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 <linux/module.h>
  10 #include <linux/device-mapper.h>
  11 #include <linux/dm-io.h>
  12 #include <linux/vmalloc.h>
  13 #include <linux/sort.h>
  14 #include <linux/rbtree.h>
  15 #include <linux/delay.h>
  16 #include <linux/random.h>
  17 #include <crypto/hash.h>
  18 #include <crypto/skcipher.h>
  19 #include <linux/async_tx.h>
  20 #include "dm-bufio.h"
  21
  22 #define DM_MSG_PREFIX "integrity"
  23
  24 #define DEFAULT_INTERLEAVE_SECTORS      32768
  25 #define DEFAULT_JOURNAL_SIZE_FACTOR     7
  26 #define DEFAULT_BUFFER_SECTORS          128
  27 #define DEFAULT_JOURNAL_WATERMARK       50
  28 #define DEFAULT_SYNC_MSEC               10000
  29 #define DEFAULT_MAX_JOURNAL_SECTORS     131072
  30 #define MIN_LOG2_INTERLEAVE_SECTORS     3
  31 #define MAX_LOG2_INTERLEAVE_SECTORS     31
  32 #define METADATA_WORKQUEUE_MAX_ACTIVE   16
  33
  34 /*
  35  * Warning - DEBUG_PRINT prints security-sensitive data to the log,
  36  * so it should not be enabled in the official kernel
  37  */
  38 //#define DEBUG_PRINT
  39 //#define INTERNAL_VERIFY
  40
  41 /*
  42  * On disk structures
  43  */
  44
  45 #define SB_MAGIC                        "integrt"
  46 #define SB_VERSION                      1
  47 #define SB_SECTORS                      8
  48 #define MAX_SECTORS_PER_BLOCK           8
  49
  50 struct superblock {
  51         __u8 magic[8];
  52         __u8 version;
  53         __u8 log2_interleave_sectors;
  54         __u16 integrity_tag_size;
  55         __u32 journal_sections;
  56         __u64 provided_data_sectors;    /* userspace uses this value */
  57         __u32 flags;
  58         __u8 log2_sectors_per_block;
  59 };
  60
  61 #define SB_FLAG_HAVE_JOURNAL_MAC        0x1
  62
  63 #define JOURNAL_ENTRY_ROUNDUP           8
  64
  65 typedef __u64 commit_id_t;
  66 #define JOURNAL_MAC_PER_SECTOR          8
  67
  68 struct journal_entry {
  69         union {
  70                 struct {
  71                         __u32 sector_lo;
  72                         __u32 sector_hi;
  73                 } s;
  74                 __u64 sector;
  75         } u;
  76         commit_id_t last_bytes[0];
  77         /* __u8 tag[0]; */
  78 };
  79
  80 #define journal_entry_tag(ic, je)               ((__u8 *)&(je)->last_bytes[(ic)->sectors_per_block])
  81
  82 #if BITS_PER_LONG == 64
  83 #define journal_entry_set_sector(je, x)         do { smp_wmb(); ACCESS_ONCE((je)->u.sector) = cpu_to_le64(x); } while (0)
  84 #define journal_entry_get_sector(je)            le64_to_cpu((je)->u.sector)
  85 #elif defined(CONFIG_LBDAF)
  86 #define journal_entry_set_sector(je, x)         do { (je)->u.s.sector_lo = cpu_to_le32(x); smp_wmb(); ACCESS_ONCE((je)->u.s.sector_hi) = cpu_to_le32((x) >> 32); } while (0)
  87 #define journal_entry_get_sector(je)            le64_to_cpu((je)->u.sector)
  88 #else
  89 #define journal_entry_set_sector(je, x)         do { (je)->u.s.sector_lo = cpu_to_le32(x); smp_wmb(); ACCESS_ONCE((je)->u.s.sector_hi) = cpu_to_le32(0); } while (0)
  90 #define journal_entry_get_sector(je)            le32_to_cpu((je)->u.s.sector_lo)
  91 #endif
  92 #define journal_entry_is_unused(je)             ((je)->u.s.sector_hi == cpu_to_le32(-1))
  93 #define journal_entry_set_unused(je)            do { ((je)->u.s.sector_hi = cpu_to_le32(-1)); } while (0)
  94 #define journal_entry_is_inprogress(je)         ((je)->u.s.sector_hi == cpu_to_le32(-2))
  95 #define journal_entry_set_inprogress(je)        do { ((je)->u.s.sector_hi = cpu_to_le32(-2)); } while (0)
  96
  97 #define JOURNAL_BLOCK_SECTORS           8
  98 #define JOURNAL_SECTOR_DATA             ((1 << SECTOR_SHIFT) - sizeof(commit_id_t))
  99 #define JOURNAL_MAC_SIZE                (JOURNAL_MAC_PER_SECTOR * JOURNAL_BLOCK_SECTORS)
 100
 101 struct journal_sector {
 102         __u8 entries[JOURNAL_SECTOR_DATA - JOURNAL_MAC_PER_SECTOR];
 103         __u8 mac[JOURNAL_MAC_PER_SECTOR];
 104         commit_id_t commit_id;
 105 };
 106
 107 #define MAX_TAG_SIZE                    (JOURNAL_SECTOR_DATA - JOURNAL_MAC_PER_SECTOR - offsetof(struct journal_entry, last_bytes[MAX_SECTORS_PER_BLOCK]))
 108
 109 #define METADATA_PADDING_SECTORS        8
 110
 111 #define N_COMMIT_IDS                    4
 112
 113 static unsigned char prev_commit_seq(unsigned char seq)
 114 {
 115         return (seq + N_COMMIT_IDS - 1) % N_COMMIT_IDS;
 116 }
 117
 118 static unsigned char next_commit_seq(unsigned char seq)
 119 {
 120         return (seq + 1) % N_COMMIT_IDS;
 121 }
 122
 123 /*
 124  * In-memory structures
 125  */
 126
 127 struct journal_node {
 128         struct rb_node node;
 129         sector_t sector;
 130 };
 131
 132 struct alg_spec {
 133         char *alg_string;
 134         char *key_string;
 135         __u8 *key;
 136         unsigned key_size;
 137 };
 138
 139 struct dm_integrity_c {
 140         struct dm_dev *dev;
 141         unsigned tag_size;
 142         __s8 log2_tag_size;
 143         sector_t start;
 144         mempool_t *journal_io_mempool;
 145         struct dm_io_client *io;
 146         struct dm_bufio_client *bufio;
 147         struct workqueue_struct *metadata_wq;
 148         struct superblock *sb;
 149         unsigned journal_pages;
 150         struct page_list *journal;
 151         struct page_list *journal_io;
 152         struct page_list *journal_xor;
 153
 154         struct crypto_skcipher *journal_crypt;
 155         struct scatterlist **journal_scatterlist;
 156         struct scatterlist **journal_io_scatterlist;
 157         struct skcipher_request **sk_requests;
 158
 159         struct crypto_shash *journal_mac;
 160
 161         struct journal_node *journal_tree;
 162         struct rb_root journal_tree_root;
 163
 164         sector_t provided_data_sectors;
 165
 166         unsigned short journal_entry_size;
 167         unsigned char journal_entries_per_sector;
 168         unsigned char journal_section_entries;
 169         unsigned short journal_section_sectors;
 170         unsigned journal_sections;
 171         unsigned journal_entries;
 172         sector_t device_sectors;
 173         unsigned initial_sectors;
 174         unsigned metadata_run;
 175         __s8 log2_metadata_run;
 176         __u8 log2_buffer_sectors;
 177         __u8 sectors_per_block;
 178
 179         unsigned char mode;
 180         bool suspending;
 181
 182         int failed;
 183
 184         struct crypto_shash *internal_hash;
 185
 186         /* these variables are locked with endio_wait.lock */
 187         struct rb_root in_progress;
 188         wait_queue_head_t endio_wait;
 189         struct workqueue_struct *wait_wq;
 190
 191         unsigned char commit_seq;
 192         commit_id_t commit_ids[N_COMMIT_IDS];
 193
 194         unsigned committed_section;
 195         unsigned n_committed_sections;
 196
 197         unsigned uncommitted_section;
 198         unsigned n_uncommitted_sections;
 199
 200         unsigned free_section;
 201         unsigned char free_section_entry;
 202         unsigned free_sectors;
 203
 204         unsigned free_sectors_threshold;
 205
 206         struct workqueue_struct *commit_wq;
 207         struct work_struct commit_work;
 208
 209         struct workqueue_struct *writer_wq;
 210         struct work_struct writer_work;
 211
 212         struct bio_list flush_bio_list;
 213
 214         unsigned long autocommit_jiffies;
 215         struct timer_list autocommit_timer;
 216         unsigned autocommit_msec;
 217
 218         wait_queue_head_t copy_to_journal_wait;
 219
 220         struct completion crypto_backoff;
 221
 222         bool journal_uptodate;
 223         bool just_formatted;
 224
 225         struct alg_spec internal_hash_alg;
 226         struct alg_spec journal_crypt_alg;
 227         struct alg_spec journal_mac_alg;
 228
 229         atomic64_t number_of_mismatches;
 230 };
 231
 232 struct dm_integrity_range {
 233         sector_t logical_sector;
 234         unsigned n_sectors;
 235         struct rb_node node;
 236 };
 237
 238 struct dm_integrity_io {
 239         struct work_struct work;
 240
 241         struct dm_integrity_c *ic;
 242         bool write;
 243         bool fua;
 244
 245         struct dm_integrity_range range;
 246
 247         sector_t metadata_block;
 248         unsigned metadata_offset;
 249
 250         atomic_t in_flight;
 251         blk_status_t bi_status;
 252
 253         struct completion *completion;
 254
 255         struct gendisk *orig_bi_disk;
 256         u8 orig_bi_partno;
 257         bio_end_io_t *orig_bi_end_io;
 258         struct bio_integrity_payload *orig_bi_integrity;
 259         struct bvec_iter orig_bi_iter;
 260 };
 261
 262 struct journal_completion {
 263         struct dm_integrity_c *ic;
 264         atomic_t in_flight;
 265         struct completion comp;
 266 };
 267
 268 struct journal_io {
 269         struct dm_integrity_range range;
 270         struct journal_completion *comp;
 271 };
 272
 273 static struct kmem_cache *journal_io_cache;
 274
 275 #define JOURNAL_IO_MEMPOOL      32
 276
 277 #ifdef DEBUG_PRINT
 278 #define DEBUG_print(x, ...)     printk(KERN_DEBUG x, ##__VA_ARGS__)
 279 static void __DEBUG_bytes(__u8 *bytes, size_t len, const char *msg, ...)
 280 {
 281         va_list args;
 282         va_start(args, msg);
 283         vprintk(msg, args);
 284         va_end(args);
 285         if (len)
 286                 pr_cont(":");
 287         while (len) {
 288                 pr_cont(" %02x", *bytes);
 289                 bytes++;
 290                 len--;
 291         }
 292         pr_cont("\n");
 293 }
 294 #define DEBUG_bytes(bytes, len, msg, ...)       __DEBUG_bytes(bytes, len, KERN_DEBUG msg, ##__VA_ARGS__)
 295 #else
 296 #define DEBUG_print(x, ...)                     do { } while (0)
 297 #define DEBUG_bytes(bytes, len, msg, ...)       do { } while (0)
 298 #endif
 299
 300 /*
 301  * DM Integrity profile, protection is performed layer above (dm-crypt)
 302  */
 303 static const struct blk_integrity_profile dm_integrity_profile = {
 304         .name                   = "DM-DIF-EXT-TAG",
 305         .generate_fn            = NULL,
 306         .verify_fn              = NULL,
 307 };
 308
 309 static void dm_integrity_map_continue(struct dm_integrity_io *dio, bool from_map);
 310 static void integrity_bio_wait(struct work_struct *w);
 311 static void dm_integrity_dtr(struct dm_target *ti);
 312
 313 static void dm_integrity_io_error(struct dm_integrity_c *ic, const char *msg, int err)
 314 {
 315         if (err == -EILSEQ)
 316                 atomic64_inc(&ic->number_of_mismatches);
 317         if (!cmpxchg(&ic->failed, 0, err))
 318                 DMERR("Error on %s: %d", msg, err);
 319 }
 320
 321 static int dm_integrity_failed(struct dm_integrity_c *ic)
 322 {
 323         return ACCESS_ONCE(ic->failed);
 324 }
 325
 326 static commit_id_t dm_integrity_commit_id(struct dm_integrity_c *ic, unsigned i,
 327                                           unsigned j, unsigned char seq)
 328 {
 329         /*
 330          * Xor the number with section and sector, so that if a piece of
 331          * journal is written at wrong place, it is detected.
 332          */
 333         return ic->commit_ids[seq] ^ cpu_to_le64(((__u64)i << 32) ^ j);
 334 }
 335
 336 static void get_area_and_offset(struct dm_integrity_c *ic, sector_t data_sector,
 337                                 sector_t *area, sector_t *offset)
 338 {
 339         __u8 log2_interleave_sectors = ic->sb->log2_interleave_sectors;
 340
 341         *area = data_sector >> log2_interleave_sectors;
 342         *offset = (unsigned)data_sector & ((1U << log2_interleave_sectors) - 1);
 343 }
 344
 345 #define sector_to_block(ic, n)                                          \
 346 do {                                                                    \
 347         BUG_ON((n) & (unsigned)((ic)->sectors_per_block - 1));          \
 348         (n) >>= (ic)->sb->log2_sectors_per_block;                       \
 349 } while (0)
 350
 351 static __u64 get_metadata_sector_and_offset(struct dm_integrity_c *ic, sector_t area,
 352                                             sector_t offset, unsigned *metadata_offset)
 353 {
 354         __u64 ms;
 355         unsigned mo;
 356
 357         ms = area << ic->sb->log2_interleave_sectors;
 358         if (likely(ic->log2_metadata_run >= 0))
 359                 ms += area << ic->log2_metadata_run;
 360         else
 361                 ms += area * ic->metadata_run;
 362         ms >>= ic->log2_buffer_sectors;
 363
 364         sector_to_block(ic, offset);
 365
 366         if (likely(ic->log2_tag_size >= 0)) {
 367                 ms += offset >> (SECTOR_SHIFT + ic->log2_buffer_sectors - ic->log2_tag_size);
 368                 mo = (offset << ic->log2_tag_size) & ((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - 1);
 369         } else {
 370                 ms += (__u64)offset * ic->tag_size >> (SECTOR_SHIFT + ic->log2_buffer_sectors);
 371                 mo = (offset * ic->tag_size) & ((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - 1);
 372         }
 373         *metadata_offset = mo;
 374         return ms;
 375 }
 376
 377 static sector_t get_data_sector(struct dm_integrity_c *ic, sector_t area, sector_t offset)
 378 {
 379         sector_t result;
 380
 381         result = area << ic->sb->log2_interleave_sectors;
 382         if (likely(ic->log2_metadata_run >= 0))
 383                 result += (area + 1) << ic->log2_metadata_run;
 384         else
 385                 result += (area + 1) * ic->metadata_run;
 386
 387         result += (sector_t)ic->initial_sectors + offset;
 388         return result;
 389 }
 390
 391 static void wraparound_section(struct dm_integrity_c *ic, unsigned *sec_ptr)
 392 {
 393         if (unlikely(*sec_ptr >= ic->journal_sections))
 394                 *sec_ptr -= ic->journal_sections;
 395 }
 396
 397 static int sync_rw_sb(struct dm_integrity_c *ic, int op, int op_flags)
 398 {
 399         struct dm_io_request io_req;
 400         struct dm_io_region io_loc;
 401
 402         io_req.bi_op = op;
 403         io_req.bi_op_flags = op_flags;
 404         io_req.mem.type = DM_IO_KMEM;
 405         io_req.mem.ptr.addr = ic->sb;
 406         io_req.notify.fn = NULL;
 407         io_req.client = ic->io;
 408         io_loc.bdev = ic->dev->bdev;
 409         io_loc.sector = ic->start;
 410         io_loc.count = SB_SECTORS;
 411
 412         return dm_io(&io_req, 1, &io_loc, NULL);
 413 }
 414
 415 static void access_journal_check(struct dm_integrity_c *ic, unsigned section, unsigned offset,
 416                                  bool e, const char *function)
 417 {
 418 #if defined(CONFIG_DM_DEBUG) || defined(INTERNAL_VERIFY)
 419         unsigned limit = e ? ic->journal_section_entries : ic->journal_section_sectors;
 420
 421         if (unlikely(section >= ic->journal_sections) ||
 422             unlikely(offset >= limit)) {
 423                 printk(KERN_CRIT "%s: invalid access at (%u,%u), limit (%u,%u)\n",
 424                         function, section, offset, ic->journal_sections, limit);
 425                 BUG();
 426         }
 427 #endif
 428 }
 429
 430 static void page_list_location(struct dm_integrity_c *ic, unsigned section, unsigned offset,
 431                                unsigned *pl_index, unsigned *pl_offset)
 432 {
 433         unsigned sector;
 434
 435         access_journal_check(ic, section, offset, false, "page_list_location");
 436
 437         sector = section * ic->journal_section_sectors + offset;
 438
 439         *pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
 440         *pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
 441 }
 442
 443 static struct journal_sector *access_page_list(struct dm_integrity_c *ic, struct page_list *pl,
 444                                                unsigned section, unsigned offset, unsigned *n_sectors)
 445 {
 446         unsigned pl_index, pl_offset;
 447         char *va;
 448
 449         page_list_location(ic, section, offset, &pl_index, &pl_offset);
 450
 451         if (n_sectors)
 452                 *n_sectors = (PAGE_SIZE - pl_offset) >> SECTOR_SHIFT;
 453
 454         va = lowmem_page_address(pl[pl_index].page);
 455
 456         return (struct journal_sector *)(va + pl_offset);
 457 }
 458
 459 static struct journal_sector *access_journal(struct dm_integrity_c *ic, unsigned section, unsigned offset)
 460 {
 461         return access_page_list(ic, ic->journal, section, offset, NULL);
 462 }
 463
 464 static struct journal_entry *access_journal_entry(struct dm_integrity_c *ic, unsigned section, unsigned n)
 465 {
 466         unsigned rel_sector, offset;
 467         struct journal_sector *js;
 468
 469         access_journal_check(ic, section, n, true, "access_journal_entry");
 470
 471         rel_sector = n % JOURNAL_BLOCK_SECTORS;
 472         offset = n / JOURNAL_BLOCK_SECTORS;
 473
 474         js = access_journal(ic, section, rel_sector);
 475         return (struct journal_entry *)((char *)js + offset * ic->journal_entry_size);
 476 }
 477
 478 static struct journal_sector *access_journal_data(struct dm_integrity_c *ic, unsigned section, unsigned n)
 479 {
 480         n <<= ic->sb->log2_sectors_per_block;
 481
 482         n += JOURNAL_BLOCK_SECTORS;
 483
 484         access_journal_check(ic, section, n, false, "access_journal_data");
 485
 486         return access_journal(ic, section, n);
 487 }
 488
 489 static void section_mac(struct dm_integrity_c *ic, unsigned section, __u8 result[JOURNAL_MAC_SIZE])
 490 {
 491         SHASH_DESC_ON_STACK(desc, ic->journal_mac);
 492         int r;
 493         unsigned j, size;
 494
 495         desc->tfm = ic->journal_mac;
 496         desc->flags = CRYPTO_TFM_REQ_MAY_SLEEP;
 497
 498         r = crypto_shash_init(desc);
 499         if (unlikely(r)) {
 500                 dm_integrity_io_error(ic, "crypto_shash_init", r);
 501                 goto err;
 502         }
 503
 504         for (j = 0; j < ic->journal_section_entries; j++) {
 505                 struct journal_entry *je = access_journal_entry(ic, section, j);
 506                 r = crypto_shash_update(desc, (__u8 *)&je->u.sector, sizeof je->u.sector);
 507                 if (unlikely(r)) {
 508                         dm_integrity_io_error(ic, "crypto_shash_update", r);
 509                         goto err;
 510                 }
 511         }
 512
 513         size = crypto_shash_digestsize(ic->journal_mac);
 514
 515         if (likely(size <= JOURNAL_MAC_SIZE)) {
 516                 r = crypto_shash_final(desc, result);
 517                 if (unlikely(r)) {
 518                         dm_integrity_io_error(ic, "crypto_shash_final", r);
 519                         goto err;
 520                 }
 521                 memset(result + size, 0, JOURNAL_MAC_SIZE - size);
 522         } else {
 523                 __u8 digest[size];
 524                 r = crypto_shash_final(desc, digest);
 525                 if (unlikely(r)) {
 526                         dm_integrity_io_error(ic, "crypto_shash_final", r);
 527                         goto err;
 528                 }
 529                 memcpy(result, digest, JOURNAL_MAC_SIZE);
 530         }
 531
 532         return;
 533 err:
 534         memset(result, 0, JOURNAL_MAC_SIZE);
 535 }
 536
 537 static void rw_section_mac(struct dm_integrity_c *ic, unsigned section, bool wr)
 538 {
 539         __u8 result[JOURNAL_MAC_SIZE];
 540         unsigned j;
 541
 542         if (!ic->journal_mac)
 543                 return;
 544
 545         section_mac(ic, section, result);
 546
 547         for (j = 0; j < JOURNAL_BLOCK_SECTORS; j++) {
 548                 struct journal_sector *js = access_journal(ic, section, j);
 549
 550                 if (likely(wr))
 551                         memcpy(&js->mac, result + (j * JOURNAL_MAC_PER_SECTOR), JOURNAL_MAC_PER_SECTOR);
 552                 else {
 553                         if (memcmp(&js->mac, result + (j * JOURNAL_MAC_PER_SECTOR), JOURNAL_MAC_PER_SECTOR))
 554                                 dm_integrity_io_error(ic, "journal mac", -EILSEQ);
 555                 }
 556         }
 557 }
 558
 559 static void complete_journal_op(void *context)
 560 {
 561         struct journal_completion *comp = context;
 562         BUG_ON(!atomic_read(&comp->in_flight));
 563         if (likely(atomic_dec_and_test(&comp->in_flight)))
 564                 complete(&comp->comp);
 565 }
 566
 567 static void xor_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
 568                         unsigned n_sections, struct journal_completion *comp)
 569 {
 570         struct async_submit_ctl submit;
 571         size_t n_bytes = (size_t)(n_sections * ic->journal_section_sectors) << SECTOR_SHIFT;
 572         unsigned pl_index, pl_offset, section_index;
 573         struct page_list *source_pl, *target_pl;
 574
 575         if (likely(encrypt)) {
 576                 source_pl = ic->journal;
 577                 target_pl = ic->journal_io;
 578         } else {
 579                 source_pl = ic->journal_io;
 580                 target_pl = ic->journal;
 581         }
 582
 583         page_list_location(ic, section, 0, &pl_index, &pl_offset);
 584
 585         atomic_add(roundup(pl_offset + n_bytes, PAGE_SIZE) >> PAGE_SHIFT, &comp->in_flight);
 586
 587         init_async_submit(&submit, ASYNC_TX_XOR_ZERO_DST, NULL, complete_journal_op, comp, NULL);
 588
 589         section_index = pl_index;
 590
 591         do {
 592                 size_t this_step;
 593                 struct page *src_pages[2];
 594                 struct page *dst_page;
 595
 596                 while (unlikely(pl_index == section_index)) {
 597                         unsigned dummy;
 598                         if (likely(encrypt))
 599                                 rw_section_mac(ic, section, true);
 600                         section++;
 601                         n_sections--;
 602                         if (!n_sections)
 603                                 break;
 604                         page_list_location(ic, section, 0, &section_index, &dummy);
 605                 }
 606
 607                 this_step = min(n_bytes, (size_t)PAGE_SIZE - pl_offset);
 608                 dst_page = target_pl[pl_index].page;
 609                 src_pages[0] = source_pl[pl_index].page;
 610                 src_pages[1] = ic->journal_xor[pl_index].page;
 611
 612                 async_xor(dst_page, src_pages, pl_offset, 2, this_step, &submit);
 613
 614                 pl_index++;
 615                 pl_offset = 0;
 616                 n_bytes -= this_step;
 617         } while (n_bytes);
 618
 619         BUG_ON(n_sections);
 620
 621         async_tx_issue_pending_all();
 622 }
 623
 624 static void complete_journal_encrypt(struct crypto_async_request *req, int err)
 625 {
 626         struct journal_completion *comp = req->data;
 627         if (unlikely(err)) {
 628                 if (likely(err == -EINPROGRESS)) {
 629                         complete(&comp->ic->crypto_backoff);
 630                         return;
 631                 }
 632                 dm_integrity_io_error(comp->ic, "asynchronous encrypt", err);
 633         }
 634         complete_journal_op(comp);
 635 }
 636
 637 static bool do_crypt(bool encrypt, struct skcipher_request *req, struct journal_completion *comp)
 638 {
 639         int r;
 640         skcipher_request_set_callback(req, CRYPTO_TFM_REQ_MAY_BACKLOG | CRYPTO_TFM_REQ_MAY_SLEEP,
 641                                       complete_journal_encrypt, comp);
 642         if (likely(encrypt))
 643                 r = crypto_skcipher_encrypt(req);
 644         else
 645                 r = crypto_skcipher_decrypt(req);
 646         if (likely(!r))
 647                 return false;
 648         if (likely(r == -EINPROGRESS))
 649                 return true;
 650         if (likely(r == -EBUSY)) {
 651                 wait_for_completion(&comp->ic->crypto_backoff);
 652                 reinit_completion(&comp->ic->crypto_backoff);
 653                 return true;
 654         }
 655         dm_integrity_io_error(comp->ic, "encrypt", r);
 656         return false;
 657 }
 658
 659 static void crypt_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
 660                           unsigned n_sections, struct journal_completion *comp)
 661 {
 662         struct scatterlist **source_sg;
 663         struct scatterlist **target_sg;
 664
 665         atomic_add(2, &comp->in_flight);
 666
 667         if (likely(encrypt)) {
 668                 source_sg = ic->journal_scatterlist;
 669                 target_sg = ic->journal_io_scatterlist;
 670         } else {
 671                 source_sg = ic->journal_io_scatterlist;
 672                 target_sg = ic->journal_scatterlist;
 673         }
 674
 675         do {
 676                 struct skcipher_request *req;
 677                 unsigned ivsize;
 678                 char *iv;
 679
 680                 if (likely(encrypt))
 681                         rw_section_mac(ic, section, true);
 682
 683                 req = ic->sk_requests[section];
 684                 ivsize = crypto_skcipher_ivsize(ic->journal_crypt);
 685                 iv = req->iv;
 686
 687                 memcpy(iv, iv + ivsize, ivsize);
 688
 689                 req->src = source_sg[section];
 690                 req->dst = target_sg[section];
 691
 692                 if (unlikely(do_crypt(encrypt, req, comp)))
 693                         atomic_inc(&comp->in_flight);
 694
 695                 section++;
 696                 n_sections--;
 697         } while (n_sections);
 698
 699         atomic_dec(&comp->in_flight);
 700         complete_journal_op(comp);
 701 }
 702
 703 static void encrypt_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
 704                             unsigned n_sections, struct journal_completion *comp)
 705 {
 706         if (ic->journal_xor)
 707                 return xor_journal(ic, encrypt, section, n_sections, comp);
 708         else
 709                 return crypt_journal(ic, encrypt, section, n_sections, comp);
 710 }
 711
 712 static void complete_journal_io(unsigned long error, void *context)
 713 {
 714         struct journal_completion *comp = context;
 715         if (unlikely(error != 0))
 716                 dm_integrity_io_error(comp->ic, "writing journal", -EIO);
 717         complete_journal_op(comp);
 718 }
 719
 720 static void rw_journal(struct dm_integrity_c *ic, int op, int op_flags, unsigned section,
 721                        unsigned n_sections, struct journal_completion *comp)
 722 {
 723         struct dm_io_request io_req;
 724         struct dm_io_region io_loc;
 725         unsigned sector, n_sectors, pl_index, pl_offset;
 726         int r;
 727
 728         if (unlikely(dm_integrity_failed(ic))) {
 729                 if (comp)
 730                         complete_journal_io(-1UL, comp);
 731                 return;
 732         }
 733
 734         sector = section * ic->journal_section_sectors;
 735         n_sectors = n_sections * ic->journal_section_sectors;
 736
 737         pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
 738         pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
 739
 740         io_req.bi_op = op;
 741         io_req.bi_op_flags = op_flags;
 742         io_req.mem.type = DM_IO_PAGE_LIST;
 743         if (ic->journal_io)
 744                 io_req.mem.ptr.pl = &ic->journal_io[pl_index];
 745         else
 746                 io_req.mem.ptr.pl = &ic->journal[pl_index];
 747         io_req.mem.offset = pl_offset;
 748         if (likely(comp != NULL)) {
 749                 io_req.notify.fn = complete_journal_io;
 750                 io_req.notify.context = comp;
 751         } else {
 752                 io_req.notify.fn = NULL;
 753         }
 754         io_req.client = ic->io;
 755         io_loc.bdev = ic->dev->bdev;
 756         io_loc.sector = ic->start + SB_SECTORS + sector;
 757         io_loc.count = n_sectors;
 758
 759         r = dm_io(&io_req, 1, &io_loc, NULL);
 760         if (unlikely(r)) {
 761                 dm_integrity_io_error(ic, op == REQ_OP_READ ? "reading journal" : "writing journal", r);
 762                 if (comp) {
 763                         WARN_ONCE(1, "asynchronous dm_io failed: %d", r);
 764                         complete_journal_io(-1UL, comp);
 765                 }
 766         }
 767 }
 768
 769 static void write_journal(struct dm_integrity_c *ic, unsigned commit_start, unsigned commit_sections)
 770 {
 771         struct journal_completion io_comp;
 772         struct journal_completion crypt_comp_1;
 773         struct journal_completion crypt_comp_2;
 774         unsigned i;
 775
 776         io_comp.ic = ic;
 777         init_completion(&io_comp.comp);
 778
 779         if (commit_start + commit_sections <= ic->journal_sections) {
 780                 io_comp.in_flight = (atomic_t)ATOMIC_INIT(1);
 781                 if (ic->journal_io) {
 782                         crypt_comp_1.ic = ic;
 783                         init_completion(&crypt_comp_1.comp);
 784                         crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
 785                         encrypt_journal(ic, true, commit_start, commit_sections, &crypt_comp_1);
 786                         wait_for_completion_io(&crypt_comp_1.comp);
 787                 } else {
 788                         for (i = 0; i < commit_sections; i++)
 789                                 rw_section_mac(ic, commit_start + i, true);
 790                 }
 791                 rw_journal(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, commit_start,
 792                            commit_sections, &io_comp);
 793         } else {
 794                 unsigned to_end;
 795                 io_comp.in_flight = (atomic_t)ATOMIC_INIT(2);
 796                 to_end = ic->journal_sections - commit_start;
 797                 if (ic->journal_io) {
 798                         crypt_comp_1.ic = ic;
 799                         init_completion(&crypt_comp_1.comp);
 800                         crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
 801                         encrypt_journal(ic, true, commit_start, to_end, &crypt_comp_1);
 802                         if (try_wait_for_completion(&crypt_comp_1.comp)) {
 803                                 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
 804                                 reinit_completion(&crypt_comp_1.comp);
 805                                 crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
 806                                 encrypt_journal(ic, true, 0, commit_sections - to_end, &crypt_comp_1);
 807                                 wait_for_completion_io(&crypt_comp_1.comp);
 808                         } else {
 809                                 crypt_comp_2.ic = ic;
 810                                 init_completion(&crypt_comp_2.comp);
 811                                 crypt_comp_2.in_flight = (atomic_t)ATOMIC_INIT(0);
 812                                 encrypt_journal(ic, true, 0, commit_sections - to_end, &crypt_comp_2);
 813                                 wait_for_completion_io(&crypt_comp_1.comp);
 814                                 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
 815                                 wait_for_completion_io(&crypt_comp_2.comp);
 816                         }
 817                 } else {
 818                         for (i = 0; i < to_end; i++)
 819                                 rw_section_mac(ic, commit_start + i, true);
 820                         rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
 821                         for (i = 0; i < commit_sections - to_end; i++)
 822                                 rw_section_mac(ic, i, true);
 823                 }
 824                 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, 0, commit_sections - to_end, &io_comp);
 825         }
 826
 827         wait_for_completion_io(&io_comp.comp);
 828 }
 829
 830 static void copy_from_journal(struct dm_integrity_c *ic, unsigned section, unsigned offset,
 831                               unsigned n_sectors, sector_t target, io_notify_fn fn, void *data)
 832 {
 833         struct dm_io_request io_req;
 834         struct dm_io_region io_loc;
 835         int r;
 836         unsigned sector, pl_index, pl_offset;
 837
 838         BUG_ON((target | n_sectors | offset) & (unsigned)(ic->sectors_per_block - 1));
 839
 840         if (unlikely(dm_integrity_failed(ic))) {
 841                 fn(-1UL, data);
 842                 return;
 843         }
 844
 845         sector = section * ic->journal_section_sectors + JOURNAL_BLOCK_SECTORS + offset;
 846
 847         pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
 848         pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
 849
 850         io_req.bi_op = REQ_OP_WRITE;
 851         io_req.bi_op_flags = 0;
 852         io_req.mem.type = DM_IO_PAGE_LIST;
 853         io_req.mem.ptr.pl = &ic->journal[pl_index];
 854         io_req.mem.offset = pl_offset;
 855         io_req.notify.fn = fn;
 856         io_req.notify.context = data;
 857         io_req.client = ic->io;
 858         io_loc.bdev = ic->dev->bdev;
 859         io_loc.sector = ic->start + target;
 860         io_loc.count = n_sectors;
 861
 862         r = dm_io(&io_req, 1, &io_loc, NULL);
 863         if (unlikely(r)) {
 864                 WARN_ONCE(1, "asynchronous dm_io failed: %d", r);
 865                 fn(-1UL, data);
 866         }
 867 }
 868
 869 static bool add_new_range(struct dm_integrity_c *ic, struct dm_integrity_range *new_range)
 870 {
 871         struct rb_node **n = &ic->in_progress.rb_node;
 872         struct rb_node *parent;
 873
 874         BUG_ON((new_range->logical_sector | new_range->n_sectors) & (unsigned)(ic->sectors_per_block - 1));
 875
 876         parent = NULL;
 877
 878         while (*n) {
 879                 struct dm_integrity_range *range = container_of(*n, struct dm_integrity_range, node);
 880
 881                 parent = *n;
 882                 if (new_range->logical_sector + new_range->n_sectors <= range->logical_sector) {
 883                         n = &range->node.rb_left;
 884                 } else if (new_range->logical_sector >= range->logical_sector + range->n_sectors) {
 885                         n = &range->node.rb_right;
 886                 } else {
 887                         return false;
 888                 }
 889         }
 890
 891         rb_link_node(&new_range->node, parent, n);
 892         rb_insert_color(&new_range->node, &ic->in_progress);
 893
 894         return true;
 895 }
 896
 897 static void remove_range_unlocked(struct dm_integrity_c *ic, struct dm_integrity_range *range)
 898 {
 899         rb_erase(&range->node, &ic->in_progress);
 900         wake_up_locked(&ic->endio_wait);
 901 }
 902
 903 static void remove_range(struct dm_integrity_c *ic, struct dm_integrity_range *range)
 904 {
 905         unsigned long flags;
 906
 907         spin_lock_irqsave(&ic->endio_wait.lock, flags);
 908         remove_range_unlocked(ic, range);
 909         spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
 910 }
 911
 912 static void init_journal_node(struct journal_node *node)
 913 {
 914         RB_CLEAR_NODE(&node->node);
 915         node->sector = (sector_t)-1;
 916 }
 917
 918 static void add_journal_node(struct dm_integrity_c *ic, struct journal_node *node, sector_t sector)
 919 {
 920         struct rb_node **link;
 921         struct rb_node *parent;
 922
 923         node->sector = sector;
 924         BUG_ON(!RB_EMPTY_NODE(&node->node));
 925
 926         link = &ic->journal_tree_root.rb_node;
 927         parent = NULL;
 928
 929         while (*link) {
 930                 struct journal_node *j;
 931                 parent = *link;
 932                 j = container_of(parent, struct journal_node, node);
 933                 if (sector < j->sector)
 934                         link = &j->node.rb_left;
 935                 else
 936                         link = &j->node.rb_right;
 937         }
 938
 939         rb_link_node(&node->node, parent, link);
 940         rb_insert_color(&node->node, &ic->journal_tree_root);
 941 }
 942
 943 static void remove_journal_node(struct dm_integrity_c *ic, struct journal_node *node)
 944 {
 945         BUG_ON(RB_EMPTY_NODE(&node->node));
 946         rb_erase(&node->node, &ic->journal_tree_root);
 947         init_journal_node(node);
 948 }
 949
 950 #define NOT_FOUND       (-1U)
 951
 952 static unsigned find_journal_node(struct dm_integrity_c *ic, sector_t sector, sector_t *next_sector)
 953 {
 954         struct rb_node *n = ic->journal_tree_root.rb_node;
 955         unsigned found = NOT_FOUND;
 956         *next_sector = (sector_t)-1;
 957         while (n) {
 958                 struct journal_node *j = container_of(n, struct journal_node, node);
 959                 if (sector == j->sector) {
 960                         found = j - ic->journal_tree;
 961                 }
 962                 if (sector < j->sector) {
 963                         *next_sector = j->sector;
 964                         n = j->node.rb_left;
 965                 } else {
 966                         n = j->node.rb_right;
 967                 }
 968         }
 969
 970         return found;
 971 }
 972
 973 static bool test_journal_node(struct dm_integrity_c *ic, unsigned pos, sector_t sector)
 974 {
 975         struct journal_node *node, *next_node;
 976         struct rb_node *next;
 977
 978         if (unlikely(pos >= ic->journal_entries))
 979                 return false;
 980         node = &ic->journal_tree[pos];
 981         if (unlikely(RB_EMPTY_NODE(&node->node)))
 982                 return false;
 983         if (unlikely(node->sector != sector))
 984                 return false;
 985
 986         next = rb_next(&node->node);
 987         if (unlikely(!next))
 988                 return true;
 989
 990         next_node = container_of(next, struct journal_node, node);
 991         return next_node->sector != sector;
 992 }
 993
 994 static bool find_newer_committed_node(struct dm_integrity_c *ic, struct journal_node *node)
 995 {
 996         struct rb_node *next;
 997         struct journal_node *next_node;
 998         unsigned next_section;
 999
1000         BUG_ON(RB_EMPTY_NODE(&node->node));
1001
1002         next = rb_next(&node->node);
1003         if (unlikely(!next))
1004                 return false;
1005
1006         next_node = container_of(next, struct journal_node, node);
1007
1008         if (next_node->sector != node->sector)
1009                 return false;
1010
1011         next_section = (unsigned)(next_node - ic->journal_tree) / ic->journal_section_entries;
1012         if (next_section >= ic->committed_section &&
1013             next_section < ic->committed_section + ic->n_committed_sections)
1014                 return true;
1015         if (next_section + ic->journal_sections < ic->committed_section + ic->n_committed_sections)
1016                 return true;
1017
1018         return false;
1019 }
1020
1021 #define TAG_READ        0
1022 #define TAG_WRITE       1
1023 #define TAG_CMP         2
1024
1025 static int dm_integrity_rw_tag(struct dm_integrity_c *ic, unsigned char *tag, sector_t *metadata_block,
1026                                unsigned *metadata_offset, unsigned total_size, int op)
1027 {
1028         do {
1029                 unsigned char *data, *dp;
1030                 struct dm_buffer *b;
1031                 unsigned to_copy;
1032                 int r;
1033
1034                 r = dm_integrity_failed(ic);
1035                 if (unlikely(r))
1036                         return r;
1037
1038                 data = dm_bufio_read(ic->bufio, *metadata_block, &b);
1039                 if (unlikely(IS_ERR(data)))
1040                         return PTR_ERR(data);
1041
1042                 to_copy = min((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - *metadata_offset, total_size);
1043                 dp = data + *metadata_offset;
1044                 if (op == TAG_READ) {
1045                         memcpy(tag, dp, to_copy);
1046                 } else if (op == TAG_WRITE) {
1047                         memcpy(dp, tag, to_copy);
1048                         dm_bufio_mark_partial_buffer_dirty(b, *metadata_offset, *metadata_offset + to_copy);
1049                 } else  {
1050                         /* e.g.: op == TAG_CMP */
1051                         if (unlikely(memcmp(dp, tag, to_copy))) {
1052                                 unsigned i;
1053
1054                                 for (i = 0; i < to_copy; i++) {
1055                                         if (dp[i] != tag[i])
1056                                                 break;
1057                                         total_size--;
1058                                 }
1059                                 dm_bufio_release(b);
1060                                 return total_size;
1061                         }
1062                 }
1063                 dm_bufio_release(b);
1064
1065                 tag += to_copy;
1066                 *metadata_offset += to_copy;
1067                 if (unlikely(*metadata_offset == 1U << SECTOR_SHIFT << ic->log2_buffer_sectors)) {
1068                         (*metadata_block)++;
1069                         *metadata_offset = 0;
1070                 }
1071                 total_size -= to_copy;
1072         } while (unlikely(total_size));
1073
1074         return 0;
1075 }
1076
1077 static void dm_integrity_flush_buffers(struct dm_integrity_c *ic)
1078 {
1079         int r;
1080         r = dm_bufio_write_dirty_buffers(ic->bufio);
1081         if (unlikely(r))
1082                 dm_integrity_io_error(ic, "writing tags", r);
1083 }
1084
1085 static void sleep_on_endio_wait(struct dm_integrity_c *ic)
1086 {
1087         DECLARE_WAITQUEUE(wait, current);
1088         __add_wait_queue(&ic->endio_wait, &wait);
1089         __set_current_state(TASK_UNINTERRUPTIBLE);
1090         spin_unlock_irq(&ic->endio_wait.lock);
1091         io_schedule();
1092         spin_lock_irq(&ic->endio_wait.lock);
1093         __remove_wait_queue(&ic->endio_wait, &wait);
1094 }
1095
1096 static void autocommit_fn(unsigned long data)
1097 {
1098         struct dm_integrity_c *ic = (struct dm_integrity_c *)data;
1099
1100         if (likely(!dm_integrity_failed(ic)))
1101                 queue_work(ic->commit_wq, &ic->commit_work);
1102 }
1103
1104 static void schedule_autocommit(struct dm_integrity_c *ic)
1105 {
1106         if (!timer_pending(&ic->autocommit_timer))
1107                 mod_timer(&ic->autocommit_timer, jiffies + ic->autocommit_jiffies);
1108 }
1109
1110 static void submit_flush_bio(struct dm_integrity_c *ic, struct dm_integrity_io *dio)
1111 {
1112         struct bio *bio;
1113         unsigned long flags;
1114
1115         spin_lock_irqsave(&ic->endio_wait.lock, flags);
1116         bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1117         bio_list_add(&ic->flush_bio_list, bio);
1118         spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
1119
1120         queue_work(ic->commit_wq, &ic->commit_work);
1121 }
1122
1123 static void do_endio(struct dm_integrity_c *ic, struct bio *bio)
1124 {
1125         int r = dm_integrity_failed(ic);
1126         if (unlikely(r) && !bio->bi_status)
1127                 bio->bi_status = errno_to_blk_status(r);
1128         bio_endio(bio);
1129 }
1130
1131 static void do_endio_flush(struct dm_integrity_c *ic, struct dm_integrity_io *dio)
1132 {
1133         struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1134
1135         if (unlikely(dio->fua) && likely(!bio->bi_status) && likely(!dm_integrity_failed(ic)))
1136                 submit_flush_bio(ic, dio);
1137         else
1138                 do_endio(ic, bio);
1139 }
1140
1141 static void dec_in_flight(struct dm_integrity_io *dio)
1142 {
1143         if (atomic_dec_and_test(&dio->in_flight)) {
1144                 struct dm_integrity_c *ic = dio->ic;
1145                 struct bio *bio;
1146
1147                 remove_range(ic, &dio->range);
1148
1149                 if (unlikely(dio->write))
1150                         schedule_autocommit(ic);
1151
1152                 bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1153
1154                 if (unlikely(dio->bi_status) && !bio->bi_status)
1155                         bio->bi_status = dio->bi_status;
1156                 if (likely(!bio->bi_status) && unlikely(bio_sectors(bio) != dio->range.n_sectors)) {
1157                         dio->range.logical_sector += dio->range.n_sectors;
1158                         bio_advance(bio, dio->range.n_sectors << SECTOR_SHIFT);
1159                         INIT_WORK(&dio->work, integrity_bio_wait);
1160                         queue_work(ic->wait_wq, &dio->work);
1161                         return;
1162                 }
1163                 do_endio_flush(ic, dio);
1164         }
1165 }
1166
1167 static void integrity_end_io(struct bio *bio)
1168 {
1169         struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
1170
1171         bio->bi_iter = dio->orig_bi_iter;
1172         bio->bi_disk = dio->orig_bi_disk;
1173         bio->bi_partno = dio->orig_bi_partno;
1174         if (dio->orig_bi_integrity) {
1175                 bio->bi_integrity = dio->orig_bi_integrity;
1176                 bio->bi_opf |= REQ_INTEGRITY;
1177         }
1178         bio->bi_end_io = dio->orig_bi_end_io;
1179
1180         if (dio->completion)
1181                 complete(dio->completion);
1182
1183         dec_in_flight(dio);
1184 }
1185
1186 static void integrity_sector_checksum(struct dm_integrity_c *ic, sector_t sector,
1187                                       const char *data, char *result)
1188 {
1189         __u64 sector_le = cpu_to_le64(sector);
1190         SHASH_DESC_ON_STACK(req, ic->internal_hash);
1191         int r;
1192         unsigned digest_size;
1193
1194         req->tfm = ic->internal_hash;
1195         req->flags = 0;
1196
1197         r = crypto_shash_init(req);
1198         if (unlikely(r < 0)) {
1199                 dm_integrity_io_error(ic, "crypto_shash_init", r);
1200                 goto failed;
1201         }
1202
1203         r = crypto_shash_update(req, (const __u8 *)&sector_le, sizeof sector_le);
1204         if (unlikely(r < 0)) {
1205                 dm_integrity_io_error(ic, "crypto_shash_update", r);
1206                 goto failed;
1207         }
1208
1209         r = crypto_shash_update(req, data, ic->sectors_per_block << SECTOR_SHIFT);
1210         if (unlikely(r < 0)) {
1211                 dm_integrity_io_error(ic, "crypto_shash_update", r);
1212                 goto failed;
1213         }
1214
1215         r = crypto_shash_final(req, result);
1216         if (unlikely(r < 0)) {
1217                 dm_integrity_io_error(ic, "crypto_shash_final", r);
1218                 goto failed;
1219         }
1220
1221         digest_size = crypto_shash_digestsize(ic->internal_hash);
1222         if (unlikely(digest_size < ic->tag_size))
1223                 memset(result + digest_size, 0, ic->tag_size - digest_size);
1224
1225         return;
1226
1227 failed:
1228         /* this shouldn't happen anyway, the hash functions have no reason to fail */
1229         get_random_bytes(result, ic->tag_size);
1230 }
1231
1232 static void integrity_metadata(struct work_struct *w)
1233 {
1234         struct dm_integrity_io *dio = container_of(w, struct dm_integrity_io, work);
1235         struct dm_integrity_c *ic = dio->ic;
1236
1237         int r;
1238
1239         if (ic->internal_hash) {
1240                 struct bvec_iter iter;
1241                 struct bio_vec bv;
1242                 unsigned digest_size = crypto_shash_digestsize(ic->internal_hash);
1243                 struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1244                 char *checksums;
1245                 unsigned extra_space = unlikely(digest_size > ic->tag_size) ? digest_size - ic->tag_size : 0;
1246                 char checksums_onstack[ic->tag_size + extra_space];
1247                 unsigned sectors_to_process = dio->range.n_sectors;
1248                 sector_t sector = dio->range.logical_sector;
1249
1250                 if (unlikely(ic->mode == 'R'))
1251                         goto skip_io;
1252
1253                 checksums = kmalloc((PAGE_SIZE >> SECTOR_SHIFT >> ic->sb->log2_sectors_per_block) * ic->tag_size + extra_space,
1254                                     GFP_NOIO | __GFP_NORETRY | __GFP_NOWARN);
1255                 if (!checksums)
1256                         checksums = checksums_onstack;
1257
1258                 __bio_for_each_segment(bv, bio, iter, dio->orig_bi_iter) {
1259                         unsigned pos;
1260                         char *mem, *checksums_ptr;
1261
1262 again:
1263                         mem = (char *)kmap_atomic(bv.bv_page) + bv.bv_offset;
1264                         pos = 0;
1265                         checksums_ptr = checksums;
1266                         do {
1267                                 integrity_sector_checksum(ic, sector, mem + pos, checksums_ptr);
1268                                 checksums_ptr += ic->tag_size;
1269                                 sectors_to_process -= ic->sectors_per_block;
1270                                 pos += ic->sectors_per_block << SECTOR_SHIFT;
1271                                 sector += ic->sectors_per_block;
1272                         } while (pos < bv.bv_len && sectors_to_process && checksums != checksums_onstack);
1273                         kunmap_atomic(mem);
1274
1275                         r = dm_integrity_rw_tag(ic, checksums, &dio->metadata_block, &dio->metadata_offset,
1276                                                 checksums_ptr - checksums, !dio->write ? TAG_CMP : TAG_WRITE);
1277                         if (unlikely(r)) {
1278                                 if (r > 0) {
1279                                         DMERR("Checksum failed at sector 0x%llx",
1280                                               (unsigned long long)(sector - ((r + ic->tag_size - 1) / ic->tag_size)));
1281                                         r = -EILSEQ;
1282                                         atomic64_inc(&ic->number_of_mismatches);
1283                                 }
1284                                 if (likely(checksums != checksums_onstack))
1285                                         kfree(checksums);
1286                                 goto error;
1287                         }
1288
1289                         if (!sectors_to_process)
1290                                 break;
1291
1292                         if (unlikely(pos < bv.bv_len)) {
1293                                 bv.bv_offset += pos;
1294                                 bv.bv_len -= pos;
1295                                 goto again;
1296                         }
1297                 }
1298
1299                 if (likely(checksums != checksums_onstack))
1300                         kfree(checksums);
1301         } else {
1302                 struct bio_integrity_payload *bip = dio->orig_bi_integrity;
1303
1304                 if (bip) {
1305                         struct bio_vec biv;
1306                         struct bvec_iter iter;
1307                         unsigned data_to_process = dio->range.n_sectors;
1308                         sector_to_block(ic, data_to_process);
1309                         data_to_process *= ic->tag_size;
1310
1311                         bip_for_each_vec(biv, bip, iter) {
1312                                 unsigned char *tag;
1313                                 unsigned this_len;
1314
1315                                 BUG_ON(PageHighMem(biv.bv_page));
1316                                 tag = lowmem_page_address(biv.bv_page) + biv.bv_offset;
1317                                 this_len = min(biv.bv_len, data_to_process);
1318                                 r = dm_integrity_rw_tag(ic, tag, &dio->metadata_block, &dio->metadata_offset,
1319                                                         this_len, !dio->write ? TAG_READ : TAG_WRITE);
1320                                 if (unlikely(r))
1321                                         goto error;
1322                                 data_to_process -= this_len;
1323                                 if (!data_to_process)
1324                                         break;
1325                         }
1326                 }
1327         }
1328 skip_io:
1329         dec_in_flight(dio);
1330         return;
1331 error:
1332         dio->bi_status = errno_to_blk_status(r);
1333         dec_in_flight(dio);
1334 }
1335
1336 static int dm_integrity_map(struct dm_target *ti, struct bio *bio)
1337 {
1338         struct dm_integrity_c *ic = ti->private;
1339         struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
1340         struct bio_integrity_payload *bip;
1341
1342         sector_t area, offset;
1343
1344         dio->ic = ic;
1345         dio->bi_status = 0;
1346
1347         if (unlikely(bio->bi_opf & REQ_PREFLUSH)) {
1348                 submit_flush_bio(ic, dio);
1349                 return DM_MAPIO_SUBMITTED;
1350         }
1351
1352         dio->range.logical_sector = dm_target_offset(ti, bio->bi_iter.bi_sector);
1353         dio->write = bio_op(bio) == REQ_OP_WRITE;
1354         dio->fua = dio->write && bio->bi_opf & REQ_FUA;
1355         if (unlikely(dio->fua)) {
1356                 /*
1357                  * Don't pass down the FUA flag because we have to flush
1358                  * disk cache anyway.
1359                  */
1360                 bio->bi_opf &= ~REQ_FUA;
1361         }
1362         if (unlikely(dio->range.logical_sector + bio_sectors(bio) > ic->provided_data_sectors)) {
1363                 DMERR("Too big sector number: 0x%llx + 0x%x > 0x%llx",
1364                       (unsigned long long)dio->range.logical_sector, bio_sectors(bio),
1365                       (unsigned long long)ic->provided_data_sectors);
1366                 return DM_MAPIO_KILL;
1367         }
1368         if (unlikely((dio->range.logical_sector | bio_sectors(bio)) & (unsigned)(ic->sectors_per_block - 1))) {
1369                 DMERR("Bio not aligned on %u sectors: 0x%llx, 0x%x",
1370                       ic->sectors_per_block,
1371                       (unsigned long long)dio->range.logical_sector, bio_sectors(bio));
1372                 return DM_MAPIO_KILL;
1373         }
1374
1375         if (ic->sectors_per_block > 1) {
1376                 struct bvec_iter iter;
1377                 struct bio_vec bv;
1378                 bio_for_each_segment(bv, bio, iter) {
1379                         if (unlikely(bv.bv_len & ((ic->sectors_per_block << SECTOR_SHIFT) - 1))) {
1380                                 DMERR("Bio vector (%u,%u) is not aligned on %u-sector boundary",
1381                                         bv.bv_offset, bv.bv_len, ic->sectors_per_block);
1382                                 return DM_MAPIO_KILL;
1383                         }
1384                 }
1385         }
1386
1387         bip = bio_integrity(bio);
1388         if (!ic->internal_hash) {
1389                 if (bip) {
1390                         unsigned wanted_tag_size = bio_sectors(bio) >> ic->sb->log2_sectors_per_block;
1391                         if (ic->log2_tag_size >= 0)
1392                                 wanted_tag_size <<= ic->log2_tag_size;
1393                         else
1394                                 wanted_tag_size *= ic->tag_size;
1395                         if (unlikely(wanted_tag_size != bip->bip_iter.bi_size)) {
1396                                 DMERR("Invalid integrity data size %u, expected %u", bip->bip_iter.bi_size, wanted_tag_size);
1397                                 return DM_MAPIO_KILL;
1398                         }
1399                 }
1400         } else {
1401                 if (unlikely(bip != NULL)) {
1402                         DMERR("Unexpected integrity data when using internal hash");
1403                         return DM_MAPIO_KILL;
1404                 }
1405         }
1406
1407         if (unlikely(ic->mode == 'R') && unlikely(dio->write))
1408                 return DM_MAPIO_KILL;
1409
1410         get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
1411         dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset, &dio->metadata_offset);
1412         bio->bi_iter.bi_sector = get_data_sector(ic, area, offset);
1413
1414         dm_integrity_map_continue(dio, true);
1415         return DM_MAPIO_SUBMITTED;
1416 }
1417
1418 static bool __journal_read_write(struct dm_integrity_io *dio, struct bio *bio,
1419                                  unsigned journal_section, unsigned journal_entry)
1420 {
1421         struct dm_integrity_c *ic = dio->ic;
1422         sector_t logical_sector;
1423         unsigned n_sectors;
1424
1425         logical_sector = dio->range.logical_sector;
1426         n_sectors = dio->range.n_sectors;
1427         do {
1428                 struct bio_vec bv = bio_iovec(bio);
1429                 char *mem;
1430
1431                 if (unlikely(bv.bv_len >> SECTOR_SHIFT > n_sectors))
1432                         bv.bv_len = n_sectors << SECTOR_SHIFT;
1433                 n_sectors -= bv.bv_len >> SECTOR_SHIFT;
1434                 bio_advance_iter(bio, &bio->bi_iter, bv.bv_len);
1435 retry_kmap:
1436                 mem = kmap_atomic(bv.bv_page);
1437                 if (likely(dio->write))
1438                         flush_dcache_page(bv.bv_page);
1439
1440                 do {
1441                         struct journal_entry *je = access_journal_entry(ic, journal_section, journal_entry);
1442
1443                         if (unlikely(!dio->write)) {
1444                                 struct journal_sector *js;
1445                                 char *mem_ptr;
1446                                 unsigned s;
1447
1448                                 if (unlikely(journal_entry_is_inprogress(je))) {
1449                                         flush_dcache_page(bv.bv_page);
1450                                         kunmap_atomic(mem);
1451
1452                                         __io_wait_event(ic->copy_to_journal_wait, !journal_entry_is_inprogress(je));
1453                                         goto retry_kmap;
1454                                 }
1455                                 smp_rmb();
1456                                 BUG_ON(journal_entry_get_sector(je) != logical_sector);
1457                                 js = access_journal_data(ic, journal_section, journal_entry);
1458                                 mem_ptr = mem + bv.bv_offset;
1459                                 s = 0;
1460                                 do {
1461                                         memcpy(mem_ptr, js, JOURNAL_SECTOR_DATA);
1462                                         *(commit_id_t *)(mem_ptr + JOURNAL_SECTOR_DATA) = je->last_bytes[s];
1463                                         js++;
1464                                         mem_ptr += 1 << SECTOR_SHIFT;
1465                                 } while (++s < ic->sectors_per_block);
1466 #ifdef INTERNAL_VERIFY
1467                                 if (ic->internal_hash) {
1468                                         char checksums_onstack[max(crypto_shash_digestsize(ic->internal_hash), ic->tag_size)];
1469
1470                                         integrity_sector_checksum(ic, logical_sector, mem + bv.bv_offset, checksums_onstack);
1471                                         if (unlikely(memcmp(checksums_onstack, journal_entry_tag(ic, je), ic->tag_size))) {
1472                                                 DMERR("Checksum failed when reading from journal, at sector 0x%llx",
1473                                                       (unsigned long long)logical_sector);
1474                                         }
1475                                 }
1476 #endif
1477                         }
1478
1479                         if (!ic->internal_hash) {
1480                                 struct bio_integrity_payload *bip = bio_integrity(bio);
1481                                 unsigned tag_todo = ic->tag_size;
1482                                 char *tag_ptr = journal_entry_tag(ic, je);
1483
1484                                 if (bip) do {
1485                                         struct bio_vec biv = bvec_iter_bvec(bip->bip_vec, bip->bip_iter);
1486                                         unsigned tag_now = min(biv.bv_len, tag_todo);
1487                                         char *tag_addr;
1488                                         BUG_ON(PageHighMem(biv.bv_page));
1489                                         tag_addr = lowmem_page_address(biv.bv_page) + biv.bv_offset;
1490                                         if (likely(dio->write))
1491                                                 memcpy(tag_ptr, tag_addr, tag_now);
1492                                         else
1493                                                 memcpy(tag_addr, tag_ptr, tag_now);
1494                                         bvec_iter_advance(bip->bip_vec, &bip->bip_iter, tag_now);
1495                                         tag_ptr += tag_now;
1496                                         tag_todo -= tag_now;
1497                                 } while (unlikely(tag_todo)); else {
1498                                         if (likely(dio->write))
1499                                                 memset(tag_ptr, 0, tag_todo);
1500                                 }
1501                         }
1502
1503                         if (likely(dio->write)) {
1504                                 struct journal_sector *js;
1505                                 unsigned s;
1506
1507                                 js = access_journal_data(ic, journal_section, journal_entry);
1508                                 memcpy(js, mem + bv.bv_offset, ic->sectors_per_block << SECTOR_SHIFT);
1509
1510                                 s = 0;
1511                                 do {
1512                                         je->last_bytes[s] = js[s].commit_id;
1513                                 } while (++s < ic->sectors_per_block);
1514
1515                                 if (ic->internal_hash) {
1516                                         unsigned digest_size = crypto_shash_digestsize(ic->internal_hash);
1517                                         if (unlikely(digest_size > ic->tag_size)) {
1518                                                 char checksums_onstack[digest_size];
1519                                                 integrity_sector_checksum(ic, logical_sector, (char *)js, checksums_onstack);
1520                                                 memcpy(journal_entry_tag(ic, je), checksums_onstack, ic->tag_size);
1521                                         } else
1522                                                 integrity_sector_checksum(ic, logical_sector, (char *)js, journal_entry_tag(ic, je));
1523                                 }
1524
1525                                 journal_entry_set_sector(je, logical_sector);
1526                         }
1527                         logical_sector += ic->sectors_per_block;
1528
1529                         journal_entry++;
1530                         if (unlikely(journal_entry == ic->journal_section_entries)) {
1531                                 journal_entry = 0;
1532                                 journal_section++;
1533                                 wraparound_section(ic, &journal_section);
1534                         }
1535
1536                         bv.bv_offset += ic->sectors_per_block << SECTOR_SHIFT;
1537                 } while (bv.bv_len -= ic->sectors_per_block << SECTOR_SHIFT);
1538
1539                 if (unlikely(!dio->write))
1540                         flush_dcache_page(bv.bv_page);
1541                 kunmap_atomic(mem);
1542         } while (n_sectors);
1543
1544         if (likely(dio->write)) {
1545                 smp_mb();
1546                 if (unlikely(waitqueue_active(&ic->copy_to_journal_wait)))
1547                         wake_up(&ic->copy_to_journal_wait);
1548                 if (ACCESS_ONCE(ic->free_sectors) <= ic->free_sectors_threshold) {
1549                         queue_work(ic->commit_wq, &ic->commit_work);
1550                 } else {
1551                         schedule_autocommit(ic);
1552                 }
1553         } else {
1554                 remove_range(ic, &dio->range);
1555         }
1556
1557         if (unlikely(bio->bi_iter.bi_size)) {
1558                 sector_t area, offset;
1559
1560                 dio->range.logical_sector = logical_sector;
1561                 get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
1562                 dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset, &dio->metadata_offset);
1563                 return true;
1564         }
1565
1566         return false;
1567 }
1568
1569 static void dm_integrity_map_continue(struct dm_integrity_io *dio, bool from_map)
1570 {
1571         struct dm_integrity_c *ic = dio->ic;
1572         struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1573         unsigned journal_section, journal_entry;
1574         unsigned journal_read_pos;
1575         struct completion read_comp;
1576         bool need_sync_io = ic->internal_hash && !dio->write;
1577
1578         if (need_sync_io && from_map) {
1579                 INIT_WORK(&dio->work, integrity_bio_wait);
1580                 queue_work(ic->metadata_wq, &dio->work);
1581                 return;
1582         }
1583
1584 lock_retry:
1585         spin_lock_irq(&ic->endio_wait.lock);
1586 retry:
1587         if (unlikely(dm_integrity_failed(ic))) {
1588                 spin_unlock_irq(&ic->endio_wait.lock);
1589                 do_endio(ic, bio);
1590                 return;
1591         }
1592         dio->range.n_sectors = bio_sectors(bio);
1593         journal_read_pos = NOT_FOUND;
1594         if (likely(ic->mode == 'J')) {
1595                 if (dio->write) {
1596                         unsigned next_entry, i, pos;
1597                         unsigned ws, we, range_sectors;
1598
1599                         dio->range.n_sectors = min(dio->range.n_sectors,
1600                                                    ic->free_sectors << ic->sb->log2_sectors_per_block);
1601                         if (unlikely(!dio->range.n_sectors))
1602                                 goto sleep;
1603                         range_sectors = dio->range.n_sectors >> ic->sb->log2_sectors_per_block;
1604                         ic->free_sectors -= range_sectors;
1605                         journal_section = ic->free_section;
1606                         journal_entry = ic->free_section_entry;
1607
1608                         next_entry = ic->free_section_entry + range_sectors;
1609                         ic->free_section_entry = next_entry % ic->journal_section_entries;
1610                         ic->free_section += next_entry / ic->journal_section_entries;
1611                         ic->n_uncommitted_sections += next_entry / ic->journal_section_entries;
1612                         wraparound_section(ic, &ic->free_section);
1613
1614                         pos = journal_section * ic->journal_section_entries + journal_entry;
1615                         ws = journal_section;
1616                         we = journal_entry;
1617                         i = 0;
1618                         do {
1619                                 struct journal_entry *je;
1620
1621                                 add_journal_node(ic, &ic->journal_tree[pos], dio->range.logical_sector + i);
1622                                 pos++;
1623                                 if (unlikely(pos >= ic->journal_entries))
1624                                         pos = 0;
1625
1626                                 je = access_journal_entry(ic, ws, we);
1627                                 BUG_ON(!journal_entry_is_unused(je));
1628                                 journal_entry_set_inprogress(je);
1629                                 we++;
1630                                 if (unlikely(we == ic->journal_section_entries)) {
1631                                         we = 0;
1632                                         ws++;
1633                                         wraparound_section(ic, &ws);
1634                                 }
1635                         } while ((i += ic->sectors_per_block) < dio->range.n_sectors);
1636
1637                         spin_unlock_irq(&ic->endio_wait.lock);
1638                         goto journal_read_write;
1639                 } else {
1640                         sector_t next_sector;
1641                         journal_read_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
1642                         if (likely(journal_read_pos == NOT_FOUND)) {
1643                                 if (unlikely(dio->range.n_sectors > next_sector - dio->range.logical_sector))
1644                                         dio->range.n_sectors = next_sector - dio->range.logical_sector;
1645                         } else {
1646                                 unsigned i;
1647                                 unsigned jp = journal_read_pos + 1;
1648                                 for (i = ic->sectors_per_block; i < dio->range.n_sectors; i += ic->sectors_per_block, jp++) {
1649                                         if (!test_journal_node(ic, jp, dio->range.logical_sector + i))
1650                                                 break;
1651                                 }
1652                                 dio->range.n_sectors = i;
1653                         }
1654                 }
1655         }
1656         if (unlikely(!add_new_range(ic, &dio->range))) {
1657                 /*
1658                  * We must not sleep in the request routine because it could
1659                  * stall bios on current->bio_list.
1660                  * So, we offload the bio to a workqueue if we have to sleep.
1661                  */
1662 sleep:
1663                 if (from_map) {
1664                         spin_unlock_irq(&ic->endio_wait.lock);
1665                         INIT_WORK(&dio->work, integrity_bio_wait);
1666                         queue_work(ic->wait_wq, &dio->work);
1667                         return;
1668                 } else {
1669                         sleep_on_endio_wait(ic);
1670                         goto retry;
1671                 }
1672         }
1673         spin_unlock_irq(&ic->endio_wait.lock);
1674
1675         if (unlikely(journal_read_pos != NOT_FOUND)) {
1676                 journal_section = journal_read_pos / ic->journal_section_entries;
1677                 journal_entry = journal_read_pos % ic->journal_section_entries;
1678                 goto journal_read_write;
1679         }
1680
1681         dio->in_flight = (atomic_t)ATOMIC_INIT(2);
1682
1683         if (need_sync_io) {
1684                 init_completion(&read_comp);
1685                 dio->completion = &read_comp;
1686         } else
1687                 dio->completion = NULL;
1688
1689         dio->orig_bi_iter = bio->bi_iter;
1690
1691         dio->orig_bi_disk = bio->bi_disk;
1692         dio->orig_bi_partno = bio->bi_partno;
1693         bio_set_dev(bio, ic->dev->bdev);
1694
1695         dio->orig_bi_integrity = bio_integrity(bio);
1696         bio->bi_integrity = NULL;
1697         bio->bi_opf &= ~REQ_INTEGRITY;
1698
1699         dio->orig_bi_end_io = bio->bi_end_io;
1700         bio->bi_end_io = integrity_end_io;
1701
1702         bio->bi_iter.bi_size = dio->range.n_sectors << SECTOR_SHIFT;
1703         bio->bi_iter.bi_sector += ic->start;
1704         generic_make_request(bio);
1705
1706         if (need_sync_io) {
1707                 wait_for_completion_io(&read_comp);
1708                 if (likely(!bio->bi_status))
1709                         integrity_metadata(&dio->work);
1710                 else
1711                         dec_in_flight(dio);
1712
1713         } else {
1714                 INIT_WORK(&dio->work, integrity_metadata);
1715                 queue_work(ic->metadata_wq, &dio->work);
1716         }
1717
1718         return;
1719
1720 journal_read_write:
1721         if (unlikely(__journal_read_write(dio, bio, journal_section, journal_entry)))
1722                 goto lock_retry;
1723
1724         do_endio_flush(ic, dio);
1725 }
1726
1727
1728 static void integrity_bio_wait(struct work_struct *w)
1729 {
1730         struct dm_integrity_io *dio = container_of(w, struct dm_integrity_io, work);
1731
1732         dm_integrity_map_continue(dio, false);
1733 }
1734
1735 static void pad_uncommitted(struct dm_integrity_c *ic)
1736 {
1737         if (ic->free_section_entry) {
1738                 ic->free_sectors -= ic->journal_section_entries - ic->free_section_entry;
1739                 ic->free_section_entry = 0;
1740                 ic->free_section++;
1741                 wraparound_section(ic, &ic->free_section);
1742                 ic->n_uncommitted_sections++;
1743         }
1744         WARN_ON(ic->journal_sections * ic->journal_section_entries !=
1745                 (ic->n_uncommitted_sections + ic->n_committed_sections) * ic->journal_section_entries + ic->free_sectors);
1746 }
1747
1748 static void integrity_commit(struct work_struct *w)
1749 {
1750         struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, commit_work);
1751         unsigned commit_start, commit_sections;
1752         unsigned i, j, n;
1753         struct bio *flushes;
1754
1755         del_timer(&ic->autocommit_timer);
1756
1757         spin_lock_irq(&ic->endio_wait.lock);
1758         flushes = bio_list_get(&ic->flush_bio_list);
1759         if (unlikely(ic->mode != 'J')) {
1760                 spin_unlock_irq(&ic->endio_wait.lock);
1761                 dm_integrity_flush_buffers(ic);
1762                 goto release_flush_bios;
1763         }
1764
1765         pad_uncommitted(ic);
1766         commit_start = ic->uncommitted_section;
1767         commit_sections = ic->n_uncommitted_sections;
1768         spin_unlock_irq(&ic->endio_wait.lock);
1769
1770         if (!commit_sections)
1771                 goto release_flush_bios;
1772
1773         i = commit_start;
1774         for (n = 0; n < commit_sections; n++) {
1775                 for (j = 0; j < ic->journal_section_entries; j++) {
1776                         struct journal_entry *je;
1777                         je = access_journal_entry(ic, i, j);
1778                         io_wait_event(ic->copy_to_journal_wait, !journal_entry_is_inprogress(je));
1779                 }
1780                 for (j = 0; j < ic->journal_section_sectors; j++) {
1781                         struct journal_sector *js;
1782                         js = access_journal(ic, i, j);
1783                         js->commit_id = dm_integrity_commit_id(ic, i, j, ic->commit_seq);
1784                 }
1785                 i++;
1786                 if (unlikely(i >= ic->journal_sections))
1787                         ic->commit_seq = next_commit_seq(ic->commit_seq);
1788                 wraparound_section(ic, &i);
1789         }
1790         smp_rmb();
1791
1792         write_journal(ic, commit_start, commit_sections);
1793
1794         spin_lock_irq(&ic->endio_wait.lock);
1795         ic->uncommitted_section += commit_sections;
1796         wraparound_section(ic, &ic->uncommitted_section);
1797         ic->n_uncommitted_sections -= commit_sections;
1798         ic->n_committed_sections += commit_sections;
1799         spin_unlock_irq(&ic->endio_wait.lock);
1800
1801         if (ACCESS_ONCE(ic->free_sectors) <= ic->free_sectors_threshold)
1802                 queue_work(ic->writer_wq, &ic->writer_work);
1803
1804 release_flush_bios:
1805         while (flushes) {
1806                 struct bio *next = flushes->bi_next;
1807                 flushes->bi_next = NULL;
1808                 do_endio(ic, flushes);
1809                 flushes = next;
1810         }
1811 }
1812
1813 static void complete_copy_from_journal(unsigned long error, void *context)
1814 {
1815         struct journal_io *io = context;
1816         struct journal_completion *comp = io->comp;
1817         struct dm_integrity_c *ic = comp->ic;
1818         remove_range(ic, &io->range);
1819         mempool_free(io, ic->journal_io_mempool);
1820         if (unlikely(error != 0))
1821                 dm_integrity_io_error(ic, "copying from journal", -EIO);
1822         complete_journal_op(comp);
1823 }
1824
1825 static void restore_last_bytes(struct dm_integrity_c *ic, struct journal_sector *js,
1826                                struct journal_entry *je)
1827 {
1828         unsigned s = 0;
1829         do {
1830                 js->commit_id = je->last_bytes[s];
1831                 js++;
1832         } while (++s < ic->sectors_per_block);
1833 }
1834
1835 static void do_journal_write(struct dm_integrity_c *ic, unsigned write_start,
1836                              unsigned write_sections, bool from_replay)
1837 {
1838         unsigned i, j, n;
1839         struct journal_completion comp;
1840         struct blk_plug plug;
1841
1842         blk_start_plug(&plug);
1843
1844         comp.ic = ic;
1845         comp.in_flight = (atomic_t)ATOMIC_INIT(1);
1846         init_completion(&comp.comp);
1847
1848         i = write_start;
1849         for (n = 0; n < write_sections; n++, i++, wraparound_section(ic, &i)) {
1850 #ifndef INTERNAL_VERIFY
1851                 if (unlikely(from_replay))
1852 #endif
1853                         rw_section_mac(ic, i, false);
1854                 for (j = 0; j < ic->journal_section_entries; j++) {
1855                         struct journal_entry *je = access_journal_entry(ic, i, j);
1856                         sector_t sec, area, offset;
1857                         unsigned k, l, next_loop;
1858                         sector_t metadata_block;
1859                         unsigned metadata_offset;
1860                         struct journal_io *io;
1861
1862                         if (journal_entry_is_unused(je))
1863                                 continue;
1864                         BUG_ON(unlikely(journal_entry_is_inprogress(je)) && !from_replay);
1865                         sec = journal_entry_get_sector(je);
1866                         if (unlikely(from_replay)) {
1867                                 if (unlikely(sec & (unsigned)(ic->sectors_per_block - 1))) {
1868                                         dm_integrity_io_error(ic, "invalid sector in journal", -EIO);
1869                                         sec &= ~(sector_t)(ic->sectors_per_block - 1);
1870                                 }
1871                         }
1872                         get_area_and_offset(ic, sec, &area, &offset);
1873                         restore_last_bytes(ic, access_journal_data(ic, i, j), je);
1874                         for (k = j + 1; k < ic->journal_section_entries; k++) {
1875                                 struct journal_entry *je2 = access_journal_entry(ic, i, k);
1876                                 sector_t sec2, area2, offset2;
1877                                 if (journal_entry_is_unused(je2))
1878                                         break;
1879                                 BUG_ON(unlikely(journal_entry_is_inprogress(je2)) && !from_replay);
1880                                 sec2 = journal_entry_get_sector(je2);
1881                                 get_area_and_offset(ic, sec2, &area2, &offset2);
1882                                 if (area2 != area || offset2 != offset + ((k - j) << ic->sb->log2_sectors_per_block))
1883                                         break;
1884                                 restore_last_bytes(ic, access_journal_data(ic, i, k), je2);
1885                         }
1886                         next_loop = k - 1;
1887
1888                         io = mempool_alloc(ic->journal_io_mempool, GFP_NOIO);
1889                         io->comp = &comp;
1890                         io->range.logical_sector = sec;
1891                         io->range.n_sectors = (k - j) << ic->sb->log2_sectors_per_block;
1892
1893                         spin_lock_irq(&ic->endio_wait.lock);
1894                         while (unlikely(!add_new_range(ic, &io->range)))
1895                                 sleep_on_endio_wait(ic);
1896
1897                         if (likely(!from_replay)) {
1898                                 struct journal_node *section_node = &ic->journal_tree[i * ic->journal_section_entries];
1899
1900                                 /* don't write if there is newer committed sector */
1901                                 while (j < k && find_newer_committed_node(ic, &section_node[j])) {
1902                                         struct journal_entry *je2 = access_journal_entry(ic, i, j);
1903
1904                                         journal_entry_set_unused(je2);
1905                                         remove_journal_node(ic, &section_node[j]);
1906                                         j++;
1907                                         sec += ic->sectors_per_block;
1908                                         offset += ic->sectors_per_block;
1909                                 }
1910                                 while (j < k && find_newer_committed_node(ic, &section_node[k - 1])) {
1911                                         struct journal_entry *je2 = access_journal_entry(ic, i, k - 1);
1912
1913                                         journal_entry_set_unused(je2);
1914                                         remove_journal_node(ic, &section_node[k - 1]);
1915                                         k--;
1916                                 }
1917                                 if (j == k) {
1918                                         remove_range_unlocked(ic, &io->range);
1919                                         spin_unlock_irq(&ic->endio_wait.lock);
1920                                         mempool_free(io, ic->journal_io_mempool);
1921                                         goto skip_io;
1922                                 }
1923                                 for (l = j; l < k; l++) {
1924                                         remove_journal_node(ic, &section_node[l]);
1925                                 }
1926                         }
1927                         spin_unlock_irq(&ic->endio_wait.lock);
1928
1929                         metadata_block = get_metadata_sector_and_offset(ic, area, offset, &metadata_offset);
1930                         for (l = j; l < k; l++) {
1931                                 int r;
1932                                 struct journal_entry *je2 = access_journal_entry(ic, i, l);
1933
1934                                 if (
1935 #ifndef INTERNAL_VERIFY
1936                                     unlikely(from_replay) &&
1937 #endif
1938                                     ic->internal_hash) {
1939                                         char test_tag[max(crypto_shash_digestsize(ic->internal_hash), ic->tag_size)];
1940
1941                                         integrity_sector_checksum(ic, sec + ((l - j) << ic->sb->log2_sectors_per_block),
1942                                                                   (char *)access_journal_data(ic, i, l), test_tag);
1943                                         if (unlikely(memcmp(test_tag, journal_entry_tag(ic, je2), ic->tag_size)))
1944                                                 dm_integrity_io_error(ic, "tag mismatch when replaying journal", -EILSEQ);
1945                                 }
1946
1947                                 journal_entry_set_unused(je2);
1948                                 r = dm_integrity_rw_tag(ic, journal_entry_tag(ic, je2), &metadata_block, &metadata_offset,
1949                                                         ic->tag_size, TAG_WRITE);
1950                                 if (unlikely(r)) {
1951                                         dm_integrity_io_error(ic, "reading tags", r);
1952                                 }
1953                         }
1954
1955                         atomic_inc(&comp.in_flight);
1956                         copy_from_journal(ic, i, j << ic->sb->log2_sectors_per_block,
1957                                           (k - j) << ic->sb->log2_sectors_per_block,
1958                                           get_data_sector(ic, area, offset),
1959                                           complete_copy_from_journal, io);
1960 skip_io:
1961                         j = next_loop;
1962                 }
1963         }
1964
1965         dm_bufio_write_dirty_buffers_async(ic->bufio);
1966
1967         blk_finish_plug(&plug);
1968
1969         complete_journal_op(&comp);
1970         wait_for_completion_io(&comp.comp);
1971
1972         dm_integrity_flush_buffers(ic);
1973 }
1974
1975 static void integrity_writer(struct work_struct *w)
1976 {
1977         struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, writer_work);
1978         unsigned write_start, write_sections;
1979
1980         unsigned prev_free_sectors;
1981
1982         /* the following test is not needed, but it tests the replay code */
1983         if (ACCESS_ONCE(ic->suspending))
1984                 return;
1985
1986         spin_lock_irq(&ic->endio_wait.lock);
1987         write_start = ic->committed_section;
1988         write_sections = ic->n_committed_sections;
1989         spin_unlock_irq(&ic->endio_wait.lock);
1990
1991         if (!write_sections)
1992                 return;
1993
1994         do_journal_write(ic, write_start, write_sections, false);
1995
1996         spin_lock_irq(&ic->endio_wait.lock);
1997
1998         ic->committed_section += write_sections;
1999         wraparound_section(ic, &ic->committed_section);
2000         ic->n_committed_sections -= write_sections;
2001
2002         prev_free_sectors = ic->free_sectors;
2003         ic->free_sectors += write_sections * ic->journal_section_entries;
2004         if (unlikely(!prev_free_sectors))
2005                 wake_up_locked(&ic->endio_wait);
2006
2007         spin_unlock_irq(&ic->endio_wait.lock);
2008 }
2009
2010 static void init_journal(struct dm_integrity_c *ic, unsigned start_section,
2011                          unsigned n_sections, unsigned char commit_seq)
2012 {
2013         unsigned i, j, n;
2014
2015         if (!n_sections)
2016                 return;
2017
2018         for (n = 0; n < n_sections; n++) {
2019                 i = start_section + n;
2020                 wraparound_section(ic, &i);
2021                 for (j = 0; j < ic->journal_section_sectors; j++) {
2022                         struct journal_sector *js = access_journal(ic, i, j);
2023                         memset(&js->entries, 0, JOURNAL_SECTOR_DATA);
2024                         js->commit_id = dm_integrity_commit_id(ic, i, j, commit_seq);
2025                 }
2026                 for (j = 0; j < ic->journal_section_entries; j++) {
2027                         struct journal_entry *je = access_journal_entry(ic, i, j);
2028                         journal_entry_set_unused(je);
2029                 }
2030         }
2031
2032         write_journal(ic, start_section, n_sections);
2033 }
2034
2035 static int find_commit_seq(struct dm_integrity_c *ic, unsigned i, unsigned j, commit_id_t id)
2036 {
2037         unsigned char k;
2038         for (k = 0; k < N_COMMIT_IDS; k++) {
2039                 if (dm_integrity_commit_id(ic, i, j, k) == id)
2040                         return k;
2041         }
2042         dm_integrity_io_error(ic, "journal commit id", -EIO);
2043         return -EIO;
2044 }
2045
2046 static void replay_journal(struct dm_integrity_c *ic)
2047 {
2048         unsigned i, j;
2049         bool used_commit_ids[N_COMMIT_IDS];
2050         unsigned max_commit_id_sections[N_COMMIT_IDS];
2051         unsigned write_start, write_sections;
2052         unsigned continue_section;
2053         bool journal_empty;
2054         unsigned char unused, last_used, want_commit_seq;
2055
2056         if (ic->mode == 'R')
2057                 return;
2058
2059         if (ic->journal_uptodate)
2060                 return;
2061
2062         last_used = 0;
2063         write_start = 0;
2064
2065         if (!ic->just_formatted) {
2066                 DEBUG_print("reading journal\n");
2067                 rw_journal(ic, REQ_OP_READ, 0, 0, ic->journal_sections, NULL);
2068                 if (ic->journal_io)
2069                         DEBUG_bytes(lowmem_page_address(ic->journal_io[0].page), 64, "read journal");
2070                 if (ic->journal_io) {
2071                         struct journal_completion crypt_comp;
2072                         crypt_comp.ic = ic;
2073                         init_completion(&crypt_comp.comp);
2074                         crypt_comp.in_flight = (atomic_t)ATOMIC_INIT(0);
2075                         encrypt_journal(ic, false, 0, ic->journal_sections, &crypt_comp);
2076                         wait_for_completion(&crypt_comp.comp);
2077                 }
2078                 DEBUG_bytes(lowmem_page_address(ic->journal[0].page), 64, "decrypted journal");
2079         }
2080
2081         if (dm_integrity_failed(ic))
2082                 goto clear_journal;
2083
2084         journal_empty = true;
2085         memset(used_commit_ids, 0, sizeof used_commit_ids);
2086         memset(max_commit_id_sections, 0, sizeof max_commit_id_sections);
2087         for (i = 0; i < ic->journal_sections; i++) {
2088                 for (j = 0; j < ic->journal_section_sectors; j++) {
2089                         int k;
2090                         struct journal_sector *js = access_journal(ic, i, j);
2091                         k = find_commit_seq(ic, i, j, js->commit_id);
2092                         if (k < 0)
2093                                 goto clear_journal;
2094                         used_commit_ids[k] = true;
2095                         max_commit_id_sections[k] = i;
2096                 }
2097                 if (journal_empty) {
2098                         for (j = 0; j < ic->journal_section_entries; j++) {
2099                                 struct journal_entry *je = access_journal_entry(ic, i, j);
2100                                 if (!journal_entry_is_unused(je)) {
2101                                         journal_empty = false;
2102                                         break;
2103                                 }
2104                         }
2105                 }
2106         }
2107
2108         if (!used_commit_ids[N_COMMIT_IDS - 1]) {
2109                 unused = N_COMMIT_IDS - 1;
2110                 while (unused && !used_commit_ids[unused - 1])
2111                         unused--;
2112         } else {
2113                 for (unused = 0; unused < N_COMMIT_IDS; unused++)
2114                         if (!used_commit_ids[unused])
2115                                 break;
2116                 if (unused == N_COMMIT_IDS) {
2117                         dm_integrity_io_error(ic, "journal commit ids", -EIO);
2118                         goto clear_journal;
2119                 }
2120         }
2121         DEBUG_print("first unused commit seq %d [%d,%d,%d,%d]\n",
2122                     unused, used_commit_ids[0], used_commit_ids[1],
2123                     used_commit_ids[2], used_commit_ids[3]);
2124
2125         last_used = prev_commit_seq(unused);
2126         want_commit_seq = prev_commit_seq(last_used);
2127
2128         if (!used_commit_ids[want_commit_seq] && used_commit_ids[prev_commit_seq(want_commit_seq)])
2129                 journal_empty = true;
2130
2131         write_start = max_commit_id_sections[last_used] + 1;
2132         if (unlikely(write_start >= ic->journal_sections))
2133                 want_commit_seq = next_commit_seq(want_commit_seq);
2134         wraparound_section(ic, &write_start);
2135
2136         i = write_start;
2137         for (write_sections = 0; write_sections < ic->journal_sections; write_sections++) {
2138                 for (j = 0; j < ic->journal_section_sectors; j++) {
2139                         struct journal_sector *js = access_journal(ic, i, j);
2140
2141                         if (js->commit_id != dm_integrity_commit_id(ic, i, j, want_commit_seq)) {
2142                                 /*
2143                                  * This could be caused by crash during writing.
2144                                  * We won't replay the inconsistent part of the
2145                                  * journal.
2146                                  */
2147                                 DEBUG_print("commit id mismatch at position (%u, %u): %d != %d\n",
2148                                             i, j, find_commit_seq(ic, i, j, js->commit_id), want_commit_seq);
2149                                 goto brk;
2150                         }
2151                 }
2152                 i++;
2153                 if (unlikely(i >= ic->journal_sections))
2154                         want_commit_seq = next_commit_seq(want_commit_seq);
2155                 wraparound_section(ic, &i);
2156         }
2157 brk:
2158
2159         if (!journal_empty) {
2160                 DEBUG_print("replaying %u sections, starting at %u, commit seq %d\n",
2161                             write_sections, write_start, want_commit_seq);
2162                 do_journal_write(ic, write_start, write_sections, true);
2163         }
2164
2165         if (write_sections == ic->journal_sections && (ic->mode == 'J' || journal_empty)) {
2166                 continue_section = write_start;
2167                 ic->commit_seq = want_commit_seq;
2168                 DEBUG_print("continuing from section %u, commit seq %d\n", write_start, ic->commit_seq);
2169         } else {
2170                 unsigned s;
2171                 unsigned char erase_seq;
2172 clear_journal:
2173                 DEBUG_print("clearing journal\n");
2174
2175                 erase_seq = prev_commit_seq(prev_commit_seq(last_used));
2176                 s = write_start;
2177                 init_journal(ic, s, 1, erase_seq);
2178                 s++;
2179                 wraparound_section(ic, &s);
2180                 if (ic->journal_sections >= 2) {
2181                         init_journal(ic, s, ic->journal_sections - 2, erase_seq);
2182                         s += ic->journal_sections - 2;
2183                         wraparound_section(ic, &s);
2184                         init_journal(ic, s, 1, erase_seq);
2185                 }
2186
2187                 continue_section = 0;
2188                 ic->commit_seq = next_commit_seq(erase_seq);
2189         }
2190
2191         ic->committed_section = continue_section;
2192         ic->n_committed_sections = 0;
2193
2194         ic->uncommitted_section = continue_section;
2195         ic->n_uncommitted_sections = 0;
2196
2197         ic->free_section = continue_section;
2198         ic->free_section_entry = 0;
2199         ic->free_sectors = ic->journal_entries;
2200
2201         ic->journal_tree_root = RB_ROOT;
2202         for (i = 0; i < ic->journal_entries; i++)
2203                 init_journal_node(&ic->journal_tree[i]);
2204 }
2205
2206 static void dm_integrity_postsuspend(struct dm_target *ti)
2207 {
2208         struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
2209
2210         del_timer_sync(&ic->autocommit_timer);
2211
2212         ic->suspending = true;
2213
2214         queue_work(ic->commit_wq, &ic->commit_work);
2215         drain_workqueue(ic->commit_wq);
2216
2217         if (ic->mode == 'J') {
2218                 drain_workqueue(ic->writer_wq);
2219                 dm_integrity_flush_buffers(ic);
2220         }
2221
2222         ic->suspending = false;
2223
2224         BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
2225
2226         ic->journal_uptodate = true;
2227 }
2228
2229 static void dm_integrity_resume(struct dm_target *ti)
2230 {
2231         struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
2232
2233         replay_journal(ic);
2234 }
2235
2236 static void dm_integrity_status(struct dm_target *ti, status_type_t type,
2237                                 unsigned status_flags, char *result, unsigned maxlen)
2238 {
2239         struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
2240         unsigned arg_count;
2241         size_t sz = 0;
2242
2243         switch (type) {
2244         case STATUSTYPE_INFO:
2245                 DMEMIT("%llu", (unsigned long long)atomic64_read(&ic->number_of_mismatches));
2246                 break;
2247
2248         case STATUSTYPE_TABLE: {
2249                 __u64 watermark_percentage = (__u64)(ic->journal_entries - ic->free_sectors_threshold) * 100;
2250                 watermark_percentage += ic->journal_entries / 2;
2251                 do_div(watermark_percentage, ic->journal_entries);
2252                 arg_count = 5;
2253                 arg_count += ic->sectors_per_block != 1;
2254                 arg_count += !!ic->internal_hash_alg.alg_string;
2255                 arg_count += !!ic->journal_crypt_alg.alg_string;
2256                 arg_count += !!ic->journal_mac_alg.alg_string;
2257                 DMEMIT("%s %llu %u %c %u", ic->dev->name, (unsigned long long)ic->start,
2258                        ic->tag_size, ic->mode, arg_count);
2259                 DMEMIT(" journal_sectors:%u", ic->initial_sectors - SB_SECTORS);
2260                 DMEMIT(" interleave_sectors:%u", 1U << ic->sb->log2_interleave_sectors);
2261                 DMEMIT(" buffer_sectors:%u", 1U << ic->log2_buffer_sectors);
2262                 DMEMIT(" journal_watermark:%u", (unsigned)watermark_percentage);
2263                 DMEMIT(" commit_time:%u", ic->autocommit_msec);
2264                 if (ic->sectors_per_block != 1)
2265                         DMEMIT(" block_size:%u", ic->sectors_per_block << SECTOR_SHIFT);
2266
2267 #define EMIT_ALG(a, n)                                                  \
2268                 do {                                                    \
2269                         if (ic->a.alg_string) {                         \
2270                                 DMEMIT(" %s:%s", n, ic->a.alg_string);  \
2271                                 if (ic->a.key_string)                   \
2272                                         DMEMIT(":%s", ic->a.key_string);\
2273                         }                                               \
2274                 } while (0)
2275                 EMIT_ALG(internal_hash_alg, "internal_hash");
2276                 EMIT_ALG(journal_crypt_alg, "journal_crypt");
2277                 EMIT_ALG(journal_mac_alg, "journal_mac");
2278                 break;
2279         }
2280         }
2281 }
2282
2283 static int dm_integrity_iterate_devices(struct dm_target *ti,
2284                                         iterate_devices_callout_fn fn, void *data)
2285 {
2286         struct dm_integrity_c *ic = ti->private;
2287
2288         return fn(ti, ic->dev, ic->start + ic->initial_sectors + ic->metadata_run, ti->len, data);
2289 }
2290
2291 static void dm_integrity_io_hints(struct dm_target *ti, struct queue_limits *limits)
2292 {
2293         struct dm_integrity_c *ic = ti->private;
2294
2295         if (ic->sectors_per_block > 1) {
2296                 limits->logical_block_size = ic->sectors_per_block << SECTOR_SHIFT;
2297                 limits->physical_block_size = ic->sectors_per_block << SECTOR_SHIFT;
2298                 blk_limits_io_min(limits, ic->sectors_per_block << SECTOR_SHIFT);
2299         }
2300 }
2301
2302 static void calculate_journal_section_size(struct dm_integrity_c *ic)
2303 {
2304         unsigned sector_space = JOURNAL_SECTOR_DATA;
2305
2306         ic->journal_sections = le32_to_cpu(ic->sb->journal_sections);
2307         ic->journal_entry_size = roundup(offsetof(struct journal_entry, last_bytes[ic->sectors_per_block]) + ic->tag_size,
2308                                          JOURNAL_ENTRY_ROUNDUP);
2309
2310         if (ic->sb->flags & cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC))
2311                 sector_space -= JOURNAL_MAC_PER_SECTOR;
2312         ic->journal_entries_per_sector = sector_space / ic->journal_entry_size;
2313         ic->journal_section_entries = ic->journal_entries_per_sector * JOURNAL_BLOCK_SECTORS;
2314         ic->journal_section_sectors = (ic->journal_section_entries << ic->sb->log2_sectors_per_block) + JOURNAL_BLOCK_SECTORS;
2315         ic->journal_entries = ic->journal_section_entries * ic->journal_sections;
2316 }
2317
2318 static int calculate_device_limits(struct dm_integrity_c *ic)
2319 {
2320         __u64 initial_sectors;
2321         sector_t last_sector, last_area, last_offset;
2322
2323         calculate_journal_section_size(ic);
2324         initial_sectors = SB_SECTORS + (__u64)ic->journal_section_sectors * ic->journal_sections;
2325         if (initial_sectors + METADATA_PADDING_SECTORS >= ic->device_sectors || initial_sectors > UINT_MAX)
2326                 return -EINVAL;
2327         ic->initial_sectors = initial_sectors;
2328
2329         ic->metadata_run = roundup((__u64)ic->tag_size << (ic->sb->log2_interleave_sectors - ic->sb->log2_sectors_per_block),
2330                                    (__u64)(1 << SECTOR_SHIFT << METADATA_PADDING_SECTORS)) >> SECTOR_SHIFT;
2331         if (!(ic->metadata_run & (ic->metadata_run - 1)))
2332                 ic->log2_metadata_run = __ffs(ic->metadata_run);
2333         else
2334                 ic->log2_metadata_run = -1;
2335
2336         get_area_and_offset(ic, ic->provided_data_sectors - 1, &last_area, &last_offset);
2337         last_sector = get_data_sector(ic, last_area, last_offset);
2338
2339         if (ic->start + last_sector < last_sector || ic->start + last_sector >= ic->device_sectors)
2340                 return -EINVAL;
2341
2342         return 0;
2343 }
2344
2345 static int initialize_superblock(struct dm_integrity_c *ic, unsigned journal_sectors, unsigned interleave_sectors)
2346 {
2347         unsigned journal_sections;
2348         int test_bit;
2349
2350         memset(ic->sb, 0, SB_SECTORS << SECTOR_SHIFT);
2351         memcpy(ic->sb->magic, SB_MAGIC, 8);
2352         ic->sb->version = SB_VERSION;
2353         ic->sb->integrity_tag_size = cpu_to_le16(ic->tag_size);
2354         ic->sb->log2_sectors_per_block = __ffs(ic->sectors_per_block);
2355         if (ic->journal_mac_alg.alg_string)
2356                 ic->sb->flags |= cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC);
2357
2358         calculate_journal_section_size(ic);
2359         journal_sections = journal_sectors / ic->journal_section_sectors;
2360         if (!journal_sections)
2361                 journal_sections = 1;
2362         ic->sb->journal_sections = cpu_to_le32(journal_sections);
2363
2364         if (!interleave_sectors)
2365                 interleave_sectors = DEFAULT_INTERLEAVE_SECTORS;
2366         ic->sb->log2_interleave_sectors = __fls(interleave_sectors);
2367         ic->sb->log2_interleave_sectors = max((__u8)MIN_LOG2_INTERLEAVE_SECTORS, ic->sb->log2_interleave_sectors);
2368         ic->sb->log2_interleave_sectors = min((__u8)MAX_LOG2_INTERLEAVE_SECTORS, ic->sb->log2_interleave_sectors);
2369
2370         ic->provided_data_sectors = 0;
2371         for (test_bit = fls64(ic->device_sectors) - 1; test_bit >= 3; test_bit--) {
2372                 __u64 prev_data_sectors = ic->provided_data_sectors;
2373
2374                 ic->provided_data_sectors |= (sector_t)1 << test_bit;
2375                 if (calculate_device_limits(ic))
2376                         ic->provided_data_sectors = prev_data_sectors;
2377         }
2378
2379         if (!ic->provided_data_sectors)
2380                 return -EINVAL;
2381
2382         ic->sb->provided_data_sectors = cpu_to_le64(ic->provided_data_sectors);
2383
2384         return 0;
2385 }
2386
2387 static void dm_integrity_set(struct dm_target *ti, struct dm_integrity_c *ic)
2388 {
2389         struct gendisk *disk = dm_disk(dm_table_get_md(ti->table));
2390         struct blk_integrity bi;
2391
2392         memset(&bi, 0, sizeof(bi));
2393         bi.profile = &dm_integrity_profile;
2394         bi.tuple_size = ic->tag_size;
2395         bi.tag_size = bi.tuple_size;
2396         bi.interval_exp = ic->sb->log2_sectors_per_block + SECTOR_SHIFT;
2397
2398         blk_integrity_register(disk, &bi);
2399         blk_queue_max_integrity_segments(disk->queue, UINT_MAX);
2400 }
2401
2402 static void dm_integrity_free_page_list(struct dm_integrity_c *ic, struct page_list *pl)
2403 {
2404         unsigned i;
2405
2406         if (!pl)
2407                 return;
2408         for (i = 0; i < ic->journal_pages; i++)
2409                 if (pl[i].page)
2410                         __free_page(pl[i].page);
2411         kvfree(pl);
2412 }
2413
2414 static struct page_list *dm_integrity_alloc_page_list(struct dm_integrity_c *ic)
2415 {
2416         size_t page_list_desc_size = ic->journal_pages * sizeof(struct page_list);
2417         struct page_list *pl;
2418         unsigned i;
2419
2420         pl = kvmalloc(page_list_desc_size, GFP_KERNEL | __GFP_ZERO);
2421         if (!pl)
2422                 return NULL;
2423
2424         for (i = 0; i < ic->journal_pages; i++) {
2425                 pl[i].page = alloc_page(GFP_KERNEL);
2426                 if (!pl[i].page) {
2427                         dm_integrity_free_page_list(ic, pl);
2428                         return NULL;
2429                 }
2430                 if (i)
2431                         pl[i - 1].next = &pl[i];
2432         }
2433
2434         return pl;
2435 }
2436
2437 static void dm_integrity_free_journal_scatterlist(struct dm_integrity_c *ic, struct scatterlist **sl)
2438 {
2439         unsigned i;
2440         for (i = 0; i < ic->journal_sections; i++)
2441                 kvfree(sl[i]);
2442         kfree(sl);
2443 }
2444
2445 static struct scatterlist **dm_integrity_alloc_journal_scatterlist(struct dm_integrity_c *ic, struct page_list *pl)
2446 {
2447         struct scatterlist **sl;
2448         unsigned i;
2449
2450         sl = kvmalloc(ic->journal_sections * sizeof(struct scatterlist *), GFP_KERNEL | __GFP_ZERO);
2451         if (!sl)
2452                 return NULL;
2453
2454         for (i = 0; i < ic->journal_sections; i++) {
2455                 struct scatterlist *s;
2456                 unsigned start_index, start_offset;
2457                 unsigned end_index, end_offset;
2458                 unsigned n_pages;
2459                 unsigned idx;
2460
2461                 page_list_location(ic, i, 0, &start_index, &start_offset);
2462                 page_list_location(ic, i, ic->journal_section_sectors - 1, &end_index, &end_offset);
2463
2464                 n_pages = (end_index - start_index + 1);
2465
2466                 s = kvmalloc(n_pages * sizeof(struct scatterlist), GFP_KERNEL);
2467                 if (!s) {
2468                         dm_integrity_free_journal_scatterlist(ic, sl);
2469                         return NULL;
2470                 }
2471
2472                 sg_init_table(s, n_pages);
2473                 for (idx = start_index; idx <= end_index; idx++) {
2474                         char *va = lowmem_page_address(pl[idx].page);
2475                         unsigned start = 0, end = PAGE_SIZE;
2476                         if (idx == start_index)
2477                                 start = start_offset;
2478                         if (idx == end_index)
2479                                 end = end_offset + (1 << SECTOR_SHIFT);
2480                         sg_set_buf(&s[idx - start_index], va + start, end - start);
2481                 }
2482
2483                 sl[i] = s;
2484         }
2485
2486         return sl;
2487 }
2488
2489 static void free_alg(struct alg_spec *a)
2490 {
2491         kzfree(a->alg_string);
2492         kzfree(a->key);
2493         memset(a, 0, sizeof *a);
2494 }
2495
2496 static int get_alg_and_key(const char *arg, struct alg_spec *a, char **error, char *error_inval)
2497 {
2498         char *k;
2499
2500         free_alg(a);
2501
2502         a->alg_string = kstrdup(strchr(arg, ':') + 1, GFP_KERNEL);
2503         if (!a->alg_string)
2504                 goto nomem;
2505
2506         k = strchr(a->alg_string, ':');
2507         if (k) {
2508                 *k = 0;
2509                 a->key_string = k + 1;
2510                 if (strlen(a->key_string) & 1)
2511                         goto inval;
2512
2513                 a->key_size = strlen(a->key_string) / 2;
2514                 a->key = kmalloc(a->key_size, GFP_KERNEL);
2515                 if (!a->key)
2516                         goto nomem;
2517                 if (hex2bin(a->key, a->key_string, a->key_size))
2518                         goto inval;
2519         }
2520
2521         return 0;
2522 inval:
2523         *error = error_inval;
2524         return -EINVAL;
2525 nomem:
2526         *error = "Out of memory for an argument";
2527         return -ENOMEM;
2528 }
2529
2530 static int get_mac(struct crypto_shash **hash, struct alg_spec *a, char **error,
2531                    char *error_alg, char *error_key)
2532 {
2533         int r;
2534
2535         if (a->alg_string) {
2536                 *hash = crypto_alloc_shash(a->alg_string, 0, CRYPTO_ALG_ASYNC);
2537                 if (IS_ERR(*hash)) {
2538                         *error = error_alg;
2539                         r = PTR_ERR(*hash);
2540                         *hash = NULL;
2541                         return r;
2542                 }
2543
2544                 if (a->key) {
2545                         r = crypto_shash_setkey(*hash, a->key, a->key_size);
2546                         if (r) {
2547                                 *error = error_key;
2548                                 return r;
2549                         }
2550                 }
2551         }
2552
2553         return 0;
2554 }
2555
2556 static int create_journal(struct dm_integrity_c *ic, char **error)
2557 {
2558         int r = 0;
2559         unsigned i;
2560         __u64 journal_pages, journal_desc_size, journal_tree_size;
2561         unsigned char *crypt_data = NULL, *crypt_iv = NULL;
2562         struct skcipher_request *req = NULL;
2563
2564         ic->commit_ids[0] = cpu_to_le64(0x1111111111111111ULL);
2565         ic->commit_ids[1] = cpu_to_le64(0x2222222222222222ULL);
2566         ic->commit_ids[2] = cpu_to_le64(0x3333333333333333ULL);
2567         ic->commit_ids[3] = cpu_to_le64(0x4444444444444444ULL);
2568
2569         journal_pages = roundup((__u64)ic->journal_sections * ic->journal_section_sectors,
2570                                 PAGE_SIZE >> SECTOR_SHIFT) >> (PAGE_SHIFT - SECTOR_SHIFT);
2571         journal_desc_size = journal_pages * sizeof(struct page_list);
2572         if (journal_pages >= totalram_pages - totalhigh_pages || journal_desc_size > ULONG_MAX) {
2573                 *error = "Journal doesn't fit into memory";
2574                 r = -ENOMEM;
2575                 goto bad;
2576         }
2577         ic->journal_pages = journal_pages;
2578
2579         ic->journal = dm_integrity_alloc_page_list(ic);
2580         if (!ic->journal) {
2581                 *error = "Could not allocate memory for journal";
2582                 r = -ENOMEM;
2583                 goto bad;
2584         }
2585         if (ic->journal_crypt_alg.alg_string) {
2586                 unsigned ivsize, blocksize;
2587                 struct journal_completion comp;
2588
2589                 comp.ic = ic;
2590                 ic->journal_crypt = crypto_alloc_skcipher(ic->journal_crypt_alg.alg_string, 0, 0);
2591                 if (IS_ERR(ic->journal_crypt)) {
2592                         *error = "Invalid journal cipher";
2593                         r = PTR_ERR(ic->journal_crypt);
2594                         ic->journal_crypt = NULL;
2595                         goto bad;
2596                 }
2597                 ivsize = crypto_skcipher_ivsize(ic->journal_crypt);
2598                 blocksize = crypto_skcipher_blocksize(ic->journal_crypt);
2599
2600                 if (ic->journal_crypt_alg.key) {
2601                         r = crypto_skcipher_setkey(ic->journal_crypt, ic->journal_crypt_alg.key,
2602                                                    ic->journal_crypt_alg.key_size);
2603                         if (r) {
2604                                 *error = "Error setting encryption key";
2605                                 goto bad;
2606                         }
2607                 }
2608                 DEBUG_print("cipher %s, block size %u iv size %u\n",
2609                             ic->journal_crypt_alg.alg_string, blocksize, ivsize);
2610
2611                 ic->journal_io = dm_integrity_alloc_page_list(ic);
2612                 if (!ic->journal_io) {
2613                         *error = "Could not allocate memory for journal io";
2614                         r = -ENOMEM;
2615                         goto bad;
2616                 }
2617
2618                 if (blocksize == 1) {
2619                         struct scatterlist *sg;
2620
2621                         req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
2622                         if (!req) {
2623                                 *error = "Could not allocate crypt request";
2624                                 r = -ENOMEM;
2625                                 goto bad;
2626                         }
2627
2628                         crypt_iv = kmalloc(ivsize, GFP_KERNEL);
2629                         if (!crypt_iv) {
2630                                 *error = "Could not allocate iv";
2631                                 r = -ENOMEM;
2632                                 goto bad;
2633                         }
2634
2635                         ic->journal_xor = dm_integrity_alloc_page_list(ic);
2636                         if (!ic->journal_xor) {
2637                                 *error = "Could not allocate memory for journal xor";
2638                                 r = -ENOMEM;
2639                                 goto bad;
2640                         }
2641
2642                         sg = kvmalloc((ic->journal_pages + 1) * sizeof(struct scatterlist), GFP_KERNEL);
2643                         if (!sg) {
2644                                 *error = "Unable to allocate sg list";
2645                                 r = -ENOMEM;
2646                                 goto bad;
2647                         }
2648                         sg_init_table(sg, ic->journal_pages + 1);
2649                         for (i = 0; i < ic->journal_pages; i++) {
2650                                 char *va = lowmem_page_address(ic->journal_xor[i].page);
2651                                 clear_page(va);
2652                                 sg_set_buf(&sg[i], va, PAGE_SIZE);
2653                         }
2654                         sg_set_buf(&sg[i], &ic->commit_ids, sizeof ic->commit_ids);
2655                         memset(crypt_iv, 0x00, ivsize);
2656
2657                         skcipher_request_set_crypt(req, sg, sg, PAGE_SIZE * ic->journal_pages + sizeof ic->commit_ids, crypt_iv);
2658                         init_completion(&comp.comp);
2659                         comp.in_flight = (atomic_t)ATOMIC_INIT(1);
2660                         if (do_crypt(true, req, &comp))
2661                                 wait_for_completion(&comp.comp);
2662                         kvfree(sg);
2663                         r = dm_integrity_failed(ic);
2664                         if (r) {
2665                                 *error = "Unable to encrypt journal";
2666                                 goto bad;
2667                         }
2668                         DEBUG_bytes(lowmem_page_address(ic->journal_xor[0].page), 64, "xor data");
2669
2670                         crypto_free_skcipher(ic->journal_crypt);
2671                         ic->journal_crypt = NULL;
2672                 } else {
2673                         unsigned crypt_len = roundup(ivsize, blocksize);
2674
2675                         req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
2676                         if (!req) {
2677                                 *error = "Could not allocate crypt request";
2678                                 r = -ENOMEM;
2679                                 goto bad;
2680                         }
2681
2682                         crypt_iv = kmalloc(ivsize, GFP_KERNEL);
2683                         if (!crypt_iv) {
2684                                 *error = "Could not allocate iv";
2685                                 r = -ENOMEM;
2686                                 goto bad;
2687                         }
2688
2689                         crypt_data = kmalloc(crypt_len, GFP_KERNEL);
2690                         if (!crypt_data) {
2691                                 *error = "Unable to allocate crypt data";
2692                                 r = -ENOMEM;
2693                                 goto bad;
2694                         }
2695
2696                         ic->journal_scatterlist = dm_integrity_alloc_journal_scatterlist(ic, ic->journal);
2697                         if (!ic->journal_scatterlist) {
2698                                 *error = "Unable to allocate sg list";
2699                                 r = -ENOMEM;
2700                                 goto bad;
2701                         }
2702                         ic->journal_io_scatterlist = dm_integrity_alloc_journal_scatterlist(ic, ic->journal_io);
2703                         if (!ic->journal_io_scatterlist) {
2704                                 *error = "Unable to allocate sg list";
2705                                 r = -ENOMEM;
2706                                 goto bad;
2707                         }
2708                         ic->sk_requests = kvmalloc(ic->journal_sections * sizeof(struct skcipher_request *), GFP_KERNEL | __GFP_ZERO);
2709                         if (!ic->sk_requests) {
2710                                 *error = "Unable to allocate sk requests";
2711                                 r = -ENOMEM;
2712                                 goto bad;
2713                         }
2714                         for (i = 0; i < ic->journal_sections; i++) {
2715                                 struct scatterlist sg;
2716                                 struct skcipher_request *section_req;
2717                                 __u32 section_le = cpu_to_le32(i);
2718
2719                                 memset(crypt_iv, 0x00, ivsize);
2720                                 memset(crypt_data, 0x00, crypt_len);
2721                                 memcpy(crypt_data, &section_le, min((size_t)crypt_len, sizeof(section_le)));
2722
2723                                 sg_init_one(&sg, crypt_data, crypt_len);
2724                                 skcipher_request_set_crypt(req, &sg, &sg, crypt_len, crypt_iv);
2725                                 init_completion(&comp.comp);
2726                                 comp.in_flight = (atomic_t)ATOMIC_INIT(1);
2727                                 if (do_crypt(true, req, &comp))
2728                                         wait_for_completion(&comp.comp);
2729
2730                                 r = dm_integrity_failed(ic);
2731                                 if (r) {
2732                                         *error = "Unable to generate iv";
2733                                         goto bad;
2734                                 }
2735
2736                                 section_req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
2737                                 if (!section_req) {
2738                                         *error = "Unable to allocate crypt request";
2739                                         r = -ENOMEM;
2740                                         goto bad;
2741                                 }
2742                                 section_req->iv = kmalloc(ivsize * 2, GFP_KERNEL);
2743                                 if (!section_req->iv) {
2744                                         skcipher_request_free(section_req);
2745                                         *error = "Unable to allocate iv";
2746                                         r = -ENOMEM;
2747                                         goto bad;
2748                                 }
2749                                 memcpy(section_req->iv + ivsize, crypt_data, ivsize);
2750                                 section_req->cryptlen = (size_t)ic->journal_section_sectors << SECTOR_SHIFT;
2751                                 ic->sk_requests[i] = section_req;
2752                                 DEBUG_bytes(crypt_data, ivsize, "iv(%u)", i);
2753                         }
2754                 }
2755         }
2756
2757         for (i = 0; i < N_COMMIT_IDS; i++) {
2758                 unsigned j;
2759 retest_commit_id:
2760                 for (j = 0; j < i; j++) {
2761                         if (ic->commit_ids[j] == ic->commit_ids[i]) {
2762                                 ic->commit_ids[i] = cpu_to_le64(le64_to_cpu(ic->commit_ids[i]) + 1);
2763                                 goto retest_commit_id;
2764                         }
2765                 }
2766                 DEBUG_print("commit id %u: %016llx\n", i, ic->commit_ids[i]);
2767         }
2768
2769         journal_tree_size = (__u64)ic->journal_entries * sizeof(struct journal_node);
2770         if (journal_tree_size > ULONG_MAX) {
2771                 *error = "Journal doesn't fit into memory";
2772                 r = -ENOMEM;
2773                 goto bad;
2774         }
2775         ic->journal_tree = kvmalloc(journal_tree_size, GFP_KERNEL);
2776         if (!ic->journal_tree) {
2777                 *error = "Could not allocate memory for journal tree";
2778                 r = -ENOMEM;
2779         }
2780 bad:
2781         kfree(crypt_data);
2782         kfree(crypt_iv);
2783         skcipher_request_free(req);
2784
2785         return r;
2786 }
2787
2788 /*
2789  * Construct a integrity mapping
2790  *
2791  * Arguments:
2792  *      device
2793  *      offset from the start of the device
2794  *      tag size
2795  *      D - direct writes, J - journal writes, R - recovery mode
2796  *      number of optional arguments
2797  *      optional arguments:
2798  *              journal_sectors
2799  *              interleave_sectors
2800  *              buffer_sectors
2801  *              journal_watermark
2802  *              commit_time
2803  *              internal_hash
2804  *              journal_crypt
2805  *              journal_mac
2806  *              block_size
2807  */
2808 static int dm_integrity_ctr(struct dm_target *ti, unsigned argc, char **argv)
2809 {
2810         struct dm_integrity_c *ic;
2811         char dummy;
2812         int r;
2813         unsigned extra_args;
2814         struct dm_arg_set as;
2815         static const struct dm_arg _args[] = {
2816                 {0, 9, "Invalid number of feature args"},
2817         };
2818         unsigned journal_sectors, interleave_sectors, buffer_sectors, journal_watermark, sync_msec;
2819         bool should_write_sb;
2820         __u64 threshold;
2821         unsigned long long start;
2822
2823 #define DIRECT_ARGUMENTS        4
2824
2825         if (argc <= DIRECT_ARGUMENTS) {
2826                 ti->error = "Invalid argument count";
2827                 return -EINVAL;
2828         }
2829
2830         ic = kzalloc(sizeof(struct dm_integrity_c), GFP_KERNEL);
2831         if (!ic) {
2832                 ti->error = "Cannot allocate integrity context";
2833                 return -ENOMEM;
2834         }
2835         ti->private = ic;
2836         ti->per_io_data_size = sizeof(struct dm_integrity_io);
2837
2838         ic->in_progress = RB_ROOT;
2839         init_waitqueue_head(&ic->endio_wait);
2840         bio_list_init(&ic->flush_bio_list);
2841         init_waitqueue_head(&ic->copy_to_journal_wait);
2842         init_completion(&ic->crypto_backoff);
2843         atomic64_set(&ic->number_of_mismatches, 0);
2844
2845         r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &ic->dev);
2846         if (r) {
2847                 ti->error = "Device lookup failed";
2848                 goto bad;
2849         }
2850
2851         if (sscanf(argv[1], "%llu%c", &start, &dummy) != 1 || start != (sector_t)start) {
2852                 ti->error = "Invalid starting offset";
2853                 r = -EINVAL;
2854                 goto bad;
2855         }
2856         ic->start = start;
2857
2858         if (strcmp(argv[2], "-")) {
2859                 if (sscanf(argv[2], "%u%c", &ic->tag_size, &dummy) != 1 || !ic->tag_size) {
2860                         ti->error = "Invalid tag size";
2861                         r = -EINVAL;
2862                         goto bad;
2863                 }
2864         }
2865
2866         if (!strcmp(argv[3], "J") || !strcmp(argv[3], "D") || !strcmp(argv[3], "R"))
2867                 ic->mode = argv[3][0];
2868         else {
2869                 ti->error = "Invalid mode (expecting J, D, R)";
2870                 r = -EINVAL;
2871                 goto bad;
2872         }
2873
2874         ic->device_sectors = i_size_read(ic->dev->bdev->bd_inode) >> SECTOR_SHIFT;
2875         journal_sectors = min((sector_t)DEFAULT_MAX_JOURNAL_SECTORS,
2876                         ic->device_sectors >> DEFAULT_JOURNAL_SIZE_FACTOR);
2877         interleave_sectors = DEFAULT_INTERLEAVE_SECTORS;
2878         buffer_sectors = DEFAULT_BUFFER_SECTORS;
2879         journal_watermark = DEFAULT_JOURNAL_WATERMARK;
2880         sync_msec = DEFAULT_SYNC_MSEC;
2881         ic->sectors_per_block = 1;
2882
2883         as.argc = argc - DIRECT_ARGUMENTS;
2884         as.argv = argv + DIRECT_ARGUMENTS;
2885         r = dm_read_arg_group(_args, &as, &extra_args, &ti->error);
2886         if (r)
2887                 goto bad;
2888
2889         while (extra_args--) {
2890                 const char *opt_string;
2891                 unsigned val;
2892                 opt_string = dm_shift_arg(&as);
2893                 if (!opt_string) {
2894                         r = -EINVAL;
2895                         ti->error = "Not enough feature arguments";
2896                         goto bad;
2897                 }
2898                 if (sscanf(opt_string, "journal_sectors:%u%c", &val, &dummy) == 1)
2899                         journal_sectors = val;
2900                 else if (sscanf(opt_string, "interleave_sectors:%u%c", &val, &dummy) == 1)
2901                         interleave_sectors = val;
2902                 else if (sscanf(opt_string, "buffer_sectors:%u%c", &val, &dummy) == 1)
2903                         buffer_sectors = val;
2904                 else if (sscanf(opt_string, "journal_watermark:%u%c", &val, &dummy) == 1 && val <= 100)
2905                         journal_watermark = val;
2906                 else if (sscanf(opt_string, "commit_time:%u%c", &val, &dummy) == 1)
2907                         sync_msec = val;
2908                 else if (sscanf(opt_string, "block_size:%u%c", &val, &dummy) == 1) {
2909                         if (val < 1 << SECTOR_SHIFT ||
2910                             val > MAX_SECTORS_PER_BLOCK << SECTOR_SHIFT ||
2911                             (val & (val -1))) {
2912                                 r = -EINVAL;
2913                                 ti->error = "Invalid block_size argument";
2914                                 goto bad;
2915                         }
2916                         ic->sectors_per_block = val >> SECTOR_SHIFT;
2917                 } else if (!memcmp(opt_string, "internal_hash:", strlen("internal_hash:"))) {
2918                         r = get_alg_and_key(opt_string, &ic->internal_hash_alg, &ti->error,
2919                                             "Invalid internal_hash argument");
2920                         if (r)
2921                                 goto bad;
2922                 } else if (!memcmp(opt_string, "journal_crypt:", strlen("journal_crypt:"))) {
2923                         r = get_alg_and_key(opt_string, &ic->journal_crypt_alg, &ti->error,
2924                                             "Invalid journal_crypt argument");
2925                         if (r)
2926                                 goto bad;
2927                 } else if (!memcmp(opt_string, "journal_mac:", strlen("journal_mac:"))) {
2928                         r = get_alg_and_key(opt_string, &ic->journal_mac_alg,  &ti->error,
2929                                             "Invalid journal_mac argument");
2930                         if (r)
2931                                 goto bad;
2932                 } else {
2933                         r = -EINVAL;
2934                         ti->error = "Invalid argument";
2935                         goto bad;
2936                 }
2937         }
2938
2939         r = get_mac(&ic->internal_hash, &ic->internal_hash_alg, &ti->error,
2940                     "Invalid internal hash", "Error setting internal hash key");
2941         if (r)
2942                 goto bad;
2943
2944         r = get_mac(&ic->journal_mac, &ic->journal_mac_alg, &ti->error,
2945                     "Invalid journal mac", "Error setting journal mac key");
2946         if (r)
2947                 goto bad;
2948
2949         if (!ic->tag_size) {
2950                 if (!ic->internal_hash) {
2951                         ti->error = "Unknown tag size";
2952                         r = -EINVAL;
2953                         goto bad;
2954                 }
2955                 ic->tag_size = crypto_shash_digestsize(ic->internal_hash);
2956         }
2957         if (ic->tag_size > MAX_TAG_SIZE) {
2958                 ti->error = "Too big tag size";
2959                 r = -EINVAL;
2960                 goto bad;
2961         }
2962         if (!(ic->tag_size & (ic->tag_size - 1)))
2963                 ic->log2_tag_size = __ffs(ic->tag_size);
2964         else
2965                 ic->log2_tag_size = -1;
2966
2967         ic->autocommit_jiffies = msecs_to_jiffies(sync_msec);
2968         ic->autocommit_msec = sync_msec;
2969         setup_timer(&ic->autocommit_timer, autocommit_fn, (unsigned long)ic);
2970
2971         ic->io = dm_io_client_create();
2972         if (IS_ERR(ic->io)) {
2973                 r = PTR_ERR(ic->io);
2974                 ic->io = NULL;
2975                 ti->error = "Cannot allocate dm io";
2976                 goto bad;
2977         }
2978
2979         ic->journal_io_mempool = mempool_create_slab_pool(JOURNAL_IO_MEMPOOL, journal_io_cache);
2980         if (!ic->journal_io_mempool) {
2981                 r = -ENOMEM;
2982                 ti->error = "Cannot allocate mempool";
2983                 goto bad;
2984         }
2985
2986         ic->metadata_wq = alloc_workqueue("dm-integrity-metadata",
2987                                           WQ_MEM_RECLAIM, METADATA_WORKQUEUE_MAX_ACTIVE);
2988         if (!ic->metadata_wq) {
2989                 ti->error = "Cannot allocate workqueue";
2990                 r = -ENOMEM;
2991                 goto bad;
2992         }
2993
2994         /*
2995          * If this workqueue were percpu, it would cause bio reordering
2996          * and reduced performance.
2997          */
2998         ic->wait_wq = alloc_workqueue("dm-integrity-wait", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
2999         if (!ic->wait_wq) {
3000                 ti->error = "Cannot allocate workqueue";
3001                 r = -ENOMEM;
3002                 goto bad;
3003         }
3004
3005         ic->commit_wq = alloc_workqueue("dm-integrity-commit", WQ_MEM_RECLAIM, 1);
3006         if (!ic->commit_wq) {
3007                 ti->error = "Cannot allocate workqueue";
3008                 r = -ENOMEM;
3009                 goto bad;
3010         }
3011         INIT_WORK(&ic->commit_work, integrity_commit);
3012
3013         if (ic->mode == 'J') {
3014                 ic->writer_wq = alloc_workqueue("dm-integrity-writer", WQ_MEM_RECLAIM, 1);
3015                 if (!ic->writer_wq) {
3016                         ti->error = "Cannot allocate workqueue";
3017                         r = -ENOMEM;
3018                         goto bad;
3019                 }
3020                 INIT_WORK(&ic->writer_work, integrity_writer);
3021         }
3022
3023         ic->sb = alloc_pages_exact(SB_SECTORS << SECTOR_SHIFT, GFP_KERNEL);
3024         if (!ic->sb) {
3025                 r = -ENOMEM;
3026                 ti->error = "Cannot allocate superblock area";
3027                 goto bad;
3028         }
3029
3030         r = sync_rw_sb(ic, REQ_OP_READ, 0);
3031         if (r) {
3032                 ti->error = "Error reading superblock";
3033                 goto bad;
3034         }
3035         should_write_sb = false;
3036         if (memcmp(ic->sb->magic, SB_MAGIC, 8)) {
3037                 if (ic->mode != 'R') {
3038                         if (memchr_inv(ic->sb, 0, SB_SECTORS << SECTOR_SHIFT)) {
3039                                 r = -EINVAL;
3040                                 ti->error = "The device is not initialized";
3041                                 goto bad;
3042                         }
3043                 }
3044
3045                 r = initialize_superblock(ic, journal_sectors, interleave_sectors);
3046                 if (r) {
3047                         ti->error = "Could not initialize superblock";
3048                         goto bad;
3049                 }
3050                 if (ic->mode != 'R')
3051                         should_write_sb = true;
3052         }
3053
3054         if (ic->sb->version != SB_VERSION) {
3055                 r = -EINVAL;
3056                 ti->error = "Unknown version";
3057                 goto bad;
3058         }
3059         if (le16_to_cpu(ic->sb->integrity_tag_size) != ic->tag_size) {
3060                 r = -EINVAL;
3061                 ti->error = "Tag size doesn't match the information in superblock";
3062                 goto bad;
3063         }
3064         if (ic->sb->log2_sectors_per_block != __ffs(ic->sectors_per_block)) {
3065                 r = -EINVAL;
3066                 ti->error = "Block size doesn't match the information in superblock";
3067                 goto bad;
3068         }
3069         if (!le32_to_cpu(ic->sb->journal_sections)) {
3070                 r = -EINVAL;
3071                 ti->error = "Corrupted superblock, journal_sections is 0";
3072                 goto bad;
3073         }
3074         /* make sure that ti->max_io_len doesn't overflow */
3075         if (ic->sb->log2_interleave_sectors < MIN_LOG2_INTERLEAVE_SECTORS ||
3076             ic->sb->log2_interleave_sectors > MAX_LOG2_INTERLEAVE_SECTORS) {
3077                 r = -EINVAL;
3078                 ti->error = "Invalid interleave_sectors in the superblock";
3079                 goto bad;
3080         }
3081         ic->provided_data_sectors = le64_to_cpu(ic->sb->provided_data_sectors);
3082         if (ic->provided_data_sectors != le64_to_cpu(ic->sb->provided_data_sectors)) {
3083                 /* test for overflow */
3084                 r = -EINVAL;
3085                 ti->error = "The superblock has 64-bit device size, but the kernel was compiled with 32-bit sectors";
3086                 goto bad;
3087         }
3088         if (!!(ic->sb->flags & cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC)) != !!ic->journal_mac_alg.alg_string) {
3089                 r = -EINVAL;
3090                 ti->error = "Journal mac mismatch";
3091                 goto bad;
3092         }
3093         r = calculate_device_limits(ic);
3094         if (r) {
3095                 ti->error = "The device is too small";
3096                 goto bad;
3097         }
3098         if (ti->len > ic->provided_data_sectors) {
3099                 r = -EINVAL;
3100                 ti->error = "Not enough provided sectors for requested mapping size";
3101                 goto bad;
3102         }
3103
3104         if (!buffer_sectors)
3105                 buffer_sectors = 1;
3106         ic->log2_buffer_sectors = min3((int)__fls(buffer_sectors), (int)__ffs(ic->metadata_run), 31 - SECTOR_SHIFT);
3107
3108         threshold = (__u64)ic->journal_entries * (100 - journal_watermark);
3109         threshold += 50;
3110         do_div(threshold, 100);
3111         ic->free_sectors_threshold = threshold;
3112
3113         DEBUG_print("initialized:\n");
3114         DEBUG_print("   integrity_tag_size %u\n", le16_to_cpu(ic->sb->integrity_tag_size));
3115         DEBUG_print("   journal_entry_size %u\n", ic->journal_entry_size);
3116         DEBUG_print("   journal_entries_per_sector %u\n", ic->journal_entries_per_sector);
3117         DEBUG_print("   journal_section_entries %u\n", ic->journal_section_entries);
3118         DEBUG_print("   journal_section_sectors %u\n", ic->journal_section_sectors);
3119         DEBUG_print("   journal_sections %u\n", (unsigned)le32_to_cpu(ic->sb->journal_sections));
3120         DEBUG_print("   journal_entries %u\n", ic->journal_entries);
3121         DEBUG_print("   log2_interleave_sectors %d\n", ic->sb->log2_interleave_sectors);
3122         DEBUG_print("   device_sectors 0x%llx\n", (unsigned long long)ic->device_sectors);
3123         DEBUG_print("   initial_sectors 0x%x\n", ic->initial_sectors);
3124         DEBUG_print("   metadata_run 0x%x\n", ic->metadata_run);
3125         DEBUG_print("   log2_metadata_run %d\n", ic->log2_metadata_run);
3126         DEBUG_print("   provided_data_sectors 0x%llx (%llu)\n", (unsigned long long)ic->provided_data_sectors,
3127                     (unsigned long long)ic->provided_data_sectors);
3128         DEBUG_print("   log2_buffer_sectors %u\n", ic->log2_buffer_sectors);
3129
3130         ic->bufio = dm_bufio_client_create(ic->dev->bdev, 1U << (SECTOR_SHIFT + ic->log2_buffer_sectors),
3131                                            1, 0, NULL, NULL);
3132         if (IS_ERR(ic->bufio)) {
3133                 r = PTR_ERR(ic->bufio);
3134                 ti->error = "Cannot initialize dm-bufio";
3135                 ic->bufio = NULL;
3136                 goto bad;
3137         }
3138         dm_bufio_set_sector_offset(ic->bufio, ic->start + ic->initial_sectors);
3139
3140         if (ic->mode != 'R') {
3141                 r = create_journal(ic, &ti->error);
3142                 if (r)
3143                         goto bad;
3144         }
3145
3146         if (should_write_sb) {
3147                 int r;
3148
3149                 init_journal(ic, 0, ic->journal_sections, 0);
3150                 r = dm_integrity_failed(ic);
3151                 if (unlikely(r)) {
3152                         ti->error = "Error initializing journal";
3153                         goto bad;
3154                 }
3155                 r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
3156                 if (r) {
3157                         ti->error = "Error initializing superblock";
3158                         goto bad;
3159                 }
3160                 ic->just_formatted = true;
3161         }
3162
3163         r = dm_set_target_max_io_len(ti, 1U << ic->sb->log2_interleave_sectors);
3164         if (r)
3165                 goto bad;
3166
3167         if (!ic->internal_hash)
3168                 dm_integrity_set(ti, ic);
3169
3170         ti->num_flush_bios = 1;
3171         ti->flush_supported = true;
3172
3173         return 0;
3174 bad:
3175         dm_integrity_dtr(ti);
3176         return r;
3177 }
3178
3179 static void dm_integrity_dtr(struct dm_target *ti)
3180 {
3181         struct dm_integrity_c *ic = ti->private;
3182
3183         BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
3184
3185         if (ic->metadata_wq)
3186                 destroy_workqueue(ic->metadata_wq);
3187         if (ic->wait_wq)
3188                 destroy_workqueue(ic->wait_wq);
3189         if (ic->commit_wq)
3190                 destroy_workqueue(ic->commit_wq);
3191         if (ic->writer_wq)
3192                 destroy_workqueue(ic->writer_wq);
3193         if (ic->bufio)
3194                 dm_bufio_client_destroy(ic->bufio);
3195         mempool_destroy(ic->journal_io_mempool);
3196         if (ic->io)
3197                 dm_io_client_destroy(ic->io);
3198         if (ic->dev)
3199                 dm_put_device(ti, ic->dev);
3200         dm_integrity_free_page_list(ic, ic->journal);
3201         dm_integrity_free_page_list(ic, ic->journal_io);
3202         dm_integrity_free_page_list(ic, ic->journal_xor);
3203         if (ic->journal_scatterlist)
3204                 dm_integrity_free_journal_scatterlist(ic, ic->journal_scatterlist);
3205         if (ic->journal_io_scatterlist)
3206                 dm_integrity_free_journal_scatterlist(ic, ic->journal_io_scatterlist);
3207         if (ic->sk_requests) {
3208                 unsigned i;
3209
3210                 for (i = 0; i < ic->journal_sections; i++) {
3211                         struct skcipher_request *req = ic->sk_requests[i];
3212                         if (req) {
3213                                 kzfree(req->iv);
3214                                 skcipher_request_free(req);
3215                         }
3216                 }
3217                 kvfree(ic->sk_requests);
3218         }
3219         kvfree(ic->journal_tree);
3220         if (ic->sb)
3221                 free_pages_exact(ic->sb, SB_SECTORS << SECTOR_SHIFT);
3222
3223         if (ic->internal_hash)
3224                 crypto_free_shash(ic->internal_hash);
3225         free_alg(&ic->internal_hash_alg);
3226
3227         if (ic->journal_crypt)
3228                 crypto_free_skcipher(ic->journal_crypt);
3229         free_alg(&ic->journal_crypt_alg);
3230
3231         if (ic->journal_mac)
3232                 crypto_free_shash(ic->journal_mac);
3233         free_alg(&ic->journal_mac_alg);
3234
3235         kfree(ic);
3236 }
3237
3238 static struct target_type integrity_target = {
3239         .name                   = "integrity",
3240         .version                = {1, 1, 0},
3241         .module                 = THIS_MODULE,
3242         .features               = DM_TARGET_SINGLETON | DM_TARGET_INTEGRITY,
3243         .ctr                    = dm_integrity_ctr,
3244         .dtr                    = dm_integrity_dtr,
3245         .map                    = dm_integrity_map,
3246         .postsuspend            = dm_integrity_postsuspend,
3247         .resume                 = dm_integrity_resume,
3248         .status                 = dm_integrity_status,
3249         .iterate_devices        = dm_integrity_iterate_devices,
3250         .io_hints               = dm_integrity_io_hints,
3251 };
3252
3253 int __init dm_integrity_init(void)
3254 {
3255         int r;
3256
3257         journal_io_cache = kmem_cache_create("integrity_journal_io",
3258                                              sizeof(struct journal_io), 0, 0, NULL);
3259         if (!journal_io_cache) {
3260                 DMERR("can't allocate journal io cache");
3261                 return -ENOMEM;
3262         }
3263
3264         r = dm_register_target(&integrity_target);
3265
3266         if (r < 0)
3267                 DMERR("register failed %d", r);
3268
3269         return r;
3270 }
3271
3272 void dm_integrity_exit(void)
3273 {
3274         dm_unregister_target(&integrity_target);
3275         kmem_cache_destroy(journal_io_cache);
3276 }
3277
3278 module_init(dm_integrity_init);
3279 module_exit(dm_integrity_exit);
3280
3281 MODULE_AUTHOR("Milan Broz");
3282 MODULE_AUTHOR("Mikulas Patocka");
3283 MODULE_DESCRIPTION(DM_NAME " target for integrity tags extension");
3284 MODULE_LICENSE("GPL");