 struct journal_entry *je = access_journal_entry(ic, section, j);
 813
 814                 r = crypto_shash_update(desc, (__u8 *)&je->u.sector, sizeof(je->u.sector));
 815                 if (unlikely(r < 0)) {
 816                         dm_integrity_io_error(ic, "crypto_shash_update", r);
 817                         goto err;
 818                 }
 819         }
 820
 821         size = crypto_shash_digestsize(ic->journal_mac);
 822
 823         if (likely(size <= JOURNAL_MAC_SIZE)) {
 824                 r = crypto_shash_final(desc, result);
 825                 if (unlikely(r < 0)) {
 826                         dm_integrity_io_error(ic, "crypto_shash_final", r);
 827                         goto err;
 828                 }
 829                 memset(result + size, 0, JOURNAL_MAC_SIZE - size);
 830         } else {
 831                 __u8 digest[HASH_MAX_DIGESTSIZE];
 832
 833                 if (WARN_ON(size > sizeof(digest))) {
 834                         dm_integrity_io_error(ic, "digest_size", -EINVAL);
 835                         goto err;
 836                 }
 837                 r = crypto_shash_final(desc, digest);
 838                 if (unlikely(r < 0)) {
 839                         dm_integrity_io_error(ic, "crypto_shash_final", r);
 840                         goto err;
 841                 }
 842                 memcpy(result, digest, JOURNAL_MAC_SIZE);
 843         }
 844
 845         return;
 846 err:
 847         memset(result, 0, JOURNAL_MAC_SIZE);
 848 }
 849
 850 static void rw_section_mac(struct dm_integrity_c *ic, unsigned int section, bool wr)
 851 {
 852         __u8 result[JOURNAL_MAC_SIZE];
 853         unsigned int j;
 854
 855         if (!ic->journal_mac)
 856                 return;
 857
 858         section_mac(ic, section, result);
 859
 860         for (j = 0; j < JOURNAL_BLOCK_SECTORS; j++) {
 861                 struct journal_sector *js = access_journal(ic, section, j);
 862
 863                 if (likely(wr))
 864                         memcpy(&js->mac, result + (j * JOURNAL_MAC_PER_SECTOR), JOURNAL_MAC_PER_SECTOR);
 865                 else {
 866                         if (memcmp(&js->mac, result + (j * JOURNAL_MAC_PER_SECTOR), JOURNAL_MAC_PER_SECTOR)) {
 867                                 dm_integrity_io_error(ic, "journal mac", -EILSEQ);
 868                                 dm_audit_log_target(DM_MSG_PREFIX, "mac-journal", ic->ti, 0);
 869                         }
 870                 }
 871         }
 872 }
 873
 874 static void complete_journal_op(void *context)
 875 {
 876         struct journal_completion *comp = context;
 877
 878         BUG_ON(!atomic_read(&comp->in_flight));
 879         if (likely(atomic_dec_and_test(&comp->in_flight)))
 880                 complete(&comp->comp);
 881 }
 882
 883 static void xor_journal(struct dm_integrity_c *ic, bool encrypt, unsigned int section,
 884                         unsigned int n_sections, struct journal_completion *comp)
 885 {
 886         struct async_submit_ctl submit;
 887         size_t n_bytes = (size_t)(n_sections * ic->journal_section_sectors) << SECTOR_SHIFT;
 888         unsigned int pl_index, pl_offset, section_index;
 889         struct page_list *source_pl, *target_pl;
 890
 891         if (likely(encrypt)) {
 892                 source_pl = ic->journal;
 893                 target_pl = ic->journal_io;
 894         } else {
 895                 source_pl = ic->journal_io;
 896                 target_pl = ic->journal;
 897         }
 898
 899         page_list_location(ic, section, 0, &pl_index, &pl_offset);
 900
 901         atomic_add(roundup(pl_offset + n_bytes, PAGE_SIZE) >> PAGE_SHIFT, &comp->in_flight);
 902
 903         init_async_submit(&submit, ASYNC_TX_XOR_ZERO_DST, NULL, complete_journal_op, comp, NULL);
 904
 905         section_index = pl_index;
 906
 907         do {
 908                 size_t this_step;
 909                 struct page *src_pages[2];
 910                 struct page *dst_page;
 911
 912                 while (unlikely(pl_index == section_index)) {
 913                         unsigned int dummy;
 914
 915                         if (likely(encrypt))
 916                                 rw_section_mac(ic, section, true);
 917                         section++;
 918                         n_sections--;
 919                         if (!n_sections)
 920                                 break;
 921                         page_list_location(ic, section, 0, &section_index, &dummy);
 922                 }
 923
 924                 this_step = min(n_bytes, (size_t)PAGE_SIZE - pl_offset);
 925                 dst_page = target_pl[pl_index].page;
 926                 src_pages[0] = source_pl[pl_index].page;
 927                 src_pages[1] = ic->journal_xor[pl_index].page;
 928
 929                 async_xor(dst_page, src_pages, pl_offset, 2, this_step, &submit);
 930
 931                 pl_index++;
 932                 pl_offset = 0;
 933                 n_bytes -= this_step;
 934         } while (n_bytes);
 935
 936         BUG_ON(n_sections);
 937
 938         async_tx_issue_pending_all();
 939 }
 940
 941 static void complete_journal_encrypt(void *data, int err)
 942 {
 943         struct journal_completion *comp = data;
 944
 945         if (unlikely(err)) {
 946                 if (likely(err == -EINPROGRESS)) {
 947                         complete(&comp->ic->crypto_backoff);
 948                         return;
 949                 }
 950                 dm_integrity_io_error(comp->ic, "asynchronous encrypt", err);
 951         }
 952         complete_journal_op(comp);
 953 }
 954
 955 static bool do_crypt(bool encrypt, struct skcipher_request *req, struct journal_completion *comp)
 956 {
 957         int r;
 958
 959         skcipher_request_set_callback(req, CRYPTO_TFM_REQ_MAY_BACKLOG,
 960                                       complete_journal_encrypt, comp);
 961         if (likely(encrypt))
 962                 r = crypto_skcipher_encrypt(req);
 963         else
 964                 r = crypto_skcipher_decrypt(req);
 965         if (likely(!r))
 966                 return false;
 967         if (likely(r == -EINPROGRESS))
 968                 return true;
 969         if (likely(r == -EBUSY)) {
 970                 wait_for_completion(&comp->ic->crypto_backoff);
 971                 reinit_completion(&comp->ic->crypto_backoff);
 972                 return true;
 973         }
 974         dm_integrity_io_error(comp->ic, "encrypt", r);
 975         return false;
 976 }
 977
 978 static void crypt_journal(struct dm_integrity_c *ic, bool encrypt, unsigned int section,
 979                           unsigned int n_sections, struct journal_completion *comp)
 980 {
 981         struct scatterlist **source_sg;
 982         struct scatterlist **target_sg;
 983
 984         atomic_add(2, &comp->in_flight);
 985
 986         if (likely(encrypt)) {
 987                 source_sg = ic->journal_scatterlist;
 988                 target_sg = ic->journal_io_scatterlist;
 989         } else {
 990                 source_sg = ic->journal_io_scatterlist;
 991                 target_sg = ic->journal_scatterlist;
 992         }
 993
 994         do {
 995                 struct skcipher_request *req;
 996                 unsigned int ivsize;
 997                 char *iv;
 998
 999                 if (likely(encrypt))
1000                         rw_section_mac(ic, section, true);
1001
1002                 req = ic->sk_requests[section];
1003                 ivsize = crypto_skcipher_ivsize(ic->journal_crypt);
1004                 iv = req->iv;
1005
1006                 memcpy(iv, iv + ivsize, ivsize);
1007
1008                 req->src = source_sg[section];
1009                 req->dst = target_sg[section];
1010
1011                 if (unlikely(do_crypt(encrypt, req, comp)))
1012                         atomic_inc(&comp->in_flight);
1013
1014                 section++;
1015                 n_sections--;
1016         } while (n_sections);
1017
1018         atomic_dec(&comp->in_flight);
1019         complete_journal_op(comp);
1020 }
1021
1022 static void encrypt_journal(struct dm_integrity_c *ic, bool encrypt, unsigned int section,
1023                             unsigned int n_sections, struct journal_completion *comp)
1024 {
1025         if (ic->journal_xor)
1026                 return xor_journal(ic, encrypt, section, n_sections, comp);
1027         else
1028                 return crypt_journal(ic, encrypt, section, n_sections, comp);
1029 }
1030
1031 static void complete_journal_io(unsigned long error, void *context)
1032 {
1033         struct journal_completion *comp = context;
1034
1035         if (unlikely(error != 0))
1036                 dm_integrity_io_error(comp->ic, "writing journal", -EIO);
1037         complete_journal_op(comp);
1038 }
1039
1040 static void rw_journal_sectors(struct dm_integrity_c *ic, blk_opf_t opf,
1041                                unsigned int sector, unsigned int n_sectors,
1042                                struct journal_completion *comp)
1043 {
1044         struct dm_io_request io_req;
1045         struct dm_io_region io_loc;
1046         unsigned int pl_index, pl_offset;
1047         int r;
1048
1049         if (unlikely(dm_integrity_failed(ic))) {
1050                 if (comp)
1051                         complete_journal_io(-1UL, comp);
1052                 return;
1053         }
1054
1055         pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
1056         pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
1057
1058         io_req.bi_opf = opf;
1059         io_req.mem.type = DM_IO_PAGE_LIST;
1060         if (ic->journal_io)
1061                 io_req.mem.ptr.pl = &ic->journal_io[pl_index];
1062         else
1063                 io_req.mem.ptr.pl = &ic->journal[pl_index];
1064         io_req.mem.offset = pl_offset;
1065         if (likely(comp != NULL)) {
1066                 io_req.notify.fn = complete_journal_io;
1067                 io_req.notify.context = comp;
1068         } else {
1069                 io_req.notify.fn = NULL;
1070         }
1071         io_req.client = ic->io;
1072         io_loc.bdev = ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev;
1073         io_loc.sector = ic->start + SB_SECTORS + sector;
1074         io_loc.count = n_sectors;
1075
1076         r = dm_io(&io_req, 1, &io_loc, NULL);
1077         if (unlikely(r)) {
1078                 dm_integrity_io_error(ic, (opf & REQ_OP_MASK) == REQ_OP_READ ?
1079                                       "reading journal" : "writing journal", r);
1080                 if (comp) {
1081                         WARN_ONCE(1, "asynchronous dm_io failed: %d", r);
1082                         complete_journal_io(-1UL, comp);
1083                 }
1084         }
1085 }
1086
1087 static void rw_journal(struct dm_integrity_c *ic, blk_opf_t opf,
1088                        unsigned int section, unsigned int n_sections,
1089                        struct journal_completion *comp)
1090 {
1091         unsigned int sector, n_sectors;
1092
1093         sector = section * ic->journal_section_sectors;
1094         n_sectors = n_sections * ic->journal_section_sectors;
1095
1096         rw_journal_sectors(ic, opf, sector, n_sectors, comp);
1097 }
1098
1099 static void write_journal(struct dm_integrity_c *ic, unsigned int commit_start, unsigned int commit_sections)
1100 {
1101         struct journal_completion io_comp;
1102         struct journal_completion crypt_comp_1;
1103         struct journal_completion crypt_comp_2;
1104         unsigned int i;
1105
1106         io_comp.ic = ic;
1107         init_completion(&io_comp.comp);
1108
1109         if (commit_start + commit_sections <= ic->journal_sections) {
1110                 io_comp.in_flight = (atomic_t)ATOMIC_INIT(1);
1111                 if (ic->journal_io) {
1112                         crypt_comp_1.ic = ic;
1113                         init_completion(&crypt_comp_1.comp);
1114                         crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
1115                         encrypt_journal(ic, true, commit_start, commit_sections, &crypt_comp_1);
1116                         wait_for_completion_io(&crypt_comp_1.comp);
1117                 } else {
1118                         for (i = 0; i < commit_sections; i++)
1119                                 rw_section_mac(ic, commit_start + i, true);
1120                 }
1121                 rw_journal(ic, REQ_OP_WRITE | REQ_FUA | REQ_SYNC, commit_start,
1122                            commit_sections, &io_comp);
1123         } else {
1124                 unsigned int to_end;
1125
1126                 io_comp.in_flight = (atomic_t)ATOMIC_INIT(2);
1127                 to_end = ic->journal_sections - commit_start;
1128                 if (ic->journal_io) {
1129                         crypt_comp_1.ic = ic;
1130                         init_completion(&crypt_comp_1.comp);
1131                         crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
1132                         encrypt_journal(ic, true, commit_start, to_end, &crypt_comp_1);
1133                         if (try_wait_for_completion(&crypt_comp_1.comp)) {
1134                                 rw_journal(ic, REQ_OP_WRITE | REQ_FUA,
1135                                            commit_start, to_end, &io_comp);
1136                                 reinit_completion(&crypt_comp_1.comp);
1137                                 crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
1138                                 encrypt_journal(ic, true, 0, commit_sections - to_end, &crypt_comp_1);
1139                                 wait_for_completion_io(&crypt_comp_1.comp);
1140                         } else {
1141                                 crypt_comp_2.ic = ic;
1142                                 init_completion(&crypt_comp_2.comp);
1143                                 crypt_comp_2.in_flight = (atomic_t)ATOMIC_INIT(0);
1144                                 encrypt_journal(ic, true, 0, commit_sections - to_end, &crypt_comp_2);
1145                                 wait_for_completion_io(&crypt_comp_1.comp);
1146                                 rw_journal(ic, REQ_OP_WRITE | REQ_FUA, commit_start, to_end, &io_comp);
1147                                 wait_for_completion_io(&crypt_comp_2.comp);
1148                         }
1149                 } else {
1150                         for (i = 0; i < to_end; i++)
1151                                 rw_section_mac(ic, commit_start + i, true);
1152                         rw_journal(ic, REQ_OP_WRITE | REQ_FUA, commit_start, to_end, &io_comp);
1153                         for (i = 0; i < commit_sections - to_end; i++)
1154                                 rw_section_mac(ic, i, true);
1155                 }
1156                 rw_journal(ic, REQ_OP_WRITE | REQ_FUA, 0, commit_sections - to_end, &io_comp);
1157         }
1158
1159         wait_for_completion_io(&io_comp.comp);
1160 }
1161
1162 static void copy_from_journal(struct dm_integrity_c *ic, unsigned int section, unsigned int offset,
1163                               unsigned int n_sectors, sector_t target, io_notify_fn fn, void *data)
1164 {
1165         struct dm_io_request io_req;
1166         struct dm_io_region io_loc;
1167         int r;
1168         unsigned int sector, pl_index, pl_offset;
1169
1170         BUG_ON((target | n_sectors | offset) & (unsigned int)(ic->sectors_per_block - 1));
1171
1172         if (unlikely(dm_integrity_failed(ic))) {
1173                 fn(-1UL, data);
1174                 return;
1175         }
1176
1177         sector = section * ic->journal_section_sectors + JOURNAL_BLOCK_SECTORS + offset;
1178
1179         pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
1180         pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
1181
1182         io_req.bi_opf = REQ_OP_WRITE;
1183         io_req.mem.type = DM_IO_PAGE_LIST;
1184         io_req.mem.ptr.pl = &ic->journal[pl_index];
1185         io_req.mem.offset = pl_offset;
1186         io_req.notify.fn = fn;
1187         io_req.notify.context = data;
1188         io_req.client = ic->io;
1189         io_loc.bdev = ic->dev->bdev;
1190         io_loc.sector = target;
1191         io_loc.count = n_sectors;
1192
1193         r = dm_io(&io_req, 1, &io_loc, NULL);
1194         if (unlikely(r)) {
1195                 WARN_ONCE(1, "asynchronous dm_io failed: %d", r);
1196                 fn(-1UL, data);
1197         }
1198 }
1199
1200 static bool ranges_overlap(struct dm_integrity_range *range1, struct dm_integrity_range *range2)
1201 {
1202         return range1->logical_sector < range2->logical_sector + range2->n_sectors &&
1203                range1->logical_sector + range1->n_sectors > range2->logical_sector;
1204 }
1205
1206 static bool add_new_range(struct dm_integrity_c *ic, struct dm_integrity_range *new_range, bool check_waiting)
1207 {
1208         struct rb_node **n = &ic->in_progress.rb_node;
1209         struct rb_node *parent;
1210
1211         BUG_ON((new_range->logical_sector | new_range->n_sectors) & (unsigned int)(ic->sectors_per_block - 1));
1212
1213         if (likely(check_waiting)) {
1214                 struct dm_integrity_range *range;
1215
1216                 list_for_each_entry(range, &ic->wait_list, wait_entry) {
1217                         if (unlikely(ranges_overlap(range, new_range)))
1218                                 return false;
1219                 }
1220         }
1221
1222         parent = NULL;
1223
1224         while (*n) {
1225                 struct dm_integrity_range *range = container_of(*n, struct dm_integrity_range, node);
1226
1227                 parent = *n;
1228                 if (new_range->logical_sector + new_range->n_sectors <= range->logical_sector)
1229                         n = &range->node.rb_left;
1230                 else if (new_range->logical_sector >= range->logical_sector + range->n_sectors)
1231                         n = &range->node.rb_right;
1232                 else
1233                         return false;
1234         }
1235
1236         rb_link_node(&new_range->node, parent, n);
1237         rb_insert_color(&new_range->node, &ic->in_progress);
1238
1239         return true;
1240 }
1241
1242 static void remove_range_unlocked(struct dm_integrity_c *ic, struct dm_integrity_range *range)
1243 {
1244         rb_erase(&range->node, &ic->in_progress);
1245         while (unlikely(!list_empty(&ic->wait_list))) {
1246                 struct dm_integrity_range *last_range =
1247                         list_first_entry(&ic->wait_list, struct dm_integrity_range, wait_entry);
1248                 struct task_struct *last_range_task;
1249
1250                 last_range_task = last_range->task;
1251                 list_del(&last_range->wait_entry);
1252                 if (!add_new_range(ic, last_range, false)) {
1253                         last_range->task = last_range_task;
1254                         list_add(&last_range->wait_entry, &ic->wait_list);
1255                         break;
1256                 }
1257                 last_range->waiting = false;
1258                 wake_up_process(last_range_task);
1259         }
1260 }
1261
1262 static void remove_range(struct dm_integrity_c *ic, struct dm_integrity_range *range)
1263 {
1264         unsigned long flags;
1265
1266         spin_lock_irqsave(&ic->endio_wait.lock, flags);
1267         remove_range_unlocked(ic, range);
1268         spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
1269 }
1270
1271 static void wait_and_add_new_range(struct dm_integrity_c *ic, struct dm_integrity_range *new_range)
1272 {
1273         new_range->waiting = true;
1274         list_add_tail(&new_range->wait_entry, &ic->wait_list);
1275         new_range->task = current;
1276         do {
1277                 __set_current_state(TASK_UNINTERRUPTIBLE);
1278                 spin_unlock_irq(&ic->endio_wait.lock);
1279                 io_schedule();
1280                 spin_lock_irq(&ic->endio_wait.lock);
1281         } while (unlikely(new_range->waiting));
1282 }
1283
1284 static void add_new_range_and_wait(struct dm_integrity_c *ic, struct dm_integrity_range *new_range)
1285 {
1286         if (unlikely(!add_new_range(ic, new_range, true)))
1287                 wait_and_add_new_range(ic, new_range);
1288 }
1289
1290 static void init_journal_node(struct journal_node *node)
1291 {
1292         RB_CLEAR_NODE(&node->node);
1293         node->sector = (sector_t)-1;
1294 }
1295
1296 static void add_journal_node(struct dm_integrity_c *ic, struct journal_node *node, sector_t sector)
1297 {
1298         struct rb_node **link;
1299         struct rb_node *parent;
1300
1301         node->sector = sector;
1302         BUG_ON(!RB_EMPTY_NODE(&node->node));
1303
1304         link = &ic->journal_tree_root.rb_node;
1305         parent = NULL;
1306
1307         while (*link) {
1308                 struct journal_node *j;
1309
1310                 parent = *link;
1311                 j = container_of(parent, struct journal_node, node);
1312                 if (sector < j->sector)
1313                         link = &j->node.rb_left;
1314                 else
1315                         link = &j->node.rb_right;
1316         }
1317
1318         rb_link_node(&node->node, parent, link);
1319         rb_insert_color(&node->node, &ic->journal_tree_root);
1320 }
1321
1322 static void remove_journal_node(struct dm_integrity_c *ic, struct journal_node *node)
1323 {
1324         BUG_ON(RB_EMPTY_NODE(&node->node));
1325         rb_erase(&node->node, &ic->journal_tree_root);
1326         init_journal_node(node);
1327 }
1328
1329 #define NOT_FOUND       (-1U)
1330
1331 static unsigned int find_journal_node(struct dm_integrity_c *ic, sector_t sector, sector_t *next_sector)
1332 {
1333         struct rb_node *n = ic->journal_tree_root.rb_node;
1334         unsigned int found = NOT_FOUND;
1335
1336         *next_sector = (sector_t)-1;
1337         while (n) {
1338                 struct journal_node *j = container_of(n, struct journal_node, node);
1339
1340                 if (sector == j->sector)
1341                         found = j - ic->journal_tree;
1342
1343                 if (sector < j->sector) {
1344                         *next_sector = j->sector;
1345                         n = j->node.rb_left;
1346                 } else
1347                         n = j->node.rb_right;
1348         }
1349
1350         return found;
1351 }
1352
1353 static bool test_journal_node(struct dm_integrity_c *ic, unsigned int pos, sector_t sector)
1354 {
1355         struct journal_node *node, *next_node;
1356         struct rb_node *next;
1357
1358         if (unlikely(pos >= ic->journal_entries))
1359                 return false;
1360         node = &ic->journal_tree[pos];
1361         if (unlikely(RB_EMPTY_NODE(&node->node)))
1362                 return false;
1363         if (unlikely(node->sector != sector))
1364                 return false;
1365
1366         next = rb_next(&node->node);
1367         if (unlikely(!next))
1368                 return true;
1369
1370         next_node = container_of(next, struct journal_node, node);
1371         return next_node->sector != sector;
1372 }
1373
1374 static bool find_newer_committed_node(struct dm_integrity_c *ic, struct journal_node *node)
1375 {
1376         struct rb_node *next;
1377         struct journal_node *next_node;
1378         unsigned int next_section;
1379
1380         BUG_ON(RB_EMPTY_NODE(&node->node));
1381
1382         next = rb_next(&node->node);
1383         if (unlikely(!next))
1384                 return false;
1385
1386         next_node = container_of(next, struct journal_node, node);
1387
1388         if (next_node->sector != node->sector)
1389                 return false;
1390
1391         next_section = (unsigned int)(next_node - ic->journal_tree) / ic->journal_section_entries;
1392         if (next_section >= ic->committed_section &&
1393             next_section < ic->committed_section + ic->n_committed_sections)
1394                 return true;
1395         if (next_section + ic->journal_sections < ic->committed_section + ic->n_committed_sections)
1396                 return true;
1397
1398         return false;
1399 }
1400
1401 #define TAG_READ        0
1402 #define TAG_WRITE       1
1403 #define TAG_CMP         2
1404
1405 static int dm_integrity_rw_tag(struct dm_integrity_c *ic, unsigned char *tag, sector_t *metadata_block,
1406                                unsigned int *metadata_offset, unsigned int total_size, int op)
1407 {
1408 #define MAY_BE_FILLER           1
1409 #define MAY_BE_HASH             2
1410         unsigned int hash_offset = 0;
1411         unsigned int may_be = MAY_BE_HASH | (ic->discard ? MAY_BE_FILLER : 0);
1412
1413         do {
1414                 unsigned char *data, *dp;
1415                 struct dm_buffer *b;
1416                 unsigned int to_copy;
1417                 int r;
1418
1419                 r = dm_integrity_failed(ic);
1420                 if (unlikely(r))
1421                         return r;
1422
1423                 data = dm_bufio_read(ic->bufio, *metadata_block, &b);
1424                 if (IS_ERR(data))
1425                         return PTR_ERR(data);
1426
1427                 to_copy = min((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - *metadata_offset, total_size);
1428                 dp = data + *metadata_offset;
1429                 if (op == TAG_READ) {
1430                         memcpy(tag, dp, to_copy);
1431                 } else if (op == TAG_WRITE) {
1432                         if (memcmp(dp, tag, to_copy)) {
1433                                 memcpy(dp, tag, to_copy);
1434                                 dm_bufio_mark_partial_buffer_dirty(b, *metadata_offset, *metadata_offset + to_copy);
1435                         }
1436                 } else {
1437                         /* e.g.: op == TAG_CMP */
1438
1439                         if (likely(is_power_of_2(ic->tag_size))) {
1440                                 if (unlikely(memcmp(dp, tag, to_copy)))
1441                                         if (unlikely(!ic->discard) ||
1442                                             unlikely(memchr_inv(dp, DISCARD_FILLER, to_copy) != NULL)) {
1443                                                 goto thorough_test;
1444                                 }
1445                         } else {
1446                                 unsigned int i, ts;
1447 thorough_test:
1448                                 ts = total_size;
1449
1450                                 for (i = 0; i < to_copy; i++, ts--) {
1451                                         if (unlikely(dp[i] != tag[i]))
1452                                                 may_be &= ~MAY_BE_HASH;
1453                                         if (likely(dp[i] != DISCARD_FILLER))
1454                                                 may_be &= ~MAY_BE_FILLER;
1455                                         hash_offset++;
1456                                         if (unlikely(hash_offset == ic->tag_size)) {
1457                                                 if (unlikely(!may_be)) {
1458                                                         dm_bufio_release(b);
1459                                                         return ts;
1460                                                 }
1461                                                 hash_offset = 0;
1462                                                 may_be = MAY_BE_HASH | (ic->discard ? MAY_BE_FILLER : 0);
1463                                         }
1464                                 }
1465                         }
1466                 }
1467                 dm_bufio_release(b);
1468
1469                 tag += to_copy;
1470                 *metadata_offset += to_copy;
1471                 if (unlikely(*metadata_offset == 1U << SECTOR_SHIFT << ic->log2_buffer_sectors)) {
1472                         (*metadata_block)++;
1473                         *metadata_offset = 0;
1474                 }
1475
1476                 if (unlikely(!is_power_of_2(ic->tag_size)))
1477                         hash_offset = (hash_offset + to_copy) % ic->tag_size;
1478
1479                 total_size -= to_copy;
1480         } while (unlikely(total_size));
1481
1482         return 0;
1483 #undef MAY_BE_FILLER
1484 #undef MAY_BE_HASH
1485 }
1486
1487 struct flush_request {
1488         struct dm_io_request io_req;
1489         struct dm_io_region io_reg;
1490         struct dm_integrity_c *ic;
1491         struct completion comp;
1492 };
1493
1494 static void flush_notify(unsigned long error, void *fr_)
1495 {
1496         struct flush_request *fr = fr_;
1497
1498         if (unlikely(error != 0))
1499                 dm_integrity_io_error(fr->ic, "flushing disk cache", -EIO);
1500         complete(&fr->comp);
1501 }
1502
1503 static void dm_integrity_flush_buffers(struct dm_integrity_c *ic, bool flush_data)
1504 {
1505         int r;
1506         struct flush_request fr;
1507
1508         if (!ic->meta_dev)
1509                 flush_data = false;
1510         if (flush_data) {
1511                 fr.io_req.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC,
1512                 fr.io_req.mem.type = DM_IO_KMEM,
1513                 fr.io_req.mem.ptr.addr = NULL,
1514                 fr.io_req.notify.fn = flush_notify,
1515                 fr.io_req.notify.context = &fr;
1516                 fr.io_req.client = dm_bufio_get_dm_io_client(ic->bufio),
1517                 fr.io_reg.bdev = ic->dev->bdev,
1518                 fr.io_reg.sector = 0,
1519                 fr.io_reg.count = 0,
1520                 fr.ic = ic;
1521                 init_completion(&fr.comp);
1522                 r = dm_io(&fr.io_req, 1, &fr.io_reg, NULL);
1523                 BUG_ON(r);
1524         }
1525
1526         r = dm_bufio_write_dirty_buffers(ic->bufio);
1527         if (unlikely(r))
1528                 dm_integrity_io_error(ic, "writing tags", r);
1529
1530         if (flush_data)
1531                 wait_for_completion(&fr.comp);
1532 }
1533
1534 static void sleep_on_endio_wait(struct dm_integrity_c *ic)
1535 {
1536         DECLARE_WAITQUEUE(wait, current);
1537
1538         __add_wait_queue(&ic->endio_wait, &wait);
1539         __set_current_state(TASK_UNINTERRUPTIBLE);
1540         spin_unlock_irq(&ic->endio_wait.lock);
1541         io_schedule();
1542         spin_lock_irq(&ic->endio_wait.lock);
1543         __remove_wait_queue(&ic->endio_wait, &wait);
1544 }
1545
1546 static void autocommit_fn(struct timer_list *t)
1547 {
1548         struct dm_integrity_c *ic = from_timer(ic, t, autocommit_timer);
1549
1550         if (likely(!dm_integrity_failed(ic)))
1551                 queue_work(ic->commit_wq, &ic->commit_work);
1552 }
1553
1554 static void schedule_autocommit(struct dm_integrity_c *ic)
1555 {
1556         if (!timer_pending(&ic->autocommit_timer))
1557                 mod_timer(&ic->autocommit_timer, jiffies + ic->autocommit_jiffies);
1558 }
1559
1560 static void submit_flush_bio(struct dm_integrity_c *ic, struct dm_integrity_io *dio)
1561 {
1562         struct bio *bio;
1563         unsigned long flags;
1564
1565         spin_lock_irqsave(&ic->endio_wait.lock, flags);
1566         bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1567         bio_list_add(&ic->flush_bio_list, bio);
1568         spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
1569
1570         queue_work(ic->commit_wq, &ic->commit_work);
1571 }
1572
1573 static void do_endio(struct dm_integrity_c *ic, struct bio *bio)
1574 {
1575         int r;
1576
1577         r = dm_integrity_failed(ic);
1578         if (unlikely(r) && !bio->bi_status)
1579                 bio->bi_status = errno_to_blk_status(r);
1580         if (unlikely(ic->synchronous_mode) && bio_op(bio) == REQ_OP_WRITE) {
1581                 unsigned long flags;
1582
1583                 spin_lock_irqsave(&ic->endio_wait.lock, flags);
1584                 bio_list_add(&ic->synchronous_bios, bio);
1585                 queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, 0);
1586                 spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
1587                 return;
1588         }
1589         bio_endio(bio);
1590 }
1591
1592 static void do_endio_flush(struct dm_integrity_c *ic, struct dm_integrity_io *dio)
1593 {
1594         struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1595
1596         if (unlikely(dio->fua) && likely(!bio->bi_status) && likely(!dm_integrity_failed(ic)))
1597                 submit_flush_bio(ic, dio);
1598         else
1599                 do_endio(ic, bio);
1600 }
1601
1602 static void dec_in_flight(struct dm_integrity_io *dio)
1603 {
1604         if (atomic_dec_and_test(&dio->in_flight)) {
1605                 struct dm_integrity_c *ic = dio->ic;
1606                 struct bio *bio;
1607
1608                 remove_range(ic, &dio->range);
1609
1610                 if (dio->op == REQ_OP_WRITE || unlikely(dio->op == REQ_OP_DISCARD))
1611                         schedule_autocommit(ic);
1612
1613                 bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1614                 if (unlikely(dio->bi_status) && !bio->bi_status)
1615                         bio->bi_status = dio->bi_status;
1616                 if (likely(!bio->bi_status) && unlikely(bio_sectors(bio) != dio->range.n_sectors)) {
1617                         dio->range.logical_sector += dio->range.n_sectors;
1618                         bio_advance(bio, dio->range.n_sectors << SECTOR_SHIFT);
1619                         INIT_WORK(&dio->work, integrity_bio_wait);
1620                         queue_work(ic->offload_wq, &dio->work);
1621                         return;
1622                 }
1623                 do_endio_flush(ic, dio);
1624         }
1625 }
1626
1627 static void integrity_end_io(struct bio *bio)
1628 {
1629         struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
1630
1631         dm_bio_restore(&dio->bio_details, bio);
1632         if (bio->bi_integrity)
1633                 bio->bi_opf |= REQ_INTEGRITY;
1634
1635         if (dio->completion)
1636                 complete(dio->completion);
1637
1638         dec_in_flight(dio);
1639 }
1640
1641 static void integrity_sector_checksum(struct dm_integrity_c *ic, sector_t sector,
1642                                       const char *data, char *result)
1643 {
1644         __le64 sector_le = cpu_to_le64(sector);
1645         SHASH_DESC_ON_STACK(req, ic->internal_hash);
1646         int r;
1647         unsigned int digest_size;
1648
1649         req->tfm = ic->internal_hash;
1650
1651         r = crypto_shash_init(req);
1652         if (unlikely(r < 0)) {
1653                 dm_integrity_io_error(ic, "crypto_shash_init", r);
1654                 goto failed;
1655         }
1656
1657         if (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) {
1658                 r = crypto_shash_update(req, (__u8 *)&ic->sb->salt, SALT_SIZE);
1659                 if (unlikely(r < 0)) {
1660                         dm_integrity_io_error(ic, "crypto_shash_update", r);
1661                         goto failed;
1662                 }
1663         }
1664
1665         r = crypto_shash_update(req, (const __u8 *)&sector_le, sizeof(sector_le));
1666         if (unlikely(r < 0)) {
1667                 dm_integrity_io_error(ic, "crypto_shash_update", r);
1668                 goto failed;
1669         }
1670
1671         r = crypto_shash_update(req, data, ic->sectors_per_block << SECTOR_SHIFT);
1672         if (unlikely(r < 0)) {
1673                 dm_integrity_io_error(ic, "crypto_shash_update", r);
1674                 goto failed;
1675         }
1676
1677         r = crypto_shash_final(req, result);
1678         if (unlikely(r < 0)) {
1679                 dm_integrity_io_error(ic, "crypto_shash_final", r);
1680                 goto failed;
1681         }
1682
1683         digest_size = crypto_shash_digestsize(ic->internal_hash);
1684         if (unlikely(digest_size < ic->tag_size))
1685                 memset(result + digest_size, 0, ic->tag_size - digest_size);
1686
1687         return;
1688
1689 failed:
1690         /* this shouldn't happen anyway, the hash functions have no reason to fail */
1691         get_random_bytes(result, ic->tag_size);
1692 }
1693
1694 static noinline void integrity_recheck(struct dm_integrity_io *dio, char *checksum)
1695 {
1696         struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1697         struct dm_integrity_c *ic = dio->ic;
1698         struct bvec_iter iter;
1699         struct bio_vec bv;
1700         sector_t sector, logical_sector, area, offset;
1701         struct page *page;
1702         void *buffer;
1703
1704         get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
1705         dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset,
1706                                                              &dio->metadata_offset);
1707         sector = get_data_sector(ic, area, offset);
1708         logical_sector = dio->range.logical_sector;
1709
1710         page = mempool_alloc(&ic->recheck_pool, GFP_NOIO);
1711         buffer = page_to_virt(page);
1712
1713         __bio_for_each_segment(bv, bio, iter, dio->bio_details.bi_iter) {
1714                 unsigned pos = 0;
1715
1716                 do {
1717                         char *mem;
1718                         int r;
1719                         struct dm_io_request io_req;
1720                         struct dm_io_region io_loc;
1721                         io_req.bi_opf = REQ_OP_READ;
1722                         io_req.mem.type = DM_IO_KMEM;
1723                         io_req.mem.ptr.addr = buffer;
1724                         io_req.notify.fn = NULL;
1725                         io_req.client = ic->io;
1726                         io_loc.bdev = ic->dev->bdev;
1727                         io_loc.sector = sector;
1728                         io_loc.count = ic->sectors_per_block;
1729
1730                         r = dm_io(&io_req, 1, &io_loc, NULL);
1731                         if (unlikely(r)) {
1732                                 dio->bi_status = errno_to_blk_status(r);
1733                                 goto free_ret;
1734                         }
1735
1736                         integrity_sector_checksum(ic, logical_sector, buffer, checksum);
1737                         r = dm_integrity_rw_tag(ic, checksum, &dio->metadata_block,
1738                                                 &dio->metadata_offset, ic->tag_size, TAG_CMP);
1739                         if (r) {
1740                                 if (r > 0) {
1741                                         DMERR_LIMIT("%pg: Checksum failed at sector 0x%llx",
1742                                                     bio->bi_bdev, logical_sector);
1743                                         atomic64_inc(&ic->number_of_mismatches);
1744                                         dm_audit_log_bio(DM_MSG_PREFIX, "integrity-checksum",
1745                                                          bio, logical_sector, 0);
1746                                         r = -EILSEQ;
1747                                 }
1748                                 dio->bi_status = errno_to_blk_status(r);
1749                                 goto free_ret;
1750                         }
1751
1752                         mem = bvec_kmap_local(&bv);
1753                         memcpy(mem + pos, buffer, ic->sectors_per_block << SECTOR_SHIFT);
1754                         kunmap_local(mem);
1755
1756                         pos += ic->sectors_per_block << SECTOR_SHIFT;
1757                         sector += ic->sectors_per_block;
1758                         logical_sector += ic->sectors_per_block;
1759                 } while (pos < bv.bv_len);
1760         }
1761 free_ret:
1762         mempool_free(page, &ic->recheck_pool);
1763 }
1764
1765 static void integrity_metadata(struct work_struct *w)
1766 {
1767         struct dm_integrity_io *dio = container_of(w, struct dm_integrity_io, work);
1768         struct dm_integrity_c *ic = dio->ic;
1769
1770         int r;
1771
1772         if (ic->internal_hash) {
1773                 struct bvec_iter iter;
1774                 struct bio_vec bv;
1775                 unsigned int digest_size = crypto_shash_digestsize(ic->internal_hash);
1776                 struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1777                 char *checksums;
1778                 unsigned int extra_space = unlikely(digest_size > ic->tag_size) ? digest_size - ic->tag_size : 0;
1779                 char checksums_onstack[max_t(size_t, HASH_MAX_DIGESTSIZE, MAX_TAG_SIZE)];
1780                 sector_t sector;
1781                 unsigned int sectors_to_process;
1782
1783                 if (unlikely(ic->mode == 'R'))
1784                         goto skip_io;
1785
1786                 if (likely(dio->op != REQ_OP_DISCARD))
1787                         checksums = kmalloc((PAGE_SIZE >> SECTOR_SHIFT >> ic->sb->log2_sectors_per_block) * ic->tag_size + extra_space,
1788                                             GFP_NOIO | __GFP_NORETRY | __GFP_NOWARN);
1789                 else
1790                         checksums = kmalloc(PAGE_SIZE, GFP_NOIO | __GFP_NORETRY | __GFP_NOWARN);
1791                 if (!checksums) {
1792                         checksums = checksums_onstack;
1793                         if (WARN_ON(extra_space &&
1794                                     digest_size > sizeof(checksums_onstack))) {
1795                                 r = -EINVAL;
1796                                 goto error;
1797                         }
1798                 }
1799
1800                 if (unlikely(dio->op == REQ_OP_DISCARD)) {
1801                         unsigned int bi_size = dio->bio_details.bi_iter.bi_size;
1802                         unsigned int max_size = likely(checksums != checksums_onstack) ? PAGE_SIZE : HASH_MAX_DIGESTSIZE;
1803                         unsigned int max_blocks = max_size / ic->tag_size;
1804
1805                         memset(checksums, DISCARD_FILLER, max_size);
1806
1807                         while (bi_size) {
1808                                 unsigned int this_step_blocks = bi_size >> (SECTOR_SHIFT + ic->sb->log2_sectors_per_block);
1809
1810                                 this_step_blocks = min(this_step_blocks, max_blocks);
1811                                 r = dm_integrity_rw_tag(ic, checksums, &dio->metadata_block, &dio->metadata_offset,
1812                                                         this_step_blocks * ic->tag_size, TAG_WRITE);
1813                                 if (unlikely(r)) {
1814                                         if (likely(checksums != checksums_onstack))
1815                                                 kfree(checksums);
1816                                         goto error;
1817                                 }
1818
1819                                 bi_size -= this_step_blocks << (SECTOR_SHIFT + ic->sb->log2_sectors_per_block);
1820                         }
1821
1822                         if (likely(checksums != checksums_onstack))
1823                                 kfree(checksums);
1824                         goto skip_io;
1825                 }
1826
1827                 sector = dio->range.logical_sector;
1828                 sectors_to_process = dio->range.n_sectors;
1829
1830                 __bio_for_each_segment(bv, bio, iter, dio->bio_details.bi_iter) {
1831                         struct bio_vec bv_copy = bv;
1832                         unsigned int pos;
1833                         char *mem, *checksums_ptr;
1834
1835 again:
1836                         mem = bvec_kmap_local(&bv_copy);
1837                         pos = 0;
1838                         checksums_ptr = checksums;
1839                         do {
1840                                 integrity_sector_checksum(ic, sector, mem + pos, checksums_ptr);
1841                                 checksums_ptr += ic->tag_size;
1842                                 sectors_to_process -= ic->sectors_per_block;
1843                                 pos += ic->sectors_per_block << SECTOR_SHIFT;
1844                                 sector += ic->sectors_per_block;
1845                         } while (pos < bv_copy.bv_len && sectors_to_process && checksums != checksums_onstack);
1846                         kunmap_local(mem);
1847
1848                         r = dm_integrity_rw_tag(ic, checksums, &dio->metadata_block, &dio->metadata_offset,
1849                                                 checksums_ptr - checksums, dio->op == REQ_OP_READ ? TAG_CMP : TAG_WRITE);
1850                         if (unlikely(r)) {
1851                                 if (r > 0) {
1852                                         integrity_recheck(dio, checksums);
1853                                         goto skip_io;
1854                                 }
1855                                 if (likely(checksums != checksums_onstack))
1856                                         kfree(checksums);
1857                                 goto error;
1858                         }
1859
1860                         if (!sectors_to_process)
1861                                 break;
1862
1863                         if (unlikely(pos < bv_copy.bv_len)) {
1864                                 bv_copy.bv_offset += pos;
1865                                 bv_copy.bv_len -= pos;
1866                                 goto again;
1867                         }
1868                 }
1869
1870                 if (likely(checksums != checksums_onstack))
1871                         kfree(checksums);
1872         } else {
1873                 struct bio_integrity_payload *bip = dio->bio_details.bi_integrity;
1874
1875                 if (bip) {
1876                         struct bio_vec biv;
1877                         struct bvec_iter iter;
1878                         unsigned int data_to_process = dio->range.n_sectors;
1879
1880                         sector_to_block(ic, data_to_process);
1881                         data_to_process *= ic->tag_size;
1882
1883                         bip_for_each_vec(biv, bip, iter) {
1884                                 unsigned char *tag;
1885                                 unsigned int this_len;
1886
1887                                 BUG_ON(PageHighMem(biv.bv_page));
1888                                 tag = bvec_virt(&biv);
1889                                 this_len = min(biv.bv_len, data_to_process);
1890                                 r = dm_integrity_rw_tag(ic, tag, &dio->metadata_block, &dio->metadata_offset,
1891                                                         this_len, dio->op == REQ_OP_READ ? TAG_READ : TAG_WRITE);
1892                                 if (unlikely(r))
1893                                         goto error;
1894                                 data_to_process -= this_len;
1895                                 if (!data_to_process)
1896                                         break;
1897                         }
1898                 }
1899         }
1900 skip_io:
1901         dec_in_flight(dio);
1902         return;
1903 error:
1904         dio->bi_status = errno_to_blk_status(r);
1905         dec_in_flight(dio);
1906 }
1907
1908 static int dm_integrity_map(struct dm_target *ti, struct bio *bio)
1909 {
1910         struct dm_integrity_c *ic = ti->private;
1911         struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
1912         struct bio_integrity_payload *bip;
1913
1914         sector_t area, offset;
1915
1916         dio->ic = ic;
1917         dio->bi_status = 0;
1918         dio->op = bio_op(bio);
1919
1920         if (unlikely(dio->op == REQ_OP_DISCARD)) {
1921                 if (ti->max_io_len) {
1922                         sector_t sec = dm_target_offset(ti, bio->bi_iter.bi_sector);
1923                         unsigned int log2_max_io_len = __fls(ti->max_io_len);
1924                         sector_t start_boundary = sec >> log2_max_io_len;
1925                         sector_t end_boundary = (sec + bio_sectors(bio) - 1) >> log2_max_io_len;
1926
1927                         if (start_boundary < end_boundary) {
1928                                 sector_t len = ti->max_io_len - (sec & (ti->max_io_len - 1));
1929
1930                                 dm_accept_partial_bio(bio, len);
1931                         }
1932                 }
1933         }
1934
1935         if (unlikely(bio->bi_opf & REQ_PREFLUSH)) {
1936                 submit_flush_bio(ic, dio);
1937                 return DM_MAPIO_SUBMITTED;
1938         }
1939
1940         dio->range.logical_sector = dm_target_offset(ti, bio->bi_iter.bi_sector);
1941         dio->fua = dio->op == REQ_OP_WRITE && bio->bi_opf & REQ_FUA;
1942         if (unlikely(dio->fua)) {
1943                 /*
1944                  * Don't pass down the FUA flag because we have to flush
1945                  * disk cache anyway.
1946                  */
1947                 bio->bi_opf &= ~REQ_FUA;
1948         }
1949         if (unlikely(dio->range.logical_sector + bio_sectors(bio) > ic->provided_data_sectors)) {
1950                 DMERR("Too big sector number: 0x%llx + 0x%x > 0x%llx",
1951                       dio->range.logical_sector, bio_sectors(bio),
1952                       ic->provided_data_sectors);
1953                 return DM_MAPIO_KILL;
1954         }
1955         if (unlikely((dio->range.logical_sector | bio_sectors(bio)) & (unsigned int)(ic->sectors_per_block - 1))) {
1956                 DMERR("Bio not aligned on %u sectors: 0x%llx, 0x%x",
1957                       ic->sectors_per_block,
1958                       dio->range.logical_sector, bio_sectors(bio));
1959                 return DM_MAPIO_KILL;
1960         }
1961
1962         if (ic->sectors_per_block > 1 && likely(dio->op != REQ_OP_DISCARD)) {
1963                 struct bvec_iter iter;
1964                 struct bio_vec bv;
1965
1966                 bio_for_each_segment(bv, bio, iter) {
1967                         if (unlikely(bv.bv_len & ((ic->sectors_per_block << SECTOR_SHIFT) - 1))) {
1968                                 DMERR("Bio vector (%u,%u) is not aligned on %u-sector boundary",
1969                                         bv.bv_offset, bv.bv_len, ic->sectors_per_block);
1970                                 return DM_MAPIO_KILL;
1971                         }
1972                 }
1973         }
1974
1975         bip = bio_integrity(bio);
1976         if (!ic->internal_hash) {
1977                 if (bip) {
1978                         unsigned int wanted_tag_size = bio_sectors(bio) >> ic->sb->log2_sectors_per_block;
1979
1980                         if (ic->log2_tag_size >= 0)
1981                                 wanted_tag_size <<= ic->log2_tag_size;
1982                         else
1983                                 wanted_tag_size *= ic->tag_size;
1984                         if (unlikely(wanted_tag_size != bip->bip_iter.bi_size)) {
1985                                 DMERR("Invalid integrity data size %u, expected %u",
1986                                       bip->bip_iter.bi_size, wanted_tag_size);
1987                                 return DM_MAPIO_KILL;
1988                         }
1989                 }
1990         } else {
1991                 if (unlikely(bip != NULL)) {
1992                         DMERR("Unexpected integrity data when using internal hash");
1993                         return DM_MAPIO_KILL;
1994                 }
1995         }
1996
1997         if (unlikely(ic->mode == 'R') && unlikely(dio->op != REQ_OP_READ))
1998                 return DM_MAPIO_KILL;
1999
2000         get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
2001         dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset, &dio->metadata_offset);
2002         bio->bi_iter.bi_sector = get_data_sector(ic, area, offset);
2003
2004         dm_integrity_map_continue(dio, true);
2005         return DM_MAPIO_SUBMITTED;
2006 }
2007
2008 static bool __journal_read_write(struct dm_integrity_io *dio, struct bio *bio,
2009                                  unsigned int journal_section, unsigned int journal_entry)
2010 {
2011         struct dm_integrity_c *ic = dio->ic;
2012         sector_t logical_sector;
2013         unsigned int n_sectors;
2014
2015         logical_sector = dio->range.logical_sector;
2016         n_sectors = dio->range.n_sectors;
2017         do {
2018                 struct bio_vec bv = bio_iovec(bio);
2019                 char *mem;
2020
2021                 if (unlikely(bv.bv_len >> SECTOR_SHIFT > n_sectors))
2022                         bv.bv_len = n_sectors << SECTOR_SHIFT;
2023                 n_sectors -= bv.bv_len >> SECTOR_SHIFT;
2024                 bio_advance_iter(bio, &bio->bi_iter, bv.bv_len);
2025 retry_kmap:
2026                 mem = kmap_local_page(bv.bv_page);
2027                 if (likely(dio->op == REQ_OP_WRITE))
2028                         flush_dcache_page(bv.bv_page);
2029
2030                 do {
2031                         struct journal_entry *je = access_journal_entry(ic, journal_section, journal_entry);
2032
2033                         if (unlikely(dio->op == REQ_OP_READ)) {
2034                                 struct journal_sector *js;
2035                                 char *mem_ptr;
2036                                 unsigned int s;
2037
2038                                 if (unlikely(journal_entry_is_inprogress(je))) {
2039                                         flush_dcache_page(bv.bv_page);
2040                                         kunmap_local(mem);
2041
2042                                         __io_wait_event(ic->copy_to_journal_wait, !journal_entry_is_inprogress(je));
2043                                         goto retry_kmap;
2044                                 }
2045                                 smp_rmb();
2046                                 BUG_ON(journal_entry_get_sector(je) != logical_sector);
2047                                 js = access_journal_data(ic, journal_section, journal_entry);
2048                                 mem_ptr = mem + bv.bv_offset;
2049                                 s = 0;
2050                                 do {
2051                                         memcpy(mem_ptr, js, JOURNAL_SECTOR_DATA);
2052                                         *(commit_id_t *)(mem_ptr + JOURNAL_SECTOR_DATA) = je->last_bytes[s];
2053                                         js++;
2054                                         mem_ptr += 1 << SECTOR_SHIFT;
2055                                 } while (++s < ic->sectors_per_block);
2056 #ifdef INTERNAL_VERIFY
2057                                 if (ic->internal_hash) {
2058                                         char checksums_onstack[max_t(size_t, HASH_MAX_DIGESTSIZE, MAX_TAG_SIZE)];
2059
2060                                         integrity_sector_checksum(ic, logical_sector, mem + bv.bv_offset, checksums_onstack);
2061                                         if (unlikely(memcmp(checksums_onstack, journal_entry_tag(ic, je), ic->tag_size))) {
2062                                                 DMERR_LIMIT("Checksum failed when reading from journal, at sector 0x%llx",
2063                                                             logical_sector);
2064                                                 dm_audit_log_bio(DM_MSG_PREFIX, "journal-checksum",
2065                                                                  bio, logical_sector, 0);
2066                                         }
2067                                 }
2068 #endif
2069                         }
2070
2071                         if (!ic->internal_hash) {
2072                                 struct bio_integrity_payload *bip = bio_integrity(bio);
2073                                 unsigned int tag_todo = ic->tag_size;
2074                                 char *tag_ptr = journal_entry_tag(ic, je);
2075
2076                                 if (bip) {
2077                                         do {
2078                                                 struct bio_vec biv = bvec_iter_bvec(bip->bip_vec, bip->bip_iter);
2079                                                 unsigned int tag_now = min(biv.bv_len, tag_todo);
2080                                                 char *tag_addr;
2081
2082                                                 BUG_ON(PageHighMem(biv.bv_page));
2083                                                 tag_addr = bvec_virt(&biv);
2084                                                 if (likely(dio->op == REQ_OP_WRITE))
2085                                                         memcpy(tag_ptr, tag_addr, tag_now);
2086                                                 else
2087                                                         memcpy(tag_addr, tag_ptr, tag_now);
2088                                                 bvec_iter_advance(bip->bip_vec, &bip->bip_iter, tag_now);
2089                                                 tag_ptr += tag_now;
2090                                                 tag_todo -= tag_now;
2091                                         } while (unlikely(tag_todo));
2092                                 } else if (likely(dio->op == REQ_OP_WRITE))
2093                                         memset(tag_ptr, 0, tag_todo);
2094                         }
2095
2096                         if (likely(dio->op == REQ_OP_WRITE)) {
2097                                 struct journal_sector *js;
2098                                 unsigned int s;
2099
2100                                 js = access_journal_data(ic, journal_section, journal_entry);
2101                                 memcpy(js, mem + bv.bv_offset, ic->sectors_per_block << SECTOR_SHIFT);
2102
2103                                 s = 0;
2104                                 do {
2105                                         je->last_bytes[s] = js[s].commit_id;
2106                                 } while (++s < ic->sectors_per_block);
2107
2108                                 if (ic->internal_hash) {
2109                                         unsigned int digest_size = crypto_shash_digestsize(ic->internal_hash);
2110
2111                                         if (unlikely(digest_size > ic->tag_size)) {
2112                                                 char checksums_onstack[HASH_MAX_DIGESTSIZE];
2113
2114                                                 integrity_sector_checksum(ic, logical_sector, (char *)js, checksums_onstack);
2115                                                 memcpy(journal_entry_tag(ic, je), checksums_onstack, ic->tag_size);
2116                                         } else
2117                                                 integrity_sector_checksum(ic, logical_sector, (char *)js, journal_entry_tag(ic, je));
2118                                 }
2119
2120                                 journal_entry_set_sector(je, logical_sector);
2121                         }
2122                         logical_sector += ic->sectors_per_block;
2123
2124                         journal_entry++;
2125                         if (unlikely(journal_entry == ic->journal_section_entries)) {
2126                                 journal_entry = 0;
2127                                 journal_section++;
2128                                 wraparound_section(ic, &journal_section);
2129                         }
2130
2131                         bv.bv_offset += ic->sectors_per_block << SECTOR_SHIFT;
2132                 } while (bv.bv_len -= ic->sectors_per_block << SECTOR_SHIFT);
2133
2134                 if (unlikely(dio->op == REQ_OP_READ))
2135                         flush_dcache_page(bv.bv_page);
2136                 kunmap_local(mem);
2137         } while (n_sectors);
2138
2139         if (likely(dio->op == REQ_OP_WRITE)) {
2140                 smp_mb();
2141                 if (unlikely(waitqueue_active(&ic->copy_to_journal_wait)))
2142                         wake_up(&ic->copy_to_journal_wait);
2143                 if (READ_ONCE(ic->free_sectors) <= ic->free_sectors_threshold)
2144                         queue_work(ic->commit_wq, &ic->commit_work);
2145                 else
2146                         schedule_autocommit(ic);
2147         } else
2148                 remove_range(ic, &dio->range);
2149
2150         if (unlikely(bio->bi_iter.bi_size)) {
2151                 sector_t area, offset;
2152
2153                 dio->range.logical_sector = logical_sector;
2154                 get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
2155                 dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset, &dio->metadata_offset);
2156                 return true;
2157         }
2158
2159         return false;
2160 }
2161
2162 static void dm_integrity_map_continue(struct dm_integrity_io *dio, bool from_map)
2163 {
2164         struct dm_integrity_c *ic = dio->ic;
2165         struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
2166         unsigned int journal_section, journal_entry;
2167         unsigned int journal_read_pos;
2168         struct completion read_comp;
2169         bool discard_retried = false;
2170         bool need_sync_io = ic->internal_hash && dio->op == REQ_OP_READ;
2171
2172         if (unlikely(dio->op == REQ_OP_DISCARD) && ic->mode != 'D')
2173                 need_sync_io = true;
2174
2175         if (need_sync_io && from_map) {
2176                 INIT_WORK(&dio->work, integrity_bio_wait);
2177                 queue_work(ic->offload_wq, &dio->work);
2178                 return;
2179         }
2180
2181 lock_retry:
2182         spin_lock_irq(&ic->endio_wait.lock);
2183 retry:
2184         if (unlikely(dm_integrity_failed(ic))) {
2185                 spin_unlock_irq(&ic->endio_wait.lock);
2186                 do_endio(ic, bio);
2187                 return;
2188         }
2189         dio->range.n_sectors = bio_sectors(bio);
2190         journal_read_pos = NOT_FOUND;
2191         if (ic->mode == 'J' && likely(dio->op != REQ_OP_DISCARD)) {
2192                 if (dio->op == REQ_OP_WRITE) {
2193                         unsigned int next_entry, i, pos;
2194                         unsigned int ws, we, range_sectors;
2195
2196                         dio->range.n_sectors = min(dio->range.n_sectors,
2197                                                    (sector_t)ic->free_sectors << ic->sb->log2_sectors_per_block);
2198                         if (unlikely(!dio->range.n_sectors)) {
2199                                 if (from_map)
2200                                         goto offload_to_thread;
2201                                 sleep_on_endio_wait(ic);
2202                                 goto retry;
2203                         }
2204                         range_sectors = dio->range.n_sectors >> ic->sb->log2_sectors_per_block;
2205                         ic->free_sectors -= range_sectors;
2206                         journal_section = ic->free_section;
2207                         journal_entry = ic->free_section_entry;
2208
2209                         next_entry = ic->free_section_entry + range_sectors;
2210                         ic->free_section_entry = next_entry % ic->journal_section_entries;
2211                         ic->free_section += next_entry / ic->journal_section_entries;
2212                         ic->n_uncommitted_sections += next_entry / ic->journal_section_entries;
2213                         wraparound_section(ic, &ic->free_section);
2214
2215                         pos = journal_section * ic->journal_section_entries + journal_entry;
2216                         ws = journal_section;
2217                         we = journal_entry;
2218                         i = 0;
2219                         do {
2220                                 struct journal_entry *je;
2221
2222                                 add_journal_node(ic, &ic->journal_tree[pos], dio->range.logical_sector + i);
2223                                 pos++;
2224                                 if (unlikely(pos >= ic->journal_entries))
2225                                         pos = 0;
2226
2227                                 je = access_journal_entry(ic, ws, we);
2228                                 BUG_ON(!journal_entry_is_unused(je));
2229                                 journal_entry_set_inprogress(je);
2230                                 we++;
2231                                 if (unlikely(we == ic->journal_section_entries)) {
2232                                         we = 0;
2233                                         ws++;
2234                                         wraparound_section(ic, &ws);
2235                                 }
2236                         } while ((i += ic->sectors_per_block) < dio->range.n_sectors);
2237
2238                         spin_unlock_irq(&ic->endio_wait.lock);
2239                         goto journal_read_write;
2240                 } else {
2241                         sector_t next_sector;
2242
2243                         journal_read_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
2244                         if (likely(journal_read_pos == NOT_FOUND)) {
2245                                 if (unlikely(dio->range.n_sectors > next_sector - dio->range.logical_sector))
2246                                         dio->range.n_sectors = next_sector - dio->range.logical_sector;
2247                         } else {
2248                                 unsigned int i;
2249                                 unsigned int jp = journal_read_pos + 1;
2250
2251                                 for (i = ic->sectors_per_block; i < dio->range.n_sectors; i += ic->sectors_per_block, jp++) {
2252                                         if (!test_journal_node(ic, jp, dio->range.logical_sector + i))
2253                                                 break;
2254                                 }
2255                                 dio->range.n_sectors = i;
2256                         }
2257                 }
2258         }
2259         if (unlikely(!add_new_range(ic, &dio->range, true))) {
2260                 /*
2261                  * We must not sleep in the request routine because it could
2262                  * stall bios on current->bio_list.
2263                  * So, we offload the bio to a workqueue if we have to sleep.
2264                  */
2265                 if (from_map) {
2266 offload_to_thread:
2267                         spin_unlock_irq(&ic->endio_wait.lock);
2268                         INIT_WORK(&dio->work, integrity_bio_wait);
2269                         queue_work(ic->wait_wq, &dio->work);
2270                         return;
2271                 }
2272                 if (journal_read_pos != NOT_FOUND)
2273                         dio->range.n_sectors = ic->sectors_per_block;
2274                 wait_and_add_new_range(ic, &dio->range);
2275                 /*
2276                  * wait_and_add_new_range drops the spinlock, so the journal
2277                  * may have been changed arbitrarily. We need to recheck.
2278                  * To simplify the code, we restrict I/O size to just one block.
2279                  */
2280                 if (journal_read_pos != NOT_FOUND) {
2281                         sector_t next_sector;
2282                         unsigned int new_pos;
2283
2284                         new_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
2285                         if (unlikely(new_pos != journal_read_pos)) {
2286                                 remove_range_unlocked(ic, &dio->range);
2287                                 goto retry;
2288                         }
2289                 }
2290         }
2291         if (ic->mode == 'J' && likely(dio->op == REQ_OP_DISCARD) && !discard_retried) {
2292                 sector_t next_sector;
2293                 unsigned int new_pos;
2294
2295                 new_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
2296                 if (unlikely(new_pos != NOT_FOUND) ||
2297                     unlikely(next_sector < dio->range.logical_sector - dio->range.n_sectors)) {
2298                         remove_range_unlocked(ic, &dio->range);
2299                         spin_unlock_irq(&ic->endio_wait.lock);
2300                         queue_work(ic->commit_wq, &ic->commit_work);
2301                         flush_workqueue(ic->commit_wq);
2302                         queue_work(ic->writer_wq, &ic->writer_work);
2303                         flush_workqueue(ic->writer_wq);
2304                         discard_retried = true;
2305                         goto lock_retry;
2306                 }
2307         }
2308         spin_unlock_irq(&ic->endio_wait.lock);
2309
2310         if (unlikely(journal_read_pos != NOT_FOUND)) {
2311                 journal_section = journal_read_pos / ic->journal_section_entries;
2312                 journal_entry = journal_read_pos % ic->journal_section_entries;
2313                 goto journal_read_write;
2314         }
2315
2316         if (ic->mode == 'B' && (dio->op == REQ_OP_WRITE || unlikely(dio->op == REQ_OP_DISCARD))) {
2317                 if (!block_bitmap_op(ic, ic->may_write_bitmap, dio->range.logical_sector,
2318                                      dio->range.n_sectors, BITMAP_OP_TEST_ALL_SET)) {
2319                         struct bitmap_block_status *bbs;
2320
2321                         bbs = sector_to_bitmap_block(ic, dio->range.logical_sector);
2322                         spin_lock(&bbs->bio_queue_lock);
2323                         bio_list_add(&bbs->bio_queue, bio);
2324                         spin_unlock(&bbs->bio_queue_lock);
2325                         queue_work(ic->writer_wq, &bbs->work);
2326                         return;
2327                 }
2328         }
2329
2330         dio->in_flight = (atomic_t)ATOMIC_INIT(2);
2331
2332         if (need_sync_io) {
2333                 init_completion(&read_comp);
2334                 dio->completion = &read_comp;
2335         } else
2336                 dio->completion = NULL;
2337
2338         dm_bio_record(&dio->bio_details, bio);
2339         bio_set_dev(bio, ic->dev->bdev);
2340         bio->bi_integrity = NULL;
2341         bio->bi_opf &= ~REQ_INTEGRITY;
2342         bio->bi_end_io = integrity_end_io;
2343         bio->bi_iter.bi_size = dio->range.n_sectors << SECTOR_SHIFT;
2344
2345         if (unlikely(dio->op == REQ_OP_DISCARD) && likely(ic->mode != 'D')) {
2346                 integrity_metadata(&dio->work);
2347                 dm_integrity_flush_buffers(ic, false);
2348
2349                 dio->in_flight = (atomic_t)ATOMIC_INIT(1);
2350                 dio->completion = NULL;
2351
2352                 submit_bio_noacct(bio);
2353
2354                 return;
2355         }
2356
2357         submit_bio_noacct(bio);
2358
2359         if (need_sync_io) {
2360                 wait_for_completion_io(&read_comp);
2361                 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING) &&
2362                     dio->range.logical_sector + dio->range.n_sectors > le64_to_cpu(ic->sb->recalc_sector))
2363                         goto skip_check;
2364                 if (ic->mode == 'B') {
2365                         if (!block_bitmap_op(ic, ic->recalc_bitmap, dio->range.logical_sector,
2366                                              dio->range.n_sectors, BITMAP_OP_TEST_ALL_CLEAR))
2367                                 goto skip_check;
2368                 }
2369
2370                 if (likely(!bio->bi_status))
2371                         integrity_metadata(&dio->work);
2372                 else
2373 skip_check:
2374                         dec_in_flight(dio);
2375         } else {
2376                 INIT_WORK(&dio->work, integrity_metadata);
2377                 queue_work(ic->metadata_wq, &dio->work);
2378         }
2379
2380         return;
2381
2382 journal_read_write:
2383         if (unlikely(__journal_read_write(dio, bio, journal_section, journal_entry)))
2384                 goto lock_retry;
2385
2386         do_endio_flush(ic, dio);
2387 }
2388
2389
2390 static void integrity_bio_wait(struct work_struct *w)
2391 {
2392         struct dm_integrity_io *dio = container_of(w, struct dm_integrity_io, work);
2393
2394         dm_integrity_map_continue(dio, false);
2395 }
2396
2397 static void pad_uncommitted(struct dm_integrity_c *ic)
2398 {
2399         if (ic->free_section_entry) {
2400                 ic->free_sectors -= ic->journal_section_entries - ic->free_section_entry;
2401                 ic->free_section_entry = 0;
2402                 ic->free_section++;
2403                 wraparound_section(ic, &ic->free_section);
2404                 ic->n_uncommitted_sections++;
2405         }
2406         if (WARN_ON(ic->journal_sections * ic->journal_section_entries !=
2407                     (ic->n_uncommitted_sections + ic->n_committed_sections) *
2408                     ic->journal_section_entries + ic->free_sectors)) {
2409                 DMCRIT("journal_sections %u, journal_section_entries %u, "
2410                        "n_uncommitted_sections %u, n_committed_sections %u, "
2411                        "journal_section_entries %u, free_sectors %u",
2412                        ic->journal_sections, ic->journal_section_entries,
2413                        ic->n_uncommitted_sections, ic->n_committed_sections,
2414                        ic->journal_section_entries, ic->free_sectors);
2415         }
2416 }
2417
2418 static void integrity_commit(struct work_struct *w)
2419 {
2420         struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, commit_work);
2421         unsigned int commit_start, commit_sections;
2422         unsigned int i, j, n;
2423         struct bio *flushes;
2424
2425         del_timer(&ic->autocommit_timer);
2426
2427         spin_lock_irq(&ic->endio_wait.lock);
2428         flushes = bio_list_get(&ic->flush_bio_list);
2429         if (unlikely(ic->mode != 'J')) {
2430                 spin_unlock_irq(&ic->endio_wait.lock);
2431                 dm_integrity_flush_buffers(ic, true);
2432                 goto release_flush_bios;
2433         }
2434
2435         pad_uncommitted(ic);
2436         commit_start = ic->uncommitted_section;
2437         commit_sections = ic->n_uncommitted_sections;
2438         spin_unlock_irq(&ic->endio_wait.lock);
2439
2440         if (!commit_sections)
2441                 goto release_flush_bios;
2442
2443         ic->wrote_to_journal = true;
2444
2445         i = commit_start;
2446         for (n = 0; n < commit_sections; n++) {
2447                 for (j = 0; j < ic->journal_section_entries; j++) {
2448                         struct journal_entry *je;
2449
2450                         je = access_journal_entry(ic, i, j);
2451                         io_wait_event(ic->copy_to_journal_wait, !journal_entry_is_inprogress(je));
2452                 }
2453                 for (j = 0; j < ic->journal_section_sectors; j++) {
2454                         struct journal_sector *js;
2455
2456                         js = access_journal(ic, i, j);
2457                         js->commit_id = dm_integrity_commit_id(ic, i, j, ic->commit_seq);
2458                 }
2459                 i++;
2460                 if (unlikely(i >= ic->journal_sections))
2461                         ic->commit_seq = next_commit_seq(ic->commit_seq);
2462                 wraparound_section(ic, &i);
2463         }
2464         smp_rmb();
2465
2466         write_journal(ic, commit_start, commit_sections);
2467
2468         spin_lock_irq(&ic->endio_wait.lock);
2469         ic->uncommitted_section += commit_sections;
2470         wraparound_section(ic, &ic->uncommitted_section);
2471         ic->n_uncommitted_sections -= commit_sections;
2472         ic->n_committed_sections += commit_sections;
2473         spin_unlock_irq(&ic->endio_wait.lock);
2474
2475         if (READ_ONCE(ic->free_sectors) <= ic->free_sectors_threshold)
2476                 queue_work(ic->writer_wq, &ic->writer_work);
2477
2478 release_flush_bios:
2479         while (flushes) {
2480                 struct bio *next = flushes->bi_next;
2481
2482                 flushes->bi_next = NULL;
2483                 do_endio(ic, flushes);
2484                 flushes = next;
2485         }
2486 }
2487
2488 static void complete_copy_from_journal(unsigned long error, void *context)
2489 {
2490         struct journal_io *io = context;
2491         struct journal_completion *comp = io->comp;
2492         struct dm_integrity_c *ic = comp->ic;
2493
2494         remove_range(ic, &io->range);
2495         mempool_free(io, &ic->journal_io_mempool);
2496         if (unlikely(error != 0))
2497                 dm_integrity_io_error(ic, "copying from journal", -EIO);
2498         complete_journal_op(comp);
2499 }
2500
2501 static void restore_last_bytes(struct dm_integrity_c *ic, struct journal_sector *js,
2502                                struct journal_entry *je)
2503 {
2504         unsigned int s = 0;
2505
2506         do {
2507                 js->commit_id = je->last_bytes[s];
2508                 js++;
2509         } while (++s < ic->sectors_per_block);
2510 }
2511
2512 static void do_journal_write(struct dm_integrity_c *ic, unsigned int write_start,
2513                              unsigned int write_sections, bool from_replay)
2514 {
2515         unsigned int i, j, n;
2516         struct journal_completion comp;
2517         struct blk_plug plug;
2518
2519         blk_start_plug(&plug);
2520
2521         comp.ic = ic;
2522         comp.in_flight = (atomic_t)ATOMIC_INIT(1);
2523         init_completion(&comp.comp);
2524
2525         i = write_start;
2526         for (n = 0; n < write_sections; n++, i++, wraparound_section(ic, &i)) {
2527 #ifndef INTERNAL_VERIFY
2528                 if (unlikely(from_replay))
2529 #endif
2530                         rw_section_mac(ic, i, false);
2531                 for (j = 0; j < ic->journal_section_entries; j++) {
2532                         struct journal_entry *je = access_journal_entry(ic, i, j);
2533                         sector_t sec, area, offset;
2534                         unsigned int k, l, next_loop;
2535                         sector_t metadata_block;
2536                         unsigned int metadata_offset;
2537                         struct journal_io *io;
2538
2539                         if (journal_entry_is_unused(je))
2540                                 continue;
2541                         BUG_ON(unlikely(journal_entry_is_inprogress(je)) && !from_replay);
2542                         sec = journal_entry_get_sector(je);
2543                         if (unlikely(from_replay)) {
2544                                 if (unlikely(sec & (unsigned int)(ic->sectors_per_block - 1))) {
2545                                         dm_integrity_io_error(ic, "invalid sector in journal", -EIO);
2546                                         sec &= ~(sector_t)(ic->sectors_per_block - 1);
2547                                 }
2548                                 if (unlikely(sec >= ic->provided_data_sectors)) {
2549                                         journal_entry_set_unused(je);
2550                                         continue;
2551                                 }
2552                         }
2553                         get_area_and_offset(ic, sec, &area, &offset);
2554                         restore_last_bytes(ic, access_journal_data(ic, i, j), je);
2555                         for (k = j + 1; k < ic->journal_section_entries; k++) {
2556                                 struct journal_entry *je2 = access_journal_entry(ic, i, k);
2557                                 sector_t sec2, area2, offset2;
2558
2559                                 if (journal_entry_is_unused(je2))
2560                                         break;
2561                                 BUG_ON(unlikely(journal_entry_is_inprogress(je2)) && !from_replay);
2562                                 sec2 = journal_entry_get_sector(je2);
2563                                 if (unlikely(sec2 >= ic->provided_data_sectors))
2564                                         break;
2565                                 get_area_and_offset(ic, sec2, &area2, &offset2);
2566                                 if (area2 != area || offset2 != offset + ((k - j) << ic->sb->log2_sectors_per_block))
2567                                         break;
2568                                 restore_last_bytes(ic, access_journal_data(ic, i, k), je2);
2569                         }
2570                         next_loop = k - 1;
2571
2572                         io = mempool_alloc(&ic->journal_io_mempool, GFP_NOIO);
2573                         io->comp = &comp;
2574                         io->range.logical_sector = sec;
2575                         io->range.n_sectors = (k - j) << ic->sb->log2_sectors_per_block;
2576
2577                         spin_lock_irq(&ic->endio_wait.lock);
2578                         add_new_range_and_wait(ic, &io->range);
2579
2580                         if (likely(!from_replay)) {
2581                                 struct journal_node *section_node = &ic->journal_tree[i * ic->journal_section_entries];
2582
2583                                 /* don't write if there is newer committed sector */
2584                                 while (j < k && find_newer_committed_node(ic, &section_node[j])) {
2585                                         struct journal_entry *je2 = access_journal_entry(ic, i, j);
2586
2587                                         journal_entry_set_unused(je2);
2588                                         remove_journal_node(ic, &section_node[j]);
2589                                         j++;
2590                                         sec += ic->sectors_per_block;
2591                                         offset += ic->sectors_per_block;
2592                                 }
2593                                 while (j < k && find_newer_committed_node(ic, &section_node[k - 1])) {
2594                                         struct journal_entry *je2 = access_journal_entry(ic, i, k - 1);
2595
2596                                         journal_entry_set_unused(je2);
2597                                         remove_journal_node(ic, &section_node[k - 1]);
2598                                         k--;
2599                                 }
2600                                 if (j == k) {
2601                                         remove_range_unlocked(ic, &io->range);
2602                                         spin_unlock_irq(&ic->endio_wait.lock);
2603                                         mempool_free(io, &ic->journal_io_mempool);
2604                                         goto skip_io;
2605                                 }
2606                                 for (l = j; l < k; l++)
2607                                         remove_journal_node(ic, &section_node[l]);
2608                         }
2609                         spin_unlock_irq(&ic->endio_wait.lock);
2610
2611                         metadata_block = get_metadata_sector_and_offset(ic, area, offset, &metadata_offset);
2612                         for (l = j; l < k; l++) {
2613                                 int r;
2614                                 struct journal_entry *je2 = access_journal_entry(ic, i, l);
2615
2616                                 if (
2617 #ifndef INTERNAL_VERIFY
2618                                     unlikely(from_replay) &&
2619 #endif
2620                                     ic->internal_hash) {
2621                                         char test_tag[max_t(size_t, HASH_MAX_DIGESTSIZE, MAX_TAG_SIZE)];
2622
2623                                         integrity_sector_checksum(ic, sec + ((l - j) << ic->sb->log2_sectors_per_block),
2624                                                                   (char *)access_journal_data(ic, i, l), test_tag);
2625                                         if (unlikely(memcmp(test_tag, journal_entry_tag(ic, je2), ic->tag_size))) {
2626                                                 dm_integrity_io_error(ic, "tag mismatch when replaying journal", -EILSEQ);
2627                                                 dm_audit_log_target(DM_MSG_PREFIX, "integrity-replay-journal", ic->ti, 0);
2628                                         }
2629                                 }
2630
2631                                 journal_entry_set_unused(je2);
2632                                 r = dm_integrity_rw_tag(ic, journal_entry_tag(ic, je2), &metadata_block, &metadata_offset,
2633                                                         ic->tag_size, TAG_WRITE);
2634                                 if (unlikely(r))
2635                                         dm_integrity_io_error(ic, "reading tags", r);
2636                         }
2637
2638                         atomic_inc(&comp.in_flight);
2639                         copy_from_journal(ic, i, j << ic->sb->log2_sectors_per_block,
2640                                           (k - j) << ic->sb->log2_sectors_per_block,
2641                                           get_data_sector(ic, area, offset),
2642                                           complete_copy_from_journal, io);
2643 skip_io:
2644                         j = next_loop;
2645                 }
2646         }
2647
2648         dm_bufio_write_dirty_buffers_async(ic->bufio);
2649
2650         blk_finish_plug(&plug);
2651
2652         complete_journal_op(&comp);
2653         wait_for_completion_io(&comp.comp);
2654
2655         dm_integrity_flush_buffers(ic, true);
2656 }
2657
2658 static void integrity_writer(struct work_struct *w)
2659 {
2660         struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, writer_work);
2661         unsigned int write_start, write_sections;
2662         unsigned int prev_free_sectors;
2663
2664         spin_lock_irq(&ic->endio_wait.lock);
2665         write_start = ic->committed_section;
2666         write_sections = ic->n_committed_sections;
2667         spin_unlock_irq(&ic->endio_wait.lock);
2668
2669         if (!write_sections)
2670                 return;
2671
2672         do_journal_write(ic, write_start, write_sections, false);
2673
2674         spin_lock_irq(&ic->endio_wait.lock);
2675
2676         ic->committed_section += write_sections;
2677         wraparound_section(ic, &ic->committed_section);
2678         ic->n_committed_sections -= write_sections;
2679
2680         prev_free_sectors = ic->free_sectors;
2681         ic->free_sectors += write_sections * ic->journal_section_entries;
2682         if (unlikely(!prev_free_sectors))
2683                 wake_up_locked(&ic->endio_wait);
2684
2685         spin_unlock_irq(&ic->endio_wait.lock);
2686 }
2687
2688 static void recalc_write_super(struct dm_integrity_c *ic)
2689 {
2690         int r;
2691
2692         dm_integrity_flush_buffers(ic, false);
2693         if (dm_integrity_failed(ic))
2694                 return;
2695
2696         r = sync_rw_sb(ic, REQ_OP_WRITE);
2697         if (unlikely(r))
2698                 dm_integrity_io_error(ic, "writing superblock", r);
2699 }
2700
2701 static void integrity_recalc(struct work_struct *w)
2702 {
2703         struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, recalc_work);
2704         size_t recalc_tags_size;
2705         u8 *recalc_buffer = NULL;
2706         u8 *recalc_tags = NULL;
2707         struct dm_integrity_range range;
2708         struct dm_io_request io_req;
2709         struct dm_io_region io_loc;
2710         sector_t area, offset;
2711         sector_t metadata_block;
2712         unsigned int metadata_offset;
2713         sector_t logical_sector, n_sectors;
2714         __u8 *t;
2715         unsigned int i;
2716         int r;
2717         unsigned int super_counter = 0;
2718         unsigned recalc_sectors = RECALC_SECTORS;
2719
2720 retry:
2721         recalc_buffer = __vmalloc(recalc_sectors << SECTOR_SHIFT, GFP_NOIO);
2722         if (!recalc_buffer) {
2723 oom:
2724                 recalc_sectors >>= 1;
2725                 if (recalc_sectors >= 1U << ic->sb->log2_sectors_per_block)
2726                         goto retry;
2727                 DMCRIT("out of memory for recalculate buffer - recalculation disabled");
2728                 goto free_ret;
2729         }
2730         recalc_tags_size = (recalc_sectors >> ic->sb->log2_sectors_per_block) * ic->tag_size;
2731         if (crypto_shash_digestsize(ic->internal_hash) > ic->tag_size)
2732                 recalc_tags_size += crypto_shash_digestsize(ic->internal_hash) - ic->tag_size;
2733         recalc_tags = kvmalloc(recalc_tags_size, GFP_NOIO);
2734         if (!recalc_tags) {
2735                 vfree(recalc_buffer);
2736                 recalc_buffer = NULL;
2737                 goto oom;
2738         }
2739
2740         DEBUG_print("start recalculation... (position %llx)\n", le64_to_cpu(ic->sb->recalc_sector));
2741
2742         spin_lock_irq(&ic->endio_wait.lock);
2743
2744 next_chunk:
2745
2746         if (unlikely(dm_post_suspending(ic->ti)))
2747                 goto unlock_ret;
2748
2749         range.logical_sector = le64_to_cpu(ic->sb->recalc_sector);
2750         if (unlikely(range.logical_sector >= ic->provided_data_sectors)) {
2751                 if (ic->mode == 'B') {
2752                         block_bitmap_op(ic, ic->recalc_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_CLEAR);
2753                         DEBUG_print("queue_delayed_work: bitmap_flush_work\n");
2754                         queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, 0);
2755                 }
2756                 goto unlock_ret;
2757         }
2758
2759         get_area_and_offset(ic, range.logical_sector, &area, &offset);
2760         range.n_sectors = min((sector_t)recalc_sectors, ic->provided_data_sectors - range.logical_sector);
2761         if (!ic->meta_dev)
2762                 range.n_sectors = min(range.n_sectors, ((sector_t)1U << ic->sb->log2_interleave_sectors) - (unsigned int)offset);
2763
2764         add_new_range_and_wait(ic, &range);
2765         spin_unlock_irq(&ic->endio_wait.lock);
2766         logical_sector = range.logical_sector;
2767         n_sectors = range.n_sectors;
2768
2769         if (ic->mode == 'B') {
2770                 if (block_bitmap_op(ic, ic->recalc_bitmap, logical_sector, n_sectors, BITMAP_OP_TEST_ALL_CLEAR))
2771                         goto advance_and_next;
2772
2773                 while (block_bitmap_op(ic, ic->recalc_bitmap, logical_sector,
2774                                        ic->sectors_per_block, BITMAP_OP_TEST_ALL_CLEAR)) {
2775                         logical_sector += ic->sectors_per_block;
2776                         n_sectors -= ic->sectors_per_block;
2777                         cond_resched();
2778                 }
2779                 while (block_bitmap_op(ic, ic->recalc_bitmap, logical_sector + n_sectors - ic->sectors_per_block,
2780                                        ic->sectors_per_block, BITMAP_OP_TEST_ALL_CLEAR)) {
2781                         n_sectors -= ic->sectors_per_block;
2782                         cond_resched();
2783                 }
2784                 get_area_and_offset(ic, logical_sector, &area, &offset);
2785         }
2786
2787         DEBUG_print("recalculating: %llx, %llx\n", logical_sector, n_sectors);
2788
2789         if (unlikely(++super_counter == RECALC_WRITE_SUPER)) {
2790                 recalc_write_super(ic);
2791                 if (ic->mode == 'B')
2792                         queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, ic->bitmap_flush_interval);
2793
2794                 super_counter = 0;
2795         }
2796
2797         if (unlikely(dm_integrity_failed(ic)))
2798                 goto err;
2799
2800         io_req.bi_opf = REQ_OP_READ;
2801         io_req.mem.type = DM_IO_VMA;
2802         io_req.mem.ptr.addr = recalc_buffer;
2803         io_req.notify.fn = NULL;
2804         io_req.client = ic->io;
2805         io_loc.bdev = ic->dev->bdev;
2806         io_loc.sector = get_data_sector(ic, area, offset);
2807         io_loc.count = n_sectors;
2808
2809         r = dm_io(&io_req, 1, &io_loc, NULL);
2810         if (unlikely(r)) {
2811                 dm_integrity_io_error(ic, "reading data", r);
2812                 goto err;
2813         }
2814
2815         t = recalc_tags;
2816         for (i = 0; i < n_sectors; i += ic->sectors_per_block) {
2817                 integrity_sector_checksum(ic, logical_sector + i, recalc_buffer + (i << SECTOR_SHIFT), t);
2818                 t += ic->tag_size;
2819         }
2820
2821         metadata_block = get_metadata_sector_and_offset(ic, area, offset, &metadata_offset);
2822
2823         r = dm_integrity_rw_tag(ic, recalc_tags, &metadata_block, &metadata_offset, t - recalc_tags, TAG_WRITE);
2824         if (unlikely(r)) {
2825                 dm_integrity_io_error(ic, "writing tags", r);
2826                 goto err;
2827         }
2828
2829         if (ic->mode == 'B') {
2830                 sector_t start, end;
2831
2832                 start = (range.logical_sector >>
2833                          (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit)) <<
2834                         (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
2835                 end = ((range.logical_sector + range.n_sectors) >>
2836                        (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit)) <<
2837                         (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
2838                 block_bitmap_op(ic, ic->recalc_bitmap, start, end - start, BITMAP_OP_CLEAR);
2839         }
2840
2841 advance_and_next:
2842         cond_resched();
2843
2844         spin_lock_irq(&ic->endio_wait.lock);
2845         remove_range_unlocked(ic, &range);
2846         ic->sb->recalc_sector = cpu_to_le64(range.logical_sector + range.n_sectors);
2847         goto next_chunk;
2848
2849 err:
2850         remove_range(ic, &range);
2851         goto free_ret;
2852
2853 unlock_ret:
2854         spin_unlock_irq(&ic->endio_wait.lock);
2855
2856         recalc_write_super(ic);
2857
2858 free_ret:
2859         vfree(recalc_buffer);
2860         kvfree(recalc_tags);
2861 }
2862
2863 static void bitmap_block_work(struct work_struct *w)
2864 {
2865         struct bitmap_block_status *bbs = container_of(w, struct bitmap_block_status, work);
2866         struct dm_integrity_c *ic = bbs->ic;
2867         struct bio *bio;
2868         struct bio_list bio_queue;
2869         struct bio_list waiting;
2870
2871         bio_list_init(&waiting);
2872
2873         spin_lock(&bbs->bio_queue_lock);
2874         bio_queue = bbs->bio_queue;
2875         bio_list_init(&bbs->bio_queue);
2876         spin_unlock(&bbs->bio_queue_lock);
2877
2878         while ((bio = bio_list_pop(&bio_queue))) {
2879                 struct dm_integrity_io *dio;
2880
2881                 dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
2882
2883                 if (block_bitmap_op(ic, ic->may_write_bitmap, dio->range.logical_sector,
2884                                     dio->range.n_sectors, BITMAP_OP_TEST_ALL_SET)) {
2885                         remove_range(ic, &dio->range);
2886                         INIT_WORK(&dio->work, integrity_bio_wait);
2887                         queue_work(ic->offload_wq, &dio->work);
2888                 } else {
2889                         block_bitmap_op(ic, ic->journal, dio->range.logical_sector,
2890                                         dio->range.n_sectors, BITMAP_OP_SET);
2891                         bio_list_add(&waiting, bio);
2892                 }
2893         }
2894
2895         if (bio_list_empty(&waiting))
2896                 return;
2897
2898         rw_journal_sectors(ic, REQ_OP_WRITE | REQ_FUA | REQ_SYNC,
2899                            bbs->idx * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT),
2900                            BITMAP_BLOCK_SIZE >> SECTOR_SHIFT, NULL);
2901
2902         while ((bio = bio_list_pop(&waiting))) {
2903                 struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
2904
2905                 block_bitmap_op(ic, ic->may_write_bitmap, dio->range.logical_sector,
2906                                 dio->range.n_sectors, BITMAP_OP_SET);
2907
2908                 remove_range(ic, &dio->range);
2909                 INIT_WORK(&dio->work, integrity_bio_wait);
2910                 queue_work(ic->offload_wq, &dio->work);
2911         }
2912
2913         queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, ic->bitmap_flush_interval);
2914 }
2915
2916 static void bitmap_flush_work(struct work_struct *work)
2917 {
2918         struct dm_integrity_c *ic = container_of(work, struct dm_integrity_c, bitmap_flush_work.work);
2919         struct dm_integrity_range range;
2920         unsigned long limit;
2921         struct bio *bio;
2922
2923         dm_integrity_flush_buffers(ic, false);
2924
2925         range.logical_sector = 0;
2926         range.n_sectors = ic->provided_data_sectors;
2927
2928         spin_lock_irq(&ic->endio_wait.lock);
2929         add_new_range_and_wait(ic, &range);
2930         spin_unlock_irq(&ic->endio_wait.lock);
2931
2932         dm_integrity_flush_buffers(ic, true);
2933
2934         limit = ic->provided_data_sectors;
2935         if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
2936                 limit = le64_to_cpu(ic->sb->recalc_sector)
2937                         >> (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit)
2938                         << (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
2939         }
2940         /*DEBUG_print("zeroing journal\n");*/
2941         block_bitmap_op(ic, ic->journal, 0, limit, BITMAP_OP_CLEAR);
2942         block_bitmap_op(ic, ic->may_write_bitmap, 0, limit, BITMAP_OP_CLEAR);
2943
2944         rw_journal_sectors(ic, REQ_OP_WRITE | REQ_FUA | REQ_SYNC, 0,
2945                            ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
2946
2947         spin_lock_irq(&ic->endio_wait.lock);
2948         remove_range_unlocked(ic, &range);
2949         while (unlikely((bio = bio_list_pop(&ic->synchronous_bios)) != NULL)) {
2950                 bio_endio(bio);
2951                 spin_unlock_irq(&ic->endio_wait.lock);
2952                 spin_lock_irq(&ic->endio_wait.lock);
2953         }
2954         spin_unlock_irq(&ic->endio_wait.lock);
2955 }
2956
2957
2958 static void init_journal(struct dm_integrity_c *ic, unsigned int start_section,
2959                          unsigned int n_sections, unsigned char commit_seq)
2960 {
2961         unsigned int i, j, n;
2962
2963         if (!n_sections)
2964                 return;
2965
2966         for (n = 0; n < n_sections; n++) {
2967                 i = start_section + n;
2968                 wraparound_section(ic, &i);
2969                 for (j = 0; j < ic->journal_section_sectors; j++) {
2970                         struct journal_sector *js = access_journal(ic, i, j);
2971
2972                         BUILD_BUG_ON(sizeof(js->sectors) != JOURNAL_SECTOR_DATA);
2973                         memset(&js->sectors, 0, sizeof(js->sectors));
2974                         js->commit_id = dm_integrity_commit_id(ic, i, j, commit_seq);
2975                 }
2976                 for (j = 0; j < ic->journal_section_entries; j++) {
2977                         struct journal_entry *je = access_journal_entry(ic, i, j);
2978
2979                         journal_entry_set_unused(je);
2980                 }
2981         }
2982
2983         write_journal(ic, start_section, n_sections);
2984 }
2985
2986 static int find_commit_seq(struct dm_integrity_c *ic, unsigned int i, unsigned int j, commit_id_t id)
2987 {
2988         unsigned char k;
2989
2990         for (k = 0; k < N_COMMIT_IDS; k++) {
2991                 if (dm_integrity_commit_id(ic, i, j, k) == id)
2992                         return k;
2993         }
2994         dm_integrity_io_error(ic, "journal commit id", -EIO);
2995         return -EIO;
2996 }
2997
2998 static void replay_journal(struct dm_integrity_c *ic)
2999 {
3000         unsigned int i, j;
3001         bool used_commit_ids[N_COMMIT_IDS];
3002         unsigned int max_commit_id_sections[N_COMMIT_IDS];
3003         unsigned int write_start, write_sections;
3004         unsigned int continue_section;
3005         bool journal_empty;
3006         unsigned char unused, last_used, want_commit_seq;
3007
3008         if (ic->mode == 'R')
3009                 return;
3010
3011         if (ic->journal_uptodate)
3012                 return;
3013
3014         last_used = 0;
3015         write_start = 0;
3016
3017         if (!ic->just_formatted) {
3018                 DEBUG_print("reading journal\n");
3019                 rw_journal(ic, REQ_OP_READ, 0, ic->journal_sections, NULL);
3020                 if (ic->journal_io)
3021                         DEBUG_bytes(lowmem_page_address(ic->journal_io[0].page), 64, "read journal");
3022                 if (ic->journal_io) {
3023                         struct journal_completion crypt_comp;
3024
3025                         crypt_comp.ic = ic;
3026                         init_completion(&crypt_comp.comp);
3027                         crypt_comp.in_flight = (atomic_t)ATOMIC_INIT(0);
3028                         encrypt_journal(ic, false, 0, ic->journal_sections, &crypt_comp);
3029                         wait_for_completion(&crypt_comp.comp);
3030                 }
3031                 DEBUG_bytes(lowmem_page_address(ic->journal[0].page), 64, "decrypted journal");
3032         }
3033
3034         if (dm_integrity_failed(ic))
3035                 goto clear_journal;
3036
3037         journal_empty = true;
3038         memset(used_commit_ids, 0, sizeof(used_commit_ids));
3039         memset(max_commit_id_sections, 0, sizeof(max_commit_id_sections));
3040         for (i = 0; i < ic->journal_sections; i++) {
3041                 for (j = 0; j < ic->journal_section_sectors; j++) {
3042                         int k;
3043                         struct journal_sector *js = access_journal(ic, i, j);
3044
3045                         k = find_commit_seq(ic, i, j, js->commit_id);
3046                         if (k < 0)
3047                                 goto clear_journal;
3048                         used_commit_ids[k] = true;
3049                         max_commit_id_sections[k] = i;
3050                 }
3051                 if (journal_empty) {
3052                         for (j = 0; j < ic->journal_section_entries; j++) {
3053                                 struct journal_entry *je = access_journal_entry(ic, i, j);
3054
3055                                 if (!journal_entry_is_unused(je)) {
3056                                         journal_empty = false;
3057                                         break;
3058                                 }
3059                         }
3060                 }
3061         }
3062
3063         if (!used_commit_ids[N_COMMIT_IDS - 1]) {
3064                 unused = N_COMMIT_IDS - 1;
3065                 while (unused && !used_commit_ids[unused - 1])
3066                         unused--;
3067         } else {
3068                 for (unused = 0; unused < N_COMMIT_IDS; unused++)
3069                         if (!used_commit_ids[unused])
3070                                 break;
3071                 if (unused == N_COMMIT_IDS) {
3072                         dm_integrity_io_error(ic, "journal commit ids", -EIO);
3073                         goto clear_journal;
3074                 }
3075         }
3076         DEBUG_print("first unused commit seq %d [%d,%d,%d,%d]\n",
3077                     unused, used_commit_ids[0], used_commit_ids[1],
3078                     used_commit_ids[2], used_commit_ids[3]);
3079
3080         last_used = prev_commit_seq(unused);
3081         want_commit_seq = prev_commit_seq(last_used);
3082
3083         if (!used_commit_ids[want_commit_seq] && used_commit_ids[prev_commit_seq(want_commit_seq)])
3084                 journal_empty = true;
3085
3086         write_start = max_commit_id_sections[last_used] + 1;
3087         if (unlikely(write_start >= ic->journal_sections))
3088                 want_commit_seq = next_commit_seq(want_commit_seq);
3089         wraparound_section(ic, &write_start);
3090
3091         i = write_start;
3092         for (write_sections = 0; write_sections < ic->journal_sections; write_sections++) {
3093                 for (j = 0; j < ic->journal_section_sectors; j++) {
3094                         struct journal_sector *js = access_journal(ic, i, j);
3095
3096                         if (js->commit_id != dm_integrity_commit_id(ic, i, j, want_commit_seq)) {
3097                                 /*
3098                                  * This could be caused by crash during writing.
3099                                  * We won't replay the inconsistent part of the
3100                                  * journal.
3101                                  */
3102                                 DEBUG_print("commit id mismatch at position (%u, %u): %d != %d\n",
3103                                             i, j, find_commit_seq(ic, i, j, js->commit_id), want_commit_seq);
3104                                 goto brk;
3105                         }
3106                 }
3107                 i++;
3108                 if (unlikely(i >= ic->journal_sections))
3109                         want_commit_seq = next_commit_seq(want_commit_seq);
3110                 wraparound_section(ic, &i);
3111         }
3112 brk:
3113
3114         if (!journal_empty) {
3115                 DEBUG_print("replaying %u sections, starting at %u, commit seq %d\n",
3116                             write_sections, write_start, want_commit_seq);
3117                 do_journal_write(ic, write_start, write_sections, true);
3118         }
3119
3120         if (write_sections == ic->journal_sections && (ic->mode == 'J' || journal_empty)) {
3121                 continue_section = write_start;
3122                 ic->commit_seq = want_commit_seq;
3123                 DEBUG_print("continuing from section %u, commit seq %d\n", write_start, ic->commit_seq);
3124         } else {
3125                 unsigned int s;
3126                 unsigned char erase_seq;
3127
3128 clear_journal:
3129                 DEBUG_print("clearing journal\n");
3130
3131                 erase_seq = prev_commit_seq(prev_commit_seq(last_used));
3132                 s = write_start;
3133                 init_journal(ic, s, 1, erase_seq);
3134                 s++;
3135                 wraparound_section(ic, &s);
3136                 if (ic->journal_sections >= 2) {
3137                         init_journal(ic, s, ic->journal_sections - 2, erase_seq);
3138                         s += ic->journal_sections - 2;
3139                         wraparound_section(ic, &s);
3140                         init_journal(ic, s, 1, erase_seq);
3141                 }
3142
3143                 continue_section = 0;
3144                 ic->commit_seq = next_commit_seq(erase_seq);
3145         }
3146
3147         ic->committed_section = continue_section;
3148         ic->n_committed_sections = 0;
3149
3150         ic->uncommitted_section = continue_section;
3151         ic->n_uncommitted_sections = 0;
3152
3153         ic->free_section = continue_section;
3154         ic->free_section_entry = 0;
3155         ic->free_sectors = ic->journal_entries;
3156
3157         ic->journal_tree_root = RB_ROOT;
3158         for (i = 0; i < ic->journal_entries; i++)
3159                 init_journal_node(&ic->journal_tree[i]);
3160 }
3161
3162 static void dm_integrity_enter_synchronous_mode(struct dm_integrity_c *ic)
3163 {
3164         DEBUG_print("%s\n", __func__);
3165
3166         if (ic->mode == 'B') {
3167                 ic->bitmap_flush_interval = msecs_to_jiffies(10) + 1;
3168                 ic->synchronous_mode = 1;
3169
3170                 cancel_delayed_work_sync(&ic->bitmap_flush_work);
3171                 queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, 0);
3172                 flush_workqueue(ic->commit_wq);
3173         }
3174 }
3175
3176 static int dm_integrity_reboot(struct notifier_block *n, unsigned long code, void *x)
3177 {
3178         struct dm_integrity_c *ic = container_of(n, struct dm_integrity_c, reboot_notifier);
3179
3180         DEBUG_print("%s\n", __func__);
3181
3182         dm_integrity_enter_synchronous_mode(ic);
3183
3184         return NOTIFY_DONE;
3185 }
3186
3187 static void dm_integrity_postsuspend(struct dm_target *ti)
3188 {
3189         struct dm_integrity_c *ic = ti->private;
3190         int r;
3191
3192         WARN_ON(unregister_reboot_notifier(&ic->reboot_notifier));
3193
3194         del_timer_sync(&ic->autocommit_timer);
3195
3196         if (ic->recalc_wq)
3197                 drain_workqueue(ic->recalc_wq);
3198
3199         if (ic->mode == 'B')
3200                 cancel_delayed_work_sync(&ic->bitmap_flush_work);
3201
3202         queue_work(ic->commit_wq, &ic->commit_work);
3203         drain_workqueue(ic->commit_wq);
3204
3205         if (ic->mode == 'J') {
3206                 queue_work(ic->writer_wq, &ic->writer_work);
3207                 drain_workqueue(ic->writer_wq);
3208                 dm_integrity_flush_buffers(ic, true);
3209                 if (ic->wrote_to_journal) {
3210                         init_journal(ic, ic->free_section,
3211                                      ic->journal_sections - ic->free_section, ic->commit_seq);
3212                         if (ic->free_section) {
3213                                 init_journal(ic, 0, ic->free_section,
3214                                              next_commit_seq(ic->commit_seq));
3215                         }
3216                 }
3217         }
3218
3219         if (ic->mode == 'B') {
3220                 dm_integrity_flush_buffers(ic, true);
3221 #if 1
3222                 /* set to 0 to test bitmap replay code */
3223                 init_journal(ic, 0, ic->journal_sections, 0);
3224                 ic->sb->flags &= ~cpu_to_le32(SB_FLAG_DIRTY_BITMAP);
3225                 r = sync_rw_sb(ic, REQ_OP_WRITE | REQ_FUA);
3226                 if (unlikely(r))
3227                         dm_integrity_io_error(ic, "writing superblock", r);
3228 #endif
3229         }
3230
3231         BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
3232
3233         ic->journal_uptodate = true;
3234 }
3235
3236 static void dm_integrity_resume(struct dm_target *ti)
3237 {
3238         struct dm_integrity_c *ic = ti->private;
3239         __u64 old_provided_data_sectors = le64_to_cpu(ic->sb->provided_data_sectors);
3240         int r;
3241
3242         DEBUG_print("resume\n");
3243
3244         ic->wrote_to_journal = false;
3245
3246         if (ic->provided_data_sectors != old_provided_data_sectors) {
3247                 if (ic->provided_data_sectors > old_provided_data_sectors &&
3248                     ic->mode == 'B' &&
3249                     ic->sb->log2_blocks_per_bitmap_bit == ic->log2_blocks_per_bitmap_bit) {
3250                         rw_journal_sectors(ic, REQ_OP_READ, 0,
3251                                            ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3252                         block_bitmap_op(ic, ic->journal, old_provided_data_sectors,
3253                                         ic->provided_data_sectors - old_provided_data_sectors, BITMAP_OP_SET);
3254                         rw_journal_sectors(ic, REQ_OP_WRITE | REQ_FUA | REQ_SYNC, 0,
3255                                            ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3256                 }
3257
3258                 ic->sb->provided_data_sectors = cpu_to_le64(ic->provided_data_sectors);
3259                 r = sync_rw_sb(ic, REQ_OP_WRITE | REQ_FUA);
3260                 if (unlikely(r))
3261                         dm_integrity_io_error(ic, "writing superblock", r);
3262         }
3263
3264         if (ic->sb->flags & cpu_to_le32(SB_FLAG_DIRTY_BITMAP)) {
3265                 DEBUG_print("resume dirty_bitmap\n");
3266                 rw_journal_sectors(ic, REQ_OP_READ, 0,
3267                                    ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3268                 if (ic->mode == 'B') {
3269                         if (ic->sb->log2_blocks_per_bitmap_bit == ic->log2_blocks_per_bitmap_bit &&
3270                             !ic->reset_recalculate_flag) {
3271                                 block_bitmap_copy(ic, ic->recalc_bitmap, ic->journal);
3272                                 block_bitmap_copy(ic, ic->may_write_bitmap, ic->journal);
3273                                 if (!block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors,
3274                                                      BITMAP_OP_TEST_ALL_CLEAR)) {
3275                                         ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
3276                                         ic->sb->recalc_sector = cpu_to_le64(0);
3277                                 }
3278                         } else {
3279                                 DEBUG_print("non-matching blocks_per_bitmap_bit: %u, %u\n",
3280                                             ic->sb->log2_blocks_per_bitmap_bit, ic->log2_blocks_per_bitmap_bit);
3281                                 ic->sb->log2_blocks_per_bitmap_bit = ic->log2_blocks_per_bitmap_bit;
3282                                 block_bitmap_op(ic, ic->recalc_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_SET);
3283                                 block_bitmap_op(ic, ic->may_write_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_SET);
3284                                 block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors, BITMAP_OP_SET);
3285                                 rw_journal_sectors(ic, REQ_OP_WRITE | REQ_FUA | REQ_SYNC, 0,
3286                                                    ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3287                                 ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
3288                                 ic->sb->recalc_sector = cpu_to_le64(0);
3289                         }
3290                 } else {
3291                         if (!(ic->sb->log2_blocks_per_bitmap_bit == ic->log2_blocks_per_bitmap_bit &&
3292                               block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors, BITMAP_OP_TEST_ALL_CLEAR)) ||
3293                             ic->reset_recalculate_flag) {
3294                                 ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
3295                                 ic->sb->recalc_sector = cpu_to_le64(0);
3296                         }
3297                         init_journal(ic, 0, ic->journal_sections, 0);
3298                         replay_journal(ic);
3299                         ic->sb->flags &= ~cpu_to_le32(SB_FLAG_DIRTY_BITMAP);
3300                 }
3301                 r = sync_rw_sb(ic, REQ_OP_WRITE | REQ_FUA);
3302                 if (unlikely(r))
3303                         dm_integrity_io_error(ic, "writing superblock", r);
3304         } else {
3305                 replay_journal(ic);
3306                 if (ic->reset_recalculate_flag) {
3307                         ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
3308                         ic->sb->recalc_sector = cpu_to_le64(0);
3309                 }
3310                 if (ic->mode == 'B') {
3311                         ic->sb->flags |= cpu_to_le32(SB_FLAG_DIRTY_BITMAP);
3312                         ic->sb->log2_blocks_per_bitmap_bit = ic->log2_blocks_per_bitmap_bit;
3313                         r = sync_rw_sb(ic, REQ_OP_WRITE | REQ_FUA);
3314                         if (unlikely(r))
3315                                 dm_integrity_io_error(ic, "writing superblock", r);
3316
3317                         block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors, BITMAP_OP_CLEAR);
3318                         block_bitmap_op(ic, ic->recalc_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_CLEAR);
3319                         block_bitmap_op(ic, ic->may_write_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_CLEAR);
3320                         if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING) &&
3321                             le64_to_cpu(ic->sb->recalc_sector) < ic->provided_data_sectors) {
3322                                 block_bitmap_op(ic, ic->journal, le64_to_cpu(ic->sb->recalc_sector),
3323                                                 ic->provided_data_sectors - le64_to_cpu(ic->sb->recalc_sector), BITMAP_OP_SET);
3324                                 block_bitmap_op(ic, ic->recalc_bitmap, le64_to_cpu(ic->sb->recalc_sector),
3325                                                 ic->provided_data_sectors - le64_to_cpu(ic->sb->recalc_sector), BITMAP_OP_SET);
3326                                 block_bitmap_op(ic, ic->may_write_bitmap, le64_to_cpu(ic->sb->recalc_sector),
3327                                                 ic->provided_data_sectors - le64_to_cpu(ic->sb->recalc_sector), BITMAP_OP_SET);
3328                         }
3329                         rw_journal_sectors(ic, REQ_OP_WRITE | REQ_FUA | REQ_SYNC, 0,
3330                                            ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3331                 }
3332         }
3333
3334         DEBUG_print("testing recalc: %x\n", ic->sb->flags);
3335         if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
3336                 __u64 recalc_pos = le64_to_cpu(ic->sb->recalc_sector);
3337
3338                 DEBUG_print("recalc pos: %llx / %llx\n", recalc_pos, ic->provided_data_sectors);
3339                 if (recalc_pos < ic->provided_data_sectors) {
3340                         queue_work(ic->recalc_wq, &ic->recalc_work);
3341                 } else if (recalc_pos > ic->provided_data_sectors) {
3342                         ic->sb->recalc_sector = cpu_to_le64(ic->provided_data_sectors);
3343                         recalc_write_super(ic);
3344                 }
3345         }
3346
3347         ic->reboot_notifier.notifier_call = dm_integrity_reboot;
3348         ic->reboot_notifier.next = NULL;
3349         ic->reboot_notifier.priority = INT_MAX - 1;     /* be notified after md and before hardware drivers */
3350         WARN_ON(register_reboot_notifier(&ic->reboot_notifier));
3351
3352 #if 0
3353         /* set to 1 to stress test synchronous mode */
3354         dm_integrity_enter_synchronous_mode(ic);
3355 #endif
3356 }
3357
3358 static void dm_integrity_status(struct dm_target *ti, status_type_t type,
3359                                 unsigned int status_flags, char *result, unsigned int maxlen)
3360 {
3361         struct dm_integrity_c *ic = ti->private;
3362         unsigned int arg_count;
3363         size_t sz = 0;
3364
3365         switch (type) {
3366         case STATUSTYPE_INFO:
3367                 DMEMIT("%llu %llu",
3368                         (unsigned long long)atomic64_read(&ic->number_of_mismatches),
3369                         ic->provided_data_sectors);
3370                 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
3371                         DMEMIT(" %llu", le64_to_cpu(ic->sb->recalc_sector));
3372                 else
3373                         DMEMIT(" -");
3374                 break;
3375
3376         case STATUSTYPE_TABLE: {
3377                 __u64 watermark_percentage = (__u64)(ic->journal_entries - ic->free_sectors_threshold) * 100;
3378
3379                 watermark_percentage += ic->journal_entries / 2;
3380                 do_div(watermark_percentage, ic->journal_entries);
3381                 arg_count = 3;
3382                 arg_count += !!ic->meta_dev;
3383                 arg_count += ic->sectors_per_block != 1;
3384                 arg_count += !!(ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING));
3385                 arg_count += ic->reset_recalculate_flag;
3386                 arg_count += ic->discard;
3387                 arg_count += ic->mode == 'J';
3388                 arg_count += ic->mode == 'J';
3389                 arg_count += ic->mode == 'B';
3390                 arg_count += ic->mode == 'B';
3391                 arg_count += !!ic->internal_hash_alg.alg_string;
3392                 arg_count += !!ic->journal_crypt_alg.alg_string;
3393                 arg_count += !!ic->journal_mac_alg.alg_string;
3394                 arg_count += (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING)) != 0;
3395                 arg_count += (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) != 0;
3396                 arg_count += ic->legacy_recalculate;
3397                 DMEMIT("%s %llu %u %c %u", ic->dev->name, ic->start,
3398                        ic->tag_size, ic->mode, arg_count);
3399                 if (ic->meta_dev)
3400                         DMEMIT(" meta_device:%s", ic->meta_dev->name);
3401                 if (ic->sectors_per_block != 1)
3402                         DMEMIT(" block_size:%u", ic->sectors_per_block << SECTOR_SHIFT);
3403                 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
3404                         DMEMIT(" recalculate");
3405                 if (ic->reset_recalculate_flag)
3406                         DMEMIT(" reset_recalculate");
3407                 if (ic->discard)
3408                         DMEMIT(" allow_discards");
3409                 DMEMIT(" journal_sectors:%u", ic->initial_sectors - SB_SECTORS);
3410                 DMEMIT(" interleave_sectors:%u", 1U << ic->sb->log2_interleave_sectors);
3411                 DMEMIT(" buffer_sectors:%u", 1U << ic->log2_buffer_sectors);
3412                 if (ic->mode == 'J') {
3413                         DMEMIT(" journal_watermark:%u", (unsigned int)watermark_percentage);
3414                         DMEMIT(" commit_time:%u", ic->autocommit_msec);
3415                 }
3416                 if (ic->mode == 'B') {
3417                         DMEMIT(" sectors_per_bit:%llu", (sector_t)ic->sectors_per_block << ic->log2_blocks_per_bitmap_bit);
3418                         DMEMIT(" bitmap_flush_interval:%u", jiffies_to_msecs(ic->bitmap_flush_interval));
3419                 }
3420                 if ((ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING)) != 0)
3421                         DMEMIT(" fix_padding");
3422                 if ((ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) != 0)
3423                         DMEMIT(" fix_hmac");
3424                 if (ic->legacy_recalculate)
3425                         DMEMIT(" legacy_recalculate");
3426
3427 #define EMIT_ALG(a, n)                                                  \
3428                 do {                                                    \
3429                         if (ic->a.alg_string) {                         \
3430                                 DMEMIT(" %s:%s", n, ic->a.alg_string);  \
3431                                 if (ic->a.key_string)                   \
3432                                         DMEMIT(":%s", ic->a.key_string);\
3433                         }                                               \
3434                 } while (0)
3435                 EMIT_ALG(internal_hash_alg, "internal_hash");
3436                 EMIT_ALG(journal_crypt_alg, "journal_crypt");
3437                 EMIT_ALG(journal_mac_alg, "journal_mac");
3438                 break;
3439         }
3440         case STATUSTYPE_IMA:
3441                 DMEMIT_TARGET_NAME_VERSION(ti->type);
3442                 DMEMIT(",dev_name=%s,start=%llu,tag_size=%u,mode=%c",
3443                         ic->dev->name, ic->start, ic->tag_size, ic->mode);
3444
3445                 if (ic->meta_dev)
3446                         DMEMIT(",meta_device=%s", ic->meta_dev->name);
3447                 if (ic->sectors_per_block != 1)
3448                         DMEMIT(",block_size=%u", ic->sectors_per_block << SECTOR_SHIFT);
3449
3450                 DMEMIT(",recalculate=%c", (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) ?
3451                        'y' : 'n');
3452                 DMEMIT(",allow_discards=%c", ic->discard ? 'y' : 'n');
3453                 DMEMIT(",fix_padding=%c",
3454                        ((ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING)) != 0) ? 'y' : 'n');
3455                 DMEMIT(",fix_hmac=%c",
3456                        ((ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) != 0) ? 'y' : 'n');
3457                 DMEMIT(",legacy_recalculate=%c", ic->legacy_recalculate ? 'y' : 'n');
3458
3459                 DMEMIT(",journal_sectors=%u", ic->initial_sectors - SB_SECTORS);
3460                 DMEMIT(",interleave_sectors=%u", 1U << ic->sb->log2_interleave_sectors);
3461                 DMEMIT(",buffer_sectors=%u", 1U << ic->log2_buffer_sectors);
3462                 DMEMIT(";");
3463                 break;
3464         }
3465 }
3466
3467 static int dm_integrity_iterate_devices(struct dm_target *ti,
3468                                         iterate_devices_callout_fn fn, void *data)
3469 {
3470         struct dm_integrity_c *ic = ti->private;
3471
3472         if (!ic->meta_dev)
3473                 return fn(ti, ic->dev, ic->start + ic->initial_sectors + ic->metadata_run, ti->len, data);
3474         else
3475                 return fn(ti, ic->dev, 0, ti->len, data);
3476 }
3477
3478 static void dm_integrity_io_hints(struct dm_target *ti, struct queue_limits *limits)
3479 {
3480         struct dm_integrity_c *ic = ti->private;
3481
3482         if (ic->sectors_per_block > 1) {
3483                 limits->logical_block_size = ic->sectors_per_block << SECTOR_SHIFT;
3484                 limits->physical_block_size = ic->sectors_per_block << SECTOR_SHIFT;
3485                 blk_limits_io_min(limits, ic->sectors_per_block << SECTOR_SHIFT);
3486                 limits->dma_alignment = limits->logical_block_size - 1;
3487         }
3488 }
3489
3490 static void calculate_journal_section_size(struct dm_integrity_c *ic)
3491 {
3492         unsigned int sector_space = JOURNAL_SECTOR_DATA;
3493
3494         ic->journal_sections = le32_to_cpu(ic->sb->journal_sections);
3495         ic->journal_entry_size = roundup(offsetof(struct journal_entry, last_bytes[ic->sectors_per_block]) + ic->tag_size,
3496                                          JOURNAL_ENTRY_ROUNDUP);
3497
3498         if (ic->sb->flags & cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC))
3499                 sector_space -= JOURNAL_MAC_PER_SECTOR;
3500         ic->journal_entries_per_sector = sector_space / ic->journal_entry_size;
3501         ic->journal_section_entries = ic->journal_entries_per_sector * JOURNAL_BLOCK_SECTORS;
3502         ic->journal_section_sectors = (ic->journal_section_entries << ic->sb->log2_sectors_per_block) + JOURNAL_BLOCK_SECTORS;
3503         ic->journal_entries = ic->journal_section_entries * ic->journal_sections;
3504 }
3505
3506 static int calculate_device_limits(struct dm_integrity_c *ic)
3507 {
3508         __u64 initial_sectors;
3509
3510         calculate_journal_section_size(ic);
3511         initial_sectors = SB_SECTORS + (__u64)ic->journal_section_sectors * ic->journal_sections;
3512         if (initial_sectors + METADATA_PADDING_SECTORS >= ic->meta_device_sectors || initial_sectors > UINT_MAX)
3513                 return -EINVAL;
3514         ic->initial_sectors = initial_sectors;
3515
3516         if (!ic->meta_dev) {
3517                 sector_t last_sector, last_area, last_offset;
3518
3519                 /* we have to maintain excessive padding for compatibility with existing volumes */
3520                 __u64 metadata_run_padding =
3521                         ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING) ?
3522                         (__u64)(METADATA_PADDING_SECTORS << SECTOR_SHIFT) :
3523                         (__u64)(1 << SECTOR_SHIFT << METADATA_PADDING_SECTORS);
3524
3525                 ic->metadata_run = round_up((__u64)ic->tag_size << (ic->sb->log2_interleave_sectors - ic->sb->log2_sectors_per_block),
3526                                             metadata_run_padding) >> SECTOR_SHIFT;
3527                 if (!(ic->metadata_run & (ic->metadata_run - 1)))
3528                         ic->log2_metadata_run = __ffs(ic->metadata_run);
3529                 else
3530                         ic->log2_metadata_run = -1;
3531
3532                 get_area_and_offset(ic, ic->provided_data_sectors - 1, &last_area, &last_offset);
3533                 last_sector = get_data_sector(ic, last_area, last_offset);
3534                 if (last_sector < ic->start || last_sector >= ic->meta_device_sectors)
3535                         return -EINVAL;
3536         } else {
3537                 __u64 meta_size = (ic->provided_data_sectors >> ic->sb->log2_sectors_per_block) * ic->tag_size;
3538
3539                 meta_size = (meta_size + ((1U << (ic->log2_buffer_sectors + SECTOR_SHIFT)) - 1))
3540                                 >> (ic->log2_buffer_sectors + SECTOR_SHIFT);
3541                 meta_size <<= ic->log2_buffer_sectors;
3542                 if (ic->initial_sectors + meta_size < ic->initial_sectors ||
3543                     ic->initial_sectors + meta_size > ic->meta_device_sectors)
3544                         return -EINVAL;
3545                 ic->metadata_run = 1;
3546                 ic->log2_metadata_run = 0;
3547         }
3548
3549         return 0;
3550 }
3551
3552 static void get_provided_data_sectors(struct dm_integrity_c *ic)
3553 {
3554         if (!ic->meta_dev) {
3555                 int test_bit;
3556
3557                 ic->provided_data_sectors = 0;
3558                 for (test_bit = fls64(ic->meta_device_sectors) - 1; test_bit >= 3; test_bit--) {
3559                         __u64 prev_data_sectors = ic->provided_data_sectors;
3560
3561                         ic->provided_data_sectors |= (sector_t)1 << test_bit;
3562                         if (calculate_device_limits(ic))
3563                                 ic->provided_data_sectors = prev_data_sectors;
3564                 }
3565         } else {
3566                 ic->provided_data_sectors = ic->data_device_sectors;
3567                 ic->provided_data_sectors &= ~(sector_t)(ic->sectors_per_block - 1);
3568         }
3569 }
3570
3571 static int initialize_superblock(struct dm_integrity_c *ic,
3572                                  unsigned int journal_sectors, unsigned int interleave_sectors)
3573 {
3574         unsigned int journal_sections;
3575         int test_bit;
3576
3577         memset(ic->sb, 0, SB_SECTORS << SECTOR_SHIFT);
3578         memcpy(ic->sb->magic, SB_MAGIC, 8);
3579         ic->sb->integrity_tag_size = cpu_to_le16(ic->tag_size);
3580         ic->sb->log2_sectors_per_block = __ffs(ic->sectors_per_block);
3581         if (ic->journal_mac_alg.alg_string)
3582                 ic->sb->flags |= cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC);
3583
3584         calculate_journal_section_size(ic);
3585         journal_sections = journal_sectors / ic->journal_section_sectors;
3586         if (!journal_sections)
3587                 journal_sections = 1;
3588
3589         if (ic->fix_hmac && (ic->internal_hash_alg.alg_string || ic->journal_mac_alg.alg_string)) {
3590                 ic->sb->flags |= cpu_to_le32(SB_FLAG_FIXED_HMAC);
3591                 get_random_bytes(ic->sb->salt, SALT_SIZE);
3592         }
3593
3594         if (!ic->meta_dev) {
3595                 if (ic->fix_padding)
3596                         ic->sb->flags |= cpu_to_le32(SB_FLAG_FIXED_PADDING);
3597                 ic->sb->journal_sections = cpu_to_le32(journal_sections);
3598                 if (!interleave_sectors)
3599                         interleave_sectors = DEFAULT_INTERLEAVE_SECTORS;
3600                 ic->sb->log2_interleave_sectors = __fls(interleave_sectors);
3601                 ic->sb->log2_interleave_sectors = max_t(__u8, MIN_LOG2_INTERLEAVE_SECTORS, ic->sb->log2_interleave_sectors);
3602                 ic->sb->log2_interleave_sectors = min_t(__u8, MAX_LOG2_INTERLEAVE_SECTORS, ic->sb->log2_interleave_sectors);
3603
3604                 get_provided_data_sectors(ic);
3605                 if (!ic->provided_data_sectors)
3606                         return -EINVAL;
3607         } else {
3608                 ic->sb->log2_interleave_sectors = 0;
3609
3610                 get_provided_data_sectors(ic);
3611                 if (!ic->provided_data_sectors)
3612                         return -EINVAL;
3613
3614 try_smaller_buffer:
3615                 ic->sb->journal_sections = cpu_to_le32(0);
3616                 for (test_bit = fls(journal_sections) - 1; test_bit >= 0; test_bit--) {
3617                         __u32 prev_journal_sections = le32_to_cpu(ic->sb->journal_sections);
3618                         __u32 test_journal_sections = prev_journal_sections | (1U << test_bit);
3619
3620                         if (test_journal_sections > journal_sections)
3621                                 continue;
3622                         ic->sb->journal_sections = cpu_to_le32(test_journal_sections);
3623                         if (calculate_device_limits(ic))
3624                                 ic->sb->journal_sections = cpu_to_le32(prev_journal_sections);
3625
3626                 }
3627                 if (!le32_to_cpu(ic->sb->journal_sections)) {
3628                         if (ic->log2_buffer_sectors > 3) {
3629                                 ic->log2_buffer_sectors--;
3630                                 goto try_smaller_buffer;
3631                         }
3632                         return -EINVAL;
3633                 }
3634         }
3635
3636         ic->sb->provided_data_sectors = cpu_to_le64(ic->provided_data_sectors);
3637
3638         sb_set_version(ic);
3639
3640         return 0;
3641 }
3642
3643 static void dm_integrity_set(struct dm_target *ti, struct dm_integrity_c *ic)
3644 {
3645         struct gendisk *disk = dm_disk(dm_table_get_md(ti->table));
3646         struct blk_integrity bi;
3647
3648         memset(&bi, 0, sizeof(bi));
3649         bi.profile = &dm_integrity_profile;
3650         bi.tuple_size = ic->tag_size;
3651         bi.tag_size = bi.tuple_size;
3652         bi.interval_exp = ic->sb->log2_sectors_per_block + SECTOR_SHIFT;
3653
3654         blk_integrity_register(disk, &bi);
3655         blk_queue_max_integrity_segments(disk->queue, UINT_MAX);
3656 }
3657
3658 static void dm_integrity_free_page_list(struct page_list *pl)
3659 {
3660         unsigned int i;
3661
3662         if (!pl)
3663                 return;
3664         for (i = 0; pl[i].page; i++)
3665                 __free_page(pl[i].page);
3666         kvfree(pl);
3667 }
3668
3669 static struct page_list *dm_integrity_alloc_page_list(unsigned int n_pages)
3670 {
3671         struct page_list *pl;
3672         unsigned int i;
3673
3674         pl = kvmalloc_array(n_pages + 1, sizeof(struct page_list), GFP_KERNEL | __GFP_ZERO);
3675         if (!pl)
3676                 return NULL;
3677
3678         for (i = 0; i < n_pages; i++) {
3679                 pl[i].page = alloc_page(GFP_KERNEL);
3680                 if (!pl[i].page) {
3681                         dm_integrity_free_page_list(pl);
3682                         return NULL;
3683                 }
3684                 if (i)
3685                         pl[i - 1].next = &pl[i];
3686         }
3687         pl[i].page = NULL;
3688         pl[i].next = NULL;
3689
3690         return pl;
3691 }
3692
3693 static void dm_integrity_free_journal_scatterlist(struct dm_integrity_c *ic, struct scatterlist **sl)
3694 {
3695         unsigned int i;
3696
3697         for (i = 0; i < ic->journal_sections; i++)
3698                 kvfree(sl[i]);
3699         kvfree(sl);
3700 }
3701
3702 static struct scatterlist **dm_integrity_alloc_journal_scatterlist(struct dm_integrity_c *ic,
3703                                                                    struct page_list *pl)
3704 {
3705         struct scatterlist **sl;
3706         unsigned int i;
3707
3708         sl = kvmalloc_array(ic->journal_sections,
3709                             sizeof(struct scatterlist *),
3710                             GFP_KERNEL | __GFP_ZERO);
3711         if (!sl)
3712                 return NULL;
3713
3714         for (i = 0; i < ic->journal_sections; i++) {
3715                 struct scatterlist *s;
3716                 unsigned int start_index, start_offset;
3717                 unsigned int end_index, end_offset;
3718                 unsigned int n_pages;
3719                 unsigned int idx;
3720
3721                 page_list_location(ic, i, 0, &start_index, &start_offset);
3722                 page_list_location(ic, i, ic->journal_section_sectors - 1,
3723                                    &end_index, &end_offset);
3724
3725                 n_pages = (end_index - start_index + 1);
3726
3727                 s = kvmalloc_array(n_pages, sizeof(struct scatterlist),
3728                                    GFP_KERNEL);
3729                 if (!s) {
3730                         dm_integrity_free_journal_scatterlist(ic, sl);
3731                         return NULL;
3732                 }
3733
3734                 sg_init_table(s, n_pages);
3735                 for (idx = start_index; idx <= end_index; idx++) {
3736                         char *va = lowmem_page_address(pl[idx].page);
3737                         unsigned int start = 0, end = PAGE_SIZE;
3738
3739                         if (idx == start_index)
3740                                 start = start_offset;
3741                         if (idx == end_index)
3742                                 end = end_offset + (1 << SECTOR_SHIFT);
3743                         sg_set_buf(&s[idx - start_index], va + start, end - start);
3744                 }
3745
3746                 sl[i] = s;
3747         }
3748
3749         return sl;
3750 }
3751
3752 static void free_alg(struct alg_spec *a)
3753 {
3754         kfree_sensitive(a->alg_string);
3755         kfree_sensitive(a->key);
3756         memset(a, 0, sizeof(*a));
3757 }
3758
3759 static int get_alg_and_key(const char *arg, struct alg_spec *a, char **error, char *error_inval)
3760 {
3761         char *k;
3762
3763         free_alg(a);
3764
3765         a->alg_string = kstrdup(strchr(arg, ':') + 1, GFP_KERNEL);
3766         if (!a->alg_string)
3767                 goto nomem;
3768
3769         k = strchr(a->alg_string, ':');
3770         if (k) {
3771                 *k = 0;
3772                 a->key_string = k + 1;
3773                 if (strlen(a->key_string) & 1)
3774                         goto inval;
3775
3776                 a->key_size = strlen(a->key_string) / 2;
3777                 a->key = kmalloc(a->key_size, GFP_KERNEL);
3778                 if (!a->key)
3779                         goto nomem;
3780                 if (hex2bin(a->key, a->key_string, a->key_size))
3781                         goto inval;
3782         }
3783
3784         return 0;
3785 inval:
3786         *error = error_inval;
3787         return -EINVAL;
3788 nomem:
3789         *error = "Out of memory for an argument";
3790         return -ENOMEM;
3791 }
3792
3793 static int get_mac(struct crypto_shash **hash, struct alg_spec *a, char **error,
3794                    char *error_alg, char *error_key)
3795 {
3796         int r;
3797
3798         if (a->alg_string) {
3799                 *hash = crypto_alloc_shash(a->alg_string, 0, CRYPTO_ALG_ALLOCATES_MEMORY);
3800                 if (IS_ERR(*hash)) {
3801                         *error = error_alg;
3802                         r = PTR_ERR(*hash);
3803                         *hash = NULL;
3804                         return r;
3805                 }
3806
3807                 if (a->key) {
3808                         r = crypto_shash_setkey(*hash, a->key, a->key_size);
3809                         if (r) {
3810                                 *error = error_key;
3811                                 return r;
3812                         }
3813                 } else if (crypto_shash_get_flags(*hash) & CRYPTO_TFM_NEED_KEY) {
3814                         *error = error_key;
3815                         return -ENOKEY;
3816                 }
3817         }
3818
3819         return 0;
3820 }
3821
3822 static int create_journal(struct dm_integrity_c *ic, char **error)
3823 {
3824         int r = 0;
3825         unsigned int i;
3826         __u64 journal_pages, journal_desc_size, journal_tree_size;
3827         unsigned char *crypt_data = NULL, *crypt_iv = NULL;
3828         struct skcipher_request *req = NULL;
3829
3830         ic->commit_ids[0] = cpu_to_le64(0x1111111111111111ULL);
3831         ic->commit_ids[1] = cpu_to_le64(0x2222222222222222ULL);
3832         ic->commit_ids[2] = cpu_to_le64(0x3333333333333333ULL);
3833         ic->commit_ids[3] = cpu_to_le64(0x4444444444444444ULL);
3834
3835         journal_pages = roundup((__u64)ic->journal_sections * ic->journal_section_sectors,
3836                                 PAGE_SIZE >> SECTOR_SHIFT) >> (PAGE_SHIFT - SECTOR_SHIFT);
3837         journal_desc_size = journal_pages * sizeof(struct page_list);
3838         if (journal_pages >= totalram_pages() - totalhigh_pages() || journal_desc_size > ULONG_MAX) {
3839                 *error = "Journal doesn't fit into memory";
3840                 r = -ENOMEM;
3841                 goto bad;
3842         }
3843         ic->journal_pages = journal_pages;
3844
3845         ic->journal = dm_integrity_alloc_page_list(ic->journal_pages);
3846         if (!ic->journal) {
3847                 *error = "Could not allocate memory for journal";
3848                 r = -ENOMEM;
3849                 goto bad;
3850         }
3851         if (ic->journal_crypt_alg.alg_string) {
3852                 unsigned int ivsize, blocksize;
3853                 struct journal_completion comp;
3854
3855                 comp.ic = ic;
3856                 ic->journal_crypt = crypto_alloc_skcipher(ic->journal_crypt_alg.alg_string, 0, CRYPTO_ALG_ALLOCATES_MEMORY);
3857                 if (IS_ERR(ic->journal_crypt)) {
3858                         *error = "Invalid journal cipher";
3859                         r = PTR_ERR(ic->journal_crypt);
3860                         ic->journal_crypt = NULL;
3861                         goto bad;
3862                 }
3863                 ivsize = crypto_skcipher_ivsize(ic->journal_crypt);
3864                 blocksize = crypto_skcipher_blocksize(ic->journal_crypt);
3865
3866                 if (ic->journal_crypt_alg.key) {
3867                         r = crypto_skcipher_setkey(ic->journal_crypt, ic->journal_crypt_alg.key,
3868                                                    ic->journal_crypt_alg.key_size);
3869                         if (r) {
3870                                 *error = "Error setting encryption key";
3871                                 goto bad;
3872                         }
3873                 }
3874                 DEBUG_print("cipher %s, block size %u iv size %u\n",
3875                             ic->journal_crypt_alg.alg_string, blocksize, ivsize);
3876
3877                 ic->journal_io = dm_integrity_alloc_page_list(ic->journal_pages);
3878                 if (!ic->journal_io) {
3879                         *error = "Could not allocate memory for journal io";
3880                         r = -ENOMEM;
3881                         goto bad;
3882                 }
3883
3884                 if (blocksize == 1) {
3885                         struct scatterlist *sg;
3886
3887                         req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
3888                         if (!req) {
3889                                 *error = "Could not allocate crypt request";
3890                                 r = -ENOMEM;
3891                                 goto bad;
3892                         }
3893
3894                         crypt_iv = kzalloc(ivsize, GFP_KERNEL);
3895                         if (!crypt_iv) {
3896                                 *error = "Could not allocate iv";
3897                                 r = -ENOMEM;
3898                                 goto bad;
3899                         }
3900
3901                         ic->journal_xor = dm_integrity_alloc_page_list(ic->journal_pages);
3902                         if (!ic->journal_xor) {
3903                                 *error = "Could not allocate memory for journal xor";
3904                                 r = -ENOMEM;
3905                                 goto bad;
3906                         }
3907
3908                         sg = kvmalloc_array(ic->journal_pages + 1,
3909                                             sizeof(struct scatterlist),
3910                                             GFP_KERNEL);
3911                         if (!sg) {
3912                                 *error = "Unable to allocate sg list";
3913                                 r = -ENOMEM;
3914                                 goto bad;
3915                         }
3916                         sg_init_table(sg, ic->journal_pages + 1);
3917                         for (i = 0; i < ic->journal_pages; i++) {
3918                                 char *va = lowmem_page_address(ic->journal_xor[i].page);
3919
3920                                 clear_page(va);
3921                                 sg_set_buf(&sg[i], va, PAGE_SIZE);
3922                         }
3923                         sg_set_buf(&sg[i], &ic->commit_ids, sizeof(ic->commit_ids));
3924
3925                         skcipher_request_set_crypt(req, sg, sg,
3926                                                    PAGE_SIZE * ic->journal_pages + sizeof(ic->commit_ids), crypt_iv);
3927                         init_completion(&comp.comp);
3928                         comp.in_flight = (atomic_t)ATOMIC_INIT(1);
3929                         if (do_crypt(true, req, &comp))
3930                                 wait_for_completion(&comp.comp);
3931                         kvfree(sg);
3932                         r = dm_integrity_failed(ic);
3933                         if (r) {
3934                                 *error = "Unable to encrypt journal";
3935                                 goto bad;
3936                         }
3937                         DEBUG_bytes(lowmem_page_address(ic->journal_xor[0].page), 64, "xor data");
3938
3939                         crypto_free_skcipher(ic->journal_crypt);
3940                         ic->journal_crypt = NULL;
3941                 } else {
3942                         unsigned int crypt_len = roundup(ivsize, blocksize);
3943
3944                         req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
3945                         if (!req) {
3946                                 *error = "Could not allocate crypt request";
3947                                 r = -ENOMEM;
3948                                 goto bad;
3949                         }
3950
3951                         crypt_iv = kmalloc(ivsize, GFP_KERNEL);
3952                         if (!crypt_iv) {
3953                                 *error = "Could not allocate iv";
3954                                 r = -ENOMEM;
3955                                 goto bad;
3956                         }
3957
3958                         crypt_data = kmalloc(crypt_len, GFP_KERNEL);
3959                         if (!crypt_data) {
3960                                 *error = "Unable to allocate crypt data";
3961                                 r = -ENOMEM;
3962                                 goto bad;
3963                         }
3964
3965                         ic->journal_scatterlist = dm_integrity_alloc_journal_scatterlist(ic, ic->journal);
3966                         if (!ic->journal_scatterlist) {
3967                                 *error = "Unable to allocate sg list";
3968                                 r = -ENOMEM;
3969                                 goto bad;
3970                         }
3971                         ic->journal_io_scatterlist = dm_integrity_alloc_journal_scatterlist(ic, ic->journal_io);
3972                         if (!ic->journal_io_scatterlist) {
3973                                 *error = "Unable to allocate sg list";
3974                                 r = -ENOMEM;
3975                                 goto bad;
3976                         }
3977                         ic->sk_requests = kvmalloc_array(ic->journal_sections,
3978                                                          sizeof(struct skcipher_request *),
3979                                                          GFP_KERNEL | __GFP_ZERO);
3980                         if (!ic->sk_requests) {
3981                                 *error = "Unable to allocate sk requests";
3982                                 r = -ENOMEM;
3983                                 goto bad;
3984                         }
3985                         for (i = 0; i < ic->journal_sections; i++) {
3986                                 struct scatterlist sg;
3987                                 struct skcipher_request *section_req;
3988                                 __le32 section_le = cpu_to_le32(i);
3989
3990                                 memset(crypt_iv, 0x00, ivsize);
3991                                 memset(crypt_data, 0x00, crypt_len);
3992                                 memcpy(crypt_data, &section_le, min_t(size_t, crypt_len, sizeof(section_le)));
3993
3994                                 sg_init_one(&sg, crypt_data, crypt_len);
3995                                 skcipher_request_set_crypt(req, &sg, &sg, crypt_len, crypt_iv);
3996                                 init_completion(&comp.comp);
3997                                 comp.in_flight = (atomic_t)ATOMIC_INIT(1);
3998                                 if (do_crypt(true, req, &comp))
3999                                         wait_for_completion(&comp.comp);
4000
4001                                 r = dm_integrity_failed(ic);
4002                                 if (r) {
4003                                         *error = "Unable to generate iv";
4004                                         goto bad;
4005                                 }
4006
4007                                 section_req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
4008                                 if (!section_req) {
4009                                         *error = "Unable to allocate crypt request";
4010                                         r = -ENOMEM;
4011                                         goto bad;
4012                                 }
4013                                 section_req->iv = kmalloc_array(ivsize, 2,
4014                                                                 GFP_KERNEL);
4015                                 if (!section_req->iv) {
4016                                         skcipher_request_free(section_req);
4017                                         *error = "Unable to allocate iv";
4018                                         r = -ENOMEM;
4019                                         goto bad;
4020                                 }
4021                                 memcpy(section_req->iv + ivsize, crypt_data, ivsize);
4022                                 section_req->cryptlen = (size_t)ic->journal_section_sectors << SECTOR_SHIFT;
4023                                 ic->sk_requests[i] = section_req;
4024                                 DEBUG_bytes(crypt_data, ivsize, "iv(%u)", i);
4025                         }
4026                 }
4027         }
4028
4029         for (i = 0; i < N_COMMIT_IDS; i++) {
4030                 unsigned int j;
4031
4032 retest_commit_id:
4033                 for (j = 0; j < i; j++) {
4034                         if (ic->commit_ids[j] == ic->commit_ids[i]) {
4035                                 ic->commit_ids[i] = cpu_to_le64(le64_to_cpu(ic->commit_ids[i]) + 1);
4036                                 goto retest_commit_id;
4037                         }
4038                 }
4039                 DEBUG_print("commit id %u: %016llx\n", i, ic->commit_ids[i]);
4040         }
4041
4042         journal_tree_size = (__u64)ic->journal_entries * sizeof(struct journal_node);
4043         if (journal_tree_size > ULONG_MAX) {
4044                 *error = "Journal doesn't fit into memory";
4045                 r = -ENOMEM;
4046                 goto bad;
4047         }
4048         ic->journal_tree = kvmalloc(journal_tree_size, GFP_KERNEL);
4049         if (!ic->journal_tree) {
4050                 *error = "Could not allocate memory for journal tree";
4051                 r = -ENOMEM;
4052         }
4053 bad:
4054         kfree(crypt_data);
4055         kfree(crypt_iv);
4056         skcipher_request_free(req);
4057
4058         return r;
4059 }
4060
4061 /*
4062  * Construct a integrity mapping
4063  *
4064  * Arguments:
4065  *      device
4066  *      offset from the start of the device
4067  *      tag size
4068  *      D - direct writes, J - journal writes, B - bitmap mode, R - recovery mode
4069  *      number of optional arguments
4070  *      optional arguments:
4071  *              journal_sectors
4072  *              interleave_sectors
4073  *              buffer_sectors
4074  *              journal_watermark
4075  *              commit_time
4076  *              meta_device
4077  *              block_size
4078  *              sectors_per_bit
4079  *              bitmap_flush_interval
4080  *              internal_hash
4081  *              journal_crypt
4082  *              journal_mac
4083  *              recalculate
4084  */
4085 static int dm_integrity_ctr(struct dm_target *ti, unsigned int argc, char **argv)
4086 {
4087         struct dm_integrity_c *ic;
4088         char dummy;
4089         int r;
4090         unsigned int extra_args;
4091         struct dm_arg_set as;
4092         static const struct dm_arg _args[] = {
4093                 {0, 18, "Invalid number of feature args"},
4094         };
4095         unsigned int journal_sectors, interleave_sectors, buffer_sectors, journal_watermark, sync_msec;
4096         bool should_write_sb;
4097         __u64 threshold;
4098         unsigned long long start;
4099         __s8 log2_sectors_per_bitmap_bit = -1;
4100         __s8 log2_blocks_per_bitmap_bit;
4101         __u64 bits_in_journal;
4102         __u64 n_bitmap_bits;
4103
4104 #define DIRECT_ARGUMENTS        4
4105
4106         if (argc <= DIRECT_ARGUMENTS) {
4107                 ti->error = "Invalid argument count";
4108                 return -EINVAL;
4109         }
4110
4111         ic = kzalloc(sizeof(struct dm_integrity_c), GFP_KERNEL);
4112         if (!ic) {
4113                 ti->error = "Cannot allocate integrity context";
4114                 return -ENOMEM;
4115         }
4116         ti->private = ic;
4117         ti->per_io_data_size = sizeof(struct dm_integrity_io);
4118         ic->ti = ti;
4119
4120         ic->in_progress = RB_ROOT;
4121         INIT_LIST_HEAD(&ic->wait_list);
4122         init_waitqueue_head(&ic->endio_wait);
4123         bio_list_init(&ic->flush_bio_list);
4124         init_waitqueue_head(&ic->copy_to_journal_wait);
4125         init_completion(&ic->crypto_backoff);
4126         atomic64_set(&ic->number_of_mismatches, 0);
4127         ic->bitmap_flush_interval = BITMAP_FLUSH_INTERVAL;
4128
4129         r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &ic->dev);
4130         if (r) {
4131                 ti->error = "Device lookup failed";
4132                 goto bad;
4133         }
4134
4135         if (sscanf(argv[1], "%llu%c", &start, &dummy) != 1 || start != (sector_t)start) {
4136                 ti->error = "Invalid starting offset";
4137                 r = -EINVAL;
4138                 goto bad;
4139         }
4140         ic->start = start;
4141
4142         if (strcmp(argv[2], "-")) {
4143                 if (sscanf(argv[2], "%u%c", &ic->tag_size, &dummy) != 1 || !ic->tag_size) {
4144                         ti->error = "Invalid tag size";
4145                         r = -EINVAL;
4146                         goto bad;
4147                 }
4148         }
4149
4150         if (!strcmp(argv[3], "J") || !strcmp(argv[3], "B") ||
4151             !strcmp(argv[3], "D") || !strcmp(argv[3], "R")) {
4152                 ic->mode = argv[3][0];
4153         } else {
4154                 ti->error = "Invalid mode (expecting J, B, D, R)";
4155                 r = -EINVAL;
4156                 goto bad;
4157         }
4158
4159         journal_sectors = 0;
4160         interleave_sectors = DEFAULT_INTERLEAVE_SECTORS;
4161         buffer_sectors = DEFAULT_BUFFER_SECTORS;
4162         journal_watermark = DEFAULT_JOURNAL_WATERMARK;
4163         sync_msec = DEFAULT_SYNC_MSEC;
4164         ic->sectors_per_block = 1;
4165
4166         as.argc = argc - DIRECT_ARGUMENTS;
4167         as.argv = argv + DIRECT_ARGUMENTS;
4168         r = dm_read_arg_group(_args, &as, &extra_args, &ti->error);
4169         if (r)
4170                 goto bad;
4171
4172         while (extra_args--) {
4173                 const char *opt_string;
4174                 unsigned int val;
4175                 unsigned long long llval;
4176
4177                 opt_string = dm_shift_arg(&as);
4178                 if (!opt_string) {
4179                         r = -EINVAL;
4180                         ti->error = "Not enough feature arguments";
4181                         goto bad;
4182                 }
4183                 if (sscanf(opt_string, "journal_sectors:%u%c", &val, &dummy) == 1)
4184                         journal_sectors = val ? val : 1;
4185                 else if (sscanf(opt_string, "interleave_sectors:%u%c", &val, &dummy) == 1)
4186                         interleave_sectors = val;
4187                 else if (sscanf(opt_string, "buffer_sectors:%u%c", &val, &dummy) == 1)
4188                         buffer_sectors = val;
4189                 else if (sscanf(opt_string, "journal_watermark:%u%c", &val, &dummy) == 1 && val <= 100)
4190                         journal_watermark = val;
4191                 else if (sscanf(opt_string, "commit_time:%u%c", &val, &dummy) == 1)
4192                         sync_msec = val;
4193                 else if (!strncmp(opt_string, "meta_device:", strlen("meta_device:"))) {
4194                         if (ic->meta_dev) {
4195                                 dm_put_device(ti, ic->meta_dev);
4196                                 ic->meta_dev = NULL;
4197                         }
4198                         r = dm_get_device(ti, strchr(opt_string, ':') + 1,
4199                                           dm_table_get_mode(ti->table), &ic->meta_dev);
4200                         if (r) {
4201                                 ti->error = "Device lookup failed";
4202                                 goto bad;
4203                         }
4204                 } else if (sscanf(opt_string, "block_size:%u%c", &val, &dummy) == 1) {
4205                         if (val < 1 << SECTOR_SHIFT ||
4206                             val > MAX_SECTORS_PER_BLOCK << SECTOR_SHIFT ||
4207                             (val & (val - 1))) {
4208                                 r = -EINVAL;
4209                                 ti->error = "Invalid block_size argument";
4210                                 goto bad;
4211                         }
4212                         ic->sectors_per_block = val >> SECTOR_SHIFT;
4213                 } else if (sscanf(opt_string, "sectors_per_bit:%llu%c", &llval, &dummy) == 1) {
4214                         log2_sectors_per_bitmap_bit = !llval ? 0 : __ilog2_u64(llval);
4215                 } else if (sscanf(opt_string, "bitmap_flush_interval:%u%c", &val, &dummy) == 1) {
4216                         if (val >= (uint64_t)UINT_MAX * 1000 / HZ) {
4217                                 r = -EINVAL;
4218                                 ti->error = "Invalid bitmap_flush_interval argument";
4219                                 goto bad;
4220                         }
4221                         ic->bitmap_flush_interval = msecs_to_jiffies(val);
4222                 } else if (!strncmp(opt_string, "internal_hash:", strlen("internal_hash:"))) {
4223                         r = get_alg_and_key(opt_string, &ic->internal_hash_alg, &ti->error,
4224                                             "Invalid internal_hash argument");
4225                         if (r)
4226                                 goto bad;
4227                 } else if (!strncmp(opt_string, "journal_crypt:", strlen("journal_crypt:"))) {
4228                         r = get_alg_and_key(opt_string, &ic->journal_crypt_alg, &ti->error,
4229                                             "Invalid journal_crypt argument");
4230                         if (r)
4231                                 goto bad;
4232                 } else if (!strncmp(opt_string, "journal_mac:", strlen("journal_mac:"))) {
4233                         r = get_alg_and_key(opt_string, &ic->journal_mac_alg, &ti->error,
4234                                             "Invalid journal_mac argument");
4235                         if (r)
4236                                 goto bad;
4237                 } else if (!strcmp(opt_string, "recalculate")) {
4238                         ic->recalculate_flag = true;
4239                 } else if (!strcmp(opt_string, "reset_recalculate")) {
4240                         ic->recalculate_flag = true;
4241                         ic->reset_recalculate_flag = true;
4242                 } else if (!strcmp(opt_string, "allow_discards")) {
4243                         ic->discard = true;
4244                 } else if (!strcmp(opt_string, "fix_padding")) {
4245                         ic->fix_padding = true;
4246                 } else if (!strcmp(opt_string, "fix_hmac")) {
4247                         ic->fix_hmac = true;
4248                 } else if (!strcmp(opt_string, "legacy_recalculate")) {
4249                         ic->legacy_recalculate = true;
4250                 } else {
4251                         r = -EINVAL;
4252                         ti->error = "Invalid argument";
4253                         goto bad;
4254                 }
4255         }
4256
4257         ic->data_device_sectors = bdev_nr_sectors(ic->dev->bdev);
4258         if (!ic->meta_dev)
4259                 ic->meta_device_sectors = ic->data_device_sectors;
4260         else
4261                 ic->meta_device_sectors = bdev_nr_sectors(ic->meta_dev->bdev);
4262
4263         if (!journal_sectors) {
4264                 journal_sectors = min((sector_t)DEFAULT_MAX_JOURNAL_SECTORS,
4265                                       ic->data_device_sectors >> DEFAULT_JOURNAL_SIZE_FACTOR);
4266         }
4267
4268         if (!buffer_sectors)
4269                 buffer_sectors = 1;
4270         ic->log2_buffer_sectors = min((int)__fls(buffer_sectors), 31 - SECTOR_SHIFT);
4271
4272         r = get_mac(&ic->internal_hash, &ic->internal_hash_alg, &ti->error,
4273                     "Invalid internal hash", "Error setting internal hash key");
4274         if (r)
4275                 goto bad;
4276
4277         r = get_mac(&ic->journal_mac, &ic->journal_mac_alg, &ti->error,
4278                     "Invalid journal mac", "Error setting journal mac key");
4279         if (r)
4280                 goto bad;
4281
4282         if (!ic->tag_size) {
4283                 if (!ic->internal_hash) {
4284                         ti->error = "Unknown tag size";
4285                         r = -EINVAL;
4286                         goto bad;
4287                 }
4288                 ic->tag_size = crypto_shash_digestsize(ic->internal_hash);
4289         }
4290         if (ic->tag_size > MAX_TAG_SIZE) {
4291                 ti->error = "Too big tag size";
4292                 r = -EINVAL;
4293                 goto bad;
4294         }
4295         if (!(ic->tag_size & (ic->tag_size - 1)))
4296                 ic->log2_tag_size = __ffs(ic->tag_size);
4297         else
4298                 ic->log2_tag_size = -1;
4299
4300         if (ic->mode == 'B' && !ic->internal_hash) {
4301                 r = -EINVAL;
4302                 ti->error = "Bitmap mode can be only used with internal hash";
4303                 goto bad;
4304         }
4305
4306         if (ic->discard && !ic->internal_hash) {
4307                 r = -EINVAL;
4308                 ti->error = "Discard can be only used with internal hash";
4309                 goto bad;
4310         }
4311
4312         ic->autocommit_jiffies = msecs_to_jiffies(sync_msec);
4313         ic->autocommit_msec = sync_msec;
4314         timer_setup(&ic->autocommit_timer, autocommit_fn, 0);
4315
4316         ic->io = dm_io_client_create();
4317         if (IS_ERR(ic->io)) {
4318                 r = PTR_ERR(ic->io);
4319                 ic->io = NULL;
4320                 ti->error = "Cannot allocate dm io";
4321                 goto bad;
4322         }
4323
4324         r = mempool_init_slab_pool(&ic->journal_io_mempool, JOURNAL_IO_MEMPOOL, journal_io_cache);
4325         if (r) {
4326                 ti->error = "Cannot allocate mempool";
4327                 goto bad;
4328         }
4329
4330         r = mempool_init_page_pool(&ic->recheck_pool, 1, 0);
4331         if (r) {
4332                 ti->error = "Cannot allocate mempool";
4333                 goto bad;
4334         }
4335
4336         ic->metadata_wq = alloc_workqueue("dm-integrity-metadata",
4337                                           WQ_MEM_RECLAIM, METADATA_WORKQUEUE_MAX_ACTIVE);
4338         if (!ic->metadata_wq) {
4339                 ti->error = "Cannot allocate workqueue";
4340                 r = -ENOMEM;
4341                 goto bad;
4342         }
4343
4344         /*
4345          * If this workqueue weren't ordered, it would cause bio reordering
4346          * and reduced performance.
4347          */
4348         ic->wait_wq = alloc_ordered_workqueue("dm-integrity-wait", WQ_MEM_RECLAIM);
4349         if (!ic->wait_wq) {
4350                 ti->error = "Cannot allocate workqueue";
4351                 r = -ENOMEM;
4352                 goto bad;
4353         }
4354
4355         ic->offload_wq = alloc_workqueue("dm-integrity-offload", WQ_MEM_RECLAIM,
4356                                           METADATA_WORKQUEUE_MAX_ACTIVE);
4357         if (!ic->offload_wq) {
4358                 ti->error = "Cannot allocate workqueue";
4359                 r = -ENOMEM;
4360                 goto bad;
4361         }
4362
4363         ic->commit_wq = alloc_workqueue("dm-integrity-commit", WQ_MEM_RECLAIM, 1);
4364         if (!ic->commit_wq) {
4365                 ti->error = "Cannot allocate workqueue";
4366                 r = -ENOMEM;
4367                 goto bad;
4368         }
4369         INIT_WORK(&ic->commit_work, integrity_commit);
4370
4371         if (ic->mode == 'J' || ic->mode == 'B') {
4372                 ic->writer_wq = alloc_workqueue("dm-integrity-writer", WQ_MEM_RECLAIM, 1);
4373                 if (!ic->writer_wq) {
4374                         ti->error = "Cannot allocate workqueue";
4375                         r = -ENOMEM;
4376                         goto bad;
4377                 }
4378                 INIT_WORK(&ic->writer_work, integrity_writer);
4379         }
4380
4381         ic->sb = alloc_pages_exact(SB_SECTORS << SECTOR_SHIFT, GFP_KERNEL);
4382         if (!ic->sb) {
4383                 r = -ENOMEM;
4384                 ti->error = "Cannot allocate superblock area";
4385                 goto bad;
4386         }
4387
4388         r = sync_rw_sb(ic, REQ_OP_READ);
4389         if (r) {
4390                 ti->error = "Error reading superblock";
4391                 goto bad;
4392         }
4393         should_write_sb = false;
4394         if (memcmp(ic->sb->magic, SB_MAGIC, 8)) {
4395                 if (ic->mode != 'R') {
4396                         if (memchr_inv(ic->sb, 0, SB_SECTORS << SECTOR_SHIFT)) {
4397                                 r = -EINVAL;
4398                                 ti->error = "The device is not initialized";
4399                                 goto bad;
4400                         }
4401                 }
4402
4403                 r = initialize_superblock(ic, journal_sectors, interleave_sectors);
4404                 if (r) {
4405                         ti->error = "Could not initialize superblock";
4406                         goto bad;
4407                 }
4408                 if (ic->mode != 'R')
4409                         should_write_sb = true;
4410         }
4411
4412         if (!ic->sb->version || ic->sb->version > SB_VERSION_5) {
4413                 r = -EINVAL;
4414                 ti->error = "Unknown version";
4415                 goto bad;
4416         }
4417         if (le16_to_cpu(ic->sb->integrity_tag_size) != ic->tag_size) {
4418                 r = -EINVAL;
4419                 ti->error = "Tag size doesn't match the information in superblock";
4420                 goto bad;
4421         }
4422         if (ic->sb->log2_sectors_per_block != __ffs(ic->sectors_per_block)) {
4423                 r = -EINVAL;
4424                 ti->error = "Block size doesn't match the information in superblock";
4425                 goto bad;
4426         }
4427         if (!le32_to_cpu(ic->sb->journal_sections)) {
4428                 r = -EINVAL;
4429                 ti->error = "Corrupted superblock, journal_sections is 0";
4430                 goto bad;
4431         }
4432         /* make sure that ti->max_io_len doesn't overflow */
4433         if (!ic->meta_dev) {
4434                 if (ic->sb->log2_interleave_sectors < MIN_LOG2_INTERLEAVE_SECTORS ||
4435                     ic->sb->log2_interleave_sectors > MAX_LOG2_INTERLEAVE_SECTORS) {
4436                         r = -EINVAL;
4437                         ti->error = "Invalid interleave_sectors in the superblock";
4438                         goto bad;
4439                 }
4440         } else {
4441                 if (ic->sb->log2_interleave_sectors) {
4442                         r = -EINVAL;
4443                         ti->error = "Invalid interleave_sectors in the superblock";
4444                         goto bad;
4445                 }
4446         }
4447         if (!!(ic->sb->flags & cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC)) != !!ic->journal_mac_alg.alg_string) {
4448                 r = -EINVAL;
4449                 ti->error = "Journal mac mismatch";
4450                 goto bad;
4451         }
4452
4453         get_provided_data_sectors(ic);
4454         if (!ic->provided_data_sectors) {
4455                 r = -EINVAL;
4456                 ti->error = "The device is too small";
4457                 goto bad;
4458         }
4459
4460 try_smaller_buffer:
4461         r = calculate_device_limits(ic);
4462         if (r) {
4463                 if (ic->meta_dev) {
4464                         if (ic->log2_buffer_sectors > 3) {
4465                                 ic->log2_buffer_sectors--;
4466                                 goto try_smaller_buffer;
4467                         }
4468                 }
4469                 ti->error = "The device is too small";
4470                 goto bad;
4471         }
4472
4473         if (log2_sectors_per_bitmap_bit < 0)
4474                 log2_sectors_per_bitmap_bit = __fls(DEFAULT_SECTORS_PER_BITMAP_BIT);
4475         if (log2_sectors_per_bitmap_bit < ic->sb->log2_sectors_per_block)
4476                 log2_sectors_per_bitmap_bit = ic->sb->log2_sectors_per_block;
4477
4478         bits_in_journal = ((__u64)ic->journal_section_sectors * ic->journal_sections) << (SECTOR_SHIFT + 3);
4479         if (bits_in_journal > UINT_MAX)
4480                 bits_in_journal = UINT_MAX;
4481         while (bits_in_journal < (ic->provided_data_sectors + ((sector_t)1 << log2_sectors_per_bitmap_bit) - 1) >> log2_sectors_per_bitmap_bit)
4482                 log2_sectors_per_bitmap_bit++;
4483
4484         log2_blocks_per_bitmap_bit = log2_sectors_per_bitmap_bit - ic->sb->log2_sectors_per_block;
4485         ic->log2_blocks_per_bitmap_bit = log2_blocks_per_bitmap_bit;
4486         if (should_write_sb)
4487                 ic->sb->log2_blocks_per_bitmap_bit = log2_blocks_per_bitmap_bit;
4488
4489         n_bitmap_bits = ((ic->provided_data_sectors >> ic->sb->log2_sectors_per_block)
4490                                 + (((sector_t)1 << log2_blocks_per_bitmap_bit) - 1)) >> log2_blocks_per_bitmap_bit;
4491         ic->n_bitmap_blocks = DIV_ROUND_UP(n_bitmap_bits, BITMAP_BLOCK_SIZE * 8);
4492
4493         if (!ic->meta_dev)
4494                 ic->log2_buffer_sectors = min(ic->log2_buffer_sectors, (__u8)__ffs(ic->metadata_run));
4495
4496         if (ti->len > ic->provided_data_sectors) {
4497                 r = -EINVAL;
4498                 ti->error = "Not enough provided sectors for requested mapping size";
4499                 goto bad;
4500         }
4501
4502
4503         threshold = (__u64)ic->journal_entries * (100 - journal_watermark);
4504         threshold += 50;
4505         do_div(threshold, 100);
4506         ic->free_sectors_threshold = threshold;
4507
4508         DEBUG_print("initialized:\n");
4509         DEBUG_print("   integrity_tag_size %u\n", le16_to_cpu(ic->sb->integrity_tag_size));
4510         DEBUG_print("   journal_entry_size %u\n", ic->journal_entry_size);
4511         DEBUG_print("   journal_entries_per_sector %u\n", ic->journal_entries_per_sector);
4512         DEBUG_print("   journal_section_entries %u\n", ic->journal_section_entries);
4513         DEBUG_print("   journal_section_sectors %u\n", ic->journal_section_sectors);
4514         DEBUG_print("   journal_sections %u\n", (unsigned int)le32_to_cpu(ic->sb->journal_sections));
4515         DEBUG_print("   journal_entries %u\n", ic->journal_entries);
4516         DEBUG_print("   log2_interleave_sectors %d\n", ic->sb->log2_interleave_sectors);
4517         DEBUG_print("   data_device_sectors 0x%llx\n", bdev_nr_sectors(ic->dev->bdev));
4518         DEBUG_print("   initial_sectors 0x%x\n", ic->initial_sectors);
4519         DEBUG_print("   metadata_run 0x%x\n", ic->metadata_run);
4520         DEBUG_print("   log2_metadata_run %d\n", ic->log2_metadata_run);
4521         DEBUG_print("   provided_data_sectors 0x%llx (%llu)\n", ic->provided_data_sectors, ic->provided_data_sectors);
4522         DEBUG_print("   log2_buffer_sectors %u\n", ic->log2_buffer_sectors);
4523         DEBUG_print("   bits_in_journal %llu\n", bits_in_journal);
4524
4525         if (ic->recalculate_flag && !(ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))) {
4526                 ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
4527                 ic->sb->recalc_sector = cpu_to_le64(0);
4528         }
4529
4530         if (ic->internal_hash) {
4531                 ic->recalc_wq = alloc_workqueue("dm-integrity-recalc", WQ_MEM_RECLAIM, 1);
4532                 if (!ic->recalc_wq) {
4533                         ti->error = "Cannot allocate workqueue";
4534                         r = -ENOMEM;
4535                         goto bad;
4536                 }
4537                 INIT_WORK(&ic->recalc_work, integrity_recalc);
4538         } else {
4539                 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
4540                         ti->error = "Recalculate can only be specified with internal_hash";
4541                         r = -EINVAL;
4542                         goto bad;
4543                 }
4544         }
4545
4546         if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING) &&
4547             le64_to_cpu(ic->sb->recalc_sector) < ic->provided_data_sectors &&
4548             dm_integrity_disable_recalculate(ic)) {
4549                 ti->error = "Recalculating with HMAC is disabled for security reasons - if you really need it, use the argument \"legacy_recalculate\"";
4550                 r = -EOPNOTSUPP;
4551                 goto bad;
4552         }
4553
4554         ic->bufio = dm_bufio_client_create(ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev,
4555                         1U << (SECTOR_SHIFT + ic->log2_buffer_sectors), 1, 0, NULL, NULL, 0);
4556         if (IS_ERR(ic->bufio)) {
4557                 r = PTR_ERR(ic->bufio);
4558                 ti->error = "Cannot initialize dm-bufio";
4559                 ic->bufio = NULL;
4560                 goto bad;
4561         }
4562         dm_bufio_set_sector_offset(ic->bufio, ic->start + ic->initial_sectors);
4563
4564         if (ic->mode != 'R') {
4565                 r = create_journal(ic, &ti->error);
4566                 if (r)
4567                         goto bad;
4568
4569         }
4570
4571         if (ic->mode == 'B') {
4572                 unsigned int i;
4573                 unsigned int n_bitmap_pages = DIV_ROUND_UP(ic->n_bitmap_blocks, PAGE_SIZE / BITMAP_BLOCK_SIZE);
4574
4575                 ic->recalc_bitmap = dm_integrity_alloc_page_list(n_bitmap_pages);
4576                 if (!ic->recalc_bitmap) {
4577                         r = -ENOMEM;
4578                         goto bad;
4579                 }
4580                 ic->may_write_bitmap = dm_integrity_alloc_page_list(n_bitmap_pages);
4581                 if (!ic->may_write_bitmap) {
4582                         r = -ENOMEM;
4583                         goto bad;
4584                 }
4585                 ic->bbs = kvmalloc_array(ic->n_bitmap_blocks, sizeof(struct bitmap_block_status), GFP_KERNEL);
4586                 if (!ic->bbs) {
4587                         r = -ENOMEM;
4588                         goto bad;
4589                 }
4590                 INIT_DELAYED_WORK(&ic->bitmap_flush_work, bitmap_flush_work);
4591                 for (i = 0; i < ic->n_bitmap_blocks; i++) {
4592                         struct bitmap_block_status *bbs = &ic->bbs[i];
4593                         unsigned int sector, pl_index, pl_offset;
4594
4595                         INIT_WORK(&bbs->work, bitmap_block_work);
4596                         bbs->ic = ic;
4597                         bbs->idx = i;
4598                         bio_list_init(&bbs->bio_queue);
4599                         spin_lock_init(&bbs->bio_queue_lock);
4600
4601                         sector = i * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT);
4602                         pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
4603                         pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
4604
4605                         bbs->bitmap = lowmem_page_address(ic->journal[pl_index].page) + pl_offset;
4606                 }
4607         }
4608
4609         if (should_write_sb) {
4610                 init_journal(ic, 0, ic->journal_sections, 0);
4611                 r = dm_integrity_failed(ic);
4612                 if (unlikely(r)) {
4613                         ti->error = "Error initializing journal";
4614                         goto bad;
4615                 }
4616                 r = sync_rw_sb(ic, REQ_OP_WRITE | REQ_FUA);
4617                 if (r) {
4618                         ti->error = "Error initializing superblock";
4619                         goto bad;
4620                 }
4621                 ic->just_formatted = true;
4622         }
4623
4624         if (!ic->meta_dev) {
4625                 r = dm_set_target_max_io_len(ti, 1U << ic->sb->log2_interleave_sectors);
4626                 if (r)
4627                         goto bad;
4628         }
4629         if (ic->mode == 'B') {
4630                 unsigned int max_io_len;
4631
4632                 max_io_len = ((sector_t)ic->sectors_per_block << ic->log2_blocks_per_bitmap_bit) * (BITMAP_BLOCK_SIZE * 8);
4633                 if (!max_io_len)
4634                         max_io_len = 1U << 31;
4635                 DEBUG_print("max_io_len: old %u, new %u\n", ti->max_io_len, max_io_len);
4636                 if (!ti->max_io_len || ti->max_io_len > max_io_len) {
4637                         r = dm_set_target_max_io_len(ti, max_io_len);
4638                         if (r)
4639                                 goto bad;
4640                 }
4641         }
4642
4643         if (!ic->internal_hash)
4644                 dm_integrity_set(ti, ic);
4645
4646         ti->num_flush_bios = 1;
4647         ti->flush_supported = true;
4648         if (ic->discard)
4649                 ti->num_discard_bios = 1;
4650
4651         dm_audit_log_ctr(DM_MSG_PREFIX, ti, 1);
4652         return 0;
4653
4654 bad:
4655         dm_audit_log_ctr(DM_MSG_PREFIX, ti, 0);
4656         dm_integrity_dtr(ti);
4657         return r;
4658 }
4659
4660 static void dm_integrity_dtr(struct dm_target *ti)
4661 {
4662         struct dm_integrity_c *ic = ti->private;
4663
4664         BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
4665         BUG_ON(!list_empty(&ic->wait_list));
4666
4667         if (ic->mode == 'B')
4668                 cancel_delayed_work_sync(&ic->bitmap_flush_work);
4669         if (ic->metadata_wq)
4670                 destroy_workqueue(ic->metadata_wq);
4671         if (ic->wait_wq)
4672                 destroy_workqueue(ic->wait_wq);
4673         if (ic->offload_wq)
4674                 destroy_workqueue(ic->offload_wq);
4675         if (ic->commit_wq)
4676                 destroy_workqueue(ic->commit_wq);
4677         if (ic->writer_wq)
4678                 destroy_workqueue(ic->writer_wq);
4679         if (ic->recalc_wq)
4680                 destroy_workqueue(ic->recalc_wq);
4681         kvfree(ic->bbs);
4682         if (ic->bufio)
4683                 dm_bufio_client_destroy(ic->bufio);
4684         mempool_exit(&ic->recheck_pool);
4685         mempool_exit(&ic->journal_io_mempool);
4686         if (ic->io)
4687                 dm_io_client_destroy(ic->io);
4688         if (ic->dev)
4689                 dm_put_device(ti, ic->dev);
4690         if (ic->meta_dev)
4691                 dm_put_device(ti, ic->meta_dev);
4692         dm_integrity_free_page_list(ic->journal);
4693         dm_integrity_free_page_list(ic->journal_io);
4694         dm_integrity_free_page_list(ic->journal_xor);
4695         dm_integrity_free_page_list(ic->recalc_bitmap);
4696         dm_integrity_free_page_list(ic->may_write_bitmap);
4697         if (ic->journal_scatterlist)
4698                 dm_integrity_free_journal_scatterlist(ic, ic->journal_scatterlist);
4699         if (ic->journal_io_scatterlist)
4700                 dm_integrity_free_journal_scatterlist(ic, ic->journal_io_scatterlist);
4701         if (ic->sk_requests) {
4702                 unsigned int i;
4703
4704                 for (i = 0; i < ic->journal_sections; i++) {
4705                         struct skcipher_request *req;
4706
4707                         req = ic->sk_requests[i];
4708                         if (req) {
4709                                 kfree_sensitive(req->iv);
4710                                 skcipher_request_free(req);
4711                         }
4712                 }
4713                 kvfree(ic->sk_requests);
4714         }
4715         kvfree(ic->journal_tree);
4716         if (ic->sb)
4717                 free_pages_exact(ic->sb, SB_SECTORS << SECTOR_SHIFT);
4718
4719         if (ic->internal_hash)
4720                 crypto_free_shash(ic->internal_hash);
4721         free_alg(&ic->internal_hash_alg);
4722
4723         if (ic->journal_crypt)
4724                 crypto_free_skcipher(ic->journal_crypt);
4725         free_alg(&ic->journal_crypt_alg);
4726
4727         if (ic->journal_mac)
4728                 crypto_free_shash(ic->journal_mac);
4729         free_alg(&ic->journal_mac_alg);
4730
4731         kfree(ic);
4732         dm_audit_log_dtr(DM_MSG_PREFIX, ti, 1);
4733 }
4734
4735 static struct target_type integrity_target = {
4736         .name                   = "integrity",
4737         .version                = {1, 11, 0},
4738         .module                 = THIS_MODULE,
4739         .features               = DM_TARGET_SINGLETON | DM_TARGET_INTEGRITY,
4740         .ctr                    = dm_integrity_ctr,
4741         .dtr                    = dm_integrity_dtr,
4742         .map                    = dm_integrity_map,
4743         .postsuspend            = dm_integrity_postsuspend,
4744         .resume                 = dm_integrity_resume,
4745         .status                 = dm_integrity_status,
4746         .iterate_devices        = dm_integrity_iterate_devices,
4747         .io_hints               = dm_integrity_io_hints,
4748 };
4749
4750 static int __init dm_integrity_init(void)
4751 {
4752         int r;
4753
4754         journal_io_cache = kmem_cache_create("integrity_journal_io",
4755                                              sizeof(struct journal_io), 0, 0, NULL);
4756         if (!journal_io_cache) {
4757                 DMERR("can't allocate journal io cache");
4758                 return -ENOMEM;
4759         }
4760
4761         r = dm_register_target(&integrity_target);
4762         if (r < 0) {
4763                 kmem_cache_destroy(journal_io_cache);
4764                 return r;
4765         }
4766
4767         return 0;
4768 }
4769
4770 static void __exit dm_integrity_exit(void)
4771 {
4772         dm_unregister_target(&integrity_target);
4773         kmem_cache_destroy(journal_io_cache);
4774 }
4775
4776 module_init(dm_integrity_init);
4777 module_exit(dm_integrity_exit);
4778
4779 MODULE_AUTHOR("Milan Broz");
4780 MODULE_AUTHOR("Mikulas Patocka");
4781 MODULE_DESCRIPTION(DM_NAME " target for integrity tags extension");
4782 MODULE_LICENSE("GPL");