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