drivers/md/dm-integrity.c

   1 /*
   2  * Copyright (C) 2016-2017 Red Hat, Inc. All rights reserved.
   3  * Copyright (C) 2016-2017 Milan Broz
   4  * Copyright (C) 2016-2017 Mikulas Patocka
   5  *
   6  * This file is released under the GPL.
   7  */
   8
   9 #include "dm-bio-record.h"
  10
  11 #include <linux/compiler.h>
  12 #include <linux/module.h>
  13 #include <linux/device-mapper.h>
  14 #include <linux/dm-io.h>
  15 #include <linux/vmalloc.h>
  16 #include <linux/sort.h>
  17 #include <linux/rbtree.h>
  18 #include <linux/delay.h>
  19 #include <linux/random.h>
  20 #include <linux/reboot.h>
  21 #include <crypto/hash.h>
  22 #include <crypto/skcipher.h>
  23 #include <linux/async_tx.h>
  24 #include <linux/dm-bufio.h>
  25
  26 #include "dm-audit.h"
  27
  28 #define DM_MSG_PREFIX "integrity"
  29
  30 #define DEFAULT_INTERLEAVE_SECTORS      32768
  31 #define DEFAULT_JOURNAL_SIZE_FACTOR     7
  32 #define DEFAULT_SECTORS_PER_BITMAP_BIT  32768
  33 #define DEFAULT_BUFFER_SECTORS          128
  34 #define DEFAULT_JOURNAL_WATERMARK       50
  35 #define DEFAULT_SYNC_MSEC               10000
  36 #define DEFAULT_MAX_JOURNAL_SECTORS     131072
  37 #define MIN_LOG2_INTERLEAVE_SECTORS     3
  38 #define MAX_LOG2_INTERLEAVE_SECTORS     31
  39 #define METADATA_WORKQUEUE_MAX_ACTIVE   16
  40 #define RECALC_SECTORS                  32768
  41 #define RECALC_WRITE_SUPER              16
  42 #define BITMAP_BLOCK_SIZE               4096    /* don't change it */
  43 #define BITMAP_FLUSH_INTERVAL           (10 * HZ)
  44 #define DISCARD_FILLER                  0xf6
  45 #define SALT_SIZE                       16
  46
  47 /*
  48  * Warning - DEBUG_PRINT prints security-sensitive data to the log,
  49  * so it should not be enabled in the official kernel
  50  */
  51 //#define DEBUG_PRINT
  52 //#define INTERNAL_VERIFY
  53
  54 /*
  55  * On disk structures
  56  */
  57
  58 #define SB_MAGIC                        "integrt"
  59 #define SB_VERSION_1                    1
  60 #define SB_VERSION_2                    2
  61 #define SB_VERSION_3                    3
  62 #define SB_VERSION_4                    4
  63 #define SB_VERSION_5                    5
  64 #define SB_SECTORS                      8
  65 #define MAX_SECTORS_PER_BLOCK           8
  66
  67 struct superblock {
  68         __u8 magic[8];
  69         __u8 version;
  70         __u8 log2_interleave_sectors;
  71         __le16 integrity_tag_size;
  72         __le32 journal_sections;
  73         __le64 provided_data_sectors;   /* userspace uses this value */
  74         __le32 flags;
  75         __u8 log2_sectors_per_block;
  76         __u8 log2_blocks_per_bitmap_bit;
  77         __u8 pad[2];
  78         __le64 recalc_sector;
  79         __u8 pad2[8];
  80         __u8 salt[SALT_SIZE];
  81 };
  82
  83 #define SB_FLAG_HAVE_JOURNAL_MAC        0x1
  84 #define SB_FLAG_RECALCULATING           0x2
  85 #define SB_FLAG_DIRTY_BITMAP            0x4
  86 #define SB_FLAG_FIXED_PADDING           0x8
  87 #define SB_FLAG_FIXED_HMAC              0x10
  88
  89 #define JOURNAL_ENTRY_ROUNDUP           8
  90
  91 typedef __le64 commit_id_t;
  92 #define JOURNAL_MAC_PER_SECTOR          8
  93
  94 struct journal_entry {
  95         union {
  96                 struct {
  97                         __le32 sector_lo;
  98                         __le32 sector_hi;
  99                 } s;
 100                 __le64 sector;
 101         } u;
 102         commit_id_t last_bytes[];
 103         /* __u8 tag[0]; */
 104 };
 105
 106 #define journal_entry_tag(ic, je)               ((__u8 *)&(je)->last_bytes[(ic)->sectors_per_block])
 107
 108 #if BITS_PER_LONG == 64
 109 #define journal_entry_set_sector(je, x)         do { smp_wmb(); WRITE_ONCE((je)->u.sector, cpu_to_le64(x)); } while (0)
 110 #else
 111 #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)
 112 #endif
 113 #define journal_entry_get_sector(je)            le64_to_cpu((je)->u.sector)
 114 #define journal_entry_is_unused(je)             ((je)->u.s.sector_hi == cpu_to_le32(-1))
 115 #define journal_entry_set_unused(je)            do { ((je)->u.s.sector_hi = cpu_to_le32(-1)); } while (0)
 116 #define journal_entry_is_inprogress(je)         ((je)->u.s.sector_hi == cpu_to_le32(-2))
 117 #define journal_entry_set_inprogress(je)        do { ((je)->u.s.sector_hi = cpu_to_le32(-2)); } while (0)
 118
 119 #define JOURNAL_BLOCK_SECTORS           8
 120 #define JOURNAL_SECTOR_DATA             ((1 << SECTOR_SHIFT) - sizeof(commit_id_t))
 121 #define JOURNAL_MAC_SIZE                (JOURNAL_MAC_PER_SECTOR * JOURNAL_BLOCK_SECTORS)
 122
 123 struct journal_sector {
 124         struct_group(sectors,
 125                 __u8 entries[JOURNAL_SECTOR_DATA - JOURNAL_MAC_PER_SECTOR];
 126                 __u8 mac[JOURNAL_MAC_PER_SECTOR];
 127         );
 128         commit_id_t commit_id;
 129 };
 130
 131 #define MAX_TAG_SIZE                    (JOURNAL_SECTOR_DATA - JOURNAL_MAC_PER_SECTOR - offsetof(struct journal_entry, last_bytes[MAX_SECTORS_PER_BLOCK]))
 132
 133 #define METADATA_PADDING_SECTORS        8
 134
 135 #define N_COMMIT_IDS                    4
 136
 137 static unsigned char prev_commit_seq(unsigned char seq)
 138 {
 139         return (seq + N_COMMIT_IDS - 1) % N_COMMIT_IDS;
 140 }
 141
 142 static unsigned char next_commit_seq(unsigned char seq)
 143 {
 144         return (seq + 1) % N_COMMIT_IDS;
 145 }
 146
 147 /*
 148  * In-memory structures
 149  */
 150
 151 struct journal_node {
 152         struct rb_node node;
 153         sector_t sector;
 154 };
 155
 156 struct alg_spec {
 157         char *alg_string;
 158         char *key_string;
 159         __u8 *key;
 160         unsigned key_size;
 161 };
 162
 163 struct dm_integrity_c {
 164         struct dm_dev *dev;
 165         struct dm_dev *meta_dev;
 166         unsigned tag_size;
 167         __s8 log2_tag_size;
 168         sector_t start;
 169         mempool_t journal_io_mempool;
 170         struct dm_io_client *io;
 171         struct dm_bufio_client *bufio;
 172         struct workqueue_struct *metadata_wq;
 173         struct superblock *sb;
 174         unsigned journal_pages;
 175         unsigned n_bitmap_blocks;
 176
 177         struct page_list *journal;
 178         struct page_list *journal_io;
 179         struct page_list *journal_xor;
 180         struct page_list *recalc_bitmap;
 181         struct page_list *may_write_bitmap;
 182         struct bitmap_block_status *bbs;
 183         unsigned bitmap_flush_interval;
 184         int synchronous_mode;
 185         struct bio_list synchronous_bios;
 186         struct delayed_work bitmap_flush_work;
 187
 188         struct crypto_skcipher *journal_crypt;
 189         struct scatterlist **journal_scatterlist;
 190         struct scatterlist **journal_io_scatterlist;
 191         struct skcipher_request **sk_requests;
 192
 193         struct crypto_shash *journal_mac;
 194
 195         struct journal_node *journal_tree;
 196         struct rb_root journal_tree_root;
 197
 198         sector_t provided_data_sectors;
 199
 200         unsigned short journal_entry_size;
 201         unsigned char journal_entries_per_sector;
 202         unsigned char journal_section_entries;
 203         unsigned short journal_section_sectors;
 204         unsigned journal_sections;
 205         unsigned journal_entries;
 206         sector_t data_device_sectors;
 207         sector_t meta_device_sectors;
 208         unsigned initial_sectors;
 209         unsigned metadata_run;
 210         __s8 log2_metadata_run;
 211         __u8 log2_buffer_sectors;
 212         __u8 sectors_per_block;
 213         __u8 log2_blocks_per_bitmap_bit;
 214
 215         unsigned char mode;
 216
 217         int failed;
 218
 219         struct crypto_shash *internal_hash;
 220
 221         struct dm_target *ti;
 222
 223         /* these variables are locked with endio_wait.lock */
 224         struct rb_root in_progress;
 225         struct list_head wait_list;
 226         wait_queue_head_t endio_wait;
 227         struct workqueue_struct *wait_wq;
 228         struct workqueue_struct *offload_wq;
 229
 230         unsigned char commit_seq;
 231         commit_id_t commit_ids[N_COMMIT_IDS];
 232
 233         unsigned committed_section;
 234         unsigned n_committed_sections;
 235
 236         unsigned uncommitted_section;
 237         unsigned n_uncommitted_sections;
 238
 239         unsigned free_section;
 240         unsigned char free_section_entry;
 241         unsigned free_sectors;
 242
 243         unsigned free_sectors_threshold;
 244
 245         struct workqueue_struct *commit_wq;
 246         struct work_struct commit_work;
 247
 248         struct workqueue_struct *writer_wq;
 249         struct work_struct writer_work;
 250
 251         struct workqueue_struct *recalc_wq;
 252         struct work_struct recalc_work;
 253         u8 *recalc_buffer;
 254         u8 *recalc_tags;
 255
 256         struct bio_list flush_bio_list;
 257
 258         unsigned long autocommit_jiffies;
 259         struct timer_list autocommit_timer;
 260         unsigned autocommit_msec;
 261
 262         wait_queue_head_t copy_to_journal_wait;
 263
 264         struct completion crypto_backoff;
 265
 266         bool wrote_to_journal;
 267         bool journal_uptodate;
 268         bool just_formatted;
 269         bool recalculate_flag;
 270         bool reset_recalculate_flag;
 271         bool discard;
 272         bool fix_padding;
 273         bool fix_hmac;
 274         bool legacy_recalculate;
 275
 276         struct alg_spec internal_hash_alg;
 277         struct alg_spec journal_crypt_alg;
 278         struct alg_spec journal_mac_alg;
 279
 280         atomic64_t number_of_mismatches;
 281
 282         struct notifier_block reboot_notifier;
 283 };
 284
 285 struct dm_integrity_range {
 286         sector_t logical_sector;
 287         sector_t n_sectors;
 288         bool waiting;
 289         union {
 290                 struct rb_node node;
 291                 struct {
 292                         struct task_struct *task;
 293                         struct list_head wait_entry;
 294                 };
 295         };
 296 };
 297
 298 struct dm_integrity_io {
 299         struct work_struct work;
 300
 301         struct dm_integrity_c *ic;
 302         enum req_op op;
 303         bool fua;
 304
 305         struct dm_integrity_range range;
 306
 307         sector_t metadata_block;
 308         unsigned metadata_offset;
 309
 310         atomic_t in_flight;
 311         blk_status_t bi_status;
 312
 313         struct completion *completion;
 314
 315         struct dm_bio_details bio_details;
 316 };
 317
 318 struct journal_completion {
 319         struct dm_integrity_c *ic;
 320         atomic_t in_flight;
 321         struct completion comp;
 322 };
 323
 324 struct journal_io {
 325         struct dm_integrity_range range;
 326         struct journal_completion *comp;
 327 };
 328
 329 struct bitmap_block_status {
 330         struct work_struct work;
 331         struct dm_integrity_c *ic;
 332         unsigned idx;
 333         unsigned long *bitmap;
 334         struct bio_list bio_queue;
 335         spinlock_t bio_queue_lock;
 336
 337 };
 338
 339 static struct kmem_cache *journal_io_cache;
 340
 341 #define JOURNAL_IO_MEMPOOL      32
 342
 343 #ifdef DEBUG_PRINT
 344 #define DEBUG_print(x, ...)     printk(KERN_DEBUG x, ##__VA_ARGS__)
 345 static void __DEBUG_bytes(__u8 *bytes, size_t len, const char *msg, ...)
 346 {
 347         va_list args;
 348         va_start(args, msg);
 349         vprintk(msg, args);
 350         va_end(args);
 351         if (len)
 352                 pr_cont(":");
 353         while (len) {
 354                 pr_cont(" %02x", *bytes);
 355                 bytes++;
 356                 len--;
 357         }
 358         pr_cont("\n");
 359 }
 360 #define DEBUG_bytes(bytes, len, msg, ...)       __DEBUG_bytes(bytes, len, KERN_DEBUG msg, ##__VA_ARGS__)
 361 #else
 362 #define DEBUG_print(x, ...)                     do { } while (0)
 363 #define DEBUG_bytes(bytes, len, msg, ...)       do { } while (0)
 364 #endif
 365
 366 static void dm_integrity_prepare(struct request *rq)
 367 {
 368 }
 369
 370 static void dm_integrity_complete(struct request *rq, unsigned int nr_bytes)
 371 {
 372 }
 373
 374 /*
 375  * DM Integrity profile, protection is performed layer above (dm-crypt)
 376  */
 377 static const struct blk_integrity_profile dm_integrity_profile = {
 378         .name                   = "DM-DIF-EXT-TAG",
 379         .generate_fn            = NULL,
 380         .verify_fn              = NULL,
 381         .prepare_fn             = dm_integrity_prepare,
 382         .complete_fn            = dm_integrity_complete,
 383 };
 384
 385 static void dm_integrity_map_continue(struct dm_integrity_io *dio, bool from_map);
 386 static void integrity_bio_wait(struct work_struct *w);
 387 static void dm_integrity_dtr(struct dm_target *ti);
 388
 389 static void dm_integrity_io_error(struct dm_integrity_c *ic, const char *msg, int err)
 390 {
 391         if (err == -EILSEQ)
 392                 atomic64_inc(&ic->number_of_mismatches);
 393         if (!cmpxchg(&ic->failed, 0, err))
 394                 DMERR("Error on %s: %d", msg, err);
 395 }
 396
 397 static int dm_integrity_failed(struct dm_integrity_c *ic)
 398 {
 399         return READ_ONCE(ic->failed);
 400 }
 401
 402 static bool dm_integrity_disable_recalculate(struct dm_integrity_c *ic)
 403 {
 404         if (ic->legacy_recalculate)
 405                 return false;
 406         if (!(ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) ?
 407             ic->internal_hash_alg.key || ic->journal_mac_alg.key :
 408             ic->internal_hash_alg.key && !ic->journal_mac_alg.key)
 409                 return true;
 410         return false;
 411 }
 412
 413 static commit_id_t dm_integrity_commit_id(struct dm_integrity_c *ic, unsigned i,
 414                                           unsigned j, unsigned char seq)
 415 {
 416         /*
 417          * Xor the number with section and sector, so that if a piece of
 418          * journal is written at wrong place, it is detected.
 419          */
 420         return ic->commit_ids[seq] ^ cpu_to_le64(((__u64)i << 32) ^ j);
 421 }
 422
 423 static void get_area_and_offset(struct dm_integrity_c *ic, sector_t data_sector,
 424                                 sector_t *area, sector_t *offset)
 425 {
 426         if (!ic->meta_dev) {
 427                 __u8 log2_interleave_sectors = ic->sb->log2_interleave_sectors;
 428                 *area = data_sector >> log2_interleave_sectors;
 429                 *offset = (unsigned)data_sector & ((1U << log2_interleave_sectors) - 1);
 430         } else {
 431                 *area = 0;
 432                 *offset = data_sector;
 433         }
 434 }
 435
 436 #define sector_to_block(ic, n)                                          \
 437 do {                                                                    \
 438         BUG_ON((n) & (unsigned)((ic)->sectors_per_block - 1));          \
 439         (n) >>= (ic)->sb->log2_sectors_per_block;                       \
 440 } while (0)
 441
 442 static __u64 get_metadata_sector_and_offset(struct dm_integrity_c *ic, sector_t area,
 443                                             sector_t offset, unsigned *metadata_offset)
 444 {
 445         __u64 ms;
 446         unsigned mo;
 447
 448         ms = area << ic->sb->log2_interleave_sectors;
 449         if (likely(ic->log2_metadata_run >= 0))
 450                 ms += area << ic->log2_metadata_run;
 451         else
 452                 ms += area * ic->metadata_run;
 453         ms >>= ic->log2_buffer_sectors;
 454
 455         sector_to_block(ic, offset);
 456
 457         if (likely(ic->log2_tag_size >= 0)) {
 458                 ms += offset >> (SECTOR_SHIFT + ic->log2_buffer_sectors - ic->log2_tag_size);
 459                 mo = (offset << ic->log2_tag_size) & ((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - 1);
 460         } else {
 461                 ms += (__u64)offset * ic->tag_size >> (SECTOR_SHIFT + ic->log2_buffer_sectors);
 462                 mo = (offset * ic->tag_size) & ((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - 1);
 463         }
 464         *metadata_offset = mo;
 465         return ms;
 466 }
 467
 468 static sector_t get_data_sector(struct dm_integrity_c *ic, sector_t area, sector_t offset)
 469 {
 470         sector_t result;
 471
 472         if (ic->meta_dev)
 473                 return offset;
 474
 475         result = area << ic->sb->log2_interleave_sectors;
 476         if (likely(ic->log2_metadata_run >= 0))
 477                 result += (area + 1) << ic->log2_metadata_run;
 478         else
 479                 result += (area + 1) * ic->metadata_run;
 480
 481         result += (sector_t)ic->initial_sectors + offset;
 482         result += ic->start;
 483
 484         return result;
 485 }
 486
 487 static void wraparound_section(struct dm_integrity_c *ic, unsigned *sec_ptr)
 488 {
 489         if (unlikely(*sec_ptr >= ic->journal_sections))
 490                 *sec_ptr -= ic->journal_sections;
 491 }
 492
 493 static void sb_set_version(struct dm_integrity_c *ic)
 494 {
 495         if (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC))
 496                 ic->sb->version = SB_VERSION_5;
 497         else if (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING))
 498                 ic->sb->version = SB_VERSION_4;
 499         else if (ic->mode == 'B' || ic->sb->flags & cpu_to_le32(SB_FLAG_DIRTY_BITMAP))
 500                 ic->sb->version = SB_VERSION_3;
 501         else if (ic->meta_dev || ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
 502                 ic->sb->version = SB_VERSION_2;
 503         else
 504                 ic->sb->version = SB_VERSION_1;
 505 }
 506
 507 static int sb_mac(struct dm_integrity_c *ic, bool wr)
 508 {
 509         SHASH_DESC_ON_STACK(desc, ic->journal_mac);
 510         int r;
 511         unsigned size = crypto_shash_digestsize(ic->journal_mac);
 512
 513         if (sizeof(struct superblock) + size > 1 << SECTOR_SHIFT) {
 514                 dm_integrity_io_error(ic, "digest is too long", -EINVAL);
 515                 return -EINVAL;
 516         }
 517
 518         desc->tfm = ic->journal_mac;
 519
 520         r = crypto_shash_init(desc);
 521         if (unlikely(r < 0)) {
 522                 dm_integrity_io_error(ic, "crypto_shash_init", r);
 523                 return r;
 524         }
 525
 526         r = crypto_shash_update(desc, (__u8 *)ic->sb, (1 << SECTOR_SHIFT) - size);
 527         if (unlikely(r < 0)) {
 528                 dm_integrity_io_error(ic, "crypto_shash_update", r);
 529                 return r;
 530         }
 531
 532         if (likely(wr)) {
 533                 r = crypto_shash_final(desc, (__u8 *)ic->sb + (1 << SECTOR_SHIFT) - size);
 534                 if (unlikely(r < 0)) {
 535                         dm_integrity_io_error(ic, "crypto_shash_final", r);
 536                         return r;
 537                 }
 538         } else {
 539                 __u8 result[HASH_MAX_DIGESTSIZE];
 540                 r = crypto_shash_final(desc, result);
 541                 if (unlikely(r < 0)) {
 542                         dm_integrity_io_error(ic, "crypto_shash_final", r);
 543                         return r;
 544                 }
 545                 if (memcmp((__u8 *)ic->sb + (1 << SECTOR_SHIFT) - size, result, size)) {
 546                         dm_integrity_io_error(ic, "superblock mac", -EILSEQ);
 547                         dm_audit_log_target(DM_MSG_PREFIX, "mac-superblock", ic->ti, 0);
 548                         return -EILSEQ;
 549                 }
 550         }
 551
 552         return 0;
 553 }
 554
 555 static int sync_rw_sb(struct dm_integrity_c *ic, blk_opf_t opf)
 556 {
 557         struct dm_io_request io_req;
 558         struct dm_io_region io_loc;
 559         const enum req_op op = opf & REQ_OP_MASK;
 560         int r;
 561
 562         io_req.bi_opf = opf;
 563         io_req.mem.type = DM_IO_KMEM;
 564         io_req.mem.ptr.addr = ic->sb;
 565         io_req.notify.fn = NULL;
 566         io_req.client = ic->io;
 567         io_loc.bdev = ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev;
 568         io_loc.sector = ic->start;
 569         io_loc.count = SB_SECTORS;
 570
 571         if (op == REQ_OP_WRITE) {
 572                 sb_set_version(ic);
 573                 if (ic->journal_mac && ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) {
 574                         r = sb_mac(ic, true);
 575                         if (unlikely(r))
 576                                 return r;
 577                 }
 578         }
 579
 580         r = dm_io(&io_req, 1, &io_loc, NULL);
 581         if (unlikely(r))
 582                 return r;
 583
 584         if (op == REQ_OP_READ) {
 585                 if (ic->mode != 'R' && ic->journal_mac && ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) {
 586                         r = sb_mac(ic, false);
 587                         if (unlikely(r))
 588                                 return r;
 589                 }
 590         }
 591
 592         return 0;
 593 }
 594
 595 #define BITMAP_OP_TEST_ALL_SET          0
 596 #define BITMAP_OP_TEST_ALL_CLEAR        1
 597 #define BITMAP_OP_SET                   2
 598 #define BITMAP_OP_CLEAR                 3
 599
 600 static bool block_bitmap_op(struct dm_integrity_c *ic, struct page_list *bitmap,
 601                             sector_t sector, sector_t n_sectors, int mode)
 602 {
 603         unsigned long bit, end_bit, this_end_bit, page, end_page;
 604         unsigned long *data;
 605
 606         if (unlikely(((sector | n_sectors) & ((1 << ic->sb->log2_sectors_per_block) - 1)) != 0)) {
 607                 DMCRIT("invalid bitmap access (%llx,%llx,%d,%d,%d)",
 608                         sector,
 609                         n_sectors,
 610                         ic->sb->log2_sectors_per_block,
 611                         ic->log2_blocks_per_bitmap_bit,
 612                         mode);
 613                 BUG();
 614         }
 615
 616         if (unlikely(!n_sectors))
 617                 return true;
 618
 619         bit = sector >> (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
 620         end_bit = (sector + n_sectors - 1) >>
 621                 (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
 622
 623         page = bit / (PAGE_SIZE * 8);
 624         bit %= PAGE_SIZE * 8;
 625
 626         end_page = end_bit / (PAGE_SIZE * 8);
 627         end_bit %= PAGE_SIZE * 8;
 628
 629 repeat:
 630         if (page < end_page) {
 631                 this_end_bit = PAGE_SIZE * 8 - 1;
 632         } else {
 633                 this_end_bit = end_bit;
 634         }
 635
 636         data = lowmem_page_address(bitmap[page].page);
 637
 638         if (mode == BITMAP_OP_TEST_ALL_SET) {
 639                 while (bit <= this_end_bit) {
 640                         if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
 641                                 do {
 642                                         if (data[bit / BITS_PER_LONG] != -1)
 643                                                 return false;
 644                                         bit += BITS_PER_LONG;
 645                                 } while (this_end_bit >= bit + BITS_PER_LONG - 1);
 646                                 continue;
 647                         }
 648                         if (!test_bit(bit, data))
 649                                 return false;
 650                         bit++;
 651                 }
 652         } else if (mode == BITMAP_OP_TEST_ALL_CLEAR) {
 653                 while (bit <= this_end_bit) {
 654                         if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
 655                                 do {
 656                                         if (data[bit / BITS_PER_LONG] != 0)
 657                                                 return false;
 658                                         bit += BITS_PER_LONG;
 659                                 } while (this_end_bit >= bit + BITS_PER_LONG - 1);
 660                                 continue;
 661                         }
 662                         if (test_bit(bit, data))
 663                                 return false;
 664                         bit++;
 665                 }
 666         } else if (mode == BITMAP_OP_SET) {
 667                 while (bit <= this_end_bit) {
 668                         if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
 669                                 do {
 670                                         data[bit / BITS_PER_LONG] = -1;
 671                                         bit += BITS_PER_LONG;
 672                                 } while (this_end_bit >= bit + BITS_PER_LONG - 1);
 673                                 continue;
 674                         }
 675                         __set_bit(bit, data);
 676                         bit++;
 677                 }
 678         } else if (mode == BITMAP_OP_CLEAR) {
 679                 if (!bit && this_end_bit == PAGE_SIZE * 8 - 1)
 680                         clear_page(data);
 681                 else while (bit <= this_end_bit) {
 682                         if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
 683                                 do {
 684                                         data[bit / BITS_PER_LONG] = 0;
 685                                         bit += BITS_PER_LONG;
 686                                 } while (this_end_bit >= bit + BITS_PER_LONG - 1);
 687                                 continue;
 688                         }
 689                         __clear_bit(bit, data);
 690                         bit++;
 691                 }
 692         } else {
 693                 BUG();
 694         }
 695
 696         if (unlikely(page < end_page)) {
 697                 bit = 0;
 698                 page++;
 699                 goto repeat;
 700         }
 701
 702         return true;
 703 }
 704
 705 static void block_bitmap_copy(struct dm_integrity_c *ic, struct page_list *dst, struct page_list *src)
 706 {
 707         unsigned n_bitmap_pages = DIV_ROUND_UP(ic->n_bitmap_blocks, PAGE_SIZE / BITMAP_BLOCK_SIZE);
 708         unsigned i;
 709
 710         for (i = 0; i < n_bitmap_pages; i++) {
 711                 unsigned long *dst_data = lowmem_page_address(dst[i].page);
 712                 unsigned long *src_data = lowmem_page_address(src[i].page);
 713                 copy_page(dst_data, src_data);
 714         }
 715 }
 716
 717 static struct bitmap_block_status *sector_to_bitmap_block(struct dm_integrity_c *ic, sector_t sector)
 718 {
 719         unsigned bit = sector >> (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
 720         unsigned bitmap_block = bit / (BITMAP_BLOCK_SIZE * 8);
 721
 722         BUG_ON(bitmap_block >= ic->n_bitmap_blocks);
 723         return &ic->bbs[bitmap_block];
 724 }
 725
 726 static void access_journal_check(struct dm_integrity_c *ic, unsigned section, unsigned offset,
 727                                  bool e, const char *function)
 728 {
 729 #if defined(CONFIG_DM_DEBUG) || defined(INTERNAL_VERIFY)
 730         unsigned limit = e ? ic->journal_section_entries : ic->journal_section_sectors;
 731
 732         if (unlikely(section >= ic->journal_sections) ||
 733             unlikely(offset >= limit)) {
 734                 DMCRIT("%s: invalid access at (%u,%u), limit (%u,%u)",
 735                        function, section, offset, ic->journal_sections, limit);
 736                 BUG();
 737         }
 738 #endif
 739 }
 740
 741 static void page_list_location(struct dm_integrity_c *ic, unsigned section, unsigned offset,
 742                                unsigned *pl_index, unsigned *pl_offset)
 743 {
 744         unsigned sector;
 745
 746         access_journal_check(ic, section, offset, false, "page_list_location");
 747
 748         sector = section * ic->journal_section_sectors + offset;
 749
 750         *pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
 751         *pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
 752 }
 753
 754 static struct journal_sector *access_page_list(struct dm_integrity_c *ic, struct page_list *pl,
 755                                                unsigned section, unsigned offset, unsigned *n_sectors)
 756 {
 757         unsigned pl_index, pl_offset;
 758         char *va;
 759
 760         page_list_location(ic, section, offset, &pl_index, &pl_offset);
 761
 762         if (n_sectors)
 763                 *n_sectors = (PAGE_SIZE - pl_offset) >> SECTOR_SHIFT;
 764
 765         va = lowmem_page_address(pl[pl_index].page);
 766
 767         return (struct journal_sector *)(va + pl_offset);
 768 }
 769
 770 static struct journal_sector *access_journal(struct dm_integrity_c *ic, unsigned section, unsigned offset)
 771 {
 772         return access_page_list(ic, ic->journal, section, offset, NULL);
 773 }
 774
 775 static struct journal_entry *access_journal_entry(struct dm_integrity_c *ic, unsigned section, unsigned n)
 776 {
 777         unsigned rel_sector, offset;
 778         struct journal_sector *js;
 779
 780         access_journal_check(ic, section, n, true, "access_journal_entry");
 781
 782         rel_sector = n % JOURNAL_BLOCK_SECTORS;
 783         offset = n / JOURNAL_BLOCK_SECTORS;
 784
 785         js = access_journal(ic, section, rel_sector);
 786         return (struct journal_entry *)((char *)js + offset * ic->journal_entry_size);
 787 }
 788
 789 static struct journal_sector *access_journal_data(struct dm_integrity_c *ic, unsigned section, unsigned n)
 790 {
 791         n <<= ic->sb->log2_sectors_per_block;
 792
 793         n += JOURNAL_BLOCK_SECTORS;
 794
 795         access_journal_check(ic, section, n, false, "access_journal_data");
 796
 797         return access_journal(ic, section, n);
 798 }
 799
 800 static void section_mac(struct dm_integrity_c *ic, unsigned section, __u8 result[JOURNAL_MAC_SIZE])
 801 {
 802         SHASH_DESC_ON_STACK(desc, ic->journal_mac);
 803         int r;
 804         unsigned j, size;
 805
 806         desc->tfm = ic->journal_mac;
 807
 808         r = crypto_shash_init(desc);
 809         if (unlikely(r < 0)) {
 810                 dm_integrity_io_error(ic, "crypto_shash_init", r);
 811                 goto err;
 812         }
 813
 814         if (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) {
 815                 __le64 section_le;
 816
 817                 r = crypto_shash_update(desc, (__u8 *)&ic->sb->salt, SALT_SIZE);
 818                 if (unlikely(r < 0)) {
 819                         dm_integrity_io_error(ic, "crypto_shash_update", r);
 820                         goto err;
 821                 }
 822
 823                 section_le = cpu_to_le64(section);
 824                 r = crypto_shash_update(desc, (__u8 *)&section_le, sizeof section_le);
 825                 if (unlikely(r < 0)) {
 826                         dm_integrity_io_error(ic, "crypto_shash_update", r);
 827                         goto err;
 828                 }
 829         }
 830
 831         for (j = 0; j < ic->journal_section_entries; j++) {
 832                 struct journal_entry *je = access_journal_entry(ic, section, j);
 833                 r = crypto_shash_update(desc, (__u8 *)&je->u.sector, sizeof je->u.sector);
 834                 if (unlikely(r < 0)) {
 835                         dm_integrity_io_error(ic, "crypto_shash_update", r);
 836                         goto err;
 837                 }
 838         }
 839
 840         size = crypto_shash_digestsize(ic->journal_mac);
 841
 842         if (likely(size <= JOURNAL_MAC_SIZE)) {
 843                 r = crypto_shash_final(desc, result);
 844                 if (unlikely(r < 0)) {
 845                         dm_integrity_io_error(ic, "crypto_shash_final", r);
 846                         goto err;
 847                 }
 848                 memset(result + size, 0, JOURNAL_MAC_SIZE - size);
 849         } else {
 850                 __u8 digest[HASH_MAX_DIGESTSIZE];
 851
 852                 if (WARN_ON(size > sizeof(digest))) {
 853                         dm_integrity_io_error(ic, "digest_size", -EINVAL);
 854                         goto err;
 855                 }
 856                 r = crypto_shash_final(desc, digest);
 857                 if (unlikely(r < 0)) {
 858                         dm_integrity_io_error(ic, "crypto_shash_final", r);
 859                         goto err;
 860                 }
 861                 memcpy(result, digest, JOURNAL_MAC_SIZE);
 862         }
 863
 864         return;
 865 err:
 866         memset(result, 0, JOURNAL_MAC_SIZE);
 867 }
 868
 869 static void rw_section_mac(struct dm_integrity_c *ic, unsigned section, bool wr)
 870 {
 871         __u8 result[JOURNAL_MAC_SIZE];
 872         unsigned j;
 873
 874         if (!ic->journal_mac)
 875                 return;
 876
 877         section_mac(ic, section, result);
 878
 879         for (j = 0; j < JOURNAL_BLOCK_SECTORS; j++) {
 880                 struct journal_sector *js = access_journal(ic, section, j);
 881
 882                 if (likely(wr))
 883                         memcpy(&js->mac, result + (j * JOURNAL_MAC_PER_SECTOR), JOURNAL_MAC_PER_SECTOR);
 884                 else {
 885                         if (memcmp(&js->mac, result + (j * JOURNAL_MAC_PER_SECTOR), JOURNAL_MAC_PER_SECTOR)) {
 886                                 dm_integrity_io_error(ic, "journal mac", -EILSEQ);
 887                                 dm_audit_log_target(DM_MSG_PREFIX, "mac-journal", ic->ti, 0);
 888                         }
 889                 }
 890         }
 891 }
 892
 893 static void complete_journal_op(void *context)
 894 {
 895         struct journal_completion *comp = context;
 896         BUG_ON(!atomic_read(&comp->in_flight));
 897         if (likely(atomic_dec_and_test(&comp->in_flight)))
 898                 complete(&comp->comp);
 899 }
 900
 901 static void xor_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
 902                         unsigned n_sections, struct journal_completion *comp)
 903 {
 904         struct async_submit_ctl submit;
 905         size_t n_bytes = (size_t)(n_sections * ic->journal_section_sectors) << SECTOR_SHIFT;
 906         unsigned pl_index, pl_offset, section_index;
 907         struct page_list *source_pl, *target_pl;
 908
 909         if (likely(encrypt)) {
 910                 source_pl = ic->journal;
 911                 target_pl = ic->journal_io;
 912         } else {
 913                 source_pl = ic->journal_io;
 914                 target_pl = ic->journal;
 915         }
 916
 917         page_list_location(ic, section, 0, &pl_index, &pl_offset);
 918
 919         atomic_add(roundup(pl_offset + n_bytes, PAGE_SIZE) >> PAGE_SHIFT, &comp->in_flight);
 920
 921         init_async_submit(&submit, ASYNC_TX_XOR_ZERO_DST, NULL, complete_journal_op, comp, NULL);
 922
 923         section_index = pl_index;
 924
 925         do {
 926                 size_t this_step;
 927                 struct page *src_pages[2];
 928                 struct page *dst_page;
 929
 930                 while (unlikely(pl_index == section_index)) {
 931                         unsigned dummy;
 932                         if (likely(encrypt))
 933                                 rw_section_mac(ic, section, true);
 934                         section++;
 935                         n_sections--;
 936                         if (!n_sections)
 937                                 break;
 938                         page_list_location(ic, section, 0, &section_index, &dummy);
 939                 }
 940
 941                 this_step = min(n_bytes, (size_t)PAGE_SIZE - pl_offset);
 942                 dst_page = target_pl[pl_index].page;
 943                 src_pages[0] = source_pl[pl_index].page;
 944                 src_pages[1] = ic->journal_xor[pl_index].page;
 945
 946                 async_xor(dst_page, src_pages, pl_offset, 2, this_step, &submit);
 947
 948                 pl_index++;
 949                 pl_offset = 0;
 950                 n_bytes -= this_step;
 951         } while (n_bytes);
 952
 953         BUG_ON(n_sections);
 954
 955         async_tx_issue_pending_all();
 956 }
 957
 958 static void complete_journal_encrypt(struct crypto_async_request *req, int err)
 959 {
 960         struct journal_completion *comp = req->data;
 961         if (unlikely(err)) {
 962                 if (likely(err == -EINPROGRESS)) {
 963                         complete(&comp->ic->crypto_backoff);
 964                         return;
 965                 }
 966                 dm_integrity_io_error(comp->ic, "asynchronous encrypt", err);
 967         }
 968         complete_journal_op(comp);
 969 }
 970
 971 static bool do_crypt(bool encrypt, struct skcipher_request *req, struct journal_completion *comp)
 972 {
 973         int r;
 974         skcipher_request_set_callback(req, CRYPTO_TFM_REQ_MAY_BACKLOG,
 975                                       complete_journal_encrypt, comp);
 976         if (likely(encrypt))
 977                 r = crypto_skcipher_encrypt(req);
 978         else
 979                 r = crypto_skcipher_decrypt(req);
 980         if (likely(!r))
 981                 return false;
 982         if (likely(r == -EINPROGRESS))
 983                 return true;
 984         if (likely(r == -EBUSY)) {
 985                 wait_for_completion(&comp->ic->crypto_backoff);
 986                 reinit_completion(&comp->ic->crypto_backoff);
 987                 return true;
 988         }
 989         dm_integrity_io_error(comp->ic, "encrypt", r);
 990         return false;
 991 }
 992
 993 static void crypt_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
 994                           unsigned n_sections, struct journal_completion *comp)
 995 {
 996         struct scatterlist **source_sg;
 997         struct scatterlist **target_sg;
 998
 999         atomic_add(2, &comp->in_flight);
1000
1001         if (likely(encrypt)) {
1002                 source_sg = ic->journal_scatterlist;
1003                 target_sg = ic->journal_io_scatterlist;
1004         } else {
1005                 source_sg = ic->journal_io_scatterlist;
1006                 target_sg = ic->journal_scatterlist;
1007         }
1008
1009         do {
1010                 struct skcipher_request *req;
1011                 unsigned ivsize;
1012                 char *iv;
1013
1014                 if (likely(encrypt))
1015                         rw_section_mac(ic, section, true);
1016
1017                 req = ic->sk_requests[section];
1018                 ivsize = crypto_skcipher_ivsize(ic->journal_crypt);
1019                 iv = req->iv;
1020
1021                 memcpy(iv, iv + ivsize, ivsize);
1022
1023                 req->src = source_sg[section];
1024                 req->dst = target_sg[section];
1025
1026                 if (unlikely(do_crypt(encrypt, req, comp)))
1027                         atomic_inc(&comp->in_flight);
1028
1029                 section++;
1030                 n_sections--;
1031         } while (n_sections);
1032
1033         atomic_dec(&comp->in_flight);
1034         complete_journal_op(comp);
1035 }
1036
1037 static void encrypt_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
1038                             unsigned n_sections, struct journal_completion *comp)
1039 {
1040         if (ic->journal_xor)
1041                 return xor_journal(ic, encrypt, section, n_sections, comp);
1042         else
1043                 return crypt_journal(ic, encrypt, section, n_sections, comp);
1044 }
1045
1046 static void complete_journal_io(unsigned long error, void *context)
1047 {
1048         struct journal_completion *comp = context;
1049         if (unlikely(error != 0))
1050                 dm_integrity_io_error(comp->ic, "writing journal", -EIO);
1051         complete_journal_op(comp);
1052 }
1053
1054 static void rw_journal_sectors(struct dm_integrity_c *ic, blk_opf_t opf,
1055                                unsigned sector, unsigned n_sectors,
1056                                struct journal_completion *comp)
1057 {
1058         struct dm_io_request io_req;
1059         struct dm_io_region io_loc;
1060         unsigned pl_index, pl_offset;
1061         int r;
1062
1063         if (unlikely(dm_integrity_failed(ic))) {
1064                 if (comp)
1065                         complete_journal_io(-1UL, comp);
1066                 return;
1067         }
1068
1069         pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
1070         pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
1071
1072         io_req.bi_opf = opf;
1073         io_req.mem.type = DM_IO_PAGE_LIST;
1074         if (ic->journal_io)
1075                 io_req.mem.ptr.pl = &ic->journal_io[pl_index];
1076         else
1077                 io_req.mem.ptr.pl = &ic->journal[pl_index];
1078         io_req.mem.offset = pl_offset;
1079         if (likely(comp != NULL)) {
1080                 io_req.notify.fn = complete_journal_io;
1081                 io_req.notify.context = comp;
1082         } else {
1083                 io_req.notify.fn = NULL;
1084         }
1085         io_req.client = ic->io;
1086         io_loc.bdev = ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev;
1087         io_loc.sector = ic->start + SB_SECTORS + sector;
1088         io_loc.count = n_sectors;
1089
1090         r = dm_io(&io_req, 1, &io_loc, NULL);
1091         if (unlikely(r)) {
1092                 dm_integrity_io_error(ic, (opf & REQ_OP_MASK) == REQ_OP_READ ?
1093                                       "reading journal" : "writing journal", r);
1094                 if (comp) {
1095                         WARN_ONCE(1, "asynchronous dm_io failed: %d", r);
1096                         complete_journal_io(-1UL, comp);
1097                 }
1098         }
1099 }
1100
1101 static void rw_journal(struct dm_integrity_c *ic, blk_opf_t opf,
1102                        unsigned section, unsigned n_sections,
1103                        struct journal_completion *comp)
1104 {
1105         unsigned sector, n_sectors;
1106
1107         sector = section * ic->journal_section_sectors;
1108         n_sectors = n_sections * ic->journal_section_sectors;
1109
1110         rw_journal_sectors(ic, opf, sector, n_sectors, comp);
1111 }
1112
1113 static void write_journal(struct dm_integrity_c *ic, unsigned commit_start, unsigned commit_sections)
1114 {
1115         struct journal_completion io_comp;
1116         struct journal_completion crypt_comp_1;
1117         struct journal_completion crypt_comp_2;
1118         unsigned i;
1119
1120         io_comp.ic = ic;
1121         init_completion(&io_comp.comp);
1122
1123         if (commit_start + commit_sections <= ic->journal_sections) {
1124                 io_comp.in_flight = (atomic_t)ATOMIC_INIT(1);
1125                 if (ic->journal_io) {
1126                         crypt_comp_1.ic = ic;
1127                         init_completion(&crypt_comp_1.comp);
1128                         crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
1129                         encrypt_journal(ic, true, commit_start, commit_sections, &crypt_comp_1);
1130                         wait_for_completion_io(&crypt_comp_1.comp);
1131                 } else {
1132                         for (i = 0; i < commit_sections; i++)
1133                                 rw_section_mac(ic, commit_start + i, true);
1134                 }
1135                 rw_journal(ic, REQ_OP_WRITE | REQ_FUA | REQ_SYNC, commit_start,
1136                            commit_sections, &io_comp);
1137         } else {
1138                 unsigned to_end;
1139                 io_comp.in_flight = (atomic_t)ATOMIC_INIT(2);
1140                 to_end = ic->journal_sections - commit_start;
1141                 if (ic->journal_io) {
1142                         crypt_comp_1.ic = ic;
1143                         init_completion(&crypt_comp_1.comp);
1144                         crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
1145                         encrypt_journal(ic, true, commit_start, to_end, &crypt_comp_1);
1146                         if (try_wait_for_completion(&crypt_comp_1.comp)) {
1147                                 rw_journal(ic, REQ_OP_WRITE | REQ_FUA,
1148                                            commit_start, to_end, &io_comp);
1149                                 reinit_completion(&crypt_comp_1.comp);
1150                                 crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
1151                                 encrypt_journal(ic, true, 0, commit_sections - to_end, &crypt_comp_1);
1152                                 wait_for_completion_io(&crypt_comp_1.comp);
1153                         } else {
1154                                 crypt_comp_2.ic = ic;
1155                                 init_completion(&crypt_comp_2.comp);
1156                                 crypt_comp_2.in_flight = (atomic_t)ATOMIC_INIT(0);
1157                                 encrypt_journal(ic, true, 0, commit_sections - to_end, &crypt_comp_2);
1158                                 wait_for_completion_io(&crypt_comp_1.comp);
1159                                 rw_journal(ic, REQ_OP_WRITE | REQ_FUA, commit_start, to_end, &io_comp);
1160                                 wait_for_completion_io(&crypt_comp_2.comp);
1161                         }
1162                 } else {
1163                         for (i = 0; i < to_end; i++)
1164                                 rw_section_mac(ic, commit_start + i, true);
1165                         rw_journal(ic, REQ_OP_WRITE | REQ_FUA, commit_start, to_end, &io_comp);
1166                         for (i = 0; i < commit_sections - to_end; i++)
1167                                 rw_section_mac(ic, i, true);
1168                 }
1169                 rw_journal(ic, REQ_OP_WRITE | REQ_FUA, 0, commit_sections - to_end, &io_comp);
1170         }
1171
1172         wait_for_completion_io(&io_comp.comp);
1173 }
1174
1175 static void copy_from_journal(struct dm_integrity_c *ic, unsigned section, unsigned offset,
1176                               unsigned n_sectors, sector_t target, io_notify_fn fn, void *data)
1177 {
1178         struct dm_io_request io_req;
1179         struct dm_io_region io_loc;
1180         int r;
1181         unsigned sector, pl_index, pl_offset;
1182
1183         BUG_ON((target | n_sectors | offset) & (unsigned)(ic->sectors_per_block - 1));
1184
1185         if (unlikely(dm_integrity_failed(ic))) {
1186                 fn(-1UL, data);
1187                 return;
1188         }
1189
1190         sector = section * ic->journal_section_sectors + JOURNAL_BLOCK_SECTORS + offset;
1191
1192         pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
1193         pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
1194
1195         io_req.bi_opf = REQ_OP_WRITE;
1196         io_req.mem.type = DM_IO_PAGE_LIST;
1197         io_req.mem.ptr.pl = &ic->journal[pl_index];
1198         io_req.mem.offset = pl_offset;
1199         io_req.notify.fn = fn;
1200         io_req.notify.context = data;
1201         io_req.client = ic->io;
1202         io_loc.bdev = ic->dev->bdev;
1203         io_loc.sector = target;
1204         io_loc.count = n_sectors;
1205
1206         r = dm_io(&io_req, 1, &io_loc, NULL);
1207         if (unlikely(r)) {
1208                 WARN_ONCE(1, "asynchronous dm_io failed: %d", r);
1209                 fn(-1UL, data);
1210         }
1211 }
1212
1213 static bool ranges_overlap(struct dm_integrity_range *range1, struct dm_integrity_range *range2)
1214 {
1215         return range1->logical_sector < range2->logical_sector + range2->n_sectors &&
1216                range1->logical_sector + range1->n_sectors > range2->logical_sector;
1217 }
1218
1219 static bool add_new_range(struct dm_integrity_c *ic, struct dm_integrity_range *new_range, bool check_waiting)
1220 {
1221         struct rb_node **n = &ic->in_progress.rb_node;
1222         struct rb_node *parent;
1223
1224         BUG_ON((new_range->logical_sector | new_range->n_sectors) & (unsigned)(ic->sectors_per_block - 1));
1225
1226         if (likely(check_waiting)) {
1227                 struct dm_integrity_range *range;
1228                 list_for_each_entry(range, &ic->wait_list, wait_entry) {
1229                         if (unlikely(ranges_overlap(range, new_range)))
1230                                 return false;
1231                 }
1232         }
1233
1234         parent = NULL;
1235
1236         while (*n) {
1237                 struct dm_integrity_range *range = container_of(*n, struct dm_integrity_range, node);
1238
1239                 parent = *n;
1240                 if (new_range->logical_sector + new_range->n_sectors <= range->logical_sector) {
1241                         n = &range->node.rb_left;
1242                 } else if (new_range->logical_sector >= range->logical_sector + range->n_sectors) {
1243                         n = &range->node.rb_right;
1244                 } else {
1245                         return false;
1246                 }
1247         }
1248
1249         rb_link_node(&new_range->node, parent, n);
1250         rb_insert_color(&new_range->node, &ic->in_progress);
1251
1252         return true;
1253 }
1254
1255 static void remove_range_unlocked(struct dm_integrity_c *ic, struct dm_integrity_range *range)
1256 {
1257         rb_erase(&range->node, &ic->in_progress);
1258         while (unlikely(!list_empty(&ic->wait_list))) {
1259                 struct dm_integrity_range *last_range =
1260                         list_first_entry(&ic->wait_list, struct dm_integrity_range, wait_entry);
1261                 struct task_struct *last_range_task;
1262                 last_range_task = last_range->task;
1263                 list_del(&last_range->wait_entry);
1264                 if (!add_new_range(ic, last_range, false)) {
1265                         last_range->task = last_range_task;
1266                         list_add(&last_range->wait_entry, &ic->wait_list);
1267                         break;
1268                 }
1269                 last_range->waiting = false;
1270                 wake_up_process(last_range_task);
1271         }
1272 }
1273
1274 static void remove_range(struct dm_integrity_c *ic, struct dm_integrity_range *range)
1275 {
1276         unsigned long flags;
1277
1278         spin_lock_irqsave(&ic->endio_wait.lock, flags);
1279         remove_range_unlocked(ic, range);
1280         spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
1281 }
1282
1283 static void wait_and_add_new_range(struct dm_integrity_c *ic, struct dm_integrity_range *new_range)
1284 {
1285         new_range->waiting = true;
1286         list_add_tail(&new_range->wait_entry, &ic->wait_list);
1287         new_range->task = current;
1288         do {
1289                 __set_current_state(TASK_UNINTERRUPTIBLE);
1290                 spin_unlock_irq(&ic->endio_wait.lock);
1291                 io_schedule();
1292                 spin_lock_irq(&ic->endio_wait.lock);
1293         } while (unlikely(new_range->waiting));
1294 }
1295
1296 static void add_new_range_and_wait(struct dm_integrity_c *ic, struct dm_integrity_range *new_range)
1297 {
1298         if (unlikely(!add_new_range(ic, new_range, true)))
1299                 wait_and_add_new_range(ic, new_range);
1300 }
1301
1302 static void init_journal_node(struct journal_node *node)
1303 {
1304         RB_CLEAR_NODE(&node->node);
1305         node->sector = (sector_t)-1;
1306 }
1307
1308 static void add_journal_node(struct dm_integrity_c *ic, struct journal_node *node, sector_t sector)
1309 {
1310         struct rb_node **link;
1311         struct rb_node *parent;
1312
1313         node->sector = sector;
1314         BUG_ON(!RB_EMPTY_NODE(&node->node));
1315
1316         link = &ic->journal_tree_root.rb_node;
1317         parent = NULL;
1318
1319         while (*link) {
1320                 struct journal_node *j;
1321                 parent = *link;
1322                 j = container_of(parent, struct journal_node, node);
1323                 if (sector < j->sector)
1324                         link = &j->node.rb_left;
1325                 else
1326                         link = &j->node.rb_right;
1327         }
1328
1329         rb_link_node(&node->node, parent, link);
1330         rb_insert_color(&node->node, &ic->journal_tree_root);
1331 }
1332
1333 static void remove_journal_node(struct dm_integrity_c *ic, struct journal_node *node)
1334 {
1335         BUG_ON(RB_EMPTY_NODE(&node->node));
1336         rb_erase(&node->node, &ic->journal_tree_root);
1337         init_journal_node(node);
1338 }
1339
1340 #define NOT_FOUND       (-1U)
1341
1342 static unsigned find_journal_node(struct dm_integrity_c *ic, sector_t sector, sector_t *next_sector)
1343 {
1344         struct rb_node *n = ic->journal_tree_root.rb_node;
1345         unsigned found = NOT_FOUND;
1346         *next_sector = (sector_t)-1;
1347         while (n) {
1348                 struct journal_node *j = container_of(n, struct journal_node, node);
1349                 if (sector == j->sector) {
1350                         found = j - ic->journal_tree;
1351                 }
1352                 if (sector < j->sector) {
1353                         *next_sector = j->sector;
1354                         n = j->node.rb_left;
1355                 } else {
1356                         n = j->node.rb_right;
1357                 }
1358         }
1359
1360         return found;
1361 }
1362
1363 static bool test_journal_node(struct dm_integrity_c *ic, unsigned pos, sector_t sector)
1364 {
1365         struct journal_node *node, *next_node;
1366         struct rb_node *next;
1367
1368         if (unlikely(pos >= ic->journal_entries))
1369                 return false;
1370         node = &ic->journal_tree[pos];
1371         if (unlikely(RB_EMPTY_NODE(&node->node)))
1372                 return false;
1373         if (unlikely(node->sector != sector))
1374                 return false;
1375
1376         next = rb_next(&node->node);
1377         if (unlikely(!next))
1378                 return true;
1379
1380         next_node = container_of(next, struct journal_node, node);
1381         return next_node->sector != sector;
1382 }
1383
1384 static bool find_newer_committed_node(struct dm_integrity_c *ic, struct journal_node *node)
1385 {
1386         struct rb_node *next;
1387         struct journal_node *next_node;
1388         unsigned next_section;
1389
1390         BUG_ON(RB_EMPTY_NODE(&node->node));
1391
1392         next = rb_next(&node->node);
1393         if (unlikely(!next))
1394                 return false;
1395
1396         next_node = container_of(next, struct journal_node, node);
1397
1398         if (next_node->sector != node->sector)
1399                 return false;
1400
1401         next_section = (unsigned)(next_node - ic->journal_tree) / ic->journal_section_entries;
1402         if (next_section >= ic->committed_section &&
1403             next_section < ic->committed_section + ic->n_committed_sections)
1404                 return true;
1405         if (next_section + ic->journal_sections < ic->committed_section + ic->n_committed_sections)
1406                 return true;
1407
1408         return false;
1409 }
1410
1411 #define TAG_READ        0
1412 #define TAG_WRITE       1
1413 #define TAG_CMP         2
1414
1415 static int dm_integrity_rw_tag(struct dm_integrity_c *ic, unsigned char *tag, sector_t *metadata_block,
1416                                unsigned *metadata_offset, unsigned total_size, int op)
1417 {
1418 #define MAY_BE_FILLER           1
1419 #define MAY_BE_HASH             2
1420         unsigned hash_offset = 0;
1421         unsigned may_be = MAY_BE_HASH | (ic->discard ? MAY_BE_FILLER : 0);
1422
1423         do {
1424                 unsigned char *data, *dp;
1425                 struct dm_buffer *b;
1426                 unsigned to_copy;
1427                 int r;
1428
1429                 r = dm_integrity_failed(ic);
1430                 if (unlikely(r))
1431                         return r;
1432
1433                 data = dm_bufio_read(ic->bufio, *metadata_block, &b);
1434                 if (IS_ERR(data))
1435                         return PTR_ERR(data);
1436
1437                 to_copy = min((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - *metadata_offset, total_size);
1438                 dp = data + *metadata_offset;
1439                 if (op == TAG_READ) {
1440                         memcpy(tag, dp, to_copy);
1441                 } else if (op == TAG_WRITE) {
1442                         if (memcmp(dp, tag, to_copy)) {
1443                                 memcpy(dp, tag, to_copy);
1444                                 dm_bufio_mark_partial_buffer_dirty(b, *metadata_offset, *metadata_offset + to_copy);
1445                         }
1446                 } else {
1447                         /* e.g.: op == TAG_CMP */
1448
1449                         if (likely(is_power_of_2(ic->tag_size))) {
1450                                 if (unlikely(memcmp(dp, tag, to_copy)))
1451                                         if (unlikely(!ic->discard) ||
1452                                             unlikely(memchr_inv(dp, DISCARD_FILLER, to_copy) != NULL)) {
1453                                                 goto thorough_test;
1454                                 }
1455                         } else {
1456                                 unsigned i, ts;
1457 thorough_test:
1458                                 ts = total_size;
1459
1460                                 for (i = 0; i < to_copy; i++, ts--) {
1461                                         if (unlikely(dp[i] != tag[i]))
1462                                                 may_be &= ~MAY_BE_HASH;
1463                                         if (likely(dp[i] != DISCARD_FILLER))
1464                                                 may_be &= ~MAY_BE_FILLER;
1465                                         hash_offset++;
1466                                         if (unlikely(hash_offset == ic->tag_size)) {
1467                                                 if (unlikely(!may_be)) {
1468                                                         dm_bufio_release(b);
1469                                                         return ts;
1470                                                 }
1471                                                 hash_offset = 0;
1472                                                 may_be = MAY_BE_HASH | (ic->discard ? MAY_BE_FILLER : 0);
1473                                         }
1474                                 }
1475                         }
1476                 }
1477                 dm_bufio_release(b);
1478
1479                 tag += to_copy;
1480                 *metadata_offset += to_copy;
1481                 if (unlikely(*metadata_offset == 1U << SECTOR_SHIFT << ic->log2_buffer_sectors)) {
1482                         (*metadata_block)++;
1483                         *metadata_offset = 0;
1484                 }
1485
1486                 if (unlikely(!is_power_of_2(ic->tag_size))) {
1487                         hash_offset = (hash_offset + to_copy) % ic->tag_size;
1488                 }
1489
1490                 total_size -= to_copy;
1491         } while (unlikely(total_size));
1492
1493         return 0;
1494 #undef MAY_BE_FILLER
1495 #undef MAY_BE_HASH
1496 }
1497
1498 struct flush_request {
1499         struct dm_io_request io_req;
1500         struct dm_io_region io_reg;
1501         struct dm_integrity_c *ic;
1502         struct completion comp;
1503 };
1504
1505 static void flush_notify(unsigned long error, void *fr_)
1506 {
1507         struct flush_request *fr = fr_;
1508         if (unlikely(error != 0))
1509                 dm_integrity_io_error(fr->ic, "flushing disk cache", -EIO);
1510         complete(&fr->comp);
1511 }
1512
1513 static void dm_integrity_flush_buffers(struct dm_integrity_c *ic, bool flush_data)
1514 {
1515         int r;
1516
1517         struct flush_request fr;
1518
1519         if (!ic->meta_dev)
1520                 flush_data = false;
1521         if (flush_data) {
1522                 fr.io_req.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC,
1523                 fr.io_req.mem.type = DM_IO_KMEM,
1524                 fr.io_req.mem.ptr.addr = NULL,
1525                 fr.io_req.notify.fn = flush_notify,
1526                 fr.io_req.notify.context = &fr;
1527                 fr.io_req.client = dm_bufio_get_dm_io_client(ic->bufio),
1528                 fr.io_reg.bdev = ic->dev->bdev,
1529                 fr.io_reg.sector = 0,
1530                 fr.io_reg.count = 0,
1531                 fr.ic = ic;
1532                 init_completion(&fr.comp);
1533                 r = dm_io(&fr.io_req, 1, &fr.io_reg, NULL);
1534                 BUG_ON(r);
1535         }
1536
1537         r = dm_bufio_write_dirty_buffers(ic->bufio);
1538         if (unlikely(r))
1539                 dm_integrity_io_error(ic, "writing tags", r);
1540
1541         if (flush_data)
1542                 wait_for_completion(&fr.comp);
1543 }
1544
1545 static void sleep_on_endio_wait(struct dm_integrity_c *ic)
1546 {
1547         DECLARE_WAITQUEUE(wait, current);
1548         __add_wait_queue(&ic->endio_wait, &wait);
1549         __set_current_state(TASK_UNINTERRUPTIBLE);
1550         spin_unlock_irq(&ic->endio_wait.lock);
1551         io_schedule();
1552         spin_lock_irq(&ic->endio_wait.lock);
1553         __remove_wait_queue(&ic->endio_wait, &wait);
1554 }
1555
1556 static void autocommit_fn(struct timer_list *t)
1557 {
1558         struct dm_integrity_c *ic = from_timer(ic, t, autocommit_timer);
1559
1560         if (likely(!dm_integrity_failed(ic)))
1561                 queue_work(ic->commit_wq, &ic->commit_work);
1562 }
1563
1564 static void schedule_autocommit(struct dm_integrity_c *ic)
1565 {
1566         if (!timer_pending(&ic->autocommit_timer))
1567                 mod_timer(&ic->autocommit_timer, jiffies + ic->autocommit_jiffies);
1568 }
1569
1570 static void submit_flush_bio(struct dm_integrity_c *ic, struct dm_integrity_io *dio)
1571 {
1572         struct bio *bio;
1573         unsigned long flags;
1574
1575         spin_lock_irqsave(&ic->endio_wait.lock, flags);
1576         bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1577         bio_list_add(&ic->flush_bio_list, bio);
1578         spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
1579
1580         queue_work(ic->commit_wq, &ic->commit_work);
1581 }
1582
1583 static void do_endio(struct dm_integrity_c *ic, struct bio *bio)
1584 {
1585         int r = dm_integrity_failed(ic);
1586         if (unlikely(r) && !bio->bi_status)
1587                 bio->bi_status = errno_to_blk_status(r);
1588         if (unlikely(ic->synchronous_mode) && bio_op(bio) == REQ_OP_WRITE) {
1589                 unsigned long flags;
1590                 spin_lock_irqsave(&ic->endio_wait.lock, flags);
1591                 bio_list_add(&ic->synchronous_bios, bio);
1592                 queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, 0);
1593                 spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
1594                 return;
1595         }
1596         bio_endio(bio);
1597 }
1598
1599 static void do_endio_flush(struct dm_integrity_c *ic, struct dm_integrity_io *dio)
1600 {
1601         struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1602
1603         if (unlikely(dio->fua) && likely(!bio->bi_status) && likely(!dm_integrity_failed(ic)))
1604                 submit_flush_bio(ic, dio);
1605         else
1606                 do_endio(ic, bio);
1607 }
1608
1609 static void dec_in_flight(struct dm_integrity_io *dio)
1610 {
1611         if (atomic_dec_and_test(&dio->in_flight)) {
1612                 struct dm_integrity_c *ic = dio->ic;
1613                 struct bio *bio;
1614
1615                 remove_range(ic, &dio->range);
1616
1617                 if (dio->op == REQ_OP_WRITE || unlikely(dio->op == REQ_OP_DISCARD))
1618                         schedule_autocommit(ic);
1619
1620                 bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1621
1622                 if (unlikely(dio->bi_status) && !bio->bi_status)
1623                         bio->bi_status = dio->bi_status;
1624                 if (likely(!bio->bi_status) && unlikely(bio_sectors(bio) != dio->range.n_sectors)) {
1625                         dio->range.logical_sector += dio->range.n_sectors;
1626                         bio_advance(bio, dio->range.n_sectors << SECTOR_SHIFT);
1627                         INIT_WORK(&dio->work, integrity_bio_wait);
1628                         queue_work(ic->offload_wq, &dio->work);
1629                         return;
1630                 }
1631                 do_endio_flush(ic, dio);
1632         }
1633 }
1634
1635 static void integrity_end_io(struct bio *bio)
1636 {
1637         struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
1638
1639         dm_bio_restore(&dio->bio_details, bio);
1640         if (bio->bi_integrity)
1641                 bio->bi_opf |= REQ_INTEGRITY;
1642
1643         if (dio->completion)
1644                 complete(dio->completion);
1645
1646         dec_in_flight(dio);
1647 }
1648
1649 static void integrity_sector_checksum(struct dm_integrity_c *ic, sector_t sector,
1650                                       const char *data, char *result)
1651 {
1652         __le64 sector_le = cpu_to_le64(sector);
1653         SHASH_DESC_ON_STACK(req, ic->internal_hash);
1654         int r;
1655         unsigned digest_size;
1656
1657         req->tfm = ic->internal_hash;
1658
1659         r = crypto_shash_init(req);
1660         if (unlikely(r < 0)) {
1661                 dm_integrity_io_error(ic, "crypto_shash_init", r);
1662                 goto failed;
1663         }
1664
1665         if (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) {
1666                 r = crypto_shash_update(req, (__u8 *)&ic->sb->salt, SALT_SIZE);
1667                 if (unlikely(r < 0)) {
1668                         dm_integrity_io_error(ic, "crypto_shash_update", r);
1669                         goto failed;
1670                 }
1671         }
1672
1673         r = crypto_shash_update(req, (const __u8 *)&sector_le, sizeof sector_le);
1674         if (unlikely(r < 0)) {
1675                 dm_integrity_io_error(ic, "crypto_shash_update", r);
1676                 goto failed;
1677         }
1678
1679         r = crypto_shash_update(req, data, ic->sectors_per_block << SECTOR_SHIFT);
1680         if (unlikely(r < 0)) {
1681                 dm_integrity_io_error(ic, "crypto_shash_update", r);
1682                 goto failed;
1683         }
1684
1685         r = crypto_shash_final(req, result);
1686         if (unlikely(r < 0)) {
1687                 dm_integrity_io_error(ic, "crypto_shash_final", r);
1688                 goto failed;
1689         }
1690
1691         digest_size = crypto_shash_digestsize(ic->internal_hash);
1692         if (unlikely(digest_size < ic->tag_size))
1693                 memset(result + digest_size, 0, ic->tag_size - digest_size);
1694
1695         return;
1696
1697 failed:
1698         /* this shouldn't happen anyway, the hash functions have no reason to fail */
1699         get_random_bytes(result, ic->tag_size);
1700 }
1701
1702 static void integrity_metadata(struct work_struct *w)
1703 {
1704         struct dm_integrity_io *dio = container_of(w, struct dm_integrity_io, work);
1705         struct dm_integrity_c *ic = dio->ic;
1706
1707         int r;
1708
1709         if (ic->internal_hash) {
1710                 struct bvec_iter iter;
1711                 struct bio_vec bv;
1712                 unsigned digest_size = crypto_shash_digestsize(ic->internal_hash);
1713                 struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1714                 char *checksums;
1715                 unsigned extra_space = unlikely(digest_size > ic->tag_size) ? digest_size - ic->tag_size : 0;
1716                 char checksums_onstack[max((size_t)HASH_MAX_DIGESTSIZE, MAX_TAG_SIZE)];
1717                 sector_t sector;
1718                 unsigned sectors_to_process;
1719
1720                 if (unlikely(ic->mode == 'R'))
1721                         goto skip_io;
1722
1723                 if (likely(dio->op != REQ_OP_DISCARD))
1724                         checksums = kmalloc((PAGE_SIZE >> SECTOR_SHIFT >> ic->sb->log2_sectors_per_block) * ic->tag_size + extra_space,
1725                                             GFP_NOIO | __GFP_NORETRY | __GFP_NOWARN);
1726                 else
1727                         checksums = kmalloc(PAGE_SIZE, GFP_NOIO | __GFP_NORETRY | __GFP_NOWARN);
1728                 if (!checksums) {
1729                         checksums = checksums_onstack;
1730                         if (WARN_ON(extra_space &&
1731                                     digest_size > sizeof(checksums_onstack))) {
1732                                 r = -EINVAL;
1733                                 goto error;
1734                         }
1735                 }
1736
1737                 if (unlikely(dio->op == REQ_OP_DISCARD)) {
1738                         sector_t bi_sector = dio->bio_details.bi_iter.bi_sector;
1739                         unsigned bi_size = dio->bio_details.bi_iter.bi_size;
1740                         unsigned max_size = likely(checksums != checksums_onstack) ? PAGE_SIZE : HASH_MAX_DIGESTSIZE;
1741                         unsigned max_blocks = max_size / ic->tag_size;
1742                         memset(checksums, DISCARD_FILLER, max_size);
1743
1744                         while (bi_size) {
1745                                 unsigned this_step_blocks = bi_size >> (SECTOR_SHIFT + ic->sb->log2_sectors_per_block);
1746                                 this_step_blocks = min(this_step_blocks, max_blocks);
1747                                 r = dm_integrity_rw_tag(ic, checksums, &dio->metadata_block, &dio->metadata_offset,
1748                                                         this_step_blocks * ic->tag_size, TAG_WRITE);
1749                                 if (unlikely(r)) {
1750                                         if (likely(checksums != checksums_onstack))
1751                                                 kfree(checksums);
1752                                         goto error;
1753                                 }
1754
1755                                 /*if (bi_size < this_step_blocks << (SECTOR_SHIFT + ic->sb->log2_sectors_per_block)) {
1756                                         printk("BUGG: bi_sector: %llx, bi_size: %u\n", bi_sector, bi_size);
1757                                         printk("BUGG: this_step_blocks: %u\n", this_step_blocks);
1758                                         BUG();
1759                                 }*/
1760                                 bi_size -= this_step_blocks << (SECTOR_SHIFT + ic->sb->log2_sectors_per_block);
1761                                 bi_sector += this_step_blocks << ic->sb->log2_sectors_per_block;
1762                         }
1763
1764                         if (likely(checksums != checksums_onstack))
1765                                 kfree(checksums);
1766                         goto skip_io;
1767                 }
1768
1769                 sector = dio->range.logical_sector;
1770                 sectors_to_process = dio->range.n_sectors;
1771
1772                 __bio_for_each_segment(bv, bio, iter, dio->bio_details.bi_iter) {
1773                         unsigned pos;
1774                         char *mem, *checksums_ptr;
1775
1776 again:
1777                         mem = bvec_kmap_local(&bv);
1778                         pos = 0;
1779                         checksums_ptr = checksums;
1780                         do {
1781                                 integrity_sector_checksum(ic, sector, mem + pos, checksums_ptr);
1782                                 checksums_ptr += ic->tag_size;
1783                                 sectors_to_process -= ic->sectors_per_block;
1784                                 pos += ic->sectors_per_block << SECTOR_SHIFT;
1785                                 sector += ic->sectors_per_block;
1786                         } while (pos < bv.bv_len && sectors_to_process && checksums != checksums_onstack);
1787                         kunmap_local(mem);
1788
1789                         r = dm_integrity_rw_tag(ic, checksums, &dio->metadata_block, &dio->metadata_offset,
1790                                                 checksums_ptr - checksums, dio->op == REQ_OP_READ ? TAG_CMP : TAG_WRITE);
1791                         if (unlikely(r)) {
1792                                 if (r > 0) {
1793                                         sector_t s;
1794
1795                                         s = sector - ((r + ic->tag_size - 1) / ic->tag_size);
1796                                         DMERR_LIMIT("%pg: Checksum failed at sector 0x%llx",
1797                                                     bio->bi_bdev, s);
1798                                         r = -EILSEQ;
1799                                         atomic64_inc(&ic->number_of_mismatches);
1800                                         dm_audit_log_bio(DM_MSG_PREFIX, "integrity-checksum",
1801                                                          bio, s, 0);
1802                                 }
1803                                 if (likely(checksums != checksums_onstack))
1804                                         kfree(checksums);
1805                                 goto error;
1806                         }
1807
1808                         if (!sectors_to_process)
1809                                 break;
1810
1811                         if (unlikely(pos < bv.bv_len)) {
1812                                 bv.bv_offset += pos;
1813                                 bv.bv_len -= pos;
1814                                 goto again;
1815                         }
1816                 }
1817
1818                 if (likely(checksums != checksums_onstack))
1819                         kfree(checksums);
1820         } else {
1821                 struct bio_integrity_payload *bip = dio->bio_details.bi_integrity;
1822
1823                 if (bip) {
1824                         struct bio_vec biv;
1825                         struct bvec_iter iter;
1826                         unsigned data_to_process = dio->range.n_sectors;
1827                         sector_to_block(ic, data_to_process);
1828                         data_to_process *= ic->tag_size;
1829
1830                         bip_for_each_vec(biv, bip, iter) {
1831                                 unsigned char *tag;
1832                                 unsigned this_len;
1833
1834                                 BUG_ON(PageHighMem(biv.bv_page));
1835                                 tag = bvec_virt(&biv);
1836                                 this_len = min(biv.bv_len, data_to_process);
1837                                 r = dm_integrity_rw_tag(ic, tag, &dio->metadata_block, &dio->metadata_offset,
1838                                                         this_len, dio->op == REQ_OP_READ ? TAG_READ : TAG_WRITE);
1839                                 if (unlikely(r))
1840                                         goto error;
1841                                 data_to_process -= this_len;
1842                                 if (!data_to_process)
1843                                         break;
1844                         }
1845                 }
1846         }
1847 skip_io:
1848         dec_in_flight(dio);
1849         return;
1850 error:
1851         dio->bi_status = errno_to_blk_status(r);
1852         dec_in_flight(dio);
1853 }
1854
1855 static int dm_integrity_map(struct dm_target *ti, struct bio *bio)
1856 {
1857         struct dm_integrity_c *ic = ti->private;
1858         struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
1859         struct bio_integrity_payload *bip;
1860
1861         sector_t area, offset;
1862
1863         dio->ic = ic;
1864         dio->bi_status = 0;
1865         dio->op = bio_op(bio);
1866
1867         if (unlikely(dio->op == REQ_OP_DISCARD)) {
1868                 if (ti->max_io_len) {
1869                         sector_t sec = dm_target_offset(ti, bio->bi_iter.bi_sector);
1870                         unsigned log2_max_io_len = __fls(ti->max_io_len);
1871                         sector_t start_boundary = sec >> log2_max_io_len;
1872                         sector_t end_boundary = (sec + bio_sectors(bio) - 1) >> log2_max_io_len;
1873                         if (start_boundary < end_boundary) {
1874                                 sector_t len = ti->max_io_len - (sec & (ti->max_io_len - 1));
1875                                 dm_accept_partial_bio(bio, len);
1876                         }
1877                 }
1878         }
1879
1880         if (unlikely(bio->bi_opf & REQ_PREFLUSH)) {
1881                 submit_flush_bio(ic, dio);
1882                 return DM_MAPIO_SUBMITTED;
1883         }
1884
1885         dio->range.logical_sector = dm_target_offset(ti, bio->bi_iter.bi_sector);
1886         dio->fua = dio->op == REQ_OP_WRITE && bio->bi_opf & REQ_FUA;
1887         if (unlikely(dio->fua)) {
1888                 /*
1889                  * Don't pass down the FUA flag because we have to flush
1890                  * disk cache anyway.
1891                  */
1892                 bio->bi_opf &= ~REQ_FUA;
1893         }
1894         if (unlikely(dio->range.logical_sector + bio_sectors(bio) > ic->provided_data_sectors)) {
1895                 DMERR("Too big sector number: 0x%llx + 0x%x > 0x%llx",
1896                       dio->range.logical_sector, bio_sectors(bio),
1897                       ic->provided_data_sectors);
1898                 return DM_MAPIO_KILL;
1899         }
1900         if (unlikely((dio->range.logical_sector | bio_sectors(bio)) & (unsigned)(ic->sectors_per_block - 1))) {
1901                 DMERR("Bio not aligned on %u sectors: 0x%llx, 0x%x",
1902                       ic->sectors_per_block,
1903                       dio->range.logical_sector, bio_sectors(bio));
1904                 return DM_MAPIO_KILL;
1905         }
1906
1907         if (ic->sectors_per_block > 1 && likely(dio->op != REQ_OP_DISCARD)) {
1908                 struct bvec_iter iter;
1909                 struct bio_vec bv;
1910                 bio_for_each_segment(bv, bio, iter) {
1911                         if (unlikely(bv.bv_len & ((ic->sectors_per_block << SECTOR_SHIFT) - 1))) {
1912                                 DMERR("Bio vector (%u,%u) is not aligned on %u-sector boundary",
1913                                         bv.bv_offset, bv.bv_len, ic->sectors_per_block);
1914                                 return DM_MAPIO_KILL;
1915                         }
1916                 }
1917         }
1918
1919         bip = bio_integrity(bio);
1920         if (!ic->internal_hash) {
1921                 if (bip) {
1922                         unsigned wanted_tag_size = bio_sectors(bio) >> ic->sb->log2_sectors_per_block;
1923                         if (ic->log2_tag_size >= 0)
1924                                 wanted_tag_size <<= ic->log2_tag_size;
1925                         else
1926                                 wanted_tag_size *= ic->tag_size;
1927                         if (unlikely(wanted_tag_size != bip->bip_iter.bi_size)) {
1928                                 DMERR("Invalid integrity data size %u, expected %u",
1929                                       bip->bip_iter.bi_size, wanted_tag_size);
1930                                 return DM_MAPIO_KILL;
1931                         }
1932                 }
1933         } else {
1934                 if (unlikely(bip != NULL)) {
1935                         DMERR("Unexpected integrity data when using internal hash");
1936                         return DM_MAPIO_KILL;
1937                 }
1938         }
1939
1940         if (unlikely(ic->mode == 'R') && unlikely(dio->op != REQ_OP_READ))
1941                 return DM_MAPIO_KILL;
1942
1943         get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
1944         dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset, &dio->metadata_offset);
1945         bio->bi_iter.bi_sector = get_data_sector(ic, area, offset);
1946
1947         dm_integrity_map_continue(dio, true);
1948         return DM_MAPIO_SUBMITTED;
1949 }
1950
1951 static bool __journal_read_write(struct dm_integrity_io *dio, struct bio *bio,
1952                                  unsigned journal_section, unsigned journal_entry)
1953 {
1954         struct dm_integrity_c *ic = dio->ic;
1955         sector_t logical_sector;
1956         unsigned n_sectors;
1957
1958         logical_sector = dio->range.logical_sector;
1959         n_sectors = dio->range.n_sectors;
1960         do {
1961                 struct bio_vec bv = bio_iovec(bio);
1962                 char *mem;
1963
1964                 if (unlikely(bv.bv_len >> SECTOR_SHIFT > n_sectors))
1965                         bv.bv_len = n_sectors << SECTOR_SHIFT;
1966                 n_sectors -= bv.bv_len >> SECTOR_SHIFT;
1967                 bio_advance_iter(bio, &bio->bi_iter, bv.bv_len);
1968 retry_kmap:
1969                 mem = kmap_local_page(bv.bv_page);
1970                 if (likely(dio->op == REQ_OP_WRITE))
1971                         flush_dcache_page(bv.bv_page);
1972
1973                 do {
1974                         struct journal_entry *je = access_journal_entry(ic, journal_section, journal_entry);
1975
1976                         if (unlikely(dio->op == REQ_OP_READ)) {
1977                                 struct journal_sector *js;
1978                                 char *mem_ptr;
1979                                 unsigned s;
1980
1981                                 if (unlikely(journal_entry_is_inprogress(je))) {
1982                                         flush_dcache_page(bv.bv_page);
1983                                         kunmap_local(mem);
1984
1985                                         __io_wait_event(ic->copy_to_journal_wait, !journal_entry_is_inprogress(je));
1986                                         goto retry_kmap;
1987                                 }
1988                                 smp_rmb();
1989                                 BUG_ON(journal_entry_get_sector(je) != logical_sector);
1990                                 js = access_journal_data(ic, journal_section, journal_entry);
1991                                 mem_ptr = mem + bv.bv_offset;
1992                                 s = 0;
1993                                 do {
1994                                         memcpy(mem_ptr, js, JOURNAL_SECTOR_DATA);
1995                                         *(commit_id_t *)(mem_ptr + JOURNAL_SECTOR_DATA) = je->last_bytes[s];
1996                                         js++;
1997                                         mem_ptr += 1 << SECTOR_SHIFT;
1998                                 } while (++s < ic->sectors_per_block);
1999 #ifdef INTERNAL_VERIFY
2000                                 if (ic->internal_hash) {
2001                                         char checksums_onstack[max((size_t)HASH_MAX_DIGESTSIZE, MAX_TAG_SIZE)];
2002
2003                                         integrity_sector_checksum(ic, logical_sector, mem + bv.bv_offset, checksums_onstack);
2004                                         if (unlikely(memcmp(checksums_onstack, journal_entry_tag(ic, je), ic->tag_size))) {
2005                                                 DMERR_LIMIT("Checksum failed when reading from journal, at sector 0x%llx",
2006                                                             logical_sector);
2007                                                 dm_audit_log_bio(DM_MSG_PREFIX, "journal-checksum",
2008                                                                  bio, logical_sector, 0);
2009                                         }
2010                                 }
2011 #endif
2012                         }
2013
2014                         if (!ic->internal_hash) {
2015                                 struct bio_integrity_payload *bip = bio_integrity(bio);
2016                                 unsigned tag_todo = ic->tag_size;
2017                                 char *tag_ptr = journal_entry_tag(ic, je);
2018
2019                                 if (bip) do {
2020                                         struct bio_vec biv = bvec_iter_bvec(bip->bip_vec, bip->bip_iter);
2021                                         unsigned tag_now = min(biv.bv_len, tag_todo);
2022                                         char *tag_addr;
2023                                         BUG_ON(PageHighMem(biv.bv_page));
2024                                         tag_addr = bvec_virt(&biv);
2025                                         if (likely(dio->op == REQ_OP_WRITE))
2026                                                 memcpy(tag_ptr, tag_addr, tag_now);
2027                                         else
2028                                                 memcpy(tag_addr, tag_ptr, tag_now);
2029                                         bvec_iter_advance(bip->bip_vec, &bip->bip_iter, tag_now);
2030                                         tag_ptr += tag_now;
2031                                         tag_todo -= tag_now;
2032                                 } while (unlikely(tag_todo)); else {
2033                                         if (likely(dio->op == REQ_OP_WRITE))
2034                                                 memset(tag_ptr, 0, tag_todo);
2035                                 }
2036                         }
2037
2038                         if (likely(dio->op == REQ_OP_WRITE)) {
2039                                 struct journal_sector *js;
2040                                 unsigned s;
2041
2042                                 js = access_journal_data(ic, journal_section, journal_entry);
2043                                 memcpy(js, mem + bv.bv_offset, ic->sectors_per_block << SECTOR_SHIFT);
2044
2045                                 s = 0;
2046                                 do {
2047                                         je->last_bytes[s] = js[s].commit_id;
2048                                 } while (++s < ic->sectors_per_block);
2049
2050                                 if (ic->internal_hash) {
2051                                         unsigned digest_size = crypto_shash_digestsize(ic->internal_hash);
2052                                         if (unlikely(digest_size > ic->tag_size)) {
2053                                                 char checksums_onstack[HASH_MAX_DIGESTSIZE];
2054                                                 integrity_sector_checksum(ic, logical_sector, (char *)js, checksums_onstack);
2055                                                 memcpy(journal_entry_tag(ic, je), checksums_onstack, ic->tag_size);
2056                                         } else
2057                                                 integrity_sector_checksum(ic, logical_sector, (char *)js, journal_entry_tag(ic, je));
2058                                 }
2059
2060                                 journal_entry_set_sector(je, logical_sector);
2061                         }
2062                         logical_sector += ic->sectors_per_block;
2063
2064                         journal_entry++;
2065                         if (unlikely(journal_entry == ic->journal_section_entries)) {
2066                                 journal_entry = 0;
2067                                 journal_section++;
2068                                 wraparound_section(ic, &journal_section);
2069                         }
2070
2071                         bv.bv_offset += ic->sectors_per_block << SECTOR_SHIFT;
2072                 } while (bv.bv_len -= ic->sectors_per_block << SECTOR_SHIFT);
2073
2074                 if (unlikely(dio->op == REQ_OP_READ))
2075                         flush_dcache_page(bv.bv_page);
2076                 kunmap_local(mem);
2077         } while (n_sectors);
2078
2079         if (likely(dio->op == REQ_OP_WRITE)) {
2080                 smp_mb();
2081                 if (unlikely(waitqueue_active(&ic->copy_to_journal_wait)))
2082                         wake_up(&ic->copy_to_journal_wait);
2083                 if (READ_ONCE(ic->free_sectors) <= ic->free_sectors_threshold) {
2084                         queue_work(ic->commit_wq, &ic->commit_work);
2085                 } else {
2086                         schedule_autocommit(ic);
2087                 }
2088         } else {
2089                 remove_range(ic, &dio->range);
2090         }
2091
2092         if (unlikely(bio->bi_iter.bi_size)) {
2093                 sector_t area, offset;
2094
2095                 dio->range.logical_sector = logical_sector;
2096                 get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
2097                 dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset, &dio->metadata_offset);
2098                 return true;
2099         }
2100
2101         return false;
2102 }
2103
2104 static void dm_integrity_map_continue(struct dm_integrity_io *dio, bool from_map)
2105 {
2106         struct dm_integrity_c *ic = dio->ic;
2107         struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
2108         unsigned journal_section, journal_entry;
2109         unsigned journal_read_pos;
2110         struct completion read_comp;
2111         bool discard_retried = false;
2112         bool need_sync_io = ic->internal_hash && dio->op == REQ_OP_READ;
2113         if (unlikely(dio->op == REQ_OP_DISCARD) && ic->mode != 'D')
2114                 need_sync_io = true;
2115
2116         if (need_sync_io && from_map) {
2117                 INIT_WORK(&dio->work, integrity_bio_wait);
2118                 queue_work(ic->offload_wq, &dio->work);
2119                 return;
2120         }
2121
2122 lock_retry:
2123         spin_lock_irq(&ic->endio_wait.lock);
2124 retry:
2125         if (unlikely(dm_integrity_failed(ic))) {
2126                 spin_unlock_irq(&ic->endio_wait.lock);
2127                 do_endio(ic, bio);
2128                 return;
2129         }
2130         dio->range.n_sectors = bio_sectors(bio);
2131         journal_read_pos = NOT_FOUND;
2132         if (ic->mode == 'J' && likely(dio->op != REQ_OP_DISCARD)) {
2133                 if (dio->op == REQ_OP_WRITE) {
2134                         unsigned next_entry, i, pos;
2135                         unsigned ws, we, range_sectors;
2136
2137                         dio->range.n_sectors = min(dio->range.n_sectors,
2138                                                    (sector_t)ic->free_sectors << ic->sb->log2_sectors_per_block);
2139                         if (unlikely(!dio->range.n_sectors)) {
2140                                 if (from_map)
2141                                         goto offload_to_thread;
2142                                 sleep_on_endio_wait(ic);
2143                                 goto retry;
2144                         }
2145                         range_sectors = dio->range.n_sectors >> ic->sb->log2_sectors_per_block;
2146                         ic->free_sectors -= range_sectors;
2147                         journal_section = ic->free_section;
2148                         journal_entry = ic->free_section_entry;
2149
2150                         next_entry = ic->free_section_entry + range_sectors;
2151                         ic->free_section_entry = next_entry % ic->journal_section_entries;
2152                         ic->free_section += next_entry / ic->journal_section_entries;
2153                         ic->n_uncommitted_sections += next_entry / ic->journal_section_entries;
2154                         wraparound_section(ic, &ic->free_section);
2155
2156                         pos = journal_section * ic->journal_section_entries + journal_entry;
2157                         ws = journal_section;
2158                         we = journal_entry;
2159                         i = 0;
2160                         do {
2161                                 struct journal_entry *je;
2162
2163                                 add_journal_node(ic, &ic->journal_tree[pos], dio->range.logical_sector + i);
2164                                 pos++;
2165                                 if (unlikely(pos >= ic->journal_entries))
2166                                         pos = 0;
2167
2168                                 je = access_journal_entry(ic, ws, we);
2169                                 BUG_ON(!journal_entry_is_unused(je));
2170                                 journal_entry_set_inprogress(je);
2171                                 we++;
2172                                 if (unlikely(we == ic->journal_section_entries)) {
2173                                         we = 0;
2174                                         ws++;
2175                                         wraparound_section(ic, &ws);
2176                                 }
2177                         } while ((i += ic->sectors_per_block) < dio->range.n_sectors);
2178
2179                         spin_unlock_irq(&ic->endio_wait.lock);
2180                         goto journal_read_write;
2181                 } else {
2182                         sector_t next_sector;
2183                         journal_read_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
2184                         if (likely(journal_read_pos == NOT_FOUND)) {
2185                                 if (unlikely(dio->range.n_sectors > next_sector - dio->range.logical_sector))
2186                                         dio->range.n_sectors = next_sector - dio->range.logical_sector;
2187                         } else {
2188                                 unsigned i;
2189                                 unsigned jp = journal_read_pos + 1;
2190                                 for (i = ic->sectors_per_block; i < dio->range.n_sectors; i += ic->sectors_per_block, jp++) {
2191                                         if (!test_journal_node(ic, jp, dio->range.logical_sector + i))
2192                                                 break;
2193                                 }
2194                                 dio->range.n_sectors = i;
2195                         }
2196                 }
2197         }
2198         if (unlikely(!add_new_range(ic, &dio->range, true))) {
2199                 /*
2200                  * We must not sleep in the request routine because it could
2201                  * stall bios on current->bio_list.
2202                  * So, we offload the bio to a workqueue if we have to sleep.
2203                  */
2204                 if (from_map) {
2205 offload_to_thread:
2206                         spin_unlock_irq(&ic->endio_wait.lock);
2207                         INIT_WORK(&dio->work, integrity_bio_wait);
2208                         queue_work(ic->wait_wq, &dio->work);
2209                         return;
2210                 }
2211                 if (journal_read_pos != NOT_FOUND)
2212                         dio->range.n_sectors = ic->sectors_per_block;
2213                 wait_and_add_new_range(ic, &dio->range);
2214                 /*
2215                  * wait_and_add_new_range drops the spinlock, so the journal
2216                  * may have been changed arbitrarily. We need to recheck.
2217                  * To simplify the code, we restrict I/O size to just one block.
2218                  */
2219                 if (journal_read_pos != NOT_FOUND) {
2220                         sector_t next_sector;
2221                         unsigned new_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
2222                         if (unlikely(new_pos != journal_read_pos)) {
2223                                 remove_range_unlocked(ic, &dio->range);
2224                                 goto retry;
2225                         }
2226                 }
2227         }
2228         if (ic->mode == 'J' && likely(dio->op == REQ_OP_DISCARD) && !discard_retried) {
2229                 sector_t next_sector;
2230                 unsigned new_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
2231                 if (unlikely(new_pos != NOT_FOUND) ||
2232                     unlikely(next_sector < dio->range.logical_sector - dio->range.n_sectors)) {
2233                         remove_range_unlocked(ic, &dio->range);
2234                         spin_unlock_irq(&ic->endio_wait.lock);
2235                         queue_work(ic->commit_wq, &ic->commit_work);
2236                         flush_workqueue(ic->commit_wq);
2237                         queue_work(ic->writer_wq, &ic->writer_work);
2238                         flush_workqueue(ic->writer_wq);
2239                         discard_retried = true;
2240                         goto lock_retry;
2241                 }
2242         }
2243         spin_unlock_irq(&ic->endio_wait.lock);
2244
2245         if (unlikely(journal_read_pos != NOT_FOUND)) {
2246                 journal_section = journal_read_pos / ic->journal_section_entries;
2247                 journal_entry = journal_read_pos % ic->journal_section_entries;
2248                 goto journal_read_write;
2249         }
2250
2251         if (ic->mode == 'B' && (dio->op == REQ_OP_WRITE || unlikely(dio->op == REQ_OP_DISCARD))) {
2252                 if (!block_bitmap_op(ic, ic->may_write_bitmap, dio->range.logical_sector,
2253                                      dio->range.n_sectors, BITMAP_OP_TEST_ALL_SET)) {
2254                         struct bitmap_block_status *bbs;
2255
2256                         bbs = sector_to_bitmap_block(ic, dio->range.logical_sector);
2257                         spin_lock(&bbs->bio_queue_lock);
2258                         bio_list_add(&bbs->bio_queue, bio);
2259                         spin_unlock(&bbs->bio_queue_lock);
2260                         queue_work(ic->writer_wq, &bbs->work);
2261                         return;
2262                 }
2263         }
2264
2265         dio->in_flight = (atomic_t)ATOMIC_INIT(2);
2266
2267         if (need_sync_io) {
2268                 init_completion(&read_comp);
2269                 dio->completion = &read_comp;
2270         } else
2271                 dio->completion = NULL;
2272
2273         dm_bio_record(&dio->bio_details, bio);
2274         bio_set_dev(bio, ic->dev->bdev);
2275         bio->bi_integrity = NULL;
2276         bio->bi_opf &= ~REQ_INTEGRITY;
2277         bio->bi_end_io = integrity_end_io;
2278         bio->bi_iter.bi_size = dio->range.n_sectors << SECTOR_SHIFT;
2279
2280         if (unlikely(dio->op == REQ_OP_DISCARD) && likely(ic->mode != 'D')) {
2281                 integrity_metadata(&dio->work);
2282                 dm_integrity_flush_buffers(ic, false);
2283
2284                 dio->in_flight = (atomic_t)ATOMIC_INIT(1);
2285                 dio->completion = NULL;
2286
2287                 submit_bio_noacct(bio);
2288
2289                 return;
2290         }
2291
2292         submit_bio_noacct(bio);
2293
2294         if (need_sync_io) {
2295                 wait_for_completion_io(&read_comp);
2296                 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING) &&
2297                     dio->range.logical_sector + dio->range.n_sectors > le64_to_cpu(ic->sb->recalc_sector))
2298                         goto skip_check;
2299                 if (ic->mode == 'B') {
2300                         if (!block_bitmap_op(ic, ic->recalc_bitmap, dio->range.logical_sector,
2301                                              dio->range.n_sectors, BITMAP_OP_TEST_ALL_CLEAR))
2302                                 goto skip_check;
2303                 }
2304
2305                 if (likely(!bio->bi_status))
2306                         integrity_metadata(&dio->work);
2307                 else
2308 skip_check:
2309                         dec_in_flight(dio);
2310
2311         } else {
2312                 INIT_WORK(&dio->work, integrity_metadata);
2313                 queue_work(ic->metadata_wq, &dio->work);
2314         }
2315
2316         return;
2317
2318 journal_read_write:
2319         if (unlikely(__journal_read_write(dio, bio, journal_section, journal_entry)))
2320                 goto lock_retry;
2321
2322         do_endio_flush(ic, dio);
2323 }
2324
2325
2326 static void integrity_bio_wait(struct work_struct *w)
2327 {
2328         struct dm_integrity_io *dio = container_of(w, struct dm_integrity_io, work);
2329
2330         dm_integrity_map_continue(dio, false);
2331 }
2332
2333 static void pad_uncommitted(struct dm_integrity_c *ic)
2334 {
2335         if (ic->free_section_entry) {
2336                 ic->free_sectors -= ic->journal_section_entries - ic->free_section_entry;
2337                 ic->free_section_entry = 0;
2338                 ic->free_section++;
2339                 wraparound_section(ic, &ic->free_section);
2340                 ic->n_uncommitted_sections++;
2341         }
2342         if (WARN_ON(ic->journal_sections * ic->journal_section_entries !=
2343                     (ic->n_uncommitted_sections + ic->n_committed_sections) *
2344                     ic->journal_section_entries + ic->free_sectors)) {
2345                 DMCRIT("journal_sections %u, journal_section_entries %u, "
2346                        "n_uncommitted_sections %u, n_committed_sections %u, "
2347                        "journal_section_entries %u, free_sectors %u",
2348                        ic->journal_sections, ic->journal_section_entries,
2349                        ic->n_uncommitted_sections, ic->n_committed_sections,
2350                        ic->journal_section_entries, ic->free_sectors);
2351         }
2352 }
2353
2354 static void integrity_commit(struct work_struct *w)
2355 {
2356         struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, commit_work);
2357         unsigned commit_start, commit_sections;
2358         unsigned i, j, n;
2359         struct bio *flushes;
2360
2361         del_timer(&ic->autocommit_timer);
2362
2363         spin_lock_irq(&ic->endio_wait.lock);
2364         flushes = bio_list_get(&ic->flush_bio_list);
2365         if (unlikely(ic->mode != 'J')) {
2366                 spin_unlock_irq(&ic->endio_wait.lock);
2367                 dm_integrity_flush_buffers(ic, true);
2368                 goto release_flush_bios;
2369         }
2370
2371         pad_uncommitted(ic);
2372         commit_start = ic->uncommitted_section;
2373         commit_sections = ic->n_uncommitted_sections;
2374         spin_unlock_irq(&ic->endio_wait.lock);
2375
2376         if (!commit_sections)
2377                 goto release_flush_bios;
2378
2379         ic->wrote_to_journal = true;
2380
2381         i = commit_start;
2382         for (n = 0; n < commit_sections; n++) {
2383                 for (j = 0; j < ic->journal_section_entries; j++) {
2384                         struct journal_entry *je;
2385                         je = access_journal_entry(ic, i, j);
2386                         io_wait_event(ic->copy_to_journal_wait, !journal_entry_is_inprogress(je));
2387                 }
2388                 for (j = 0; j < ic->journal_section_sectors; j++) {
2389                         struct journal_sector *js;
2390                         js = access_journal(ic, i, j);
2391                         js->commit_id = dm_integrity_commit_id(ic, i, j, ic->commit_seq);
2392                 }
2393                 i++;
2394                 if (unlikely(i >= ic->journal_sections))
2395                         ic->commit_seq = next_commit_seq(ic->commit_seq);
2396                 wraparound_section(ic, &i);
2397         }
2398         smp_rmb();
2399
2400         write_journal(ic, commit_start, commit_sections);
2401
2402         spin_lock_irq(&ic->endio_wait.lock);
2403         ic->uncommitted_section += commit_sections;
2404         wraparound_section(ic, &ic->uncommitted_section);
2405         ic->n_uncommitted_sections -= commit_sections;
2406         ic->n_committed_sections += commit_sections;
2407         spin_unlock_irq(&ic->endio_wait.lock);
2408
2409         if (READ_ONCE(ic->free_sectors) <= ic->free_sectors_threshold)
2410                 queue_work(ic->writer_wq, &ic->writer_work);
2411
2412 release_flush_bios:
2413         while (flushes) {
2414                 struct bio *next = flushes->bi_next;
2415                 flushes->bi_next = NULL;
2416                 do_endio(ic, flushes);
2417                 flushes = next;
2418         }
2419 }
2420
2421 static void complete_copy_from_journal(unsigned long error, void *context)
2422 {
2423         struct journal_io *io = context;
2424         struct journal_completion *comp = io->comp;
2425         struct dm_integrity_c *ic = comp->ic;
2426         remove_range(ic, &io->range);
2427         mempool_free(io, &ic->journal_io_mempool);
2428         if (unlikely(error != 0))
2429                 dm_integrity_io_error(ic, "copying from journal", -EIO);
2430         complete_journal_op(comp);
2431 }
2432
2433 static void restore_last_bytes(struct dm_integrity_c *ic, struct journal_sector *js,
2434                                struct journal_entry *je)
2435 {
2436         unsigned s = 0;
2437         do {
2438                 js->commit_id = je->last_bytes[s];
2439                 js++;
2440         } while (++s < ic->sectors_per_block);
2441 }
2442
2443 static void do_journal_write(struct dm_integrity_c *ic, unsigned write_start,
2444                              unsigned write_sections, bool from_replay)
2445 {
2446         unsigned i, j, n;
2447         struct journal_completion comp;
2448         struct blk_plug plug;
2449
2450         blk_start_plug(&plug);
2451
2452         comp.ic = ic;
2453         comp.in_flight = (atomic_t)ATOMIC_INIT(1);
2454         init_completion(&comp.comp);
2455
2456         i = write_start;
2457         for (n = 0; n < write_sections; n++, i++, wraparound_section(ic, &i)) {
2458 #ifndef INTERNAL_VERIFY
2459                 if (unlikely(from_replay))
2460 #endif
2461                         rw_section_mac(ic, i, false);
2462                 for (j = 0; j < ic->journal_section_entries; j++) {
2463                         struct journal_entry *je = access_journal_entry(ic, i, j);
2464                         sector_t sec, area, offset;
2465                         unsigned k, l, next_loop;
2466                         sector_t metadata_block;
2467                         unsigned metadata_offset;
2468                         struct journal_io *io;
2469
2470                         if (journal_entry_is_unused(je))
2471                                 continue;
2472                         BUG_ON(unlikely(journal_entry_is_inprogress(je)) && !from_replay);
2473                         sec = journal_entry_get_sector(je);
2474                         if (unlikely(from_replay)) {
2475                                 if (unlikely(sec & (unsigned)(ic->sectors_per_block - 1))) {
2476                                         dm_integrity_io_error(ic, "invalid sector in journal", -EIO);
2477                                         sec &= ~(sector_t)(ic->sectors_per_block - 1);
2478                                 }
2479                                 if (unlikely(sec >= ic->provided_data_sectors)) {
2480                                         journal_entry_set_unused(je);
2481                                         continue;
2482                                 }
2483                         }
2484                         get_area_and_offset(ic, sec, &area, &offset);
2485                         restore_last_bytes(ic, access_journal_data(ic, i, j), je);
2486                         for (k = j + 1; k < ic->journal_section_entries; k++) {
2487                                 struct journal_entry *je2 = access_journal_entry(ic, i, k);
2488                                 sector_t sec2, area2, offset2;
2489                                 if (journal_entry_is_unused(je2))
2490                                         break;
2491                                 BUG_ON(unlikely(journal_entry_is_inprogress(je2)) && !from_replay);
2492                                 sec2 = journal_entry_get_sector(je2);
2493                                 if (unlikely(sec2 >= ic->provided_data_sectors))
2494                                         break;
2495                                 get_area_and_offset(ic, sec2, &area2, &offset2);
2496                                 if (area2 != area || offset2 != offset + ((k - j) << ic->sb->log2_sectors_per_block))
2497                                         break;
2498                                 restore_last_bytes(ic, access_journal_data(ic, i, k), je2);
2499                         }
2500                         next_loop = k - 1;
2501
2502                         io = mempool_alloc(&ic->journal_io_mempool, GFP_NOIO);
2503                         io->comp = &comp;
2504                         io->range.logical_sector = sec;
2505                         io->range.n_sectors = (k - j) << ic->sb->log2_sectors_per_block;
2506
2507                         spin_lock_irq(&ic->endio_wait.lock);
2508                         add_new_range_and_wait(ic, &io->range);
2509
2510                         if (likely(!from_replay)) {
2511                                 struct journal_node *section_node = &ic->journal_tree[i * ic->journal_section_entries];
2512
2513                                 /* don't write if there is newer committed sector */
2514                                 while (j < k && find_newer_committed_node(ic, &section_node[j])) {
2515                                         struct journal_entry *je2 = access_journal_entry(ic, i, j);
2516
2517                                         journal_entry_set_unused(je2);
2518                                         remove_journal_node(ic, &section_node[j]);
2519                                         j++;
2520                                         sec += ic->sectors_per_block;
2521                                         offset += ic->sectors_per_block;
2522                                 }
2523                                 while (j < k && find_newer_committed_node(ic, &section_node[k - 1])) {
2524                                         struct journal_entry *je2 = access_journal_entry(ic, i, k - 1);
2525
2526                                         journal_entry_set_unused(je2);
2527                                         remove_journal_node(ic, &section_node[k - 1]);
2528                                         k--;
2529                                 }
2530                                 if (j == k) {
2531                                         remove_range_unlocked(ic, &io->range);
2532                                         spin_unlock_irq(&ic->endio_wait.lock);
2533                                         mempool_free(io, &ic->journal_io_mempool);
2534                                         goto skip_io;
2535                                 }
2536                                 for (l = j; l < k; l++) {
2537                                         remove_journal_node(ic, &section_node[l]);
2538                                 }
2539                         }
2540                         spin_unlock_irq(&ic->endio_wait.lock);
2541
2542                         metadata_block = get_metadata_sector_and_offset(ic, area, offset, &metadata_offset);
2543                         for (l = j; l < k; l++) {
2544                                 int r;
2545                                 struct journal_entry *je2 = access_journal_entry(ic, i, l);
2546
2547                                 if (
2548 #ifndef INTERNAL_VERIFY
2549                                     unlikely(from_replay) &&
2550 #endif
2551                                     ic->internal_hash) {
2552                                         char test_tag[max_t(size_t, HASH_MAX_DIGESTSIZE, MAX_TAG_SIZE)];
2553
2554                                         integrity_sector_checksum(ic, sec + ((l - j) << ic->sb->log2_sectors_per_block),
2555                                                                   (char *)access_journal_data(ic, i, l), test_tag);
2556                                         if (unlikely(memcmp(test_tag, journal_entry_tag(ic, je2), ic->tag_size))) {
2557                                                 dm_integrity_io_error(ic, "tag mismatch when replaying journal", -EILSEQ);
2558                                                 dm_audit_log_target(DM_MSG_PREFIX, "integrity-replay-journal", ic->ti, 0);
2559                                         }
2560                                 }
2561
2562                                 journal_entry_set_unused(je2);
2563                                 r = dm_integrity_rw_tag(ic, journal_entry_tag(ic, je2), &metadata_block, &metadata_offset,
2564                                                         ic->tag_size, TAG_WRITE);
2565                                 if (unlikely(r)) {
2566                                         dm_integrity_io_error(ic, "reading tags", r);
2567                                 }
2568                         }
2569
2570                         atomic_inc(&comp.in_flight);
2571                         copy_from_journal(ic, i, j << ic->sb->log2_sectors_per_block,
2572                                           (k - j) << ic->sb->log2_sectors_per_block,
2573                                           get_data_sector(ic, area, offset),
2574                                           complete_copy_from_journal, io);
2575 skip_io:
2576                         j = next_loop;
2577                 }
2578         }
2579
2580         dm_bufio_write_dirty_buffers_async(ic->bufio);
2581
2582         blk_finish_plug(&plug);
2583
2584         complete_journal_op(&comp);
2585         wait_for_completion_io(&comp.comp);
2586
2587         dm_integrity_flush_buffers(ic, true);
2588 }
2589
2590 static void integrity_writer(struct work_struct *w)
2591 {
2592         struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, writer_work);
2593         unsigned write_start, write_sections;
2594
2595         unsigned prev_free_sectors;
2596
2597         spin_lock_irq(&ic->endio_wait.lock);
2598         write_start = ic->committed_section;
2599         write_sections = ic->n_committed_sections;
2600         spin_unlock_irq(&ic->endio_wait.lock);
2601
2602         if (!write_sections)
2603                 return;
2604
2605         do_journal_write(ic, write_start, write_sections, false);
2606
2607         spin_lock_irq(&ic->endio_wait.lock);
2608
2609         ic->committed_section += write_sections;
2610         wraparound_section(ic, &ic->committed_section);
2611         ic->n_committed_sections -= write_sections;
2612
2613         prev_free_sectors = ic->free_sectors;
2614         ic->free_sectors += write_sections * ic->journal_section_entries;
2615         if (unlikely(!prev_free_sectors))
2616                 wake_up_locked(&ic->endio_wait);
2617
2618         spin_unlock_irq(&ic->endio_wait.lock);
2619 }
2620
2621 static void recalc_write_super(struct dm_integrity_c *ic)
2622 {
2623         int r;
2624
2625         dm_integrity_flush_buffers(ic, false);
2626         if (dm_integrity_failed(ic))
2627                 return;
2628
2629         r = sync_rw_sb(ic, REQ_OP_WRITE);
2630         if (unlikely(r))
2631                 dm_integrity_io_error(ic, "writing superblock", r);
2632 }
2633
2634 static void integrity_recalc(struct work_struct *w)
2635 {
2636         struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, recalc_work);
2637         struct dm_integrity_range range;
2638         struct dm_io_request io_req;
2639         struct dm_io_region io_loc;
2640         sector_t area, offset;
2641         sector_t metadata_block;
2642         unsigned metadata_offset;
2643         sector_t logical_sector, n_sectors;
2644         __u8 *t;
2645         unsigned i;
2646         int r;
2647         unsigned super_counter = 0;
2648
2649         DEBUG_print("start recalculation... (position %llx)\n", le64_to_cpu(ic->sb->recalc_sector));
2650
2651         spin_lock_irq(&ic->endio_wait.lock);
2652
2653 next_chunk:
2654
2655         if (unlikely(dm_post_suspending(ic->ti)))
2656                 goto unlock_ret;
2657
2658         range.logical_sector = le64_to_cpu(ic->sb->recalc_sector);
2659         if (unlikely(range.logical_sector >= ic->provided_data_sectors)) {
2660                 if (ic->mode == 'B') {
2661                         block_bitmap_op(ic, ic->recalc_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_CLEAR);
2662                         DEBUG_print("queue_delayed_work: bitmap_flush_work\n");
2663                         queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, 0);
2664                 }
2665                 goto unlock_ret;
2666         }
2667
2668         get_area_and_offset(ic, range.logical_sector, &area, &offset);
2669         range.n_sectors = min((sector_t)RECALC_SECTORS, ic->provided_data_sectors - range.logical_sector);
2670         if (!ic->meta_dev)
2671                 range.n_sectors = min(range.n_sectors, ((sector_t)1U << ic->sb->log2_interleave_sectors) - (unsigned)offset);
2672
2673         add_new_range_and_wait(ic, &range);
2674         spin_unlock_irq(&ic->endio_wait.lock);
2675         logical_sector = range.logical_sector;
2676         n_sectors = range.n_sectors;
2677
2678         if (ic->mode == 'B') {
2679                 if (block_bitmap_op(ic, ic->recalc_bitmap, logical_sector, n_sectors, BITMAP_OP_TEST_ALL_CLEAR)) {
2680                         goto advance_and_next;
2681                 }
2682                 while (block_bitmap_op(ic, ic->recalc_bitmap, logical_sector,
2683                                        ic->sectors_per_block, BITMAP_OP_TEST_ALL_CLEAR)) {
2684                         logical_sector += ic->sectors_per_block;
2685                         n_sectors -= ic->sectors_per_block;
2686                         cond_resched();
2687                 }
2688                 while (block_bitmap_op(ic, ic->recalc_bitmap, logical_sector + n_sectors - ic->sectors_per_block,
2689                                        ic->sectors_per_block, BITMAP_OP_TEST_ALL_CLEAR)) {
2690                         n_sectors -= ic->sectors_per_block;
2691                         cond_resched();
2692                 }
2693                 get_area_and_offset(ic, logical_sector, &area, &offset);
2694         }
2695
2696         DEBUG_print("recalculating: %llx, %llx\n", logical_sector, n_sectors);
2697
2698         if (unlikely(++super_counter == RECALC_WRITE_SUPER)) {
2699                 recalc_write_super(ic);
2700                 if (ic->mode == 'B') {
2701                         queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, ic->bitmap_flush_interval);
2702                 }
2703                 super_counter = 0;
2704         }
2705
2706         if (unlikely(dm_integrity_failed(ic)))
2707                 goto err;
2708
2709         io_req.bi_opf = REQ_OP_READ;
2710         io_req.mem.type = DM_IO_VMA;
2711         io_req.mem.ptr.addr = ic->recalc_buffer;
2712         io_req.notify.fn = NULL;
2713         io_req.client = ic->io;
2714         io_loc.bdev = ic->dev->bdev;
2715         io_loc.sector = get_data_sector(ic, area, offset);
2716         io_loc.count = n_sectors;
2717
2718         r = dm_io(&io_req, 1, &io_loc, NULL);
2719         if (unlikely(r)) {
2720                 dm_integrity_io_error(ic, "reading data", r);
2721                 goto err;
2722         }
2723
2724         t = ic->recalc_tags;
2725         for (i = 0; i < n_sectors; i += ic->sectors_per_block) {
2726                 integrity_sector_checksum(ic, logical_sector + i, ic->recalc_buffer + (i << SECTOR_SHIFT), t);
2727                 t += ic->tag_size;
2728         }
2729
2730         metadata_block = get_metadata_sector_and_offset(ic, area, offset, &metadata_offset);
2731
2732         r = dm_integrity_rw_tag(ic, ic->recalc_tags, &metadata_block, &metadata_offset, t - ic->recalc_tags, TAG_WRITE);
2733         if (unlikely(r)) {
2734                 dm_integrity_io_error(ic, "writing tags", r);
2735                 goto err;
2736         }
2737
2738         if (ic->mode == 'B') {
2739                 sector_t start, end;
2740                 start = (range.logical_sector >>
2741                          (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit)) <<
2742                         (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
2743                 end = ((range.logical_sector + range.n_sectors) >>
2744                        (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit)) <<
2745                         (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
2746                 block_bitmap_op(ic, ic->recalc_bitmap, start, end - start, BITMAP_OP_CLEAR);
2747         }
2748
2749 advance_and_next:
2750         cond_resched();
2751
2752         spin_lock_irq(&ic->endio_wait.lock);
2753         remove_range_unlocked(ic, &range);
2754         ic->sb->recalc_sector = cpu_to_le64(range.logical_sector + range.n_sectors);
2755         goto next_chunk;
2756
2757 err:
2758         remove_range(ic, &range);
2759         return;
2760
2761 unlock_ret:
2762         spin_unlock_irq(&ic->endio_wait.lock);
2763
2764         recalc_write_super(ic);
2765 }
2766
2767 static void bitmap_block_work(struct work_struct *w)
2768 {
2769         struct bitmap_block_status *bbs = container_of(w, struct bitmap_block_status, work);
2770         struct dm_integrity_c *ic = bbs->ic;
2771         struct bio *bio;
2772         struct bio_list bio_queue;
2773         struct bio_list waiting;
2774
2775         bio_list_init(&waiting);
2776
2777         spin_lock(&bbs->bio_queue_lock);
2778         bio_queue = bbs->bio_queue;
2779         bio_list_init(&bbs->bio_queue);
2780         spin_unlock(&bbs->bio_queue_lock);
2781
2782         while ((bio = bio_list_pop(&bio_queue))) {
2783                 struct dm_integrity_io *dio;
2784
2785                 dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
2786
2787                 if (block_bitmap_op(ic, ic->may_write_bitmap, dio->range.logical_sector,
2788                                     dio->range.n_sectors, BITMAP_OP_TEST_ALL_SET)) {
2789                         remove_range(ic, &dio->range);
2790                         INIT_WORK(&dio->work, integrity_bio_wait);
2791                         queue_work(ic->offload_wq, &dio->work);
2792                 } else {
2793                         block_bitmap_op(ic, ic->journal, dio->range.logical_sector,
2794                                         dio->range.n_sectors, BITMAP_OP_SET);
2795                         bio_list_add(&waiting, bio);
2796                 }
2797         }
2798
2799         if (bio_list_empty(&waiting))
2800                 return;
2801
2802         rw_journal_sectors(ic, REQ_OP_WRITE | REQ_FUA | REQ_SYNC,
2803                            bbs->idx * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT),
2804                            BITMAP_BLOCK_SIZE >> SECTOR_SHIFT, NULL);
2805
2806         while ((bio = bio_list_pop(&waiting))) {
2807                 struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
2808
2809                 block_bitmap_op(ic, ic->may_write_bitmap, dio->range.logical_sector,
2810                                 dio->range.n_sectors, BITMAP_OP_SET);
2811
2812                 remove_range(ic, &dio->range);
2813                 INIT_WORK(&dio->work, integrity_bio_wait);
2814                 queue_work(ic->offload_wq, &dio->work);
2815         }
2816
2817         queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, ic->bitmap_flush_interval);
2818 }
2819
2820 static void bitmap_flush_work(struct work_struct *work)
2821 {
2822         struct dm_integrity_c *ic = container_of(work, struct dm_integrity_c, bitmap_flush_work.work);
2823         struct dm_integrity_range range;
2824         unsigned long limit;
2825         struct bio *bio;
2826
2827         dm_integrity_flush_buffers(ic, false);
2828
2829         range.logical_sector = 0;
2830         range.n_sectors = ic->provided_data_sectors;
2831
2832         spin_lock_irq(&ic->endio_wait.lock);
2833         add_new_range_and_wait(ic, &range);
2834         spin_unlock_irq(&ic->endio_wait.lock);
2835
2836         dm_integrity_flush_buffers(ic, true);
2837
2838         limit = ic->provided_data_sectors;
2839         if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
2840                 limit = le64_to_cpu(ic->sb->recalc_sector)
2841                         >> (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit)
2842                         << (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
2843         }
2844         /*DEBUG_print("zeroing journal\n");*/
2845         block_bitmap_op(ic, ic->journal, 0, limit, BITMAP_OP_CLEAR);
2846         block_bitmap_op(ic, ic->may_write_bitmap, 0, limit, BITMAP_OP_CLEAR);
2847
2848         rw_journal_sectors(ic, REQ_OP_WRITE | REQ_FUA | REQ_SYNC, 0,
2849                            ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
2850
2851         spin_lock_irq(&ic->endio_wait.lock);
2852         remove_range_unlocked(ic, &range);
2853         while (unlikely((bio = bio_list_pop(&ic->synchronous_bios)) != NULL)) {
2854                 bio_endio(bio);
2855                 spin_unlock_irq(&ic->endio_wait.lock);
2856                 spin_lock_irq(&ic->endio_wait.lock);
2857         }
2858         spin_unlock_irq(&ic->endio_wait.lock);
2859 }
2860
2861
2862 static void init_journal(struct dm_integrity_c *ic, unsigned start_section,
2863                          unsigned n_sections, unsigned char commit_seq)
2864 {
2865         unsigned i, j, n;
2866
2867         if (!n_sections)
2868                 return;
2869
2870         for (n = 0; n < n_sections; n++) {
2871                 i = start_section + n;
2872                 wraparound_section(ic, &i);
2873                 for (j = 0; j < ic->journal_section_sectors; j++) {
2874                         struct journal_sector *js = access_journal(ic, i, j);
2875                         BUILD_BUG_ON(sizeof(js->sectors) != JOURNAL_SECTOR_DATA);
2876                         memset(&js->sectors, 0, sizeof(js->sectors));
2877                         js->commit_id = dm_integrity_commit_id(ic, i, j, commit_seq);
2878                 }
2879                 for (j = 0; j < ic->journal_section_entries; j++) {
2880                         struct journal_entry *je = access_journal_entry(ic, i, j);
2881                         journal_entry_set_unused(je);
2882                 }
2883         }
2884
2885         write_journal(ic, start_section, n_sections);
2886 }
2887
2888 static int find_commit_seq(struct dm_integrity_c *ic, unsigned i, unsigned j, commit_id_t id)
2889 {
2890         unsigned char k;
2891         for (k = 0; k < N_COMMIT_IDS; k++) {
2892                 if (dm_integrity_commit_id(ic, i, j, k) == id)
2893                         return k;
2894         }
2895         dm_integrity_io_error(ic, "journal commit id", -EIO);
2896         return -EIO;
2897 }
2898
2899 static void replay_journal(struct dm_integrity_c *ic)
2900 {
2901         unsigned i, j;
2902         bool used_commit_ids[N_COMMIT_IDS];
2903         unsigned max_commit_id_sections[N_COMMIT_IDS];
2904         unsigned write_start, write_sections;
2905         unsigned continue_section;
2906         bool journal_empty;
2907         unsigned char unused, last_used, want_commit_seq;
2908
2909         if (ic->mode == 'R')
2910                 return;
2911
2912         if (ic->journal_uptodate)
2913                 return;
2914
2915         last_used = 0;
2916         write_start = 0;
2917
2918         if (!ic->just_formatted) {
2919                 DEBUG_print("reading journal\n");
2920                 rw_journal(ic, REQ_OP_READ, 0, ic->journal_sections, NULL);
2921                 if (ic->journal_io)
2922                         DEBUG_bytes(lowmem_page_address(ic->journal_io[0].page), 64, "read journal");
2923                 if (ic->journal_io) {
2924                         struct journal_completion crypt_comp;
2925                         crypt_comp.ic = ic;
2926                         init_completion(&crypt_comp.comp);
2927                         crypt_comp.in_flight = (atomic_t)ATOMIC_INIT(0);
2928                         encrypt_journal(ic, false, 0, ic->journal_sections, &crypt_comp);
2929                         wait_for_completion(&crypt_comp.comp);
2930                 }
2931                 DEBUG_bytes(lowmem_page_address(ic->journal[0].page), 64, "decrypted journal");
2932         }
2933
2934         if (dm_integrity_failed(ic))
2935                 goto clear_journal;
2936
2937         journal_empty = true;
2938         memset(used_commit_ids, 0, sizeof used_commit_ids);
2939         memset(max_commit_id_sections, 0, sizeof max_commit_id_sections);
2940         for (i = 0; i < ic->journal_sections; i++) {
2941                 for (j = 0; j < ic->journal_section_sectors; j++) {
2942                         int k;
2943                         struct journal_sector *js = access_journal(ic, i, j);
2944                         k = find_commit_seq(ic, i, j, js->commit_id);
2945                         if (k < 0)
2946                                 goto clear_journal;
2947                         used_commit_ids[k] = true;
2948                         max_commit_id_sections[k] = i;
2949                 }
2950                 if (journal_empty) {
2951                         for (j = 0; j < ic->journal_section_entries; j++) {
2952                                 struct journal_entry *je = access_journal_entry(ic, i, j);
2953                                 if (!journal_entry_is_unused(je)) {
2954                                         journal_empty = false;
2955                                         break;
2956                                 }
2957                         }
2958                 }
2959         }
2960
2961         if (!used_commit_ids[N_COMMIT_IDS - 1]) {
2962                 unused = N_COMMIT_IDS - 1;
2963                 while (unused && !used_commit_ids[unused - 1])
2964                         unused--;
2965         } else {
2966                 for (unused = 0; unused < N_COMMIT_IDS; unused++)
2967                         if (!used_commit_ids[unused])
2968                                 break;
2969                 if (unused == N_COMMIT_IDS) {
2970                         dm_integrity_io_error(ic, "journal commit ids", -EIO);
2971                         goto clear_journal;
2972                 }
2973         }
2974         DEBUG_print("first unused commit seq %d [%d,%d,%d,%d]\n",
2975                     unused, used_commit_ids[0], used_commit_ids[1],
2976                     used_commit_ids[2], used_commit_ids[3]);
2977
2978         last_used = prev_commit_seq(unused);
2979         want_commit_seq = prev_commit_seq(last_used);
2980
2981         if (!used_commit_ids[want_commit_seq] && used_commit_ids[prev_commit_seq(want_commit_seq)])
2982                 journal_empty = true;
2983
2984         write_start = max_commit_id_sections[last_used] + 1;
2985         if (unlikely(write_start >= ic->journal_sections))
2986                 want_commit_seq = next_commit_seq(want_commit_seq);
2987         wraparound_section(ic, &write_start);
2988
2989         i = write_start;
2990         for (write_sections = 0; write_sections < ic->journal_sections; write_sections++) {
2991                 for (j = 0; j < ic->journal_section_sectors; j++) {
2992                         struct journal_sector *js = access_journal(ic, i, j);
2993
2994                         if (js->commit_id != dm_integrity_commit_id(ic, i, j, want_commit_seq)) {
2995                                 /*
2996                                  * This could be caused by crash during writing.
2997                                  * We won't replay the inconsistent part of the
2998                                  * journal.
2999                                  */
3000                                 DEBUG_print("commit id mismatch at position (%u, %u): %d != %d\n",
3001                                             i, j, find_commit_seq(ic, i, j, js->commit_id), want_commit_seq);
3002                                 goto brk;
3003                         }
3004                 }
3005                 i++;
3006                 if (unlikely(i >= ic->journal_sections))
3007                         want_commit_seq = next_commit_seq(want_commit_seq);
3008                 wraparound_section(ic, &i);
3009         }
3010 brk:
3011
3012         if (!journal_empty) {
3013                 DEBUG_print("replaying %u sections, starting at %u, commit seq %d\n",
3014                             write_sections, write_start, want_commit_seq);
3015                 do_journal_write(ic, write_start, write_sections, true);
3016         }
3017
3018         if (write_sections == ic->journal_sections && (ic->mode == 'J' || journal_empty)) {
3019                 continue_section = write_start;
3020                 ic->commit_seq = want_commit_seq;
3021                 DEBUG_print("continuing from section %u, commit seq %d\n", write_start, ic->commit_seq);
3022         } else {
3023                 unsigned s;
3024                 unsigned char erase_seq;
3025 clear_journal:
3026                 DEBUG_print("clearing journal\n");
3027
3028                 erase_seq = prev_commit_seq(prev_commit_seq(last_used));
3029                 s = write_start;
3030                 init_journal(ic, s, 1, erase_seq);
3031                 s++;
3032                 wraparound_section(ic, &s);
3033                 if (ic->journal_sections >= 2) {
3034                         init_journal(ic, s, ic->journal_sections - 2, erase_seq);
3035                         s += ic->journal_sections - 2;
3036                         wraparound_section(ic, &s);
3037                         init_journal(ic, s, 1, erase_seq);
3038                 }
3039
3040                 continue_section = 0;
3041                 ic->commit_seq = next_commit_seq(erase_seq);
3042         }
3043
3044         ic->committed_section = continue_section;
3045         ic->n_committed_sections = 0;
3046
3047         ic->uncommitted_section = continue_section;
3048         ic->n_uncommitted_sections = 0;
3049
3050         ic->free_section = continue_section;
3051         ic->free_section_entry = 0;
3052         ic->free_sectors = ic->journal_entries;
3053
3054         ic->journal_tree_root = RB_ROOT;
3055         for (i = 0; i < ic->journal_entries; i++)
3056                 init_journal_node(&ic->journal_tree[i]);
3057 }
3058
3059 static void dm_integrity_enter_synchronous_mode(struct dm_integrity_c *ic)
3060 {
3061         DEBUG_print("dm_integrity_enter_synchronous_mode\n");
3062
3063         if (ic->mode == 'B') {
3064                 ic->bitmap_flush_interval = msecs_to_jiffies(10) + 1;
3065                 ic->synchronous_mode = 1;
3066
3067                 cancel_delayed_work_sync(&ic->bitmap_flush_work);
3068                 queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, 0);
3069                 flush_workqueue(ic->commit_wq);
3070         }
3071 }
3072
3073 static int dm_integrity_reboot(struct notifier_block *n, unsigned long code, void *x)
3074 {
3075         struct dm_integrity_c *ic = container_of(n, struct dm_integrity_c, reboot_notifier);
3076
3077         DEBUG_print("dm_integrity_reboot\n");
3078
3079         dm_integrity_enter_synchronous_mode(ic);
3080
3081         return NOTIFY_DONE;
3082 }
3083
3084 static void dm_integrity_postsuspend(struct dm_target *ti)
3085 {
3086         struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
3087         int r;
3088
3089         WARN_ON(unregister_reboot_notifier(&ic->reboot_notifier));
3090
3091         del_timer_sync(&ic->autocommit_timer);
3092
3093         if (ic->recalc_wq)
3094                 drain_workqueue(ic->recalc_wq);
3095
3096         if (ic->mode == 'B')
3097                 cancel_delayed_work_sync(&ic->bitmap_flush_work);
3098
3099         queue_work(ic->commit_wq, &ic->commit_work);
3100         drain_workqueue(ic->commit_wq);
3101
3102         if (ic->mode == 'J') {
3103                 queue_work(ic->writer_wq, &ic->writer_work);
3104                 drain_workqueue(ic->writer_wq);
3105                 dm_integrity_flush_buffers(ic, true);
3106                 if (ic->wrote_to_journal) {
3107                         init_journal(ic, ic->free_section,
3108                                      ic->journal_sections - ic->free_section, ic->commit_seq);
3109                         if (ic->free_section) {
3110                                 init_journal(ic, 0, ic->free_section,
3111                                              next_commit_seq(ic->commit_seq));
3112                         }
3113                 }
3114         }
3115
3116         if (ic->mode == 'B') {
3117                 dm_integrity_flush_buffers(ic, true);
3118 #if 1
3119                 /* set to 0 to test bitmap replay code */
3120                 init_journal(ic, 0, ic->journal_sections, 0);
3121                 ic->sb->flags &= ~cpu_to_le32(SB_FLAG_DIRTY_BITMAP);
3122                 r = sync_rw_sb(ic, REQ_OP_WRITE | REQ_FUA);
3123                 if (unlikely(r))
3124                         dm_integrity_io_error(ic, "writing superblock", r);
3125 #endif
3126         }
3127
3128         BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
3129
3130         ic->journal_uptodate = true;
3131 }
3132
3133 static void dm_integrity_resume(struct dm_target *ti)
3134 {
3135         struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
3136         __u64 old_provided_data_sectors = le64_to_cpu(ic->sb->provided_data_sectors);
3137         int r;
3138
3139         DEBUG_print("resume\n");
3140
3141         ic->wrote_to_journal = false;
3142
3143         if (ic->provided_data_sectors != old_provided_data_sectors) {
3144                 if (ic->provided_data_sectors > old_provided_data_sectors &&
3145                     ic->mode == 'B' &&
3146                     ic->sb->log2_blocks_per_bitmap_bit == ic->log2_blocks_per_bitmap_bit) {
3147                         rw_journal_sectors(ic, REQ_OP_READ, 0,
3148                                            ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3149                         block_bitmap_op(ic, ic->journal, old_provided_data_sectors,
3150                                         ic->provided_data_sectors - old_provided_data_sectors, BITMAP_OP_SET);
3151                         rw_journal_sectors(ic, REQ_OP_WRITE | REQ_FUA | REQ_SYNC, 0,
3152                                            ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3153                 }
3154
3155                 ic->sb->provided_data_sectors = cpu_to_le64(ic->provided_data_sectors);
3156                 r = sync_rw_sb(ic, REQ_OP_WRITE | REQ_FUA);
3157                 if (unlikely(r))
3158                         dm_integrity_io_error(ic, "writing superblock", r);
3159         }
3160
3161         if (ic->sb->flags & cpu_to_le32(SB_FLAG_DIRTY_BITMAP)) {
3162                 DEBUG_print("resume dirty_bitmap\n");
3163                 rw_journal_sectors(ic, REQ_OP_READ, 0,
3164                                    ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3165                 if (ic->mode == 'B') {
3166                         if (ic->sb->log2_blocks_per_bitmap_bit == ic->log2_blocks_per_bitmap_bit &&
3167                             !ic->reset_recalculate_flag) {
3168                                 block_bitmap_copy(ic, ic->recalc_bitmap, ic->journal);
3169                                 block_bitmap_copy(ic, ic->may_write_bitmap, ic->journal);
3170                                 if (!block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors,
3171                                                      BITMAP_OP_TEST_ALL_CLEAR)) {
3172                                         ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
3173                                         ic->sb->recalc_sector = cpu_to_le64(0);
3174                                 }
3175                         } else {
3176                                 DEBUG_print("non-matching blocks_per_bitmap_bit: %u, %u\n",
3177                                             ic->sb->log2_blocks_per_bitmap_bit, ic->log2_blocks_per_bitmap_bit);
3178                                 ic->sb->log2_blocks_per_bitmap_bit = ic->log2_blocks_per_bitmap_bit;
3179                                 block_bitmap_op(ic, ic->recalc_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_SET);
3180                                 block_bitmap_op(ic, ic->may_write_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_SET);
3181                                 block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors, BITMAP_OP_SET);
3182                                 rw_journal_sectors(ic, REQ_OP_WRITE | REQ_FUA | REQ_SYNC, 0,
3183                                                    ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3184                                 ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
3185                                 ic->sb->recalc_sector = cpu_to_le64(0);
3186                         }
3187                 } else {
3188                         if (!(ic->sb->log2_blocks_per_bitmap_bit == ic->log2_blocks_per_bitmap_bit &&
3189                               block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors, BITMAP_OP_TEST_ALL_CLEAR)) ||
3190                             ic->reset_recalculate_flag) {
3191                                 ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
3192                                 ic->sb->recalc_sector = cpu_to_le64(0);
3193                         }
3194                         init_journal(ic, 0, ic->journal_sections, 0);
3195                         replay_journal(ic);
3196                         ic->sb->flags &= ~cpu_to_le32(SB_FLAG_DIRTY_BITMAP);
3197                 }
3198                 r = sync_rw_sb(ic, REQ_OP_WRITE | REQ_FUA);
3199                 if (unlikely(r))
3200                         dm_integrity_io_error(ic, "writing superblock", r);
3201         } else {
3202                 replay_journal(ic);
3203                 if (ic->reset_recalculate_flag) {
3204                         ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
3205                         ic->sb->recalc_sector = cpu_to_le64(0);
3206                 }
3207                 if (ic->mode == 'B') {
3208                         ic->sb->flags |= cpu_to_le32(SB_FLAG_DIRTY_BITMAP);
3209                         ic->sb->log2_blocks_per_bitmap_bit = ic->log2_blocks_per_bitmap_bit;
3210                         r = sync_rw_sb(ic, REQ_OP_WRITE | REQ_FUA);
3211                         if (unlikely(r))
3212                                 dm_integrity_io_error(ic, "writing superblock", r);
3213
3214                         block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors, BITMAP_OP_CLEAR);
3215                         block_bitmap_op(ic, ic->recalc_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_CLEAR);
3216                         block_bitmap_op(ic, ic->may_write_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_CLEAR);
3217                         if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING) &&
3218                             le64_to_cpu(ic->sb->recalc_sector) < ic->provided_data_sectors) {
3219                                 block_bitmap_op(ic, ic->journal, le64_to_cpu(ic->sb->recalc_sector),
3220                                                 ic->provided_data_sectors - le64_to_cpu(ic->sb->recalc_sector), BITMAP_OP_SET);
3221                                 block_bitmap_op(ic, ic->recalc_bitmap, le64_to_cpu(ic->sb->recalc_sector),
3222                                                 ic->provided_data_sectors - le64_to_cpu(ic->sb->recalc_sector), BITMAP_OP_SET);
3223                                 block_bitmap_op(ic, ic->may_write_bitmap, le64_to_cpu(ic->sb->recalc_sector),
3224                                                 ic->provided_data_sectors - le64_to_cpu(ic->sb->recalc_sector), BITMAP_OP_SET);
3225                         }
3226                         rw_journal_sectors(ic, REQ_OP_WRITE | REQ_FUA | REQ_SYNC, 0,
3227                                            ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3228                 }
3229         }
3230
3231         DEBUG_print("testing recalc: %x\n", ic->sb->flags);
3232         if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
3233                 __u64 recalc_pos = le64_to_cpu(ic->sb->recalc_sector);
3234                 DEBUG_print("recalc pos: %llx / %llx\n", recalc_pos, ic->provided_data_sectors);
3235                 if (recalc_pos < ic->provided_data_sectors) {
3236                         queue_work(ic->recalc_wq, &ic->recalc_work);
3237                 } else if (recalc_pos > ic->provided_data_sectors) {
3238                         ic->sb->recalc_sector = cpu_to_le64(ic->provided_data_sectors);
3239                         recalc_write_super(ic);
3240                 }
3241         }
3242
3243         ic->reboot_notifier.notifier_call = dm_integrity_reboot;
3244         ic->reboot_notifier.next = NULL;
3245         ic->reboot_notifier.priority = INT_MAX - 1;     /* be notified after md and before hardware drivers */
3246         WARN_ON(register_reboot_notifier(&ic->reboot_notifier));
3247
3248 #if 0
3249         /* set to 1 to stress test synchronous mode */
3250         dm_integrity_enter_synchronous_mode(ic);
3251 #endif
3252 }
3253
3254 static void dm_integrity_status(struct dm_target *ti, status_type_t type,
3255                                 unsigned status_flags, char *result, unsigned maxlen)
3256 {
3257         struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
3258         unsigned arg_count;
3259         size_t sz = 0;
3260
3261         switch (type) {
3262         case STATUSTYPE_INFO:
3263                 DMEMIT("%llu %llu",
3264                         (unsigned long long)atomic64_read(&ic->number_of_mismatches),
3265                         ic->provided_data_sectors);
3266                 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
3267                         DMEMIT(" %llu", le64_to_cpu(ic->sb->recalc_sector));
3268                 else
3269                         DMEMIT(" -");
3270                 break;
3271
3272         case STATUSTYPE_TABLE: {
3273                 __u64 watermark_percentage = (__u64)(ic->journal_entries - ic->free_sectors_threshold) * 100;
3274                 watermark_percentage += ic->journal_entries / 2;
3275                 do_div(watermark_percentage, ic->journal_entries);
3276                 arg_count = 3;
3277                 arg_count += !!ic->meta_dev;
3278                 arg_count += ic->sectors_per_block != 1;
3279                 arg_count += !!(ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING));
3280                 arg_count += ic->reset_recalculate_flag;
3281                 arg_count += ic->discard;
3282                 arg_count += ic->mode == 'J';
3283                 arg_count += ic->mode == 'J';
3284                 arg_count += ic->mode == 'B';
3285                 arg_count += ic->mode == 'B';
3286                 arg_count += !!ic->internal_hash_alg.alg_string;
3287                 arg_count += !!ic->journal_crypt_alg.alg_string;
3288                 arg_count += !!ic->journal_mac_alg.alg_string;
3289                 arg_count += (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING)) != 0;
3290                 arg_count += (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) != 0;
3291                 arg_count += ic->legacy_recalculate;
3292                 DMEMIT("%s %llu %u %c %u", ic->dev->name, ic->start,
3293                        ic->tag_size, ic->mode, arg_count);
3294                 if (ic->meta_dev)
3295                         DMEMIT(" meta_device:%s", ic->meta_dev->name);
3296                 if (ic->sectors_per_block != 1)
3297                         DMEMIT(" block_size:%u", ic->sectors_per_block << SECTOR_SHIFT);
3298                 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
3299                         DMEMIT(" recalculate");
3300                 if (ic->reset_recalculate_flag)
3301                         DMEMIT(" reset_recalculate");
3302                 if (ic->discard)
3303                         DMEMIT(" allow_discards");
3304                 DMEMIT(" journal_sectors:%u", ic->initial_sectors - SB_SECTORS);
3305                 DMEMIT(" interleave_sectors:%u", 1U << ic->sb->log2_interleave_sectors);
3306                 DMEMIT(" buffer_sectors:%u", 1U << ic->log2_buffer_sectors);
3307                 if (ic->mode == 'J') {
3308                         DMEMIT(" journal_watermark:%u", (unsigned)watermark_percentage);
3309                         DMEMIT(" commit_time:%u", ic->autocommit_msec);
3310                 }
3311                 if (ic->mode == 'B') {
3312                         DMEMIT(" sectors_per_bit:%llu", (sector_t)ic->sectors_per_block << ic->log2_blocks_per_bitmap_bit);
3313                         DMEMIT(" bitmap_flush_interval:%u", jiffies_to_msecs(ic->bitmap_flush_interval));
3314                 }
3315                 if ((ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING)) != 0)
3316                         DMEMIT(" fix_padding");
3317                 if ((ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) != 0)
3318                         DMEMIT(" fix_hmac");
3319                 if (ic->legacy_recalculate)
3320                         DMEMIT(" legacy_recalculate");
3321
3322 #define EMIT_ALG(a, n)                                                  \
3323                 do {                                                    \
3324                         if (ic->a.alg_string) {                         \
3325                                 DMEMIT(" %s:%s", n, ic->a.alg_string);  \
3326                                 if (ic->a.key_string)                   \
3327                                         DMEMIT(":%s", ic->a.key_string);\
3328                         }                                               \
3329                 } while (0)
3330                 EMIT_ALG(internal_hash_alg, "internal_hash");
3331                 EMIT_ALG(journal_crypt_alg, "journal_crypt");
3332                 EMIT_ALG(journal_mac_alg, "journal_mac");
3333                 break;
3334         }
3335         case STATUSTYPE_IMA:
3336                 DMEMIT_TARGET_NAME_VERSION(ti->type);
3337                 DMEMIT(",dev_name=%s,start=%llu,tag_size=%u,mode=%c",
3338                         ic->dev->name, ic->start, ic->tag_size, ic->mode);
3339
3340                 if (ic->meta_dev)
3341                         DMEMIT(",meta_device=%s", ic->meta_dev->name);
3342                 if (ic->sectors_per_block != 1)
3343                         DMEMIT(",block_size=%u", ic->sectors_per_block << SECTOR_SHIFT);
3344
3345                 DMEMIT(",recalculate=%c", (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) ?
3346                        'y' : 'n');
3347                 DMEMIT(",allow_discards=%c", ic->discard ? 'y' : 'n');
3348                 DMEMIT(",fix_padding=%c",
3349                        ((ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING)) != 0) ? 'y' : 'n');
3350                 DMEMIT(",fix_hmac=%c",
3351                        ((ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) != 0) ? 'y' : 'n');
3352                 DMEMIT(",legacy_recalculate=%c", ic->legacy_recalculate ? 'y' : 'n');
3353
3354                 DMEMIT(",journal_sectors=%u", ic->initial_sectors - SB_SECTORS);
3355                 DMEMIT(",interleave_sectors=%u", 1U << ic->sb->log2_interleave_sectors);
3356                 DMEMIT(",buffer_sectors=%u", 1U << ic->log2_buffer_sectors);
3357                 DMEMIT(";");
3358                 break;
3359         }
3360 }
3361
3362 static int dm_integrity_iterate_devices(struct dm_target *ti,
3363                                         iterate_devices_callout_fn fn, void *data)
3364 {
3365         struct dm_integrity_c *ic = ti->private;
3366
3367         if (!ic->meta_dev)
3368                 return fn(ti, ic->dev, ic->start + ic->initial_sectors + ic->metadata_run, ti->len, data);
3369         else
3370                 return fn(ti, ic->dev, 0, ti->len, data);
3371 }
3372
3373 static void dm_integrity_io_hints(struct dm_target *ti, struct queue_limits *limits)
3374 {
3375         struct dm_integrity_c *ic = ti->private;
3376
3377         if (ic->sectors_per_block > 1) {
3378                 limits->logical_block_size = ic->sectors_per_block << SECTOR_SHIFT;
3379                 limits->physical_block_size = ic->sectors_per_block << SECTOR_SHIFT;
3380                 blk_limits_io_min(limits, ic->sectors_per_block << SECTOR_SHIFT);
3381                 limits->dma_alignment = limits->logical_block_size - 1;
3382         }
3383 }
3384
3385 static void calculate_journal_section_size(struct dm_integrity_c *ic)
3386 {
3387         unsigned sector_space = JOURNAL_SECTOR_DATA;
3388
3389         ic->journal_sections = le32_to_cpu(ic->sb->journal_sections);
3390         ic->journal_entry_size = roundup(offsetof(struct journal_entry, last_bytes[ic->sectors_per_block]) + ic->tag_size,
3391                                          JOURNAL_ENTRY_ROUNDUP);
3392
3393         if (ic->sb->flags & cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC))
3394                 sector_space -= JOURNAL_MAC_PER_SECTOR;
3395         ic->journal_entries_per_sector = sector_space / ic->journal_entry_size;
3396         ic->journal_section_entries = ic->journal_entries_per_sector * JOURNAL_BLOCK_SECTORS;
3397         ic->journal_section_sectors = (ic->journal_section_entries << ic->sb->log2_sectors_per_block) + JOURNAL_BLOCK_SECTORS;
3398         ic->journal_entries = ic->journal_section_entries * ic->journal_sections;
3399 }
3400
3401 static int calculate_device_limits(struct dm_integrity_c *ic)
3402 {
3403         __u64 initial_sectors;
3404
3405         calculate_journal_section_size(ic);
3406         initial_sectors = SB_SECTORS + (__u64)ic->journal_section_sectors * ic->journal_sections;
3407         if (initial_sectors + METADATA_PADDING_SECTORS >= ic->meta_device_sectors || initial_sectors > UINT_MAX)
3408                 return -EINVAL;
3409         ic->initial_sectors = initial_sectors;
3410
3411         if (!ic->meta_dev) {
3412                 sector_t last_sector, last_area, last_offset;
3413
3414                 /* we have to maintain excessive padding for compatibility with existing volumes */
3415                 __u64 metadata_run_padding =
3416                         ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING) ?
3417                         (__u64)(METADATA_PADDING_SECTORS << SECTOR_SHIFT) :
3418                         (__u64)(1 << SECTOR_SHIFT << METADATA_PADDING_SECTORS);
3419
3420                 ic->metadata_run = round_up((__u64)ic->tag_size << (ic->sb->log2_interleave_sectors - ic->sb->log2_sectors_per_block),
3421                                             metadata_run_padding) >> SECTOR_SHIFT;
3422                 if (!(ic->metadata_run & (ic->metadata_run - 1)))
3423                         ic->log2_metadata_run = __ffs(ic->metadata_run);
3424                 else
3425                         ic->log2_metadata_run = -1;
3426
3427                 get_area_and_offset(ic, ic->provided_data_sectors - 1, &last_area, &last_offset);
3428                 last_sector = get_data_sector(ic, last_area, last_offset);
3429                 if (last_sector < ic->start || last_sector >= ic->meta_device_sectors)
3430                         return -EINVAL;
3431         } else {
3432                 __u64 meta_size = (ic->provided_data_sectors >> ic->sb->log2_sectors_per_block) * ic->tag_size;
3433                 meta_size = (meta_size + ((1U << (ic->log2_buffer_sectors + SECTOR_SHIFT)) - 1))
3434                                 >> (ic->log2_buffer_sectors + SECTOR_SHIFT);
3435                 meta_size <<= ic->log2_buffer_sectors;
3436                 if (ic->initial_sectors + meta_size < ic->initial_sectors ||
3437                     ic->initial_sectors + meta_size > ic->meta_device_sectors)
3438                         return -EINVAL;
3439                 ic->metadata_run = 1;
3440                 ic->log2_metadata_run = 0;
3441         }
3442
3443         return 0;
3444 }
3445
3446 static void get_provided_data_sectors(struct dm_integrity_c *ic)
3447 {
3448         if (!ic->meta_dev) {
3449                 int test_bit;
3450                 ic->provided_data_sectors = 0;
3451                 for (test_bit = fls64(ic->meta_device_sectors) - 1; test_bit >= 3; test_bit--) {
3452                         __u64 prev_data_sectors = ic->provided_data_sectors;
3453
3454                         ic->provided_data_sectors |= (sector_t)1 << test_bit;
3455                         if (calculate_device_limits(ic))
3456                                 ic->provided_data_sectors = prev_data_sectors;
3457                 }
3458         } else {
3459                 ic->provided_data_sectors = ic->data_device_sectors;
3460                 ic->provided_data_sectors &= ~(sector_t)(ic->sectors_per_block - 1);
3461         }
3462 }
3463
3464 static int initialize_superblock(struct dm_integrity_c *ic, unsigned journal_sectors, unsigned interleave_sectors)
3465 {
3466         unsigned journal_sections;
3467         int test_bit;
3468
3469         memset(ic->sb, 0, SB_SECTORS << SECTOR_SHIFT);
3470         memcpy(ic->sb->magic, SB_MAGIC, 8);
3471         ic->sb->integrity_tag_size = cpu_to_le16(ic->tag_size);
3472         ic->sb->log2_sectors_per_block = __ffs(ic->sectors_per_block);
3473         if (ic->journal_mac_alg.alg_string)
3474                 ic->sb->flags |= cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC);
3475
3476         calculate_journal_section_size(ic);
3477         journal_sections = journal_sectors / ic->journal_section_sectors;
3478         if (!journal_sections)
3479                 journal_sections = 1;
3480
3481         if (ic->fix_hmac && (ic->internal_hash_alg.alg_string || ic->journal_mac_alg.alg_string)) {
3482                 ic->sb->flags |= cpu_to_le32(SB_FLAG_FIXED_HMAC);
3483                 get_random_bytes(ic->sb->salt, SALT_SIZE);
3484         }
3485
3486         if (!ic->meta_dev) {
3487                 if (ic->fix_padding)
3488                         ic->sb->flags |= cpu_to_le32(SB_FLAG_FIXED_PADDING);
3489                 ic->sb->journal_sections = cpu_to_le32(journal_sections);
3490                 if (!interleave_sectors)
3491                         interleave_sectors = DEFAULT_INTERLEAVE_SECTORS;
3492                 ic->sb->log2_interleave_sectors = __fls(interleave_sectors);
3493                 ic->sb->log2_interleave_sectors = max((__u8)MIN_LOG2_INTERLEAVE_SECTORS, ic->sb->log2_interleave_sectors);
3494                 ic->sb->log2_interleave_sectors = min((__u8)MAX_LOG2_INTERLEAVE_SECTORS, ic->sb->log2_interleave_sectors);
3495
3496                 get_provided_data_sectors(ic);
3497                 if (!ic->provided_data_sectors)
3498                         return -EINVAL;
3499         } else {
3500                 ic->sb->log2_interleave_sectors = 0;
3501
3502                 get_provided_data_sectors(ic);
3503                 if (!ic->provided_data_sectors)
3504                         return -EINVAL;
3505
3506 try_smaller_buffer:
3507                 ic->sb->journal_sections = cpu_to_le32(0);
3508                 for (test_bit = fls(journal_sections) - 1; test_bit >= 0; test_bit--) {
3509                         __u32 prev_journal_sections = le32_to_cpu(ic->sb->journal_sections);
3510                         __u32 test_journal_sections = prev_journal_sections | (1U << test_bit);
3511                         if (test_journal_sections > journal_sections)
3512                                 continue;
3513                         ic->sb->journal_sections = cpu_to_le32(test_journal_sections);
3514                         if (calculate_device_limits(ic))
3515                                 ic->sb->journal_sections = cpu_to_le32(prev_journal_sections);
3516
3517                 }
3518                 if (!le32_to_cpu(ic->sb->journal_sections)) {
3519                         if (ic->log2_buffer_sectors > 3) {
3520                                 ic->log2_buffer_sectors--;
3521                                 goto try_smaller_buffer;
3522                         }
3523                         return -EINVAL;
3524                 }
3525         }
3526
3527         ic->sb->provided_data_sectors = cpu_to_le64(ic->provided_data_sectors);
3528
3529         sb_set_version(ic);
3530
3531         return 0;
3532 }
3533
3534 static void dm_integrity_set(struct dm_target *ti, struct dm_integrity_c *ic)
3535 {
3536         struct gendisk *disk = dm_disk(dm_table_get_md(ti->table));
3537         struct blk_integrity bi;
3538
3539         memset(&bi, 0, sizeof(bi));
3540         bi.profile = &dm_integrity_profile;
3541         bi.tuple_size = ic->tag_size;
3542         bi.tag_size = bi.tuple_size;
3543         bi.interval_exp = ic->sb->log2_sectors_per_block + SECTOR_SHIFT;
3544
3545         blk_integrity_register(disk, &bi);
3546         blk_queue_max_integrity_segments(disk->queue, UINT_MAX);
3547 }
3548
3549 static void dm_integrity_free_page_list(struct page_list *pl)
3550 {
3551         unsigned i;
3552
3553         if (!pl)
3554                 return;
3555         for (i = 0; pl[i].page; i++)
3556                 __free_page(pl[i].page);
3557         kvfree(pl);
3558 }
3559
3560 static struct page_list *dm_integrity_alloc_page_list(unsigned n_pages)
3561 {
3562         struct page_list *pl;
3563         unsigned i;
3564
3565         pl = kvmalloc_array(n_pages + 1, sizeof(struct page_list), GFP_KERNEL | __GFP_ZERO);
3566         if (!pl)
3567                 return NULL;
3568
3569         for (i = 0; i < n_pages; i++) {
3570                 pl[i].page = alloc_page(GFP_KERNEL);
3571                 if (!pl[i].page) {
3572                         dm_integrity_free_page_list(pl);
3573                         return NULL;
3574                 }
3575                 if (i)
3576                         pl[i - 1].next = &pl[i];
3577         }
3578         pl[i].page = NULL;
3579         pl[i].next = NULL;
3580
3581         return pl;
3582 }
3583
3584 static void dm_integrity_free_journal_scatterlist(struct dm_integrity_c *ic, struct scatterlist **sl)
3585 {
3586         unsigned i;
3587         for (i = 0; i < ic->journal_sections; i++)
3588                 kvfree(sl[i]);
3589         kvfree(sl);
3590 }
3591
3592 static struct scatterlist **dm_integrity_alloc_journal_scatterlist(struct dm_integrity_c *ic,
3593                                                                    struct page_list *pl)
3594 {
3595         struct scatterlist **sl;
3596         unsigned i;
3597
3598         sl = kvmalloc_array(ic->journal_sections,
3599                             sizeof(struct scatterlist *),
3600                             GFP_KERNEL | __GFP_ZERO);
3601         if (!sl)
3602                 return NULL;
3603
3604         for (i = 0; i < ic->journal_sections; i++) {
3605                 struct scatterlist *s;
3606                 unsigned start_index, start_offset;
3607                 unsigned end_index, end_offset;
3608                 unsigned n_pages;
3609                 unsigned idx;
3610
3611                 page_list_location(ic, i, 0, &start_index, &start_offset);
3612                 page_list_location(ic, i, ic->journal_section_sectors - 1,
3613                                    &end_index, &end_offset);
3614
3615                 n_pages = (end_index - start_index + 1);
3616
3617                 s = kvmalloc_array(n_pages, sizeof(struct scatterlist),
3618                                    GFP_KERNEL);
3619                 if (!s) {
3620                         dm_integrity_free_journal_scatterlist(ic, sl);
3621                         return NULL;
3622                 }
3623
3624                 sg_init_table(s, n_pages);
3625                 for (idx = start_index; idx <= end_index; idx++) {
3626                         char *va = lowmem_page_address(pl[idx].page);
3627                         unsigned start = 0, end = PAGE_SIZE;
3628                         if (idx == start_index)
3629                                 start = start_offset;
3630                         if (idx == end_index)
3631                                 end = end_offset + (1 << SECTOR_SHIFT);
3632                         sg_set_buf(&s[idx - start_index], va + start, end - start);
3633                 }
3634
3635                 sl[i] = s;
3636         }
3637
3638         return sl;
3639 }
3640
3641 static void free_alg(struct alg_spec *a)
3642 {
3643         kfree_sensitive(a->alg_string);
3644         kfree_sensitive(a->key);
3645         memset(a, 0, sizeof *a);
3646 }
3647
3648 static int get_alg_and_key(const char *arg, struct alg_spec *a, char **error, char *error_inval)
3649 {
3650         char *k;
3651
3652         free_alg(a);
3653
3654         a->alg_string = kstrdup(strchr(arg, ':') + 1, GFP_KERNEL);
3655         if (!a->alg_string)
3656                 goto nomem;
3657
3658         k = strchr(a->alg_string, ':');
3659         if (k) {
3660                 *k = 0;
3661                 a->key_string = k + 1;
3662                 if (strlen(a->key_string) & 1)
3663                         goto inval;
3664
3665                 a->key_size = strlen(a->key_string) / 2;
3666                 a->key = kmalloc(a->key_size, GFP_KERNEL);
3667                 if (!a->key)
3668                         goto nomem;
3669                 if (hex2bin(a->key, a->key_string, a->key_size))
3670                         goto inval;
3671         }
3672
3673         return 0;
3674 inval:
3675         *error = error_inval;
3676         return -EINVAL;
3677 nomem:
3678         *error = "Out of memory for an argument";
3679         return -ENOMEM;
3680 }
3681
3682 static int get_mac(struct crypto_shash **hash, struct alg_spec *a, char **error,
3683                    char *error_alg, char *error_key)
3684 {
3685         int r;
3686
3687         if (a->alg_string) {
3688                 *hash = crypto_alloc_shash(a->alg_string, 0, CRYPTO_ALG_ALLOCATES_MEMORY);
3689                 if (IS_ERR(*hash)) {
3690                         *error = error_alg;
3691                         r = PTR_ERR(*hash);
3692                         *hash = NULL;
3693                         return r;
3694                 }
3695
3696                 if (a->key) {
3697                         r = crypto_shash_setkey(*hash, a->key, a->key_size);
3698                         if (r) {
3699                                 *error = error_key;
3700                                 return r;
3701                         }
3702                 } else if (crypto_shash_get_flags(*hash) & CRYPTO_TFM_NEED_KEY) {
3703                         *error = error_key;
3704                         return -ENOKEY;
3705                 }
3706         }
3707
3708         return 0;
3709 }
3710
3711 static int create_journal(struct dm_integrity_c *ic, char **error)
3712 {
3713         int r = 0;
3714         unsigned i;
3715         __u64 journal_pages, journal_desc_size, journal_tree_size;
3716         unsigned char *crypt_data = NULL, *crypt_iv = NULL;
3717         struct skcipher_request *req = NULL;
3718
3719         ic->commit_ids[0] = cpu_to_le64(0x1111111111111111ULL);
3720         ic->commit_ids[1] = cpu_to_le64(0x2222222222222222ULL);
3721         ic->commit_ids[2] = cpu_to_le64(0x3333333333333333ULL);
3722         ic->commit_ids[3] = cpu_to_le64(0x4444444444444444ULL);
3723
3724         journal_pages = roundup((__u64)ic->journal_sections * ic->journal_section_sectors,
3725                                 PAGE_SIZE >> SECTOR_SHIFT) >> (PAGE_SHIFT - SECTOR_SHIFT);
3726         journal_desc_size = journal_pages * sizeof(struct page_list);
3727         if (journal_pages >= totalram_pages() - totalhigh_pages() || journal_desc_size > ULONG_MAX) {
3728                 *error = "Journal doesn't fit into memory";
3729                 r = -ENOMEM;
3730                 goto bad;
3731         }
3732         ic->journal_pages = journal_pages;
3733
3734         ic->journal = dm_integrity_alloc_page_list(ic->journal_pages);
3735         if (!ic->journal) {
3736                 *error = "Could not allocate memory for journal";
3737                 r = -ENOMEM;
3738                 goto bad;
3739         }
3740         if (ic->journal_crypt_alg.alg_string) {
3741                 unsigned ivsize, blocksize;
3742                 struct journal_completion comp;
3743
3744                 comp.ic = ic;
3745                 ic->journal_crypt = crypto_alloc_skcipher(ic->journal_crypt_alg.alg_string, 0, CRYPTO_ALG_ALLOCATES_MEMORY);
3746                 if (IS_ERR(ic->journal_crypt)) {
3747                         *error = "Invalid journal cipher";
3748                         r = PTR_ERR(ic->journal_crypt);
3749                         ic->journal_crypt = NULL;
3750                         goto bad;
3751                 }
3752                 ivsize = crypto_skcipher_ivsize(ic->journal_crypt);
3753                 blocksize = crypto_skcipher_blocksize(ic->journal_crypt);
3754
3755                 if (ic->journal_crypt_alg.key) {
3756                         r = crypto_skcipher_setkey(ic->journal_crypt, ic->journal_crypt_alg.key,
3757                                                    ic->journal_crypt_alg.key_size);
3758                         if (r) {
3759                                 *error = "Error setting encryption key";
3760                                 goto bad;
3761                         }
3762                 }
3763                 DEBUG_print("cipher %s, block size %u iv size %u\n",
3764                             ic->journal_crypt_alg.alg_string, blocksize, ivsize);
3765
3766                 ic->journal_io = dm_integrity_alloc_page_list(ic->journal_pages);
3767                 if (!ic->journal_io) {
3768                         *error = "Could not allocate memory for journal io";
3769                         r = -ENOMEM;
3770                         goto bad;
3771                 }
3772
3773                 if (blocksize == 1) {
3774                         struct scatterlist *sg;
3775
3776                         req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
3777                         if (!req) {
3778                                 *error = "Could not allocate crypt request";
3779                                 r = -ENOMEM;
3780                                 goto bad;
3781                         }
3782
3783                         crypt_iv = kzalloc(ivsize, GFP_KERNEL);
3784                         if (!crypt_iv) {
3785                                 *error = "Could not allocate iv";
3786                                 r = -ENOMEM;
3787                                 goto bad;
3788                         }
3789
3790                         ic->journal_xor = dm_integrity_alloc_page_list(ic->journal_pages);
3791                         if (!ic->journal_xor) {
3792                                 *error = "Could not allocate memory for journal xor";
3793                                 r = -ENOMEM;
3794                                 goto bad;
3795                         }
3796
3797                         sg = kvmalloc_array(ic->journal_pages + 1,
3798                                             sizeof(struct scatterlist),
3799                                             GFP_KERNEL);
3800                         if (!sg) {
3801                                 *error = "Unable to allocate sg list";
3802                                 r = -ENOMEM;
3803                                 goto bad;
3804                         }
3805                         sg_init_table(sg, ic->journal_pages + 1);
3806                         for (i = 0; i < ic->journal_pages; i++) {
3807                                 char *va = lowmem_page_address(ic->journal_xor[i].page);
3808                                 clear_page(va);
3809                                 sg_set_buf(&sg[i], va, PAGE_SIZE);
3810                         }
3811                         sg_set_buf(&sg[i], &ic->commit_ids, sizeof ic->commit_ids);
3812
3813                         skcipher_request_set_crypt(req, sg, sg,
3814                                                    PAGE_SIZE * ic->journal_pages + sizeof ic->commit_ids, crypt_iv);
3815                         init_completion(&comp.comp);
3816                         comp.in_flight = (atomic_t)ATOMIC_INIT(1);
3817                         if (do_crypt(true, req, &comp))
3818                                 wait_for_completion(&comp.comp);
3819                         kvfree(sg);
3820                         r = dm_integrity_failed(ic);
3821                         if (r) {
3822                                 *error = "Unable to encrypt journal";
3823                                 goto bad;
3824                         }
3825                         DEBUG_bytes(lowmem_page_address(ic->journal_xor[0].page), 64, "xor data");
3826
3827                         crypto_free_skcipher(ic->journal_crypt);
3828                         ic->journal_crypt = NULL;
3829                 } else {
3830                         unsigned crypt_len = roundup(ivsize, blocksize);
3831
3832                         req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
3833                         if (!req) {
3834                                 *error = "Could not allocate crypt request";
3835                                 r = -ENOMEM;
3836                                 goto bad;
3837                         }
3838
3839                         crypt_iv = kmalloc(ivsize, GFP_KERNEL);
3840                         if (!crypt_iv) {
3841                                 *error = "Could not allocate iv";
3842                                 r = -ENOMEM;
3843                                 goto bad;
3844                         }
3845
3846                         crypt_data = kmalloc(crypt_len, GFP_KERNEL);
3847                         if (!crypt_data) {
3848                                 *error = "Unable to allocate crypt data";
3849                                 r = -ENOMEM;
3850                                 goto bad;
3851                         }
3852
3853                         ic->journal_scatterlist = dm_integrity_alloc_journal_scatterlist(ic, ic->journal);
3854                         if (!ic->journal_scatterlist) {
3855                                 *error = "Unable to allocate sg list";
3856                                 r = -ENOMEM;
3857                                 goto bad;
3858                         }
3859                         ic->journal_io_scatterlist = dm_integrity_alloc_journal_scatterlist(ic, ic->journal_io);
3860                         if (!ic->journal_io_scatterlist) {
3861                                 *error = "Unable to allocate sg list";
3862                                 r = -ENOMEM;
3863                                 goto bad;
3864                         }
3865                         ic->sk_requests = kvmalloc_array(ic->journal_sections,
3866                                                          sizeof(struct skcipher_request *),
3867                                                          GFP_KERNEL | __GFP_ZERO);
3868                         if (!ic->sk_requests) {
3869                                 *error = "Unable to allocate sk requests";
3870                                 r = -ENOMEM;
3871                                 goto bad;
3872                         }
3873                         for (i = 0; i < ic->journal_sections; i++) {
3874                                 struct scatterlist sg;
3875                                 struct skcipher_request *section_req;
3876                                 __le32 section_le = cpu_to_le32(i);
3877
3878                                 memset(crypt_iv, 0x00, ivsize);
3879                                 memset(crypt_data, 0x00, crypt_len);
3880                                 memcpy(crypt_data, &section_le, min((size_t)crypt_len, sizeof(section_le)));
3881
3882                                 sg_init_one(&sg, crypt_data, crypt_len);
3883                                 skcipher_request_set_crypt(req, &sg, &sg, crypt_len, crypt_iv);
3884                                 init_completion(&comp.comp);
3885                                 comp.in_flight = (atomic_t)ATOMIC_INIT(1);
3886                                 if (do_crypt(true, req, &comp))
3887                                         wait_for_completion(&comp.comp);
3888
3889                                 r = dm_integrity_failed(ic);
3890                                 if (r) {
3891                                         *error = "Unable to generate iv";
3892                                         goto bad;
3893                                 }
3894
3895                                 section_req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
3896                                 if (!section_req) {
3897                                         *error = "Unable to allocate crypt request";
3898                                         r = -ENOMEM;
3899                                         goto bad;
3900                                 }
3901                                 section_req->iv = kmalloc_array(ivsize, 2,
3902                                                                 GFP_KERNEL);
3903                                 if (!section_req->iv) {
3904                                         skcipher_request_free(section_req);
3905                                         *error = "Unable to allocate iv";
3906                                         r = -ENOMEM;
3907                                         goto bad;
3908                                 }
3909                                 memcpy(section_req->iv + ivsize, crypt_data, ivsize);
3910                                 section_req->cryptlen = (size_t)ic->journal_section_sectors << SECTOR_SHIFT;
3911                                 ic->sk_requests[i] = section_req;
3912                                 DEBUG_bytes(crypt_data, ivsize, "iv(%u)", i);
3913                         }
3914                 }
3915         }
3916
3917         for (i = 0; i < N_COMMIT_IDS; i++) {
3918                 unsigned j;
3919 retest_commit_id:
3920                 for (j = 0; j < i; j++) {
3921                         if (ic->commit_ids[j] == ic->commit_ids[i]) {
3922                                 ic->commit_ids[i] = cpu_to_le64(le64_to_cpu(ic->commit_ids[i]) + 1);
3923                                 goto retest_commit_id;
3924                         }
3925                 }
3926                 DEBUG_print("commit id %u: %016llx\n", i, ic->commit_ids[i]);
3927         }
3928
3929         journal_tree_size = (__u64)ic->journal_entries * sizeof(struct journal_node);
3930         if (journal_tree_size > ULONG_MAX) {
3931                 *error = "Journal doesn't fit into memory";
3932                 r = -ENOMEM;
3933                 goto bad;
3934         }
3935         ic->journal_tree = kvmalloc(journal_tree_size, GFP_KERNEL);
3936         if (!ic->journal_tree) {
3937                 *error = "Could not allocate memory for journal tree";
3938                 r = -ENOMEM;
3939         }
3940 bad:
3941         kfree(crypt_data);
3942         kfree(crypt_iv);
3943         skcipher_request_free(req);
3944
3945         return r;
3946 }
3947
3948 /*
3949  * Construct a integrity mapping
3950  *
3951  * Arguments:
3952  *      device
3953  *      offset from the start of the device
3954  *      tag size
3955  *      D - direct writes, J - journal writes, B - bitmap mode, R - recovery mode
3956  *      number of optional arguments
3957  *      optional arguments:
3958  *              journal_sectors
3959  *              interleave_sectors
3960  *              buffer_sectors
3961  *              journal_watermark
3962  *              commit_time
3963  *              meta_device
3964  *              block_size
3965  *              sectors_per_bit
3966  *              bitmap_flush_interval
3967  *              internal_hash
3968  *              journal_crypt
3969  *              journal_mac
3970  *              recalculate
3971  */
3972 static int dm_integrity_ctr(struct dm_target *ti, unsigned argc, char **argv)
3973 {
3974         struct dm_integrity_c *ic;
3975         char dummy;
3976         int r;
3977         unsigned extra_args;
3978         struct dm_arg_set as;
3979         static const struct dm_arg _args[] = {
3980                 {0, 18, "Invalid number of feature args"},
3981         };
3982         unsigned journal_sectors, interleave_sectors, buffer_sectors, journal_watermark, sync_msec;
3983         bool should_write_sb;
3984         __u64 threshold;
3985         unsigned long long start;
3986         __s8 log2_sectors_per_bitmap_bit = -1;
3987         __s8 log2_blocks_per_bitmap_bit;
3988         __u64 bits_in_journal;
3989         __u64 n_bitmap_bits;
3990
3991 #define DIRECT_ARGUMENTS        4
3992
3993         if (argc <= DIRECT_ARGUMENTS) {
3994                 ti->error = "Invalid argument count";
3995                 return -EINVAL;
3996         }
3997
3998         ic = kzalloc(sizeof(struct dm_integrity_c), GFP_KERNEL);
3999         if (!ic) {
4000                 ti->error = "Cannot allocate integrity context";
4001                 return -ENOMEM;
4002         }
4003         ti->private = ic;
4004         ti->per_io_data_size = sizeof(struct dm_integrity_io);
4005         ic->ti = ti;
4006
4007         ic->in_progress = RB_ROOT;
4008         INIT_LIST_HEAD(&ic->wait_list);
4009         init_waitqueue_head(&ic->endio_wait);
4010         bio_list_init(&ic->flush_bio_list);
4011         init_waitqueue_head(&ic->copy_to_journal_wait);
4012         init_completion(&ic->crypto_backoff);
4013         atomic64_set(&ic->number_of_mismatches, 0);
4014         ic->bitmap_flush_interval = BITMAP_FLUSH_INTERVAL;
4015
4016         r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &ic->dev);
4017         if (r) {
4018                 ti->error = "Device lookup failed";
4019                 goto bad;
4020         }
4021
4022         if (sscanf(argv[1], "%llu%c", &start, &dummy) != 1 || start != (sector_t)start) {
4023                 ti->error = "Invalid starting offset";
4024                 r = -EINVAL;
4025                 goto bad;
4026         }
4027         ic->start = start;
4028
4029         if (strcmp(argv[2], "-")) {
4030                 if (sscanf(argv[2], "%u%c", &ic->tag_size, &dummy) != 1 || !ic->tag_size) {
4031                         ti->error = "Invalid tag size";
4032                         r = -EINVAL;
4033                         goto bad;
4034                 }
4035         }
4036
4037         if (!strcmp(argv[3], "J") || !strcmp(argv[3], "B") ||
4038             !strcmp(argv[3], "D") || !strcmp(argv[3], "R")) {
4039                 ic->mode = argv[3][0];
4040         } else {
4041                 ti->error = "Invalid mode (expecting J, B, D, R)";
4042                 r = -EINVAL;
4043                 goto bad;
4044         }
4045
4046         journal_sectors = 0;
4047         interleave_sectors = DEFAULT_INTERLEAVE_SECTORS;
4048         buffer_sectors = DEFAULT_BUFFER_SECTORS;
4049         journal_watermark = DEFAULT_JOURNAL_WATERMARK;
4050         sync_msec = DEFAULT_SYNC_MSEC;
4051         ic->sectors_per_block = 1;
4052
4053         as.argc = argc - DIRECT_ARGUMENTS;
4054         as.argv = argv + DIRECT_ARGUMENTS;
4055         r = dm_read_arg_group(_args, &as, &extra_args, &ti->error);
4056         if (r)
4057                 goto bad;
4058
4059         while (extra_args--) {
4060                 const char *opt_string;
4061                 unsigned val;
4062                 unsigned long long llval;
4063                 opt_string = dm_shift_arg(&as);
4064                 if (!opt_string) {
4065                         r = -EINVAL;
4066                         ti->error = "Not enough feature arguments";
4067                         goto bad;
4068                 }
4069                 if (sscanf(opt_string, "journal_sectors:%u%c", &val, &dummy) == 1)
4070                         journal_sectors = val ? val : 1;
4071                 else if (sscanf(opt_string, "interleave_sectors:%u%c", &val, &dummy) == 1)
4072                         interleave_sectors = val;
4073                 else if (sscanf(opt_string, "buffer_sectors:%u%c", &val, &dummy) == 1)
4074                         buffer_sectors = val;
4075                 else if (sscanf(opt_string, "journal_watermark:%u%c", &val, &dummy) == 1 && val <= 100)
4076                         journal_watermark = val;
4077                 else if (sscanf(opt_string, "commit_time:%u%c", &val, &dummy) == 1)
4078                         sync_msec = val;
4079                 else if (!strncmp(opt_string, "meta_device:", strlen("meta_device:"))) {
4080                         if (ic->meta_dev) {
4081                                 dm_put_device(ti, ic->meta_dev);
4082                                 ic->meta_dev = NULL;
4083                         }
4084                         r = dm_get_device(ti, strchr(opt_string, ':') + 1,
4085                                           dm_table_get_mode(ti->table), &ic->meta_dev);
4086                         if (r) {
4087                                 ti->error = "Device lookup failed";
4088                                 goto bad;
4089                         }
4090                 } else if (sscanf(opt_string, "block_size:%u%c", &val, &dummy) == 1) {
4091                         if (val < 1 << SECTOR_SHIFT ||
4092                             val > MAX_SECTORS_PER_BLOCK << SECTOR_SHIFT ||
4093                             (val & (val -1))) {
4094                                 r = -EINVAL;
4095                                 ti->error = "Invalid block_size argument";
4096                                 goto bad;
4097                         }
4098                         ic->sectors_per_block = val >> SECTOR_SHIFT;
4099                 } else if (sscanf(opt_string, "sectors_per_bit:%llu%c", &llval, &dummy) == 1) {
4100                         log2_sectors_per_bitmap_bit = !llval ? 0 : __ilog2_u64(llval);
4101                 } else if (sscanf(opt_string, "bitmap_flush_interval:%u%c", &val, &dummy) == 1) {
4102                         if (val >= (uint64_t)UINT_MAX * 1000 / HZ) {
4103                                 r = -EINVAL;
4104                                 ti->error = "Invalid bitmap_flush_interval argument";
4105                                 goto bad;
4106                         }
4107                         ic->bitmap_flush_interval = msecs_to_jiffies(val);
4108                 } else if (!strncmp(opt_string, "internal_hash:", strlen("internal_hash:"))) {
4109                         r = get_alg_and_key(opt_string, &ic->internal_hash_alg, &ti->error,
4110                                             "Invalid internal_hash argument");
4111                         if (r)
4112                                 goto bad;
4113                 } else if (!strncmp(opt_string, "journal_crypt:", strlen("journal_crypt:"))) {
4114                         r = get_alg_and_key(opt_string, &ic->journal_crypt_alg, &ti->error,
4115                                             "Invalid journal_crypt argument");
4116                         if (r)
4117                                 goto bad;
4118                 } else if (!strncmp(opt_string, "journal_mac:", strlen("journal_mac:"))) {
4119                         r = get_alg_and_key(opt_string, &ic->journal_mac_alg, &ti->error,
4120                                             "Invalid journal_mac argument");
4121                         if (r)
4122                                 goto bad;
4123                 } else if (!strcmp(opt_string, "recalculate")) {
4124                         ic->recalculate_flag = true;
4125                 } else if (!strcmp(opt_string, "reset_recalculate")) {
4126                         ic->recalculate_flag = true;
4127                         ic->reset_recalculate_flag = true;
4128                 } else if (!strcmp(opt_string, "allow_discards")) {
4129                         ic->discard = true;
4130                 } else if (!strcmp(opt_string, "fix_padding")) {
4131                         ic->fix_padding = true;
4132                 } else if (!strcmp(opt_string, "fix_hmac")) {
4133                         ic->fix_hmac = true;
4134                 } else if (!strcmp(opt_string, "legacy_recalculate")) {
4135                         ic->legacy_recalculate = true;
4136                 } else {
4137                         r = -EINVAL;
4138                         ti->error = "Invalid argument";
4139                         goto bad;
4140                 }
4141         }
4142
4143         ic->data_device_sectors = bdev_nr_sectors(ic->dev->bdev);
4144         if (!ic->meta_dev)
4145                 ic->meta_device_sectors = ic->data_device_sectors;
4146         else
4147                 ic->meta_device_sectors = bdev_nr_sectors(ic->meta_dev->bdev);
4148
4149         if (!journal_sectors) {
4150                 journal_sectors = min((sector_t)DEFAULT_MAX_JOURNAL_SECTORS,
4151                                       ic->data_device_sectors >> DEFAULT_JOURNAL_SIZE_FACTOR);
4152         }
4153
4154         if (!buffer_sectors)
4155                 buffer_sectors = 1;
4156         ic->log2_buffer_sectors = min((int)__fls(buffer_sectors), 31 - SECTOR_SHIFT);
4157
4158         r = get_mac(&ic->internal_hash, &ic->internal_hash_alg, &ti->error,
4159                     "Invalid internal hash", "Error setting internal hash key");
4160         if (r)
4161                 goto bad;
4162
4163         r = get_mac(&ic->journal_mac, &ic->journal_mac_alg, &ti->error,
4164                     "Invalid journal mac", "Error setting journal mac key");
4165         if (r)
4166                 goto bad;
4167
4168         if (!ic->tag_size) {
4169                 if (!ic->internal_hash) {
4170                         ti->error = "Unknown tag size";
4171                         r = -EINVAL;
4172                         goto bad;
4173                 }
4174                 ic->tag_size = crypto_shash_digestsize(ic->internal_hash);
4175         }
4176         if (ic->tag_size > MAX_TAG_SIZE) {
4177                 ti->error = "Too big tag size";
4178                 r = -EINVAL;
4179                 goto bad;
4180         }
4181         if (!(ic->tag_size & (ic->tag_size - 1)))
4182                 ic->log2_tag_size = __ffs(ic->tag_size);
4183         else
4184                 ic->log2_tag_size = -1;
4185
4186         if (ic->mode == 'B' && !ic->internal_hash) {
4187                 r = -EINVAL;
4188                 ti->error = "Bitmap mode can be only used with internal hash";
4189                 goto bad;
4190         }
4191
4192         if (ic->discard && !ic->internal_hash) {
4193                 r = -EINVAL;
4194                 ti->error = "Discard can be only used with internal hash";
4195                 goto bad;
4196         }
4197
4198         ic->autocommit_jiffies = msecs_to_jiffies(sync_msec);
4199         ic->autocommit_msec = sync_msec;
4200         timer_setup(&ic->autocommit_timer, autocommit_fn, 0);
4201
4202         ic->io = dm_io_client_create();
4203         if (IS_ERR(ic->io)) {
4204                 r = PTR_ERR(ic->io);
4205                 ic->io = NULL;
4206                 ti->error = "Cannot allocate dm io";
4207                 goto bad;
4208         }
4209
4210         r = mempool_init_slab_pool(&ic->journal_io_mempool, JOURNAL_IO_MEMPOOL, journal_io_cache);
4211         if (r) {
4212                 ti->error = "Cannot allocate mempool";
4213                 goto bad;
4214         }
4215
4216         ic->metadata_wq = alloc_workqueue("dm-integrity-metadata",
4217                                           WQ_MEM_RECLAIM, METADATA_WORKQUEUE_MAX_ACTIVE);
4218         if (!ic->metadata_wq) {
4219                 ti->error = "Cannot allocate workqueue";
4220                 r = -ENOMEM;
4221                 goto bad;
4222         }
4223
4224         /*
4225          * If this workqueue were percpu, it would cause bio reordering
4226          * and reduced performance.
4227          */
4228         ic->wait_wq = alloc_workqueue("dm-integrity-wait", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
4229         if (!ic->wait_wq) {
4230                 ti->error = "Cannot allocate workqueue";
4231                 r = -ENOMEM;
4232                 goto bad;
4233         }
4234
4235         ic->offload_wq = alloc_workqueue("dm-integrity-offload", WQ_MEM_RECLAIM,
4236                                           METADATA_WORKQUEUE_MAX_ACTIVE);
4237         if (!ic->offload_wq) {
4238                 ti->error = "Cannot allocate workqueue";
4239                 r = -ENOMEM;
4240                 goto bad;
4241         }
4242
4243         ic->commit_wq = alloc_workqueue("dm-integrity-commit", WQ_MEM_RECLAIM, 1);
4244         if (!ic->commit_wq) {
4245                 ti->error = "Cannot allocate workqueue";
4246                 r = -ENOMEM;
4247                 goto bad;
4248         }
4249         INIT_WORK(&ic->commit_work, integrity_commit);
4250
4251         if (ic->mode == 'J' || ic->mode == 'B') {
4252                 ic->writer_wq = alloc_workqueue("dm-integrity-writer", WQ_MEM_RECLAIM, 1);
4253                 if (!ic->writer_wq) {
4254                         ti->error = "Cannot allocate workqueue";
4255                         r = -ENOMEM;
4256                         goto bad;
4257                 }
4258                 INIT_WORK(&ic->writer_work, integrity_writer);
4259         }
4260
4261         ic->sb = alloc_pages_exact(SB_SECTORS << SECTOR_SHIFT, GFP_KERNEL);
4262         if (!ic->sb) {
4263                 r = -ENOMEM;
4264                 ti->error = "Cannot allocate superblock area";
4265                 goto bad;
4266         }
4267
4268         r = sync_rw_sb(ic, REQ_OP_READ);
4269         if (r) {
4270                 ti->error = "Error reading superblock";
4271                 goto bad;
4272         }
4273         should_write_sb = false;
4274         if (memcmp(ic->sb->magic, SB_MAGIC, 8)) {
4275                 if (ic->mode != 'R') {
4276                         if (memchr_inv(ic->sb, 0, SB_SECTORS << SECTOR_SHIFT)) {
4277                                 r = -EINVAL;
4278                                 ti->error = "The device is not initialized";
4279                                 goto bad;
4280                         }
4281                 }
4282
4283                 r = initialize_superblock(ic, journal_sectors, interleave_sectors);
4284                 if (r) {
4285                         ti->error = "Could not initialize superblock";
4286                         goto bad;
4287                 }
4288                 if (ic->mode != 'R')
4289                         should_write_sb = true;
4290         }
4291
4292         if (!ic->sb->version || ic->sb->version > SB_VERSION_5) {
4293                 r = -EINVAL;
4294                 ti->error = "Unknown version";
4295                 goto bad;
4296         }
4297         if (le16_to_cpu(ic->sb->integrity_tag_size) != ic->tag_size) {
4298                 r = -EINVAL;
4299                 ti->error = "Tag size doesn't match the information in superblock";
4300                 goto bad;
4301         }
4302         if (ic->sb->log2_sectors_per_block != __ffs(ic->sectors_per_block)) {
4303                 r = -EINVAL;
4304                 ti->error = "Block size doesn't match the information in superblock";
4305                 goto bad;
4306         }
4307         if (!le32_to_cpu(ic->sb->journal_sections)) {
4308                 r = -EINVAL;
4309                 ti->error = "Corrupted superblock, journal_sections is 0";
4310                 goto bad;
4311         }
4312         /* make sure that ti->max_io_len doesn't overflow */
4313         if (!ic->meta_dev) {
4314                 if (ic->sb->log2_interleave_sectors < MIN_LOG2_INTERLEAVE_SECTORS ||
4315                     ic->sb->log2_interleave_sectors > MAX_LOG2_INTERLEAVE_SECTORS) {
4316                         r = -EINVAL;
4317                         ti->error = "Invalid interleave_sectors in the superblock";
4318                         goto bad;
4319                 }
4320         } else {
4321                 if (ic->sb->log2_interleave_sectors) {
4322                         r = -EINVAL;
4323                         ti->error = "Invalid interleave_sectors in the superblock";
4324                         goto bad;
4325                 }
4326         }
4327         if (!!(ic->sb->flags & cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC)) != !!ic->journal_mac_alg.alg_string) {
4328                 r = -EINVAL;
4329                 ti->error = "Journal mac mismatch";
4330                 goto bad;
4331         }
4332
4333         get_provided_data_sectors(ic);
4334         if (!ic->provided_data_sectors) {
4335                 r = -EINVAL;
4336                 ti->error = "The device is too small";
4337                 goto bad;
4338         }
4339
4340 try_smaller_buffer:
4341         r = calculate_device_limits(ic);
4342         if (r) {
4343                 if (ic->meta_dev) {
4344                         if (ic->log2_buffer_sectors > 3) {
4345                                 ic->log2_buffer_sectors--;
4346                                 goto try_smaller_buffer;
4347                         }
4348                 }
4349                 ti->error = "The device is too small";
4350                 goto bad;
4351         }
4352
4353         if (log2_sectors_per_bitmap_bit < 0)
4354                 log2_sectors_per_bitmap_bit = __fls(DEFAULT_SECTORS_PER_BITMAP_BIT);
4355         if (log2_sectors_per_bitmap_bit < ic->sb->log2_sectors_per_block)
4356                 log2_sectors_per_bitmap_bit = ic->sb->log2_sectors_per_block;
4357
4358         bits_in_journal = ((__u64)ic->journal_section_sectors * ic->journal_sections) << (SECTOR_SHIFT + 3);
4359         if (bits_in_journal > UINT_MAX)
4360                 bits_in_journal = UINT_MAX;
4361         while (bits_in_journal < (ic->provided_data_sectors + ((sector_t)1 << log2_sectors_per_bitmap_bit) - 1) >> log2_sectors_per_bitmap_bit)
4362                 log2_sectors_per_bitmap_bit++;
4363
4364         log2_blocks_per_bitmap_bit = log2_sectors_per_bitmap_bit - ic->sb->log2_sectors_per_block;
4365         ic->log2_blocks_per_bitmap_bit = log2_blocks_per_bitmap_bit;
4366         if (should_write_sb) {
4367                 ic->sb->log2_blocks_per_bitmap_bit = log2_blocks_per_bitmap_bit;
4368         }
4369         n_bitmap_bits = ((ic->provided_data_sectors >> ic->sb->log2_sectors_per_block)
4370                                 + (((sector_t)1 << log2_blocks_per_bitmap_bit) - 1)) >> log2_blocks_per_bitmap_bit;
4371         ic->n_bitmap_blocks = DIV_ROUND_UP(n_bitmap_bits, BITMAP_BLOCK_SIZE * 8);
4372
4373         if (!ic->meta_dev)
4374                 ic->log2_buffer_sectors = min(ic->log2_buffer_sectors, (__u8)__ffs(ic->metadata_run));
4375
4376         if (ti->len > ic->provided_data_sectors) {
4377                 r = -EINVAL;
4378                 ti->error = "Not enough provided sectors for requested mapping size";
4379                 goto bad;
4380         }
4381
4382
4383         threshold = (__u64)ic->journal_entries * (100 - journal_watermark);
4384         threshold += 50;
4385         do_div(threshold, 100);
4386         ic->free_sectors_threshold = threshold;
4387
4388         DEBUG_print("initialized:\n");
4389         DEBUG_print("   integrity_tag_size %u\n", le16_to_cpu(ic->sb->integrity_tag_size));
4390         DEBUG_print("   journal_entry_size %u\n", ic->journal_entry_size);
4391         DEBUG_print("   journal_entries_per_sector %u\n", ic->journal_entries_per_sector);
4392         DEBUG_print("   journal_section_entries %u\n", ic->journal_section_entries);
4393         DEBUG_print("   journal_section_sectors %u\n", ic->journal_section_sectors);
4394         DEBUG_print("   journal_sections %u\n", (unsigned)le32_to_cpu(ic->sb->journal_sections));
4395         DEBUG_print("   journal_entries %u\n", ic->journal_entries);
4396         DEBUG_print("   log2_interleave_sectors %d\n", ic->sb->log2_interleave_sectors);
4397         DEBUG_print("   data_device_sectors 0x%llx\n", bdev_nr_sectors(ic->dev->bdev));
4398         DEBUG_print("   initial_sectors 0x%x\n", ic->initial_sectors);
4399         DEBUG_print("   metadata_run 0x%x\n", ic->metadata_run);
4400         DEBUG_print("   log2_metadata_run %d\n", ic->log2_metadata_run);
4401         DEBUG_print("   provided_data_sectors 0x%llx (%llu)\n", ic->provided_data_sectors, ic->provided_data_sectors);
4402         DEBUG_print("   log2_buffer_sectors %u\n", ic->log2_buffer_sectors);
4403         DEBUG_print("   bits_in_journal %llu\n", bits_in_journal);
4404
4405         if (ic->recalculate_flag && !(ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))) {
4406                 ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
4407                 ic->sb->recalc_sector = cpu_to_le64(0);
4408         }
4409
4410         if (ic->internal_hash) {
4411                 size_t recalc_tags_size;
4412                 ic->recalc_wq = alloc_workqueue("dm-integrity-recalc", WQ_MEM_RECLAIM, 1);
4413                 if (!ic->recalc_wq ) {
4414                         ti->error = "Cannot allocate workqueue";
4415                         r = -ENOMEM;
4416                         goto bad;
4417                 }
4418                 INIT_WORK(&ic->recalc_work, integrity_recalc);
4419                 ic->recalc_buffer = vmalloc(RECALC_SECTORS << SECTOR_SHIFT);
4420                 if (!ic->recalc_buffer) {
4421                         ti->error = "Cannot allocate buffer for recalculating";
4422                         r = -ENOMEM;
4423                         goto bad;
4424                 }
4425                 recalc_tags_size = (RECALC_SECTORS >> ic->sb->log2_sectors_per_block) * ic->tag_size;
4426                 if (crypto_shash_digestsize(ic->internal_hash) > ic->tag_size)
4427                         recalc_tags_size += crypto_shash_digestsize(ic->internal_hash) - ic->tag_size;
4428                 ic->recalc_tags = kvmalloc(recalc_tags_size, GFP_KERNEL);
4429                 if (!ic->recalc_tags) {
4430                         ti->error = "Cannot allocate tags for recalculating";
4431                         r = -ENOMEM;
4432                         goto bad;
4433                 }
4434         } else {
4435                 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
4436                         ti->error = "Recalculate can only be specified with internal_hash";
4437                         r = -EINVAL;
4438                         goto bad;
4439                 }
4440         }
4441
4442         if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING) &&
4443             le64_to_cpu(ic->sb->recalc_sector) < ic->provided_data_sectors &&
4444             dm_integrity_disable_recalculate(ic)) {
4445                 ti->error = "Recalculating with HMAC is disabled for security reasons - if you really need it, use the argument \"legacy_recalculate\"";
4446                 r = -EOPNOTSUPP;
4447                 goto bad;
4448         }
4449
4450         ic->bufio = dm_bufio_client_create(ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev,
4451                         1U << (SECTOR_SHIFT + ic->log2_buffer_sectors), 1, 0, NULL, NULL, 0);
4452         if (IS_ERR(ic->bufio)) {
4453                 r = PTR_ERR(ic->bufio);
4454                 ti->error = "Cannot initialize dm-bufio";
4455                 ic->bufio = NULL;
4456                 goto bad;
4457         }
4458         dm_bufio_set_sector_offset(ic->bufio, ic->start + ic->initial_sectors);
4459
4460         if (ic->mode != 'R') {
4461                 r = create_journal(ic, &ti->error);
4462                 if (r)
4463                         goto bad;
4464
4465         }
4466
4467         if (ic->mode == 'B') {
4468                 unsigned i;
4469                 unsigned n_bitmap_pages = DIV_ROUND_UP(ic->n_bitmap_blocks, PAGE_SIZE / BITMAP_BLOCK_SIZE);
4470
4471                 ic->recalc_bitmap = dm_integrity_alloc_page_list(n_bitmap_pages);
4472                 if (!ic->recalc_bitmap) {
4473                         r = -ENOMEM;
4474                         goto bad;
4475                 }
4476                 ic->may_write_bitmap = dm_integrity_alloc_page_list(n_bitmap_pages);
4477                 if (!ic->may_write_bitmap) {
4478                         r = -ENOMEM;
4479                         goto bad;
4480                 }
4481                 ic->bbs = kvmalloc_array(ic->n_bitmap_blocks, sizeof(struct bitmap_block_status), GFP_KERNEL);
4482                 if (!ic->bbs) {
4483                         r = -ENOMEM;
4484                         goto bad;
4485                 }
4486                 INIT_DELAYED_WORK(&ic->bitmap_flush_work, bitmap_flush_work);
4487                 for (i = 0; i < ic->n_bitmap_blocks; i++) {
4488                         struct bitmap_block_status *bbs = &ic->bbs[i];
4489                         unsigned sector, pl_index, pl_offset;
4490
4491                         INIT_WORK(&bbs->work, bitmap_block_work);
4492                         bbs->ic = ic;
4493                         bbs->idx = i;
4494                         bio_list_init(&bbs->bio_queue);
4495                         spin_lock_init(&bbs->bio_queue_lock);
4496
4497                         sector = i * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT);
4498                         pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
4499                         pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
4500
4501                         bbs->bitmap = lowmem_page_address(ic->journal[pl_index].page) + pl_offset;
4502                 }
4503         }
4504
4505         if (should_write_sb) {
4506                 init_journal(ic, 0, ic->journal_sections, 0);
4507                 r = dm_integrity_failed(ic);
4508                 if (unlikely(r)) {
4509                         ti->error = "Error initializing journal";
4510                         goto bad;
4511                 }
4512                 r = sync_rw_sb(ic, REQ_OP_WRITE | REQ_FUA);
4513                 if (r) {
4514                         ti->error = "Error initializing superblock";
4515                         goto bad;
4516                 }
4517                 ic->just_formatted = true;
4518         }
4519
4520         if (!ic->meta_dev) {
4521                 r = dm_set_target_max_io_len(ti, 1U << ic->sb->log2_interleave_sectors);
4522                 if (r)
4523                         goto bad;
4524         }
4525         if (ic->mode == 'B') {
4526                 unsigned max_io_len = ((sector_t)ic->sectors_per_block << ic->log2_blocks_per_bitmap_bit) * (BITMAP_BLOCK_SIZE * 8);
4527                 if (!max_io_len)
4528                         max_io_len = 1U << 31;
4529                 DEBUG_print("max_io_len: old %u, new %u\n", ti->max_io_len, max_io_len);
4530                 if (!ti->max_io_len || ti->max_io_len > max_io_len) {
4531                         r = dm_set_target_max_io_len(ti, max_io_len);
4532                         if (r)
4533                                 goto bad;
4534                 }
4535         }
4536
4537         if (!ic->internal_hash)
4538                 dm_integrity_set(ti, ic);
4539
4540         ti->num_flush_bios = 1;
4541         ti->flush_supported = true;
4542         if (ic->discard)
4543                 ti->num_discard_bios = 1;
4544
4545         dm_audit_log_ctr(DM_MSG_PREFIX, ti, 1);
4546         return 0;
4547
4548 bad:
4549         dm_audit_log_ctr(DM_MSG_PREFIX, ti, 0);
4550         dm_integrity_dtr(ti);
4551         return r;
4552 }
4553
4554 static void dm_integrity_dtr(struct dm_target *ti)
4555 {
4556         struct dm_integrity_c *ic = ti->private;
4557
4558         BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
4559         BUG_ON(!list_empty(&ic->wait_list));
4560
4561         if (ic->mode == 'B')
4562                 cancel_delayed_work_sync(&ic->bitmap_flush_work);
4563         if (ic->metadata_wq)
4564                 destroy_workqueue(ic->metadata_wq);
4565         if (ic->wait_wq)
4566                 destroy_workqueue(ic->wait_wq);
4567         if (ic->offload_wq)
4568                 destroy_workqueue(ic->offload_wq);
4569         if (ic->commit_wq)
4570                 destroy_workqueue(ic->commit_wq);
4571         if (ic->writer_wq)
4572                 destroy_workqueue(ic->writer_wq);
4573         if (ic->recalc_wq)
4574                 destroy_workqueue(ic->recalc_wq);
4575         vfree(ic->recalc_buffer);
4576         kvfree(ic->recalc_tags);
4577         kvfree(ic->bbs);
4578         if (ic->bufio)
4579                 dm_bufio_client_destroy(ic->bufio);
4580         mempool_exit(&ic->journal_io_mempool);
4581         if (ic->io)
4582                 dm_io_client_destroy(ic->io);
4583         if (ic->dev)
4584                 dm_put_device(ti, ic->dev);
4585         if (ic->meta_dev)
4586                 dm_put_device(ti, ic->meta_dev);
4587         dm_integrity_free_page_list(ic->journal);
4588         dm_integrity_free_page_list(ic->journal_io);
4589         dm_integrity_free_page_list(ic->journal_xor);
4590         dm_integrity_free_page_list(ic->recalc_bitmap);
4591         dm_integrity_free_page_list(ic->may_write_bitmap);
4592         if (ic->journal_scatterlist)
4593                 dm_integrity_free_journal_scatterlist(ic, ic->journal_scatterlist);
4594         if (ic->journal_io_scatterlist)
4595                 dm_integrity_free_journal_scatterlist(ic, ic->journal_io_scatterlist);
4596         if (ic->sk_requests) {
4597                 unsigned i;
4598
4599                 for (i = 0; i < ic->journal_sections; i++) {
4600                         struct skcipher_request *req = ic->sk_requests[i];
4601                         if (req) {
4602                                 kfree_sensitive(req->iv);
4603                                 skcipher_request_free(req);
4604                         }
4605                 }
4606                 kvfree(ic->sk_requests);
4607         }
4608         kvfree(ic->journal_tree);
4609         if (ic->sb)
4610                 free_pages_exact(ic->sb, SB_SECTORS << SECTOR_SHIFT);
4611
4612         if (ic->internal_hash)
4613                 crypto_free_shash(ic->internal_hash);
4614         free_alg(&ic->internal_hash_alg);
4615
4616         if (ic->journal_crypt)
4617                 crypto_free_skcipher(ic->journal_crypt);
4618         free_alg(&ic->journal_crypt_alg);
4619
4620         if (ic->journal_mac)
4621                 crypto_free_shash(ic->journal_mac);
4622         free_alg(&ic->journal_mac_alg);
4623
4624         kfree(ic);
4625         dm_audit_log_dtr(DM_MSG_PREFIX, ti, 1);
4626 }
4627
4628 static struct target_type integrity_target = {
4629         .name                   = "integrity",
4630         .version                = {1, 10, 0},
4631         .module                 = THIS_MODULE,
4632         .features               = DM_TARGET_SINGLETON | DM_TARGET_INTEGRITY,
4633         .ctr                    = dm_integrity_ctr,
4634         .dtr                    = dm_integrity_dtr,
4635         .map                    = dm_integrity_map,
4636         .postsuspend            = dm_integrity_postsuspend,
4637         .resume                 = dm_integrity_resume,
4638         .status                 = dm_integrity_status,
4639         .iterate_devices        = dm_integrity_iterate_devices,
4640         .io_hints               = dm_integrity_io_hints,
4641 };
4642
4643 static int __init dm_integrity_init(void)
4644 {
4645         int r;
4646
4647         journal_io_cache = kmem_cache_create("integrity_journal_io",
4648                                              sizeof(struct journal_io), 0, 0, NULL);
4649         if (!journal_io_cache) {
4650                 DMERR("can't allocate journal io cache");
4651                 return -ENOMEM;
4652         }
4653
4654         r = dm_register_target(&integrity_target);
4655
4656         if (r < 0)
4657                 DMERR("register failed %d", r);
4658
4659         return r;
4660 }
4661
4662 static void __exit dm_integrity_exit(void)
4663 {
4664         dm_unregister_target(&integrity_target);
4665         kmem_cache_destroy(journal_io_cache);
4666 }
4667
4668 module_init(dm_integrity_init);
4669 module_exit(dm_integrity_exit);
4670
4671 MODULE_AUTHOR("Milan Broz");
4672 MODULE_AUTHOR("Mikulas Patocka");
4673 MODULE_DESCRIPTION(DM_NAME " target for integrity tags extension");
4674 MODULE_LICENSE("GPL");