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
  49 struct superblock {
  50         __u8 magic[8];
  51         __u8 version;
  52         __u8 log2_interleave_sectors;
  53         __u16 integrity_tag_size;
  54         __u32 journal_sections;
  55         __u64 provided_data_sectors;    /* userspace uses this value */
  56         __u32 flags;
  57 };
  58
  59 #define SB_FLAG_HAVE_JOURNAL_MAC        0x1
  60
  61 #define JOURNAL_ENTRY_ROUNDUP           8
  62
  63 typedef __u64 commit_id_t;
  64 #define JOURNAL_MAC_PER_SECTOR          8
  65
  66 struct journal_entry {
  67         union {
  68                 struct {
  69                         __u32 sector_lo;
  70                         __u32 sector_hi;
  71                 } s;
  72                 __u64 sector;
  73         } u;
  74         commit_id_t last_bytes;
  75         __u8 tag[0];
  76 };
  77
  78 #if BITS_PER_LONG == 64
  79 #define journal_entry_set_sector(je, x)         do { smp_wmb(); ACCESS_ONCE((je)->u.sector) = cpu_to_le64(x); } while (0)
  80 #define journal_entry_get_sector(je)            le64_to_cpu((je)->u.sector)
  81 #elif defined(CONFIG_LBDAF)
  82 #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)
  83 #define journal_entry_get_sector(je)            le64_to_cpu((je)->u.sector)
  84 #else
  85 #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)
  86 #define journal_entry_get_sector(je)            le32_to_cpu((je)->u.s.sector_lo)
  87 #endif
  88 #define journal_entry_is_unused(je)             ((je)->u.s.sector_hi == cpu_to_le32(-1))
  89 #define journal_entry_set_unused(je)            do { ((je)->u.s.sector_hi = cpu_to_le32(-1)); } while (0)
  90 #define journal_entry_is_inprogress(je)         ((je)->u.s.sector_hi == cpu_to_le32(-2))
  91 #define journal_entry_set_inprogress(je)        do { ((je)->u.s.sector_hi = cpu_to_le32(-2)); } while (0)
  92
  93 #define JOURNAL_BLOCK_SECTORS           8
  94 #define JOURNAL_SECTOR_DATA             ((1 << SECTOR_SHIFT) - sizeof(commit_id_t))
  95 #define JOURNAL_MAC_SIZE                (JOURNAL_MAC_PER_SECTOR * JOURNAL_BLOCK_SECTORS)
  96
  97 struct journal_sector {
  98         __u8 entries[JOURNAL_SECTOR_DATA - JOURNAL_MAC_PER_SECTOR];
  99         __u8 mac[JOURNAL_MAC_PER_SECTOR];
 100         commit_id_t commit_id;
 101 };
 102
 103 #define MAX_TAG_SIZE                    (JOURNAL_SECTOR_DATA - JOURNAL_MAC_PER_SECTOR - offsetof(struct journal_entry, tag))
 104
 105 #define METADATA_PADDING_SECTORS        8
 106
 107 #define N_COMMIT_IDS                    4
 108
 109 static unsigned char prev_commit_seq(unsigned char seq)
 110 {
 111         return (seq + N_COMMIT_IDS - 1) % N_COMMIT_IDS;
 112 }
 113
 114 static unsigned char next_commit_seq(unsigned char seq)
 115 {
 116         return (seq + 1) % N_COMMIT_IDS;
 117 }
 118
 119 /*
 120  * In-memory structures
 121  */
 122
 123 struct journal_node {
 124         struct rb_node node;
 125         sector_t sector;
 126 };
 127
 128 struct alg_spec {
 129         char *alg_string;
 130         char *key_string;
 131         __u8 *key;
 132         unsigned key_size;
 133 };
 134
 135 struct dm_integrity_c {
 136         struct dm_dev *dev;
 137         unsigned tag_size;
 138         __s8 log2_tag_size;
 139         sector_t start;
 140         mempool_t *journal_io_mempool;
 141         struct dm_io_client *io;
 142         struct dm_bufio_client *bufio;
 143         struct workqueue_struct *metadata_wq;
 144         struct superblock *sb;
 145         unsigned journal_pages;
 146         struct page_list *journal;
 147         struct page_list *journal_io;
 148         struct page_list *journal_xor;
 149
 150         struct crypto_skcipher *journal_crypt;
 151         struct scatterlist **journal_scatterlist;
 152         struct scatterlist **journal_io_scatterlist;
 153         struct skcipher_request **sk_requests;
 154
 155         struct crypto_shash *journal_mac;
 156
 157         struct journal_node *journal_tree;
 158         struct rb_root journal_tree_root;
 159
 160         sector_t provided_data_sectors;
 161
 162         unsigned short journal_entry_size;
 163         unsigned char journal_entries_per_sector;
 164         unsigned char journal_section_entries;
 165         unsigned char journal_section_sectors;
 166         unsigned journal_sections;
 167         unsigned journal_entries;
 168         sector_t device_sectors;
 169         unsigned initial_sectors;
 170         unsigned metadata_run;
 171         __s8 log2_metadata_run;
 172         __u8 log2_buffer_sectors;
 173
 174         unsigned char mode;
 175         bool suspending;
 176
 177         int failed;
 178
 179         struct crypto_shash *internal_hash;
 180
 181         /* these variables are locked with endio_wait.lock */
 182         struct rb_root in_progress;
 183         wait_queue_head_t endio_wait;
 184         struct workqueue_struct *wait_wq;
 185
 186         unsigned char commit_seq;
 187         commit_id_t commit_ids[N_COMMIT_IDS];
 188
 189         unsigned committed_section;
 190         unsigned n_committed_sections;
 191
 192         unsigned uncommitted_section;
 193         unsigned n_uncommitted_sections;
 194
 195         unsigned free_section;
 196         unsigned char free_section_entry;
 197         unsigned free_sectors;
 198
 199         unsigned free_sectors_threshold;
 200
 201         struct workqueue_struct *commit_wq;
 202         struct work_struct commit_work;
 203
 204         struct workqueue_struct *writer_wq;
 205         struct work_struct writer_work;
 206
 207         struct bio_list flush_bio_list;
 208
 209         unsigned long autocommit_jiffies;
 210         struct timer_list autocommit_timer;
 211         unsigned autocommit_msec;
 212
 213         wait_queue_head_t copy_to_journal_wait;
 214
 215         struct completion crypto_backoff;
 216
 217         bool journal_uptodate;
 218         bool just_formatted;
 219
 220         struct alg_spec internal_hash_alg;
 221         struct alg_spec journal_crypt_alg;
 222         struct alg_spec journal_mac_alg;
 223 };
 224
 225 struct dm_integrity_range {
 226         sector_t logical_sector;
 227         unsigned n_sectors;
 228         struct rb_node node;
 229 };
 230
 231 struct dm_integrity_io {
 232         struct work_struct work;
 233
 234         struct dm_integrity_c *ic;
 235         bool write;
 236         bool fua;
 237
 238         struct dm_integrity_range range;
 239
 240         sector_t metadata_block;
 241         unsigned metadata_offset;
 242
 243         atomic_t in_flight;
 244         int bi_error;
 245
 246         struct completion *completion;
 247
 248         struct block_device *orig_bi_bdev;
 249         bio_end_io_t *orig_bi_end_io;
 250         struct bio_integrity_payload *orig_bi_integrity;
 251         struct bvec_iter orig_bi_iter;
 252 };
 253
 254 struct journal_completion {
 255         struct dm_integrity_c *ic;
 256         atomic_t in_flight;
 257         struct completion comp;
 258 };
 259
 260 struct journal_io {
 261         struct dm_integrity_range range;
 262         struct journal_completion *comp;
 263 };
 264
 265 static struct kmem_cache *journal_io_cache;
 266
 267 #define JOURNAL_IO_MEMPOOL      32
 268
 269 #ifdef DEBUG_PRINT
 270 #define DEBUG_print(x, ...)     printk(KERN_DEBUG x, ##__VA_ARGS__)
 271 static void __DEBUG_bytes(__u8 *bytes, size_t len, const char *msg, ...)
 272 {
 273         va_list args;
 274         va_start(args, msg);
 275         vprintk(msg, args);
 276         va_end(args);
 277         if (len)
 278                 pr_cont(":");
 279         while (len) {
 280                 pr_cont(" %02x", *bytes);
 281                 bytes++;
 282                 len--;
 283         }
 284         pr_cont("\n");
 285 }
 286 #define DEBUG_bytes(bytes, len, msg, ...)       __DEBUG_bytes(bytes, len, KERN_DEBUG msg, ##__VA_ARGS__)
 287 #else
 288 #define DEBUG_print(x, ...)                     do { } while (0)
 289 #define DEBUG_bytes(bytes, len, msg, ...)       do { } while (0)
 290 #endif
 291
 292 /*
 293  * DM Integrity profile, protection is performed layer above (dm-crypt)
 294  */
 295 static struct blk_integrity_profile dm_integrity_profile = {
 296         .name                   = "DM-DIF-EXT-TAG",
 297         .generate_fn            = NULL,
 298         .verify_fn              = NULL,
 299 };
 300
 301 static void dm_integrity_map_continue(struct dm_integrity_io *dio, bool from_map);
 302 static void integrity_bio_wait(struct work_struct *w);
 303 static void dm_integrity_dtr(struct dm_target *ti);
 304
 305 static void dm_integrity_io_error(struct dm_integrity_c *ic, const char *msg, int err)
 306 {
 307         if (!cmpxchg(&ic->failed, 0, err))
 308                 DMERR("Error on %s: %d", msg, err);
 309 }
 310
 311 static int dm_integrity_failed(struct dm_integrity_c *ic)
 312 {
 313         return ACCESS_ONCE(ic->failed);
 314 }
 315
 316 static commit_id_t dm_integrity_commit_id(struct dm_integrity_c *ic, unsigned i,
 317                                           unsigned j, unsigned char seq)
 318 {
 319         /*
 320          * Xor the number with section and sector, so that if a piece of
 321          * journal is written at wrong place, it is detected.
 322          */
 323         return ic->commit_ids[seq] ^ cpu_to_le64(((__u64)i << 32) ^ j);
 324 }
 325
 326 static void get_area_and_offset(struct dm_integrity_c *ic, sector_t data_sector,
 327                                 sector_t *area, sector_t *offset)
 328 {
 329         __u8 log2_interleave_sectors = ic->sb->log2_interleave_sectors;
 330
 331         *area = data_sector >> log2_interleave_sectors;
 332         *offset = (unsigned)data_sector & ((1U << log2_interleave_sectors) - 1);
 333 }
 334
 335 static __u64 get_metadata_sector_and_offset(struct dm_integrity_c *ic, sector_t area,
 336                                             sector_t offset, unsigned *metadata_offset)
 337 {
 338         __u64 ms;
 339         unsigned mo;
 340
 341         ms = area << ic->sb->log2_interleave_sectors;
 342         if (likely(ic->log2_metadata_run >= 0))
 343                 ms += area << ic->log2_metadata_run;
 344         else
 345                 ms += area * ic->metadata_run;
 346         ms >>= ic->log2_buffer_sectors;
 347
 348         if (likely(ic->log2_tag_size >= 0)) {
 349                 ms += offset >> (SECTOR_SHIFT + ic->log2_buffer_sectors - ic->log2_tag_size);
 350                 mo = (offset << ic->log2_tag_size) & ((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - 1);
 351         } else {
 352                 ms += (__u64)offset * ic->tag_size >> (SECTOR_SHIFT + ic->log2_buffer_sectors);
 353                 mo = (offset * ic->tag_size) & ((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - 1);
 354         }
 355         *metadata_offset = mo;
 356         return ms;
 357 }
 358
 359 static sector_t get_data_sector(struct dm_integrity_c *ic, sector_t area, sector_t offset)
 360 {
 361         sector_t result;
 362
 363         result = area << ic->sb->log2_interleave_sectors;
 364         if (likely(ic->log2_metadata_run >= 0))
 365                 result += (area + 1) << ic->log2_metadata_run;
 366         else
 367                 result += (area + 1) * ic->metadata_run;
 368
 369         result += (sector_t)ic->initial_sectors + offset;
 370         return result;
 371 }
 372
 373 static void wraparound_section(struct dm_integrity_c *ic, unsigned *sec_ptr)
 374 {
 375         if (unlikely(*sec_ptr >= ic->journal_sections))
 376                 *sec_ptr -= ic->journal_sections;
 377 }
 378
 379 static int sync_rw_sb(struct dm_integrity_c *ic, int op, int op_flags)
 380 {
 381         struct dm_io_request io_req;
 382         struct dm_io_region io_loc;
 383
 384         io_req.bi_op = op;
 385         io_req.bi_op_flags = op_flags;
 386         io_req.mem.type = DM_IO_KMEM;
 387         io_req.mem.ptr.addr = ic->sb;
 388         io_req.notify.fn = NULL;
 389         io_req.client = ic->io;
 390         io_loc.bdev = ic->dev->bdev;
 391         io_loc.sector = ic->start;
 392         io_loc.count = SB_SECTORS;
 393
 394         return dm_io(&io_req, 1, &io_loc, NULL);
 395 }
 396
 397 static void access_journal_check(struct dm_integrity_c *ic, unsigned section, unsigned offset,
 398                                  bool e, const char *function)
 399 {
 400 #if defined(CONFIG_DM_DEBUG) || defined(INTERNAL_VERIFY)
 401         unsigned limit = e ? ic->journal_section_entries : ic->journal_section_sectors;
 402
 403         if (unlikely(section >= ic->journal_sections) ||
 404             unlikely(offset >= limit)) {
 405                 printk(KERN_CRIT "%s: invalid access at (%u,%u), limit (%u,%u)\n",
 406                         function, section, offset, ic->journal_sections, limit);
 407                 BUG();
 408         }
 409 #endif
 410 }
 411
 412 static void page_list_location(struct dm_integrity_c *ic, unsigned section, unsigned offset,
 413                                unsigned *pl_index, unsigned *pl_offset)
 414 {
 415         unsigned sector;
 416
 417         access_journal_check(ic, section, offset, false, "page_list_location");
 418
 419         sector = section * ic->journal_section_sectors + offset;
 420
 421         *pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
 422         *pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
 423 }
 424
 425 static struct journal_sector *access_page_list(struct dm_integrity_c *ic, struct page_list *pl,
 426                                                unsigned section, unsigned offset, unsigned *n_sectors)
 427 {
 428         unsigned pl_index, pl_offset;
 429         char *va;
 430
 431         page_list_location(ic, section, offset, &pl_index, &pl_offset);
 432
 433         if (n_sectors)
 434                 *n_sectors = (PAGE_SIZE - pl_offset) >> SECTOR_SHIFT;
 435
 436         va = lowmem_page_address(pl[pl_index].page);
 437
 438         return (struct journal_sector *)(va + pl_offset);
 439 }
 440
 441 static struct journal_sector *access_journal(struct dm_integrity_c *ic, unsigned section, unsigned offset)
 442 {
 443         return access_page_list(ic, ic->journal, section, offset, NULL);
 444 }
 445
 446 static struct journal_entry *access_journal_entry(struct dm_integrity_c *ic, unsigned section, unsigned n)
 447 {
 448         unsigned rel_sector, offset;
 449         struct journal_sector *js;
 450
 451         access_journal_check(ic, section, n, true, "access_journal_entry");
 452
 453         rel_sector = n % JOURNAL_BLOCK_SECTORS;
 454         offset = n / JOURNAL_BLOCK_SECTORS;
 455
 456         js = access_journal(ic, section, rel_sector);
 457         return (struct journal_entry *)((char *)js + offset * ic->journal_entry_size);
 458 }
 459
 460 static struct journal_sector *access_journal_data(struct dm_integrity_c *ic, unsigned section, unsigned n)
 461 {
 462         access_journal_check(ic, section, n, true, "access_journal_data");
 463
 464         return access_journal(ic, section, n + JOURNAL_BLOCK_SECTORS);
 465 }
 466
 467 static void section_mac(struct dm_integrity_c *ic, unsigned section, __u8 result[JOURNAL_MAC_SIZE])
 468 {
 469         SHASH_DESC_ON_STACK(desc, ic->journal_mac);
 470         int r;
 471         unsigned j, size;
 472
 473         desc->tfm = ic->journal_mac;
 474         desc->flags = CRYPTO_TFM_REQ_MAY_SLEEP;
 475
 476         r = crypto_shash_init(desc);
 477         if (unlikely(r)) {
 478                 dm_integrity_io_error(ic, "crypto_shash_init", r);
 479                 goto err;
 480         }
 481
 482         for (j = 0; j < ic->journal_section_entries; j++) {
 483                 struct journal_entry *je = access_journal_entry(ic, section, j);
 484                 r = crypto_shash_update(desc, (__u8 *)&je->u.sector, sizeof je->u.sector);
 485                 if (unlikely(r)) {
 486                         dm_integrity_io_error(ic, "crypto_shash_update", r);
 487                         goto err;
 488                 }
 489         }
 490
 491         size = crypto_shash_digestsize(ic->journal_mac);
 492
 493         if (likely(size <= JOURNAL_MAC_SIZE)) {
 494                 r = crypto_shash_final(desc, result);
 495                 if (unlikely(r)) {
 496                         dm_integrity_io_error(ic, "crypto_shash_final", r);
 497                         goto err;
 498                 }
 499                 memset(result + size, 0, JOURNAL_MAC_SIZE - size);
 500         } else {
 501                 __u8 digest[size];
 502                 r = crypto_shash_final(desc, digest);
 503                 if (unlikely(r)) {
 504                         dm_integrity_io_error(ic, "crypto_shash_final", r);
 505                         goto err;
 506                 }
 507                 memcpy(result, digest, JOURNAL_MAC_SIZE);
 508         }
 509
 510         return;
 511 err:
 512         memset(result, 0, JOURNAL_MAC_SIZE);
 513 }
 514
 515 static void rw_section_mac(struct dm_integrity_c *ic, unsigned section, bool wr)
 516 {
 517         __u8 result[JOURNAL_MAC_SIZE];
 518         unsigned j;
 519
 520         if (!ic->journal_mac)
 521                 return;
 522
 523         section_mac(ic, section, result);
 524
 525         for (j = 0; j < JOURNAL_BLOCK_SECTORS; j++) {
 526                 struct journal_sector *js = access_journal(ic, section, j);
 527
 528                 if (likely(wr))
 529                         memcpy(&js->mac, result + (j * JOURNAL_MAC_PER_SECTOR), JOURNAL_MAC_PER_SECTOR);
 530                 else {
 531                         if (memcmp(&js->mac, result + (j * JOURNAL_MAC_PER_SECTOR), JOURNAL_MAC_PER_SECTOR))
 532                                 dm_integrity_io_error(ic, "journal mac", -EILSEQ);
 533                 }
 534         }
 535 }
 536
 537 static void complete_journal_op(void *context)
 538 {
 539         struct journal_completion *comp = context;
 540         BUG_ON(!atomic_read(&comp->in_flight));
 541         if (likely(atomic_dec_and_test(&comp->in_flight)))
 542                 complete(&comp->comp);
 543 }
 544
 545 static void xor_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
 546                         unsigned n_sections, struct journal_completion *comp)
 547 {
 548         struct async_submit_ctl submit;
 549         size_t n_bytes = (size_t)(n_sections * ic->journal_section_sectors) << SECTOR_SHIFT;
 550         unsigned pl_index, pl_offset, section_index;
 551         struct page_list *source_pl, *target_pl;
 552
 553         if (likely(encrypt)) {
 554                 source_pl = ic->journal;
 555                 target_pl = ic->journal_io;
 556         } else {
 557                 source_pl = ic->journal_io;
 558                 target_pl = ic->journal;
 559         }
 560
 561         page_list_location(ic, section, 0, &pl_index, &pl_offset);
 562
 563         atomic_add(roundup(pl_offset + n_bytes, PAGE_SIZE) >> PAGE_SHIFT, &comp->in_flight);
 564
 565         init_async_submit(&submit, ASYNC_TX_XOR_ZERO_DST, NULL, complete_journal_op, comp, NULL);
 566
 567         section_index = pl_index;
 568
 569         do {
 570                 size_t this_step;
 571                 struct page *src_pages[2];
 572                 struct page *dst_page;
 573
 574                 while (unlikely(pl_index == section_index)) {
 575                         unsigned dummy;
 576                         if (likely(encrypt))
 577                                 rw_section_mac(ic, section, true);
 578                         section++;
 579                         n_sections--;
 580                         if (!n_sections)
 581                                 break;
 582                         page_list_location(ic, section, 0, &section_index, &dummy);
 583                 }
 584
 585                 this_step = min(n_bytes, (size_t)PAGE_SIZE - pl_offset);
 586                 dst_page = target_pl[pl_index].page;
 587                 src_pages[0] = source_pl[pl_index].page;
 588                 src_pages[1] = ic->journal_xor[pl_index].page;
 589
 590                 async_xor(dst_page, src_pages, pl_offset, 2, this_step, &submit);
 591
 592                 pl_index++;
 593                 pl_offset = 0;
 594                 n_bytes -= this_step;
 595         } while (n_bytes);
 596
 597         BUG_ON(n_sections);
 598
 599         async_tx_issue_pending_all();
 600 }
 601
 602 static void complete_journal_encrypt(struct crypto_async_request *req, int err)
 603 {
 604         struct journal_completion *comp = req->data;
 605         if (unlikely(err)) {
 606                 if (likely(err == -EINPROGRESS)) {
 607                         complete(&comp->ic->crypto_backoff);
 608                         return;
 609                 }
 610                 dm_integrity_io_error(comp->ic, "asynchronous encrypt", err);
 611         }
 612         complete_journal_op(comp);
 613 }
 614
 615 static bool do_crypt(bool encrypt, struct skcipher_request *req, struct journal_completion *comp)
 616 {
 617         int r;
 618         skcipher_request_set_callback(req, CRYPTO_TFM_REQ_MAY_BACKLOG | CRYPTO_TFM_REQ_MAY_SLEEP,
 619                                       complete_journal_encrypt, comp);
 620         if (likely(encrypt))
 621                 r = crypto_skcipher_encrypt(req);
 622         else
 623                 r = crypto_skcipher_decrypt(req);
 624         if (likely(!r))
 625                 return false;
 626         if (likely(r == -EINPROGRESS))
 627                 return true;
 628         if (likely(r == -EBUSY)) {
 629                 wait_for_completion(&comp->ic->crypto_backoff);
 630                 reinit_completion(&comp->ic->crypto_backoff);
 631                 return true;
 632         }
 633         dm_integrity_io_error(comp->ic, "encrypt", r);
 634         return false;
 635 }
 636
 637 static void crypt_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
 638                           unsigned n_sections, struct journal_completion *comp)
 639 {
 640         struct scatterlist **source_sg;
 641         struct scatterlist **target_sg;
 642
 643         atomic_add(2, &comp->in_flight);
 644
 645         if (likely(encrypt)) {
 646                 source_sg = ic->journal_scatterlist;
 647                 target_sg = ic->journal_io_scatterlist;
 648         } else {
 649                 source_sg = ic->journal_io_scatterlist;
 650                 target_sg = ic->journal_scatterlist;
 651         }
 652
 653         do {
 654                 struct skcipher_request *req;
 655                 unsigned ivsize;
 656                 char *iv;
 657
 658                 if (likely(encrypt))
 659                         rw_section_mac(ic, section, true);
 660
 661                 req = ic->sk_requests[section];
 662                 ivsize = crypto_skcipher_ivsize(ic->journal_crypt);
 663                 iv = req->iv;
 664
 665                 memcpy(iv, iv + ivsize, ivsize);
 666
 667                 req->src = source_sg[section];
 668                 req->dst = target_sg[section];
 669
 670                 if (unlikely(do_crypt(encrypt, req, comp)))
 671                         atomic_inc(&comp->in_flight);
 672
 673                 section++;
 674                 n_sections--;
 675         } while (n_sections);
 676
 677         atomic_dec(&comp->in_flight);
 678         complete_journal_op(comp);
 679 }
 680
 681 static void encrypt_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
 682                             unsigned n_sections, struct journal_completion *comp)
 683 {
 684         if (ic->journal_xor)
 685                 return xor_journal(ic, encrypt, section, n_sections, comp);
 686         else
 687                 return crypt_journal(ic, encrypt, section, n_sections, comp);
 688 }
 689
 690 static void complete_journal_io(unsigned long error, void *context)
 691 {
 692         struct journal_completion *comp = context;
 693         if (unlikely(error != 0))
 694                 dm_integrity_io_error(comp->ic, "writing journal", -EIO);
 695         complete_journal_op(comp);
 696 }
 697
 698 static void rw_journal(struct dm_integrity_c *ic, int op, int op_flags, unsigned section,
 699                        unsigned n_sections, struct journal_completion *comp)
 700 {
 701         struct dm_io_request io_req;
 702         struct dm_io_region io_loc;
 703         unsigned sector, n_sectors, pl_index, pl_offset;
 704         int r;
 705
 706         if (unlikely(dm_integrity_failed(ic))) {
 707                 if (comp)
 708                         complete_journal_io(-1UL, comp);
 709                 return;
 710         }
 711
 712         sector = section * ic->journal_section_sectors;
 713         n_sectors = n_sections * ic->journal_section_sectors;
 714
 715         pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
 716         pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
 717
 718         io_req.bi_op = op;
 719         io_req.bi_op_flags = op_flags;
 720         io_req.mem.type = DM_IO_PAGE_LIST;
 721         if (ic->journal_io)
 722                 io_req.mem.ptr.pl = &ic->journal_io[pl_index];
 723         else
 724                 io_req.mem.ptr.pl = &ic->journal[pl_index];
 725         io_req.mem.offset = pl_offset;
 726         if (likely(comp != NULL)) {
 727                 io_req.notify.fn = complete_journal_io;
 728                 io_req.notify.context = comp;
 729         } else {
 730                 io_req.notify.fn = NULL;
 731         }
 732         io_req.client = ic->io;
 733         io_loc.bdev = ic->dev->bdev;
 734         io_loc.sector = ic->start + SB_SECTORS + sector;
 735         io_loc.count = n_sectors;
 736
 737         r = dm_io(&io_req, 1, &io_loc, NULL);
 738         if (unlikely(r)) {
 739                 dm_integrity_io_error(ic, op == REQ_OP_READ ? "reading journal" : "writing journal", r);
 740                 if (comp) {
 741                         WARN_ONCE(1, "asynchronous dm_io failed: %d", r);
 742                         complete_journal_io(-1UL, comp);
 743                 }
 744         }
 745 }
 746
 747 static void write_journal(struct dm_integrity_c *ic, unsigned commit_start, unsigned commit_sections)
 748 {
 749         struct journal_completion io_comp;
 750         struct journal_completion crypt_comp_1;
 751         struct journal_completion crypt_comp_2;
 752         unsigned i;
 753
 754         io_comp.ic = ic;
 755         io_comp.comp = COMPLETION_INITIALIZER_ONSTACK(io_comp.comp);
 756
 757         if (commit_start + commit_sections <= ic->journal_sections) {
 758                 io_comp.in_flight = (atomic_t)ATOMIC_INIT(1);
 759                 if (ic->journal_io) {
 760                         crypt_comp_1.ic = ic;
 761                         crypt_comp_1.comp = COMPLETION_INITIALIZER_ONSTACK(crypt_comp_1.comp);
 762                         crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
 763                         encrypt_journal(ic, true, commit_start, commit_sections, &crypt_comp_1);
 764                         wait_for_completion_io(&crypt_comp_1.comp);
 765                 } else {
 766                         for (i = 0; i < commit_sections; i++)
 767                                 rw_section_mac(ic, commit_start + i, true);
 768                 }
 769                 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, commit_sections, &io_comp);
 770         } else {
 771                 unsigned to_end;
 772                 io_comp.in_flight = (atomic_t)ATOMIC_INIT(2);
 773                 to_end = ic->journal_sections - commit_start;
 774                 if (ic->journal_io) {
 775                         crypt_comp_1.ic = ic;
 776                         crypt_comp_1.comp = COMPLETION_INITIALIZER_ONSTACK(crypt_comp_1.comp);
 777                         crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
 778                         encrypt_journal(ic, true, commit_start, to_end, &crypt_comp_1);
 779                         if (try_wait_for_completion(&crypt_comp_1.comp)) {
 780                                 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
 781                                 crypt_comp_1.comp = COMPLETION_INITIALIZER_ONSTACK(crypt_comp_1.comp);
 782                                 crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
 783                                 encrypt_journal(ic, true, 0, commit_sections - to_end, &crypt_comp_1);
 784                                 wait_for_completion_io(&crypt_comp_1.comp);
 785                         } else {
 786                                 crypt_comp_2.ic = ic;
 787                                 crypt_comp_2.comp = COMPLETION_INITIALIZER_ONSTACK(crypt_comp_2.comp);
 788                                 crypt_comp_2.in_flight = (atomic_t)ATOMIC_INIT(0);
 789                                 encrypt_journal(ic, true, 0, commit_sections - to_end, &crypt_comp_2);
 790                                 wait_for_completion_io(&crypt_comp_1.comp);
 791                                 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
 792                                 wait_for_completion_io(&crypt_comp_2.comp);
 793                         }
 794                 } else {
 795                         for (i = 0; i < to_end; i++)
 796                                 rw_section_mac(ic, commit_start + i, true);
 797                         rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
 798                         for (i = 0; i < commit_sections - to_end; i++)
 799                                 rw_section_mac(ic, i, true);
 800                 }
 801                 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, 0, commit_sections - to_end, &io_comp);
 802         }
 803
 804         wait_for_completion_io(&io_comp.comp);
 805 }
 806
 807 static void copy_from_journal(struct dm_integrity_c *ic, unsigned section, unsigned offset,
 808                               unsigned n_sectors, sector_t target, io_notify_fn fn, void *data)
 809 {
 810         struct dm_io_request io_req;
 811         struct dm_io_region io_loc;
 812         int r;
 813         unsigned sector, pl_index, pl_offset;
 814
 815         if (unlikely(dm_integrity_failed(ic))) {
 816                 fn(-1UL, data);
 817                 return;
 818         }
 819
 820         sector = section * ic->journal_section_sectors + JOURNAL_BLOCK_SECTORS + offset;
 821
 822         pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
 823         pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
 824
 825         io_req.bi_op = REQ_OP_WRITE;
 826         io_req.bi_op_flags = 0;
 827         io_req.mem.type = DM_IO_PAGE_LIST;
 828         io_req.mem.ptr.pl = &ic->journal[pl_index];
 829         io_req.mem.offset = pl_offset;
 830         io_req.notify.fn = fn;
 831         io_req.notify.context = data;
 832         io_req.client = ic->io;
 833         io_loc.bdev = ic->dev->bdev;
 834         io_loc.sector = ic->start + target;
 835         io_loc.count = n_sectors;
 836
 837         r = dm_io(&io_req, 1, &io_loc, NULL);
 838         if (unlikely(r)) {
 839                 WARN_ONCE(1, "asynchronous dm_io failed: %d", r);
 840                 fn(-1UL, data);
 841         }
 842 }
 843
 844 static bool add_new_range(struct dm_integrity_c *ic, struct dm_integrity_range *new_range)
 845 {
 846         struct rb_node **n = &ic->in_progress.rb_node;
 847         struct rb_node *parent;
 848
 849         parent = NULL;
 850
 851         while (*n) {
 852                 struct dm_integrity_range *range = container_of(*n, struct dm_integrity_range, node);
 853
 854                 parent = *n;
 855                 if (new_range->logical_sector + new_range->n_sectors <= range->logical_sector) {
 856                         n = &range->node.rb_left;
 857                 } else if (new_range->logical_sector >= range->logical_sector + range->n_sectors) {
 858                         n = &range->node.rb_right;
 859                 } else {
 860                         return false;
 861                 }
 862         }
 863
 864         rb_link_node(&new_range->node, parent, n);
 865         rb_insert_color(&new_range->node, &ic->in_progress);
 866
 867         return true;
 868 }
 869
 870 static void remove_range_unlocked(struct dm_integrity_c *ic, struct dm_integrity_range *range)
 871 {
 872         rb_erase(&range->node, &ic->in_progress);
 873         wake_up_locked(&ic->endio_wait);
 874 }
 875
 876 static void remove_range(struct dm_integrity_c *ic, struct dm_integrity_range *range)
 877 {
 878         unsigned long flags;
 879
 880         spin_lock_irqsave(&ic->endio_wait.lock, flags);
 881         remove_range_unlocked(ic, range);
 882         spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
 883 }
 884
 885 static void init_journal_node(struct journal_node *node)
 886 {
 887         RB_CLEAR_NODE(&node->node);
 888         node->sector = (sector_t)-1;
 889 }
 890
 891 static void add_journal_node(struct dm_integrity_c *ic, struct journal_node *node, sector_t sector)
 892 {
 893         struct rb_node **link;
 894         struct rb_node *parent;
 895
 896         node->sector = sector;
 897         BUG_ON(!RB_EMPTY_NODE(&node->node));
 898
 899         link = &ic->journal_tree_root.rb_node;
 900         parent = NULL;
 901
 902         while (*link) {
 903                 struct journal_node *j;
 904                 parent = *link;
 905                 j = container_of(parent, struct journal_node, node);
 906                 if (sector < j->sector)
 907                         link = &j->node.rb_left;
 908                 else
 909                         link = &j->node.rb_right;
 910         }
 911
 912         rb_link_node(&node->node, parent, link);
 913         rb_insert_color(&node->node, &ic->journal_tree_root);
 914 }
 915
 916 static void remove_journal_node(struct dm_integrity_c *ic, struct journal_node *node)
 917 {
 918         BUG_ON(RB_EMPTY_NODE(&node->node));
 919         rb_erase(&node->node, &ic->journal_tree_root);
 920         init_journal_node(node);
 921 }
 922
 923 #define NOT_FOUND       (-1U)
 924
 925 static unsigned find_journal_node(struct dm_integrity_c *ic, sector_t sector, sector_t *next_sector)
 926 {
 927         struct rb_node *n = ic->journal_tree_root.rb_node;
 928         unsigned found = NOT_FOUND;
 929         *next_sector = (sector_t)-1;
 930         while (n) {
 931                 struct journal_node *j = container_of(n, struct journal_node, node);
 932                 if (sector == j->sector) {
 933                         found = j - ic->journal_tree;
 934                 }
 935                 if (sector < j->sector) {
 936                         *next_sector = j->sector;
 937                         n = j->node.rb_left;
 938                 } else {
 939                         n = j->node.rb_right;
 940                 }
 941         }
 942
 943         return found;
 944 }
 945
 946 static bool test_journal_node(struct dm_integrity_c *ic, unsigned pos, sector_t sector)
 947 {
 948         struct journal_node *node, *next_node;
 949         struct rb_node *next;
 950
 951         if (unlikely(pos >= ic->journal_entries))
 952                 return false;
 953         node = &ic->journal_tree[pos];
 954         if (unlikely(RB_EMPTY_NODE(&node->node)))
 955                 return false;
 956         if (unlikely(node->sector != sector))
 957                 return false;
 958
 959         next = rb_next(&node->node);
 960         if (unlikely(!next))
 961                 return true;
 962
 963         next_node = container_of(next, struct journal_node, node);
 964         return next_node->sector != sector;
 965 }
 966
 967 static bool find_newer_committed_node(struct dm_integrity_c *ic, struct journal_node *node)
 968 {
 969         struct rb_node *next;
 970         struct journal_node *next_node;
 971         unsigned next_section;
 972
 973         BUG_ON(RB_EMPTY_NODE(&node->node));
 974
 975         next = rb_next(&node->node);
 976         if (unlikely(!next))
 977                 return false;
 978
 979         next_node = container_of(next, struct journal_node, node);
 980
 981         if (next_node->sector != node->sector)
 982                 return false;
 983
 984         next_section = (unsigned)(next_node - ic->journal_tree) / ic->journal_section_entries;
 985         if (next_section >= ic->committed_section &&
 986             next_section < ic->committed_section + ic->n_committed_sections)
 987                 return true;
 988         if (next_section + ic->journal_sections < ic->committed_section + ic->n_committed_sections)
 989                 return true;
 990
 991         return false;
 992 }
 993
 994 #define TAG_READ        0
 995 #define TAG_WRITE       1
 996 #define TAG_CMP         2
 997
 998 static int dm_integrity_rw_tag(struct dm_integrity_c *ic, unsigned char *tag, sector_t *metadata_block,
 999                                unsigned *metadata_offset, unsigned total_size, int op)
1000 {
1001         do {
1002                 unsigned char *data, *dp;
1003                 struct dm_buffer *b;
1004                 unsigned to_copy;
1005                 int r;
1006
1007                 r = dm_integrity_failed(ic);
1008                 if (unlikely(r))
1009                         return r;
1010
1011                 data = dm_bufio_read(ic->bufio, *metadata_block, &b);
1012                 if (unlikely(IS_ERR(data)))
1013                         return PTR_ERR(data);
1014
1015                 to_copy = min((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - *metadata_offset, total_size);
1016                 dp = data + *metadata_offset;
1017                 if (op == TAG_READ) {
1018                         memcpy(tag, dp, to_copy);
1019                 } else if (op == TAG_WRITE) {
1020                         memcpy(dp, tag, to_copy);
1021                         dm_bufio_mark_buffer_dirty(b);
1022                 } else  {
1023                         /* e.g.: op == TAG_CMP */
1024                         if (unlikely(memcmp(dp, tag, to_copy))) {
1025                                 unsigned i;
1026
1027                                 for (i = 0; i < to_copy; i++) {
1028                                         if (dp[i] != tag[i])
1029                                                 break;
1030                                         total_size--;
1031                                 }
1032                                 dm_bufio_release(b);
1033                                 return total_size;
1034                         }
1035                 }
1036                 dm_bufio_release(b);
1037
1038                 tag += to_copy;
1039                 *metadata_offset += to_copy;
1040                 if (unlikely(*metadata_offset == 1U << SECTOR_SHIFT << ic->log2_buffer_sectors)) {
1041                         (*metadata_block)++;
1042                         *metadata_offset = 0;
1043                 }
1044                 total_size -= to_copy;
1045         } while (unlikely(total_size));
1046
1047         return 0;
1048 }
1049
1050 static void dm_integrity_flush_buffers(struct dm_integrity_c *ic)
1051 {
1052         int r;
1053         r = dm_bufio_write_dirty_buffers(ic->bufio);
1054         if (unlikely(r))
1055                 dm_integrity_io_error(ic, "writing tags", r);
1056 }
1057
1058 static void sleep_on_endio_wait(struct dm_integrity_c *ic)
1059 {
1060         DECLARE_WAITQUEUE(wait, current);
1061         __add_wait_queue(&ic->endio_wait, &wait);
1062         __set_current_state(TASK_UNINTERRUPTIBLE);
1063         spin_unlock_irq(&ic->endio_wait.lock);
1064         io_schedule();
1065         spin_lock_irq(&ic->endio_wait.lock);
1066         __remove_wait_queue(&ic->endio_wait, &wait);
1067 }
1068
1069 static void autocommit_fn(unsigned long data)
1070 {
1071         struct dm_integrity_c *ic = (struct dm_integrity_c *)data;
1072
1073         if (likely(!dm_integrity_failed(ic)))
1074                 queue_work(ic->commit_wq, &ic->commit_work);
1075 }
1076
1077 static void schedule_autocommit(struct dm_integrity_c *ic)
1078 {
1079         if (!timer_pending(&ic->autocommit_timer))
1080                 mod_timer(&ic->autocommit_timer, jiffies + ic->autocommit_jiffies);
1081 }
1082
1083 static void submit_flush_bio(struct dm_integrity_c *ic, struct dm_integrity_io *dio)
1084 {
1085         struct bio *bio;
1086         spin_lock_irq(&ic->endio_wait.lock);
1087         bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1088         bio_list_add(&ic->flush_bio_list, bio);
1089         spin_unlock_irq(&ic->endio_wait.lock);
1090         queue_work(ic->commit_wq, &ic->commit_work);
1091 }
1092
1093 static void do_endio(struct dm_integrity_c *ic, struct bio *bio)
1094 {
1095         int r = dm_integrity_failed(ic);
1096         if (unlikely(r) && !bio->bi_error)
1097                 bio->bi_error = r;
1098         bio_endio(bio);
1099 }
1100
1101 static void do_endio_flush(struct dm_integrity_c *ic, struct dm_integrity_io *dio)
1102 {
1103         struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1104
1105         if (unlikely(dio->fua) && likely(!bio->bi_error) && likely(!dm_integrity_failed(ic)))
1106                 submit_flush_bio(ic, dio);
1107         else
1108                 do_endio(ic, bio);
1109 }
1110
1111 static void dec_in_flight(struct dm_integrity_io *dio)
1112 {
1113         if (atomic_dec_and_test(&dio->in_flight)) {
1114                 struct dm_integrity_c *ic = dio->ic;
1115                 struct bio *bio;
1116
1117                 remove_range(ic, &dio->range);
1118
1119                 if (unlikely(dio->write))
1120                         schedule_autocommit(ic);
1121
1122                 bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1123
1124                 if (unlikely(dio->bi_error) && !bio->bi_error)
1125                         bio->bi_error = dio->bi_error;
1126                 if (likely(!bio->bi_error) && unlikely(bio_sectors(bio) != dio->range.n_sectors)) {
1127                         dio->range.logical_sector += dio->range.n_sectors;
1128                         bio_advance(bio, dio->range.n_sectors << SECTOR_SHIFT);
1129                         INIT_WORK(&dio->work, integrity_bio_wait);
1130                         queue_work(ic->wait_wq, &dio->work);
1131                         return;
1132                 }
1133                 do_endio_flush(ic, dio);
1134         }
1135 }
1136
1137 static void integrity_end_io(struct bio *bio)
1138 {
1139         struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
1140
1141         bio->bi_iter = dio->orig_bi_iter;
1142         bio->bi_bdev = dio->orig_bi_bdev;
1143         if (dio->orig_bi_integrity) {
1144                 bio->bi_integrity = dio->orig_bi_integrity;
1145                 bio->bi_opf |= REQ_INTEGRITY;
1146         }
1147         bio->bi_end_io = dio->orig_bi_end_io;
1148
1149         if (dio->completion)
1150                 complete(dio->completion);
1151
1152         dec_in_flight(dio);
1153 }
1154
1155 static void integrity_sector_checksum(struct dm_integrity_c *ic, sector_t sector,
1156                                       const char *data, char *result)
1157 {
1158         __u64 sector_le = cpu_to_le64(sector);
1159         SHASH_DESC_ON_STACK(req, ic->internal_hash);
1160         int r;
1161         unsigned digest_size;
1162
1163         req->tfm = ic->internal_hash;
1164         req->flags = 0;
1165
1166         r = crypto_shash_init(req);
1167         if (unlikely(r < 0)) {
1168                 dm_integrity_io_error(ic, "crypto_shash_init", r);
1169                 goto failed;
1170         }
1171
1172         r = crypto_shash_update(req, (const __u8 *)&sector_le, sizeof sector_le);
1173         if (unlikely(r < 0)) {
1174                 dm_integrity_io_error(ic, "crypto_shash_update", r);
1175                 goto failed;
1176         }
1177
1178         r = crypto_shash_update(req, data, 1 << SECTOR_SHIFT);
1179         if (unlikely(r < 0)) {
1180                 dm_integrity_io_error(ic, "crypto_shash_update", r);
1181                 goto failed;
1182         }
1183
1184         r = crypto_shash_final(req, result);
1185         if (unlikely(r < 0)) {
1186                 dm_integrity_io_error(ic, "crypto_shash_final", r);
1187                 goto failed;
1188         }
1189
1190         digest_size = crypto_shash_digestsize(ic->internal_hash);
1191         if (unlikely(digest_size < ic->tag_size))
1192                 memset(result + digest_size, 0, ic->tag_size - digest_size);
1193
1194         return;
1195
1196 failed:
1197         /* this shouldn't happen anyway, the hash functions have no reason to fail */
1198         get_random_bytes(result, ic->tag_size);
1199 }
1200
1201 static void integrity_metadata(struct work_struct *w)
1202 {
1203         struct dm_integrity_io *dio = container_of(w, struct dm_integrity_io, work);
1204         struct dm_integrity_c *ic = dio->ic;
1205
1206         int r;
1207
1208         if (ic->internal_hash) {
1209                 struct bvec_iter iter;
1210                 struct bio_vec bv;
1211                 unsigned digest_size = crypto_shash_digestsize(ic->internal_hash);
1212                 struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1213                 char *checksums;
1214                 unsigned extra_space = unlikely(digest_size > ic->tag_size) ? digest_size - ic->tag_size : 0;
1215                 char checksums_onstack[ic->tag_size + extra_space];
1216                 unsigned sectors_to_process = dio->range.n_sectors;
1217                 sector_t sector = dio->range.logical_sector;
1218
1219                 if (unlikely(ic->mode == 'R'))
1220                         goto skip_io;
1221
1222                 checksums = kmalloc((PAGE_SIZE >> SECTOR_SHIFT) * ic->tag_size + extra_space,
1223                                     GFP_NOIO | __GFP_NORETRY | __GFP_NOWARN);
1224                 if (!checksums)
1225                         checksums = checksums_onstack;
1226
1227                 __bio_for_each_segment(bv, bio, iter, dio->orig_bi_iter) {
1228                         unsigned pos;
1229                         char *mem, *checksums_ptr;
1230
1231 again:
1232                         mem = (char *)kmap_atomic(bv.bv_page) + bv.bv_offset;
1233                         pos = 0;
1234                         checksums_ptr = checksums;
1235                         do {
1236                                 integrity_sector_checksum(ic, sector, mem + pos, checksums_ptr);
1237                                 checksums_ptr += ic->tag_size;
1238                                 sectors_to_process--;
1239                                 pos += 1 << SECTOR_SHIFT;
1240                                 sector++;
1241                         } while (pos < bv.bv_len && sectors_to_process && checksums != checksums_onstack);
1242                         kunmap_atomic(mem);
1243
1244                         r = dm_integrity_rw_tag(ic, checksums, &dio->metadata_block, &dio->metadata_offset,
1245                                                 checksums_ptr - checksums, !dio->write ? TAG_CMP : TAG_WRITE);
1246                         if (unlikely(r)) {
1247                                 if (r > 0) {
1248                                         DMERR("Checksum failed at sector 0x%llx",
1249                                               (unsigned long long)(sector - ((r + ic->tag_size - 1) / ic->tag_size)));
1250                                         r = -EILSEQ;
1251                                 }
1252                                 if (likely(checksums != checksums_onstack))
1253                                         kfree(checksums);
1254                                 goto error;
1255                         }
1256
1257                         if (!sectors_to_process)
1258                                 break;
1259
1260                         if (unlikely(pos < bv.bv_len)) {
1261                                 bv.bv_offset += pos;
1262                                 bv.bv_len -= pos;
1263                                 goto again;
1264                         }
1265                 }
1266
1267                 if (likely(checksums != checksums_onstack))
1268                         kfree(checksums);
1269         } else {
1270                 struct bio_integrity_payload *bip = dio->orig_bi_integrity;
1271
1272                 if (bip) {
1273                         struct bio_vec biv;
1274                         struct bvec_iter iter;
1275                         unsigned data_to_process = dio->range.n_sectors * ic->tag_size;
1276
1277                         bip_for_each_vec(biv, bip, iter) {
1278                                 unsigned char *tag;
1279                                 unsigned this_len;
1280
1281                                 BUG_ON(PageHighMem(biv.bv_page));
1282                                 tag = lowmem_page_address(biv.bv_page) + biv.bv_offset;
1283                                 this_len = min(biv.bv_len, data_to_process);
1284                                 r = dm_integrity_rw_tag(ic, tag, &dio->metadata_block, &dio->metadata_offset,
1285                                                         this_len, !dio->write ? TAG_READ : TAG_WRITE);
1286                                 if (unlikely(r))
1287                                         goto error;
1288                                 data_to_process -= this_len;
1289                                 if (!data_to_process)
1290                                         break;
1291                         }
1292                 }
1293         }
1294 skip_io:
1295         dec_in_flight(dio);
1296         return;
1297 error:
1298         dio->bi_error = r;
1299         dec_in_flight(dio);
1300 }
1301
1302 static int dm_integrity_map(struct dm_target *ti, struct bio *bio)
1303 {
1304         struct dm_integrity_c *ic = ti->private;
1305         struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
1306
1307         sector_t area, offset;
1308
1309         dio->ic = ic;
1310         dio->bi_error = 0;
1311
1312         if (unlikely(bio->bi_opf & REQ_PREFLUSH)) {
1313                 submit_flush_bio(ic, dio);
1314                 return DM_MAPIO_SUBMITTED;
1315         }
1316
1317         dio->range.logical_sector = dm_target_offset(ti, bio->bi_iter.bi_sector);
1318         dio->write = bio_op(bio) == REQ_OP_WRITE;
1319         dio->fua = dio->write && bio->bi_opf & REQ_FUA;
1320         if (unlikely(dio->fua)) {
1321                 /*
1322                  * Don't pass down the FUA flag because we have to flush
1323                  * disk cache anyway.
1324                  */
1325                 bio->bi_opf &= ~REQ_FUA;
1326         }
1327         if (unlikely(dio->range.logical_sector + bio_sectors(bio) > ic->provided_data_sectors)) {
1328                 DMERR("Too big sector number: 0x%llx + 0x%x > 0x%llx",
1329                       (unsigned long long)dio->range.logical_sector, bio_sectors(bio),
1330                       (unsigned long long)ic->provided_data_sectors);
1331                 return -EIO;
1332         }
1333
1334         if (unlikely(ic->mode == 'R') && unlikely(dio->write))
1335                 return -EIO;
1336
1337         get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
1338         dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset, &dio->metadata_offset);
1339         bio->bi_iter.bi_sector = get_data_sector(ic, area, offset);
1340
1341         dm_integrity_map_continue(dio, true);
1342         return DM_MAPIO_SUBMITTED;
1343 }
1344
1345 static bool __journal_read_write(struct dm_integrity_io *dio, struct bio *bio,
1346                                  unsigned journal_section, unsigned journal_entry)
1347 {
1348         struct dm_integrity_c *ic = dio->ic;
1349         sector_t logical_sector;
1350         unsigned n_sectors;
1351
1352         logical_sector = dio->range.logical_sector;
1353         n_sectors = dio->range.n_sectors;
1354         do {
1355                 struct bio_vec bv = bio_iovec(bio);
1356                 char *mem;
1357
1358                 if (unlikely(bv.bv_len >> SECTOR_SHIFT > n_sectors))
1359                         bv.bv_len = n_sectors << SECTOR_SHIFT;
1360                 n_sectors -= bv.bv_len >> SECTOR_SHIFT;
1361                 bio_advance_iter(bio, &bio->bi_iter, bv.bv_len);
1362 retry_kmap:
1363                 mem = kmap_atomic(bv.bv_page);
1364                 if (likely(dio->write))
1365                         flush_dcache_page(bv.bv_page);
1366
1367                 do {
1368                         struct journal_entry *je = access_journal_entry(ic, journal_section, journal_entry);
1369
1370                         if (unlikely(!dio->write)) {
1371                                 struct journal_sector *js;
1372
1373                                 if (unlikely(journal_entry_is_inprogress(je))) {
1374                                         flush_dcache_page(bv.bv_page);
1375                                         kunmap_atomic(mem);
1376
1377                                         __io_wait_event(ic->copy_to_journal_wait, !journal_entry_is_inprogress(je));
1378                                         goto retry_kmap;
1379                                 }
1380                                 smp_rmb();
1381                                 BUG_ON(journal_entry_get_sector(je) != logical_sector);
1382                                 js = access_journal_data(ic, journal_section, journal_entry);
1383                                 memcpy(mem + bv.bv_offset, js, JOURNAL_SECTOR_DATA);
1384                                 memcpy(mem + bv.bv_offset + JOURNAL_SECTOR_DATA, &je->last_bytes, sizeof je->last_bytes);
1385 #ifdef INTERNAL_VERIFY
1386                                 if (ic->internal_hash) {
1387                                         char checksums_onstack[max(crypto_shash_digestsize(ic->internal_hash), ic->tag_size)];
1388
1389                                         integrity_sector_checksum(ic, logical_sector, mem + bv.bv_offset, checksums_onstack);
1390                                         if (unlikely(memcmp(checksums_onstack, je->tag, ic->tag_size))) {
1391                                                 DMERR("Checksum failed when reading from journal, at sector 0x%llx",
1392                                                       (unsigned long long)logical_sector);
1393                                         }
1394                                 }
1395 #endif
1396                         }
1397
1398                         if (!ic->internal_hash) {
1399                                 struct bio_integrity_payload *bip = bio_integrity(bio);
1400                                 unsigned tag_todo = ic->tag_size;
1401                                 char *tag_ptr = je->tag;
1402
1403                                 if (bip) do {
1404                                         struct bio_vec biv = bvec_iter_bvec(bip->bip_vec, bip->bip_iter);
1405                                         unsigned tag_now = min(biv.bv_len, tag_todo);
1406                                         char *tag_addr;
1407                                         BUG_ON(PageHighMem(biv.bv_page));
1408                                         tag_addr = lowmem_page_address(biv.bv_page) + biv.bv_offset;
1409                                         if (likely(dio->write))
1410                                                 memcpy(tag_ptr, tag_addr, tag_now);
1411                                         else
1412                                                 memcpy(tag_addr, tag_ptr, tag_now);
1413                                         bvec_iter_advance(bip->bip_vec, &bip->bip_iter, tag_now);
1414                                         tag_ptr += tag_now;
1415                                         tag_todo -= tag_now;
1416                                 } while (unlikely(tag_todo)); else {
1417                                         if (likely(dio->write))
1418                                                 memset(tag_ptr, 0, tag_todo);
1419                                 }
1420                         }
1421
1422                         if (likely(dio->write)) {
1423                                 struct journal_sector *js;
1424
1425                                 js = access_journal_data(ic, journal_section, journal_entry);
1426                                 memcpy(js, mem + bv.bv_offset, 1 << SECTOR_SHIFT);
1427                                 je->last_bytes = js->commit_id;
1428
1429                                 if (ic->internal_hash) {
1430                                         unsigned digest_size = crypto_shash_digestsize(ic->internal_hash);
1431                                         if (unlikely(digest_size > ic->tag_size)) {
1432                                                 char checksums_onstack[digest_size];
1433                                                 integrity_sector_checksum(ic, logical_sector, (char *)js, checksums_onstack);
1434                                                 memcpy(je->tag, checksums_onstack, ic->tag_size);
1435                                         } else
1436                                                 integrity_sector_checksum(ic, logical_sector, (char *)js, je->tag);
1437                                 }
1438
1439                                 journal_entry_set_sector(je, logical_sector);
1440                         }
1441                         logical_sector++;
1442
1443                         journal_entry++;
1444                         if (unlikely(journal_entry == ic->journal_section_entries)) {
1445                                 journal_entry = 0;
1446                                 journal_section++;
1447                                 wraparound_section(ic, &journal_section);
1448                         }
1449
1450                         bv.bv_offset += 1 << SECTOR_SHIFT;
1451                 } while (bv.bv_len -= 1 << SECTOR_SHIFT);
1452
1453                 if (unlikely(!dio->write))
1454                         flush_dcache_page(bv.bv_page);
1455                 kunmap_atomic(mem);
1456         } while (n_sectors);
1457
1458         if (likely(dio->write)) {
1459                 smp_mb();
1460                 if (unlikely(waitqueue_active(&ic->copy_to_journal_wait)))
1461                         wake_up(&ic->copy_to_journal_wait);
1462                 if (ACCESS_ONCE(ic->free_sectors) <= ic->free_sectors_threshold) {
1463                         queue_work(ic->commit_wq, &ic->commit_work);
1464                 } else {
1465                         schedule_autocommit(ic);
1466                 }
1467         } else {
1468                 remove_range(ic, &dio->range);
1469         }
1470
1471         if (unlikely(bio->bi_iter.bi_size)) {
1472                 sector_t area, offset;
1473
1474                 dio->range.logical_sector = logical_sector;
1475                 get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
1476                 dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset, &dio->metadata_offset);
1477                 return true;
1478         }
1479
1480         return false;
1481 }
1482
1483 static void dm_integrity_map_continue(struct dm_integrity_io *dio, bool from_map)
1484 {
1485         struct dm_integrity_c *ic = dio->ic;
1486         struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1487         unsigned journal_section, journal_entry;
1488         unsigned journal_read_pos;
1489         struct completion read_comp;
1490         bool need_sync_io = ic->internal_hash && !dio->write;
1491
1492         if (need_sync_io && from_map) {
1493                 INIT_WORK(&dio->work, integrity_bio_wait);
1494                 queue_work(ic->metadata_wq, &dio->work);
1495                 return;
1496         }
1497
1498 lock_retry:
1499         spin_lock_irq(&ic->endio_wait.lock);
1500 retry:
1501         if (unlikely(dm_integrity_failed(ic))) {
1502                 spin_unlock_irq(&ic->endio_wait.lock);
1503                 do_endio(ic, bio);
1504                 return;
1505         }
1506         dio->range.n_sectors = bio_sectors(bio);
1507         journal_read_pos = NOT_FOUND;
1508         if (likely(ic->mode == 'J')) {
1509                 if (dio->write) {
1510                         unsigned next_entry, i, pos;
1511                         unsigned ws, we;
1512
1513                         dio->range.n_sectors = min(dio->range.n_sectors, ic->free_sectors);
1514                         if (unlikely(!dio->range.n_sectors))
1515                                 goto sleep;
1516                         ic->free_sectors -= dio->range.n_sectors;
1517                         journal_section = ic->free_section;
1518                         journal_entry = ic->free_section_entry;
1519
1520                         next_entry = ic->free_section_entry + dio->range.n_sectors;
1521                         ic->free_section_entry = next_entry % ic->journal_section_entries;
1522                         ic->free_section += next_entry / ic->journal_section_entries;
1523                         ic->n_uncommitted_sections += next_entry / ic->journal_section_entries;
1524                         wraparound_section(ic, &ic->free_section);
1525
1526                         pos = journal_section * ic->journal_section_entries + journal_entry;
1527                         ws = journal_section;
1528                         we = journal_entry;
1529                         for (i = 0; i < dio->range.n_sectors; i++) {
1530                                 struct journal_entry *je;
1531
1532                                 add_journal_node(ic, &ic->journal_tree[pos], dio->range.logical_sector + i);
1533                                 pos++;
1534                                 if (unlikely(pos >= ic->journal_entries))
1535                                         pos = 0;
1536
1537                                 je = access_journal_entry(ic, ws, we);
1538                                 BUG_ON(!journal_entry_is_unused(je));
1539                                 journal_entry_set_inprogress(je);
1540                                 we++;
1541                                 if (unlikely(we == ic->journal_section_entries)) {
1542                                         we = 0;
1543                                         ws++;
1544                                         wraparound_section(ic, &ws);
1545                                 }
1546                         }
1547
1548                         spin_unlock_irq(&ic->endio_wait.lock);
1549                         goto journal_read_write;
1550                 } else {
1551                         sector_t next_sector;
1552                         journal_read_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
1553                         if (likely(journal_read_pos == NOT_FOUND)) {
1554                                 if (unlikely(dio->range.n_sectors > next_sector - dio->range.logical_sector))
1555                                         dio->range.n_sectors = next_sector - dio->range.logical_sector;
1556                         } else {
1557                                 unsigned i;
1558                                 for (i = 1; i < dio->range.n_sectors; i++) {
1559                                         if (!test_journal_node(ic, journal_read_pos + i, dio->range.logical_sector + i))
1560                                                 break;
1561                                 }
1562                                 dio->range.n_sectors = i;
1563                         }
1564                 }
1565         }
1566         if (unlikely(!add_new_range(ic, &dio->range))) {
1567                 /*
1568                  * We must not sleep in the request routine because it could
1569                  * stall bios on current->bio_list.
1570                  * So, we offload the bio to a workqueue if we have to sleep.
1571                  */
1572 sleep:
1573                 if (from_map) {
1574                         spin_unlock_irq(&ic->endio_wait.lock);
1575                         INIT_WORK(&dio->work, integrity_bio_wait);
1576                         queue_work(ic->wait_wq, &dio->work);
1577                         return;
1578                 } else {
1579                         sleep_on_endio_wait(ic);
1580                         goto retry;
1581                 }
1582         }
1583         spin_unlock_irq(&ic->endio_wait.lock);
1584
1585         if (unlikely(journal_read_pos != NOT_FOUND)) {
1586                 journal_section = journal_read_pos / ic->journal_section_entries;
1587                 journal_entry = journal_read_pos % ic->journal_section_entries;
1588                 goto journal_read_write;
1589         }
1590
1591         dio->in_flight = (atomic_t)ATOMIC_INIT(2);
1592
1593         if (need_sync_io) {
1594                 read_comp = COMPLETION_INITIALIZER_ONSTACK(read_comp);
1595                 dio->completion = &read_comp;
1596         } else
1597                 dio->completion = NULL;
1598
1599         dio->orig_bi_iter = bio->bi_iter;
1600
1601         dio->orig_bi_bdev = bio->bi_bdev;
1602         bio->bi_bdev = ic->dev->bdev;
1603
1604         dio->orig_bi_integrity = bio_integrity(bio);
1605         bio->bi_integrity = NULL;
1606         bio->bi_opf &= ~REQ_INTEGRITY;
1607
1608         dio->orig_bi_end_io = bio->bi_end_io;
1609         bio->bi_end_io = integrity_end_io;
1610
1611         bio->bi_iter.bi_size = dio->range.n_sectors << SECTOR_SHIFT;
1612         bio->bi_iter.bi_sector += ic->start;
1613         generic_make_request(bio);
1614
1615         if (need_sync_io) {
1616                 wait_for_completion_io(&read_comp);
1617                 integrity_metadata(&dio->work);
1618         } else {
1619                 INIT_WORK(&dio->work, integrity_metadata);
1620                 queue_work(ic->metadata_wq, &dio->work);
1621         }
1622
1623         return;
1624
1625 journal_read_write:
1626         if (unlikely(__journal_read_write(dio, bio, journal_section, journal_entry)))
1627                 goto lock_retry;
1628
1629         do_endio_flush(ic, dio);
1630 }
1631
1632
1633 static void integrity_bio_wait(struct work_struct *w)
1634 {
1635         struct dm_integrity_io *dio = container_of(w, struct dm_integrity_io, work);
1636
1637         dm_integrity_map_continue(dio, false);
1638 }
1639
1640 static void pad_uncommitted(struct dm_integrity_c *ic)
1641 {
1642         if (ic->free_section_entry) {
1643                 ic->free_sectors -= ic->journal_section_entries - ic->free_section_entry;
1644                 ic->free_section_entry = 0;
1645                 ic->free_section++;
1646                 wraparound_section(ic, &ic->free_section);
1647                 ic->n_uncommitted_sections++;
1648         }
1649 }
1650
1651 static void integrity_commit(struct work_struct *w)
1652 {
1653         struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, commit_work);
1654         unsigned commit_start, commit_sections;
1655         unsigned i, j, n;
1656         struct bio *flushes;
1657
1658         del_timer(&ic->autocommit_timer);
1659
1660         spin_lock_irq(&ic->endio_wait.lock);
1661         flushes = bio_list_get(&ic->flush_bio_list);
1662         if (unlikely(ic->mode != 'J')) {
1663                 spin_unlock_irq(&ic->endio_wait.lock);
1664                 dm_integrity_flush_buffers(ic);
1665                 goto release_flush_bios;
1666         }
1667
1668         pad_uncommitted(ic);
1669         commit_start = ic->uncommitted_section;
1670         commit_sections = ic->n_uncommitted_sections;
1671         spin_unlock_irq(&ic->endio_wait.lock);
1672
1673         if (!commit_sections)
1674                 goto release_flush_bios;
1675
1676         i = commit_start;
1677         for (n = 0; n < commit_sections; n++) {
1678                 for (j = 0; j < ic->journal_section_entries; j++) {
1679                         struct journal_entry *je;
1680                         je = access_journal_entry(ic, i, j);
1681                         io_wait_event(ic->copy_to_journal_wait, !journal_entry_is_inprogress(je));
1682                 }
1683                 for (j = 0; j < ic->journal_section_sectors; j++) {
1684                         struct journal_sector *js;
1685                         js = access_journal(ic, i, j);
1686                         js->commit_id = dm_integrity_commit_id(ic, i, j, ic->commit_seq);
1687                 }
1688                 i++;
1689                 if (unlikely(i >= ic->journal_sections))
1690                         ic->commit_seq = next_commit_seq(ic->commit_seq);
1691                 wraparound_section(ic, &i);
1692         }
1693         smp_rmb();
1694
1695         write_journal(ic, commit_start, commit_sections);
1696
1697         spin_lock_irq(&ic->endio_wait.lock);
1698         ic->uncommitted_section += commit_sections;
1699         wraparound_section(ic, &ic->uncommitted_section);
1700         ic->n_uncommitted_sections -= commit_sections;
1701         ic->n_committed_sections += commit_sections;
1702         spin_unlock_irq(&ic->endio_wait.lock);
1703
1704         if (ACCESS_ONCE(ic->free_sectors) <= ic->free_sectors_threshold)
1705                 queue_work(ic->writer_wq, &ic->writer_work);
1706
1707 release_flush_bios:
1708         while (flushes) {
1709                 struct bio *next = flushes->bi_next;
1710                 flushes->bi_next = NULL;
1711                 do_endio(ic, flushes);
1712                 flushes = next;
1713         }
1714 }
1715
1716 static void complete_copy_from_journal(unsigned long error, void *context)
1717 {
1718         struct journal_io *io = context;
1719         struct journal_completion *comp = io->comp;
1720         struct dm_integrity_c *ic = comp->ic;
1721         remove_range(ic, &io->range);
1722         mempool_free(io, ic->journal_io_mempool);
1723         if (unlikely(error != 0))
1724                 dm_integrity_io_error(ic, "copying from journal", -EIO);
1725         complete_journal_op(comp);
1726 }
1727
1728 static void do_journal_write(struct dm_integrity_c *ic, unsigned write_start,
1729                              unsigned write_sections, bool from_replay)
1730 {
1731         unsigned i, j, n;
1732         struct journal_completion comp;
1733
1734         comp.ic = ic;
1735         comp.in_flight = (atomic_t)ATOMIC_INIT(1);
1736         comp.comp = COMPLETION_INITIALIZER_ONSTACK(comp.comp);
1737
1738         i = write_start;
1739         for (n = 0; n < write_sections; n++, i++, wraparound_section(ic, &i)) {
1740 #ifndef INTERNAL_VERIFY
1741                 if (unlikely(from_replay))
1742 #endif
1743                         rw_section_mac(ic, i, false);
1744                 for (j = 0; j < ic->journal_section_entries; j++) {
1745                         struct journal_entry *je = access_journal_entry(ic, i, j);
1746                         sector_t sec, area, offset;
1747                         unsigned k, l, next_loop;
1748                         sector_t metadata_block;
1749                         unsigned metadata_offset;
1750                         struct journal_io *io;
1751
1752                         if (journal_entry_is_unused(je))
1753                                 continue;
1754                         BUG_ON(unlikely(journal_entry_is_inprogress(je)) && !from_replay);
1755                         sec = journal_entry_get_sector(je);
1756                         get_area_and_offset(ic, sec, &area, &offset);
1757                         access_journal_data(ic, i, j)->commit_id = je->last_bytes;
1758                         for (k = j + 1; k < ic->journal_section_entries; k++) {
1759                                 struct journal_entry *je2 = access_journal_entry(ic, i, k);
1760                                 sector_t sec2, area2, offset2;
1761                                 if (journal_entry_is_unused(je2))
1762                                         break;
1763                                 BUG_ON(unlikely(journal_entry_is_inprogress(je2)) && !from_replay);
1764                                 sec2 = journal_entry_get_sector(je2);
1765                                 get_area_and_offset(ic, sec2, &area2, &offset2);
1766                                 if (area2 != area || offset2 != offset + (k - j))
1767                                         break;
1768                                 access_journal_data(ic, i, k)->commit_id = je2->last_bytes;
1769                         }
1770                         next_loop = k - 1;
1771
1772                         io = mempool_alloc(ic->journal_io_mempool, GFP_NOIO);
1773                         io->comp = &comp;
1774                         io->range.logical_sector = sec;
1775                         io->range.n_sectors = k - j;
1776
1777                         spin_lock_irq(&ic->endio_wait.lock);
1778                         while (unlikely(!add_new_range(ic, &io->range)))
1779                                 sleep_on_endio_wait(ic);
1780
1781                         if (likely(!from_replay)) {
1782                                 struct journal_node *section_node = &ic->journal_tree[i * ic->journal_section_entries];
1783
1784                                 /* don't write if there is newer committed sector */
1785                                 while (j < k && find_newer_committed_node(ic, &section_node[j])) {
1786                                         struct journal_entry *je2 = access_journal_entry(ic, i, j);
1787
1788                                         journal_entry_set_unused(je2);
1789                                         remove_journal_node(ic, &section_node[j]);
1790                                         j++;
1791                                         sec++;
1792                                         offset++;
1793                                 }
1794                                 while (j < k && find_newer_committed_node(ic, &section_node[k - 1])) {
1795                                         struct journal_entry *je2 = access_journal_entry(ic, i, k - 1);
1796
1797                                         journal_entry_set_unused(je2);
1798                                         remove_journal_node(ic, &section_node[k - 1]);
1799                                         k--;
1800                                 }
1801                                 if (j == k) {
1802                                         remove_range_unlocked(ic, &io->range);
1803                                         spin_unlock_irq(&ic->endio_wait.lock);
1804                                         mempool_free(io, ic->journal_io_mempool);
1805                                         goto skip_io;
1806                                 }
1807                                 for (l = j; l < k; l++) {
1808                                         remove_journal_node(ic, &section_node[l]);
1809                                 }
1810                         }
1811                         spin_unlock_irq(&ic->endio_wait.lock);
1812
1813                         metadata_block = get_metadata_sector_and_offset(ic, area, offset, &metadata_offset);
1814                         for (l = j; l < k; l++) {
1815                                 int r;
1816                                 struct journal_entry *je2 = access_journal_entry(ic, i, l);
1817
1818                                 if (
1819 #ifndef INTERNAL_VERIFY
1820                                     unlikely(from_replay) &&
1821 #endif
1822                                     ic->internal_hash) {
1823                                         char test_tag[max(crypto_shash_digestsize(ic->internal_hash), ic->tag_size)];
1824
1825                                         integrity_sector_checksum(ic, sec + (l - j),
1826                                                                   (char *)access_journal_data(ic, i, l), test_tag);
1827                                         if (unlikely(memcmp(test_tag, je2->tag, ic->tag_size)))
1828                                                 dm_integrity_io_error(ic, "tag mismatch when replaying journal", -EILSEQ);
1829                                 }
1830
1831                                 journal_entry_set_unused(je2);
1832                                 r = dm_integrity_rw_tag(ic, je2->tag, &metadata_block, &metadata_offset,
1833                                                         ic->tag_size, TAG_WRITE);
1834                                 if (unlikely(r)) {
1835                                         dm_integrity_io_error(ic, "reading tags", r);
1836                                 }
1837                         }
1838
1839                         atomic_inc(&comp.in_flight);
1840                         copy_from_journal(ic, i, j, k - j, get_data_sector(ic, area, offset),
1841                                           complete_copy_from_journal, io);
1842 skip_io:
1843                         j = next_loop;
1844                 }
1845         }
1846
1847         dm_bufio_write_dirty_buffers_async(ic->bufio);
1848
1849         complete_journal_op(&comp);
1850         wait_for_completion_io(&comp.comp);
1851
1852         dm_integrity_flush_buffers(ic);
1853 }
1854
1855 static void integrity_writer(struct work_struct *w)
1856 {
1857         struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, writer_work);
1858         unsigned write_start, write_sections;
1859
1860         unsigned prev_free_sectors;
1861
1862         /* the following test is not needed, but it tests the replay code */
1863         if (ACCESS_ONCE(ic->suspending))
1864                 return;
1865
1866         spin_lock_irq(&ic->endio_wait.lock);
1867         write_start = ic->committed_section;
1868         write_sections = ic->n_committed_sections;
1869         spin_unlock_irq(&ic->endio_wait.lock);
1870
1871         if (!write_sections)
1872                 return;
1873
1874         do_journal_write(ic, write_start, write_sections, false);
1875
1876         spin_lock_irq(&ic->endio_wait.lock);
1877
1878         ic->committed_section += write_sections;
1879         wraparound_section(ic, &ic->committed_section);
1880         ic->n_committed_sections -= write_sections;
1881
1882         prev_free_sectors = ic->free_sectors;
1883         ic->free_sectors += write_sections * ic->journal_section_entries;
1884         if (unlikely(!prev_free_sectors))
1885                 wake_up_locked(&ic->endio_wait);
1886
1887         spin_unlock_irq(&ic->endio_wait.lock);
1888 }
1889
1890 static void init_journal(struct dm_integrity_c *ic, unsigned start_section,
1891                          unsigned n_sections, unsigned char commit_seq)
1892 {
1893         unsigned i, j, n;
1894
1895         if (!n_sections)
1896                 return;
1897
1898         for (n = 0; n < n_sections; n++) {
1899                 i = start_section + n;
1900                 wraparound_section(ic, &i);
1901                 for (j = 0; j < ic->journal_section_sectors; j++) {
1902                         struct journal_sector *js = access_journal(ic, i, j);
1903                         memset(&js->entries, 0, JOURNAL_SECTOR_DATA);
1904                         js->commit_id = dm_integrity_commit_id(ic, i, j, commit_seq);
1905                 }
1906                 for (j = 0; j < ic->journal_section_entries; j++) {
1907                         struct journal_entry *je = access_journal_entry(ic, i, j);
1908                         journal_entry_set_unused(je);
1909                 }
1910         }
1911
1912         write_journal(ic, start_section, n_sections);
1913 }
1914
1915 static int find_commit_seq(struct dm_integrity_c *ic, unsigned i, unsigned j, commit_id_t id)
1916 {
1917         unsigned char k;
1918         for (k = 0; k < N_COMMIT_IDS; k++) {
1919                 if (dm_integrity_commit_id(ic, i, j, k) == id)
1920                         return k;
1921         }
1922         dm_integrity_io_error(ic, "journal commit id", -EIO);
1923         return -EIO;
1924 }
1925
1926 static void replay_journal(struct dm_integrity_c *ic)
1927 {
1928         unsigned i, j;
1929         bool used_commit_ids[N_COMMIT_IDS];
1930         unsigned max_commit_id_sections[N_COMMIT_IDS];
1931         unsigned write_start, write_sections;
1932         unsigned continue_section;
1933         bool journal_empty;
1934         unsigned char unused, last_used, want_commit_seq;
1935
1936         if (ic->mode == 'R')
1937                 return;
1938
1939         if (ic->journal_uptodate)
1940                 return;
1941
1942         last_used = 0;
1943         write_start = 0;
1944
1945         if (!ic->just_formatted) {
1946                 DEBUG_print("reading journal\n");
1947                 rw_journal(ic, REQ_OP_READ, 0, 0, ic->journal_sections, NULL);
1948                 if (ic->journal_io)
1949                         DEBUG_bytes(lowmem_page_address(ic->journal_io[0].page), 64, "read journal");
1950                 if (ic->journal_io) {
1951                         struct journal_completion crypt_comp;
1952                         crypt_comp.ic = ic;
1953                         crypt_comp.comp = COMPLETION_INITIALIZER_ONSTACK(crypt_comp.comp);
1954                         crypt_comp.in_flight = (atomic_t)ATOMIC_INIT(0);
1955                         encrypt_journal(ic, false, 0, ic->journal_sections, &crypt_comp);
1956                         wait_for_completion(&crypt_comp.comp);
1957                 }
1958                 DEBUG_bytes(lowmem_page_address(ic->journal[0].page), 64, "decrypted journal");
1959         }
1960
1961         if (dm_integrity_failed(ic))
1962                 goto clear_journal;
1963
1964         journal_empty = true;
1965         memset(used_commit_ids, 0, sizeof used_commit_ids);
1966         memset(max_commit_id_sections, 0, sizeof max_commit_id_sections);
1967         for (i = 0; i < ic->journal_sections; i++) {
1968                 for (j = 0; j < ic->journal_section_sectors; j++) {
1969                         int k;
1970                         struct journal_sector *js = access_journal(ic, i, j);
1971                         k = find_commit_seq(ic, i, j, js->commit_id);
1972                         if (k < 0)
1973                                 goto clear_journal;
1974                         used_commit_ids[k] = true;
1975                         max_commit_id_sections[k] = i;
1976                 }
1977                 if (journal_empty) {
1978                         for (j = 0; j < ic->journal_section_entries; j++) {
1979                                 struct journal_entry *je = access_journal_entry(ic, i, j);
1980                                 if (!journal_entry_is_unused(je)) {
1981                                         journal_empty = false;
1982                                         break;
1983                                 }
1984                         }
1985                 }
1986         }
1987
1988         if (!used_commit_ids[N_COMMIT_IDS - 1]) {
1989                 unused = N_COMMIT_IDS - 1;
1990                 while (unused && !used_commit_ids[unused - 1])
1991                         unused--;
1992         } else {
1993                 for (unused = 0; unused < N_COMMIT_IDS; unused++)
1994                         if (!used_commit_ids[unused])
1995                                 break;
1996                 if (unused == N_COMMIT_IDS) {
1997                         dm_integrity_io_error(ic, "journal commit ids", -EIO);
1998                         goto clear_journal;
1999                 }
2000         }
2001         DEBUG_print("first unused commit seq %d [%d,%d,%d,%d]\n",
2002                     unused, used_commit_ids[0], used_commit_ids[1],
2003                     used_commit_ids[2], used_commit_ids[3]);
2004
2005         last_used = prev_commit_seq(unused);
2006         want_commit_seq = prev_commit_seq(last_used);
2007
2008         if (!used_commit_ids[want_commit_seq] && used_commit_ids[prev_commit_seq(want_commit_seq)])
2009                 journal_empty = true;
2010
2011         write_start = max_commit_id_sections[last_used] + 1;
2012         if (unlikely(write_start >= ic->journal_sections))
2013                 want_commit_seq = next_commit_seq(want_commit_seq);
2014         wraparound_section(ic, &write_start);
2015
2016         i = write_start;
2017         for (write_sections = 0; write_sections < ic->journal_sections; write_sections++) {
2018                 for (j = 0; j < ic->journal_section_sectors; j++) {
2019                         struct journal_sector *js = access_journal(ic, i, j);
2020
2021                         if (js->commit_id != dm_integrity_commit_id(ic, i, j, want_commit_seq)) {
2022                                 /*
2023                                  * This could be caused by crash during writing.
2024                                  * We won't replay the inconsistent part of the
2025                                  * journal.
2026                                  */
2027                                 DEBUG_print("commit id mismatch at position (%u, %u): %d != %d\n",
2028                                             i, j, find_commit_seq(ic, i, j, js->commit_id), want_commit_seq);
2029                                 goto brk;
2030                         }
2031                 }
2032                 i++;
2033                 if (unlikely(i >= ic->journal_sections))
2034                         want_commit_seq = next_commit_seq(want_commit_seq);
2035                 wraparound_section(ic, &i);
2036         }
2037 brk:
2038
2039         if (!journal_empty) {
2040                 DEBUG_print("replaying %u sections, starting at %u, commit seq %d\n",
2041                             write_sections, write_start, want_commit_seq);
2042                 do_journal_write(ic, write_start, write_sections, true);
2043         }
2044
2045         if (write_sections == ic->journal_sections && (ic->mode == 'J' || journal_empty)) {
2046                 continue_section = write_start;
2047                 ic->commit_seq = want_commit_seq;
2048                 DEBUG_print("continuing from section %u, commit seq %d\n", write_start, ic->commit_seq);
2049         } else {
2050                 unsigned s;
2051                 unsigned char erase_seq;
2052 clear_journal:
2053                 DEBUG_print("clearing journal\n");
2054
2055                 erase_seq = prev_commit_seq(prev_commit_seq(last_used));
2056                 s = write_start;
2057                 init_journal(ic, s, 1, erase_seq);
2058                 s++;
2059                 wraparound_section(ic, &s);
2060                 if (ic->journal_sections >= 2) {
2061                         init_journal(ic, s, ic->journal_sections - 2, erase_seq);
2062                         s += ic->journal_sections - 2;
2063                         wraparound_section(ic, &s);
2064                         init_journal(ic, s, 1, erase_seq);
2065                 }
2066
2067                 continue_section = 0;
2068                 ic->commit_seq = next_commit_seq(erase_seq);
2069         }
2070
2071         ic->committed_section = continue_section;
2072         ic->n_committed_sections = 0;
2073
2074         ic->uncommitted_section = continue_section;
2075         ic->n_uncommitted_sections = 0;
2076
2077         ic->free_section = continue_section;
2078         ic->free_section_entry = 0;
2079         ic->free_sectors = ic->journal_entries;
2080
2081         ic->journal_tree_root = RB_ROOT;
2082         for (i = 0; i < ic->journal_entries; i++)
2083                 init_journal_node(&ic->journal_tree[i]);
2084 }
2085
2086 static void dm_integrity_postsuspend(struct dm_target *ti)
2087 {
2088         struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
2089
2090         del_timer_sync(&ic->autocommit_timer);
2091
2092         ic->suspending = true;
2093
2094         queue_work(ic->commit_wq, &ic->commit_work);
2095         drain_workqueue(ic->commit_wq);
2096
2097         if (ic->mode == 'J') {
2098                 drain_workqueue(ic->writer_wq);
2099                 dm_integrity_flush_buffers(ic);
2100         }
2101
2102         ic->suspending = false;
2103
2104         BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
2105
2106         ic->journal_uptodate = true;
2107 }
2108
2109 static void dm_integrity_resume(struct dm_target *ti)
2110 {
2111         struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
2112
2113         replay_journal(ic);
2114 }
2115
2116 static void dm_integrity_status(struct dm_target *ti, status_type_t type,
2117                                 unsigned status_flags, char *result, unsigned maxlen)
2118 {
2119         struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
2120         unsigned arg_count;
2121         size_t sz = 0;
2122
2123         switch (type) {
2124         case STATUSTYPE_INFO:
2125                 result[0] = '\0';
2126                 break;
2127
2128         case STATUSTYPE_TABLE: {
2129                 __u64 watermark_percentage = (__u64)(ic->journal_entries - ic->free_sectors_threshold) * 100;
2130                 watermark_percentage += ic->journal_entries / 2;
2131                 do_div(watermark_percentage, ic->journal_entries);
2132                 arg_count = 5;
2133                 arg_count += !!ic->internal_hash_alg.alg_string;
2134                 arg_count += !!ic->journal_crypt_alg.alg_string;
2135                 arg_count += !!ic->journal_mac_alg.alg_string;
2136                 DMEMIT("%s %llu %u %c %u", ic->dev->name, (unsigned long long)ic->start,
2137                        ic->tag_size, ic->mode, arg_count);
2138                 DMEMIT(" journal_sectors:%u", ic->initial_sectors - SB_SECTORS);
2139                 DMEMIT(" interleave_sectors:%u", 1U << ic->sb->log2_interleave_sectors);
2140                 DMEMIT(" buffer_sectors:%u", 1U << ic->log2_buffer_sectors);
2141                 DMEMIT(" journal_watermark:%u", (unsigned)watermark_percentage);
2142                 DMEMIT(" commit_time:%u", ic->autocommit_msec);
2143
2144 #define EMIT_ALG(a, n)                                                  \
2145                 do {                                                    \
2146                         if (ic->a.alg_string) {                         \
2147                                 DMEMIT(" %s:%s", n, ic->a.alg_string);  \
2148                                 if (ic->a.key_string)                   \
2149                                         DMEMIT(":%s", ic->a.key_string);\
2150                         }                                               \
2151                 } while (0)
2152                 EMIT_ALG(internal_hash_alg, "internal_hash");
2153                 EMIT_ALG(journal_crypt_alg, "journal_crypt");
2154                 EMIT_ALG(journal_mac_alg, "journal_mac");
2155                 break;
2156         }
2157         }
2158 }
2159
2160 static int dm_integrity_iterate_devices(struct dm_target *ti,
2161                                         iterate_devices_callout_fn fn, void *data)
2162 {
2163         struct dm_integrity_c *ic = ti->private;
2164
2165         return fn(ti, ic->dev, ic->start + ic->initial_sectors + ic->metadata_run, ti->len, data);
2166 }
2167
2168 static void calculate_journal_section_size(struct dm_integrity_c *ic)
2169 {
2170         unsigned sector_space = JOURNAL_SECTOR_DATA;
2171
2172         ic->journal_sections = le32_to_cpu(ic->sb->journal_sections);
2173         ic->journal_entry_size = roundup(offsetof(struct journal_entry, tag) + ic->tag_size,
2174                                          JOURNAL_ENTRY_ROUNDUP);
2175
2176         if (ic->sb->flags & cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC))
2177                 sector_space -= JOURNAL_MAC_PER_SECTOR;
2178         ic->journal_entries_per_sector = sector_space / ic->journal_entry_size;
2179         ic->journal_section_entries = ic->journal_entries_per_sector * JOURNAL_BLOCK_SECTORS;
2180         ic->journal_section_sectors = ic->journal_section_entries + JOURNAL_BLOCK_SECTORS;
2181         ic->journal_entries = ic->journal_section_entries * ic->journal_sections;
2182 }
2183
2184 static int calculate_device_limits(struct dm_integrity_c *ic)
2185 {
2186         __u64 initial_sectors;
2187         sector_t last_sector, last_area, last_offset;
2188
2189         calculate_journal_section_size(ic);
2190         initial_sectors = SB_SECTORS + (__u64)ic->journal_section_sectors * ic->journal_sections;
2191         if (initial_sectors + METADATA_PADDING_SECTORS >= ic->device_sectors || initial_sectors > UINT_MAX)
2192                 return -EINVAL;
2193         ic->initial_sectors = initial_sectors;
2194
2195         ic->metadata_run = roundup((__u64)ic->tag_size << ic->sb->log2_interleave_sectors,
2196                                    (__u64)(1 << SECTOR_SHIFT << METADATA_PADDING_SECTORS)) >> SECTOR_SHIFT;
2197         if (!(ic->metadata_run & (ic->metadata_run - 1)))
2198                 ic->log2_metadata_run = __ffs(ic->metadata_run);
2199         else
2200                 ic->log2_metadata_run = -1;
2201
2202         get_area_and_offset(ic, ic->provided_data_sectors - 1, &last_area, &last_offset);
2203         last_sector = get_data_sector(ic, last_area, last_offset);
2204
2205         if (ic->start + last_sector < last_sector || ic->start + last_sector >= ic->device_sectors)
2206                 return -EINVAL;
2207
2208         return 0;
2209 }
2210
2211 static int initialize_superblock(struct dm_integrity_c *ic, unsigned journal_sectors, unsigned interleave_sectors)
2212 {
2213         unsigned journal_sections;
2214         int test_bit;
2215
2216         memset(ic->sb, 0, SB_SECTORS << SECTOR_SHIFT);
2217         memcpy(ic->sb->magic, SB_MAGIC, 8);
2218         ic->sb->version = SB_VERSION;
2219         ic->sb->integrity_tag_size = cpu_to_le16(ic->tag_size);
2220         if (ic->journal_mac_alg.alg_string)
2221                 ic->sb->flags |= cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC);
2222
2223         calculate_journal_section_size(ic);
2224         journal_sections = journal_sectors / ic->journal_section_sectors;
2225         if (!journal_sections)
2226                 journal_sections = 1;
2227         ic->sb->journal_sections = cpu_to_le32(journal_sections);
2228
2229         if (!interleave_sectors)
2230                 interleave_sectors = DEFAULT_INTERLEAVE_SECTORS;
2231         ic->sb->log2_interleave_sectors = __fls(interleave_sectors);
2232         ic->sb->log2_interleave_sectors = max((__u8)MIN_LOG2_INTERLEAVE_SECTORS, ic->sb->log2_interleave_sectors);
2233         ic->sb->log2_interleave_sectors = min((__u8)MAX_LOG2_INTERLEAVE_SECTORS, ic->sb->log2_interleave_sectors);
2234
2235         ic->provided_data_sectors = 0;
2236         for (test_bit = fls64(ic->device_sectors) - 1; test_bit >= 3; test_bit--) {
2237                 __u64 prev_data_sectors = ic->provided_data_sectors;
2238
2239                 ic->provided_data_sectors |= (sector_t)1 << test_bit;
2240                 if (calculate_device_limits(ic))
2241                         ic->provided_data_sectors = prev_data_sectors;
2242         }
2243
2244         if (!ic->provided_data_sectors)
2245                 return -EINVAL;
2246
2247         ic->sb->provided_data_sectors = cpu_to_le64(ic->provided_data_sectors);
2248
2249         return 0;
2250 }
2251
2252 static void dm_integrity_set(struct dm_target *ti, struct dm_integrity_c *ic)
2253 {
2254         struct gendisk *disk = dm_disk(dm_table_get_md(ti->table));
2255         struct blk_integrity bi;
2256
2257         memset(&bi, 0, sizeof(bi));
2258         bi.profile = &dm_integrity_profile;
2259         bi.tuple_size = ic->tag_size * (queue_logical_block_size(disk->queue) >> SECTOR_SHIFT);
2260         bi.tag_size = ic->tag_size;
2261
2262         blk_integrity_register(disk, &bi);
2263         blk_queue_max_integrity_segments(disk->queue, UINT_MAX);
2264 }
2265
2266 /* FIXME: use new kvmalloc */
2267 static void *dm_integrity_kvmalloc(size_t size, gfp_t gfp)
2268 {
2269         void *ptr = NULL;
2270
2271         if (size <= PAGE_SIZE)
2272                 ptr = kmalloc(size, GFP_KERNEL | gfp);
2273         if (!ptr && size <= KMALLOC_MAX_SIZE)
2274                 ptr = kmalloc(size, GFP_KERNEL | __GFP_NOWARN | __GFP_NORETRY | gfp);
2275         if (!ptr)
2276                 ptr = __vmalloc(size, GFP_KERNEL | gfp, PAGE_KERNEL);
2277
2278         return ptr;
2279 }
2280
2281 static void dm_integrity_free_page_list(struct dm_integrity_c *ic, struct page_list *pl)
2282 {
2283         unsigned i;
2284
2285         if (!pl)
2286                 return;
2287         for (i = 0; i < ic->journal_pages; i++)
2288                 if (pl[i].page)
2289                         __free_page(pl[i].page);
2290         kvfree(pl);
2291 }
2292
2293 static struct page_list *dm_integrity_alloc_page_list(struct dm_integrity_c *ic)
2294 {
2295         size_t page_list_desc_size = ic->journal_pages * sizeof(struct page_list);
2296         struct page_list *pl;
2297         unsigned i;
2298
2299         pl = dm_integrity_kvmalloc(page_list_desc_size, __GFP_ZERO);
2300         if (!pl)
2301                 return NULL;
2302
2303         for (i = 0; i < ic->journal_pages; i++) {
2304                 pl[i].page = alloc_page(GFP_KERNEL);
2305                 if (!pl[i].page) {
2306                         dm_integrity_free_page_list(ic, pl);
2307                         return NULL;
2308                 }
2309                 if (i)
2310                         pl[i - 1].next = &pl[i];
2311         }
2312
2313         return pl;
2314 }
2315
2316 static void dm_integrity_free_journal_scatterlist(struct dm_integrity_c *ic, struct scatterlist **sl)
2317 {
2318         unsigned i;
2319         for (i = 0; i < ic->journal_sections; i++)
2320                 kvfree(sl[i]);
2321         kfree(sl);
2322 }
2323
2324 static struct scatterlist **dm_integrity_alloc_journal_scatterlist(struct dm_integrity_c *ic, struct page_list *pl)
2325 {
2326         struct scatterlist **sl;
2327         unsigned i;
2328
2329         sl = dm_integrity_kvmalloc(ic->journal_sections * sizeof(struct scatterlist *), __GFP_ZERO);
2330         if (!sl)
2331                 return NULL;
2332
2333         for (i = 0; i < ic->journal_sections; i++) {
2334                 struct scatterlist *s;
2335                 unsigned start_index, start_offset;
2336                 unsigned end_index, end_offset;
2337                 unsigned n_pages;
2338                 unsigned idx;
2339
2340                 page_list_location(ic, i, 0, &start_index, &start_offset);
2341                 page_list_location(ic, i, ic->journal_section_sectors - 1, &end_index, &end_offset);
2342
2343                 n_pages = (end_index - start_index + 1);
2344
2345                 s = dm_integrity_kvmalloc(n_pages * sizeof(struct scatterlist), 0);
2346                 if (!s) {
2347                         dm_integrity_free_journal_scatterlist(ic, sl);
2348                         return NULL;
2349                 }
2350
2351                 sg_init_table(s, n_pages);
2352                 for (idx = start_index; idx <= end_index; idx++) {
2353                         char *va = lowmem_page_address(pl[idx].page);
2354                         unsigned start = 0, end = PAGE_SIZE;
2355                         if (idx == start_index)
2356                                 start = start_offset;
2357                         if (idx == end_index)
2358                                 end = end_offset + (1 << SECTOR_SHIFT);
2359                         sg_set_buf(&s[idx - start_index], va + start, end - start);
2360                 }
2361
2362                 sl[i] = s;
2363         }
2364
2365         return sl;
2366 }
2367
2368 static void free_alg(struct alg_spec *a)
2369 {
2370         kzfree(a->alg_string);
2371         kzfree(a->key);
2372         memset(a, 0, sizeof *a);
2373 }
2374
2375 static int get_alg_and_key(const char *arg, struct alg_spec *a, char **error, char *error_inval)
2376 {
2377         char *k;
2378
2379         free_alg(a);
2380
2381         a->alg_string = kstrdup(strchr(arg, ':') + 1, GFP_KERNEL);
2382         if (!a->alg_string)
2383                 goto nomem;
2384
2385         k = strchr(a->alg_string, ':');
2386         if (k) {
2387                 unsigned i;
2388
2389                 *k = 0;
2390                 a->key_string = k + 1;
2391                 if (strlen(a->key_string) & 1)
2392                         goto inval;
2393
2394                 a->key_size = strlen(a->key_string) / 2;
2395                 a->key = kmalloc(a->key_size, GFP_KERNEL);
2396                 if (!a->key)
2397                         goto nomem;
2398                 for (i = 0; i < a->key_size; i++) {
2399                         char digit[3];
2400                         digit[0] = a->key_string[i * 2];
2401                         digit[1] = a->key_string[i * 2 + 1];
2402                         digit[2] = 0;
2403                         if (strspn(digit, "0123456789abcdefABCDEF") != 2)
2404                                 goto inval;
2405                         if (kstrtou8(digit, 16, &a->key[i]))
2406                                 goto inval;
2407                 }
2408         }
2409
2410         return 0;
2411 inval:
2412         *error = error_inval;
2413         return -EINVAL;
2414 nomem:
2415         *error = "Out of memory for an argument";
2416         return -ENOMEM;
2417 }
2418
2419 static int get_mac(struct crypto_shash **hash, struct alg_spec *a, char **error,
2420                    char *error_alg, char *error_key)
2421 {
2422         int r;
2423
2424         if (a->alg_string) {
2425                 *hash = crypto_alloc_shash(a->alg_string, 0, CRYPTO_ALG_ASYNC);
2426                 if (IS_ERR(*hash)) {
2427                         *error = error_alg;
2428                         r = PTR_ERR(*hash);
2429                         *hash = NULL;
2430                         return r;
2431                 }
2432
2433                 if (a->key) {
2434                         r = crypto_shash_setkey(*hash, a->key, a->key_size);
2435                         if (r) {
2436                                 *error = error_key;
2437                                 return r;
2438                         }
2439                 }
2440         }
2441
2442         return 0;
2443 }
2444
2445 static int create_journal(struct dm_integrity_c *ic, char **error)
2446 {
2447         int r = 0;
2448         unsigned i;
2449         __u64 journal_pages, journal_desc_size, journal_tree_size;
2450         unsigned char *crypt_data = NULL;
2451
2452         ic->commit_ids[0] = cpu_to_le64(0x1111111111111111ULL);
2453         ic->commit_ids[1] = cpu_to_le64(0x2222222222222222ULL);
2454         ic->commit_ids[2] = cpu_to_le64(0x3333333333333333ULL);
2455         ic->commit_ids[3] = cpu_to_le64(0x4444444444444444ULL);
2456
2457         journal_pages = roundup((__u64)ic->journal_sections * ic->journal_section_sectors,
2458                                 PAGE_SIZE >> SECTOR_SHIFT) >> (PAGE_SHIFT - SECTOR_SHIFT);
2459         journal_desc_size = journal_pages * sizeof(struct page_list);
2460         if (journal_pages >= totalram_pages - totalhigh_pages || journal_desc_size > ULONG_MAX) {
2461                 *error = "Journal doesn't fit into memory";
2462                 r = -ENOMEM;
2463                 goto bad;
2464         }
2465         ic->journal_pages = journal_pages;
2466
2467         ic->journal = dm_integrity_alloc_page_list(ic);
2468         if (!ic->journal) {
2469                 *error = "Could not allocate memory for journal";
2470                 r = -ENOMEM;
2471                 goto bad;
2472         }
2473         if (ic->journal_crypt_alg.alg_string) {
2474                 unsigned ivsize, blocksize;
2475                 struct journal_completion comp;
2476
2477                 comp.ic = ic;
2478                 ic->journal_crypt = crypto_alloc_skcipher(ic->journal_crypt_alg.alg_string, 0, 0);
2479                 if (IS_ERR(ic->journal_crypt)) {
2480                         *error = "Invalid journal cipher";
2481                         r = PTR_ERR(ic->journal_crypt);
2482                         ic->journal_crypt = NULL;
2483                         goto bad;
2484                 }
2485                 ivsize = crypto_skcipher_ivsize(ic->journal_crypt);
2486                 blocksize = crypto_skcipher_blocksize(ic->journal_crypt);
2487
2488                 if (ic->journal_crypt_alg.key) {
2489                         r = crypto_skcipher_setkey(ic->journal_crypt, ic->journal_crypt_alg.key,
2490                                                    ic->journal_crypt_alg.key_size);
2491                         if (r) {
2492                                 *error = "Error setting encryption key";
2493                                 goto bad;
2494                         }
2495                 }
2496                 DEBUG_print("cipher %s, block size %u iv size %u\n",
2497                             ic->journal_crypt_alg.alg_string, blocksize, ivsize);
2498
2499                 ic->journal_io = dm_integrity_alloc_page_list(ic);
2500                 if (!ic->journal_io) {
2501                         *error = "Could not allocate memory for journal io";
2502                         r = -ENOMEM;
2503                         goto bad;
2504                 }
2505
2506                 if (blocksize == 1) {
2507                         struct scatterlist *sg;
2508                         SKCIPHER_REQUEST_ON_STACK(req, ic->journal_crypt);
2509                         unsigned char iv[ivsize];
2510                         skcipher_request_set_tfm(req, ic->journal_crypt);
2511
2512                         ic->journal_xor = dm_integrity_alloc_page_list(ic);
2513                         if (!ic->journal_xor) {
2514                                 *error = "Could not allocate memory for journal xor";
2515                                 r = -ENOMEM;
2516                                 goto bad;
2517                         }
2518
2519                         sg = dm_integrity_kvmalloc((ic->journal_pages + 1) * sizeof(struct scatterlist), 0);
2520                         if (!sg) {
2521                                 *error = "Unable to allocate sg list";
2522                                 r = -ENOMEM;
2523                                 goto bad;
2524                         }
2525                         sg_init_table(sg, ic->journal_pages + 1);
2526                         for (i = 0; i < ic->journal_pages; i++) {
2527                                 char *va = lowmem_page_address(ic->journal_xor[i].page);
2528                                 clear_page(va);
2529                                 sg_set_buf(&sg[i], va, PAGE_SIZE);
2530                         }
2531                         sg_set_buf(&sg[i], &ic->commit_ids, sizeof ic->commit_ids);
2532                         memset(iv, 0x00, ivsize);
2533
2534                         skcipher_request_set_crypt(req, sg, sg, PAGE_SIZE * ic->journal_pages + sizeof ic->commit_ids, iv);
2535                         comp.comp = COMPLETION_INITIALIZER_ONSTACK(comp.comp);
2536                         comp.in_flight = (atomic_t)ATOMIC_INIT(1);
2537                         if (do_crypt(true, req, &comp))
2538                                 wait_for_completion(&comp.comp);
2539                         kvfree(sg);
2540                         r = dm_integrity_failed(ic);
2541                         if (r) {
2542                                 *error = "Unable to encrypt journal";
2543                                 goto bad;
2544                         }
2545                         DEBUG_bytes(lowmem_page_address(ic->journal_xor[0].page), 64, "xor data");
2546
2547                         crypto_free_skcipher(ic->journal_crypt);
2548                         ic->journal_crypt = NULL;
2549                 } else {
2550                         SKCIPHER_REQUEST_ON_STACK(req, ic->journal_crypt);
2551                         unsigned char iv[ivsize];
2552                         unsigned crypt_len = roundup(ivsize, blocksize);
2553
2554                         crypt_data = kmalloc(crypt_len, GFP_KERNEL);
2555                         if (!crypt_data) {
2556                                 *error = "Unable to allocate crypt data";
2557                                 r = -ENOMEM;
2558                                 goto bad;
2559                         }
2560
2561                         skcipher_request_set_tfm(req, ic->journal_crypt);
2562
2563                         ic->journal_scatterlist = dm_integrity_alloc_journal_scatterlist(ic, ic->journal);
2564                         if (!ic->journal_scatterlist) {
2565                                 *error = "Unable to allocate sg list";
2566                                 r = -ENOMEM;
2567                                 goto bad;
2568                         }
2569                         ic->journal_io_scatterlist = dm_integrity_alloc_journal_scatterlist(ic, ic->journal_io);
2570                         if (!ic->journal_io_scatterlist) {
2571                                 *error = "Unable to allocate sg list";
2572                                 r = -ENOMEM;
2573                                 goto bad;
2574                         }
2575                         ic->sk_requests = dm_integrity_kvmalloc(ic->journal_sections * sizeof(struct skcipher_request *), __GFP_ZERO);
2576                         if (!ic->sk_requests) {
2577                                 *error = "Unable to allocate sk requests";
2578                                 r = -ENOMEM;
2579                                 goto bad;
2580                         }
2581                         for (i = 0; i < ic->journal_sections; i++) {
2582                                 struct scatterlist sg;
2583                                 struct skcipher_request *section_req;
2584                                 __u32 section_le = cpu_to_le32(i);
2585
2586                                 memset(iv, 0x00, ivsize);
2587                                 memset(crypt_data, 0x00, crypt_len);
2588                                 memcpy(crypt_data, &section_le, min((size_t)crypt_len, sizeof(section_le)));
2589
2590                                 sg_init_one(&sg, crypt_data, crypt_len);
2591                                 skcipher_request_set_crypt(req, &sg, &sg, crypt_len, iv);
2592                                 comp.comp = COMPLETION_INITIALIZER_ONSTACK(comp.comp);
2593                                 comp.in_flight = (atomic_t)ATOMIC_INIT(1);
2594                                 if (do_crypt(true, req, &comp))
2595                                         wait_for_completion(&comp.comp);
2596
2597                                 r = dm_integrity_failed(ic);
2598                                 if (r) {
2599                                         *error = "Unable to generate iv";
2600                                         goto bad;
2601                                 }
2602
2603                                 section_req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
2604                                 if (!section_req) {
2605                                         *error = "Unable to allocate crypt request";
2606                                         r = -ENOMEM;
2607                                         goto bad;
2608                                 }
2609                                 section_req->iv = kmalloc(ivsize * 2, GFP_KERNEL);
2610                                 if (!section_req->iv) {
2611                                         skcipher_request_free(section_req);
2612                                         *error = "Unable to allocate iv";
2613                                         r = -ENOMEM;
2614                                         goto bad;
2615                                 }
2616                                 memcpy(section_req->iv + ivsize, crypt_data, ivsize);
2617                                 section_req->cryptlen = (size_t)ic->journal_section_sectors << SECTOR_SHIFT;
2618                                 ic->sk_requests[i] = section_req;
2619                                 DEBUG_bytes(crypt_data, ivsize, "iv(%u)", i);
2620                         }
2621                 }
2622         }
2623
2624         for (i = 0; i < N_COMMIT_IDS; i++) {
2625                 unsigned j;
2626 retest_commit_id:
2627                 for (j = 0; j < i; j++) {
2628                         if (ic->commit_ids[j] == ic->commit_ids[i]) {
2629                                 ic->commit_ids[i] = cpu_to_le64(le64_to_cpu(ic->commit_ids[i]) + 1);
2630                                 goto retest_commit_id;
2631                         }
2632                 }
2633                 DEBUG_print("commit id %u: %016llx\n", i, ic->commit_ids[i]);
2634         }
2635
2636         journal_tree_size = (__u64)ic->journal_entries * sizeof(struct journal_node);
2637         if (journal_tree_size > ULONG_MAX) {
2638                 *error = "Journal doesn't fit into memory";
2639                 r = -ENOMEM;
2640                 goto bad;
2641         }
2642         ic->journal_tree = dm_integrity_kvmalloc(journal_tree_size, 0);
2643         if (!ic->journal_tree) {
2644                 *error = "Could not allocate memory for journal tree";
2645                 r = -ENOMEM;
2646         }
2647 bad:
2648         kfree(crypt_data);
2649         return r;
2650 }
2651
2652 /*
2653  * Construct a integrity mapping
2654  *
2655  * Arguments:
2656  *      device
2657  *      offset from the start of the device
2658  *      tag size
2659  *      D - direct writes, J - journal writes, R - recovery mode
2660  *      number of optional arguments
2661  *      optional arguments:
2662  *              journal_sectors
2663  *              interleave_sectors
2664  *              buffer_sectors
2665  *              journal_watermark
2666  *              commit_time
2667  *              internal_hash
2668  *              journal_crypt
2669  *              journal_mac
2670  */
2671 static int dm_integrity_ctr(struct dm_target *ti, unsigned argc, char **argv)
2672 {
2673         struct dm_integrity_c *ic;
2674         char dummy;
2675         int r;
2676         unsigned extra_args;
2677         struct dm_arg_set as;
2678         static struct dm_arg _args[] = {
2679                 {0, 8, "Invalid number of feature args"},
2680         };
2681         unsigned journal_sectors, interleave_sectors, buffer_sectors, journal_watermark, sync_msec;
2682         bool should_write_sb;
2683         __u64 threshold;
2684         unsigned long long start;
2685
2686 #define DIRECT_ARGUMENTS        4
2687
2688         if (argc <= DIRECT_ARGUMENTS) {
2689                 ti->error = "Invalid argument count";
2690                 return -EINVAL;
2691         }
2692
2693         ic = kzalloc(sizeof(struct dm_integrity_c), GFP_KERNEL);
2694         if (!ic) {
2695                 ti->error = "Cannot allocate integrity context";
2696                 return -ENOMEM;
2697         }
2698         ti->private = ic;
2699         ti->per_io_data_size = sizeof(struct dm_integrity_io);
2700
2701         ic->in_progress = RB_ROOT;
2702         init_waitqueue_head(&ic->endio_wait);
2703         bio_list_init(&ic->flush_bio_list);
2704         init_waitqueue_head(&ic->copy_to_journal_wait);
2705         init_completion(&ic->crypto_backoff);
2706
2707         r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &ic->dev);
2708         if (r) {
2709                 ti->error = "Device lookup failed";
2710                 goto bad;
2711         }
2712
2713         if (sscanf(argv[1], "%llu%c", &start, &dummy) != 1 || start != (sector_t)start) {
2714                 ti->error = "Invalid starting offset";
2715                 r = -EINVAL;
2716                 goto bad;
2717         }
2718         ic->start = start;
2719
2720         if (strcmp(argv[2], "-")) {
2721                 if (sscanf(argv[2], "%u%c", &ic->tag_size, &dummy) != 1 || !ic->tag_size) {
2722                         ti->error = "Invalid tag size";
2723                         r = -EINVAL;
2724                         goto bad;
2725                 }
2726         }
2727
2728         if (!strcmp(argv[3], "J") || !strcmp(argv[3], "D") || !strcmp(argv[3], "R"))
2729                 ic->mode = argv[3][0];
2730         else {
2731                 ti->error = "Invalid mode (expecting J, D, R)";
2732                 r = -EINVAL;
2733                 goto bad;
2734         }
2735
2736         ic->device_sectors = i_size_read(ic->dev->bdev->bd_inode) >> SECTOR_SHIFT;
2737         journal_sectors = min((sector_t)DEFAULT_MAX_JOURNAL_SECTORS,
2738                         ic->device_sectors >> DEFAULT_JOURNAL_SIZE_FACTOR);
2739         interleave_sectors = DEFAULT_INTERLEAVE_SECTORS;
2740         buffer_sectors = DEFAULT_BUFFER_SECTORS;
2741         journal_watermark = DEFAULT_JOURNAL_WATERMARK;
2742         sync_msec = DEFAULT_SYNC_MSEC;
2743
2744         as.argc = argc - DIRECT_ARGUMENTS;
2745         as.argv = argv + DIRECT_ARGUMENTS;
2746         r = dm_read_arg_group(_args, &as, &extra_args, &ti->error);
2747         if (r)
2748                 goto bad;
2749
2750         while (extra_args--) {
2751                 const char *opt_string;
2752                 unsigned val;
2753                 opt_string = dm_shift_arg(&as);
2754                 if (!opt_string) {
2755                         r = -EINVAL;
2756                         ti->error = "Not enough feature arguments";
2757                         goto bad;
2758                 }
2759                 if (sscanf(opt_string, "journal_sectors:%u%c", &val, &dummy) == 1)
2760                         journal_sectors = val;
2761                 else if (sscanf(opt_string, "interleave_sectors:%u%c", &val, &dummy) == 1)
2762                         interleave_sectors = val;
2763                 else if (sscanf(opt_string, "buffer_sectors:%u%c", &val, &dummy) == 1)
2764                         buffer_sectors = val;
2765                 else if (sscanf(opt_string, "journal_watermark:%u%c", &val, &dummy) == 1 && val <= 100)
2766                         journal_watermark = val;
2767                 else if (sscanf(opt_string, "commit_time:%u%c", &val, &dummy) == 1)
2768                         sync_msec = val;
2769                 else if (!memcmp(opt_string, "internal_hash:", strlen("internal_hash:"))) {
2770                         r = get_alg_and_key(opt_string, &ic->internal_hash_alg, &ti->error,
2771                                             "Invalid internal_hash argument");
2772                         if (r)
2773                                 goto bad;
2774                 } else if (!memcmp(opt_string, "journal_crypt:", strlen("journal_crypt:"))) {
2775                         r = get_alg_and_key(opt_string, &ic->journal_crypt_alg, &ti->error,
2776                                             "Invalid journal_crypt argument");
2777                         if (r)
2778                                 goto bad;
2779                 } else if (!memcmp(opt_string, "journal_mac:", strlen("journal_mac:"))) {
2780                         r = get_alg_and_key(opt_string, &ic->journal_mac_alg,  &ti->error,
2781                                             "Invalid journal_mac argument");
2782                         if (r)
2783                                 goto bad;
2784                 } else {
2785                         r = -EINVAL;
2786                         ti->error = "Invalid argument";
2787                         goto bad;
2788                 }
2789         }
2790
2791         r = get_mac(&ic->internal_hash, &ic->internal_hash_alg, &ti->error,
2792                     "Invalid internal hash", "Error setting internal hash key");
2793         if (r)
2794                 goto bad;
2795
2796         r = get_mac(&ic->journal_mac, &ic->journal_mac_alg, &ti->error,
2797                     "Invalid journal mac", "Error setting journal mac key");
2798         if (r)
2799                 goto bad;
2800
2801         if (!ic->tag_size) {
2802                 if (!ic->internal_hash) {
2803                         ti->error = "Unknown tag size";
2804                         r = -EINVAL;
2805                         goto bad;
2806                 }
2807                 ic->tag_size = crypto_shash_digestsize(ic->internal_hash);
2808         }
2809         if (ic->tag_size > MAX_TAG_SIZE) {
2810                 ti->error = "Too big tag size";
2811                 r = -EINVAL;
2812                 goto bad;
2813         }
2814         if (!(ic->tag_size & (ic->tag_size - 1)))
2815                 ic->log2_tag_size = __ffs(ic->tag_size);
2816         else
2817                 ic->log2_tag_size = -1;
2818
2819         ic->autocommit_jiffies = msecs_to_jiffies(sync_msec);
2820         ic->autocommit_msec = sync_msec;
2821         setup_timer(&ic->autocommit_timer, autocommit_fn, (unsigned long)ic);
2822
2823         ic->io = dm_io_client_create();
2824         if (IS_ERR(ic->io)) {
2825                 r = PTR_ERR(ic->io);
2826                 ic->io = NULL;
2827                 ti->error = "Cannot allocate dm io";
2828                 goto bad;
2829         }
2830
2831         ic->journal_io_mempool = mempool_create_slab_pool(JOURNAL_IO_MEMPOOL, journal_io_cache);
2832         if (!ic->journal_io_mempool) {
2833                 r = -ENOMEM;
2834                 ti->error = "Cannot allocate mempool";
2835                 goto bad;
2836         }
2837
2838         ic->metadata_wq = alloc_workqueue("dm-integrity-metadata",
2839                                           WQ_MEM_RECLAIM, METADATA_WORKQUEUE_MAX_ACTIVE);
2840         if (!ic->metadata_wq) {
2841                 ti->error = "Cannot allocate workqueue";
2842                 r = -ENOMEM;
2843                 goto bad;
2844         }
2845
2846         /*
2847          * If this workqueue were percpu, it would cause bio reordering
2848          * and reduced performance.
2849          */
2850         ic->wait_wq = alloc_workqueue("dm-integrity-wait", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
2851         if (!ic->wait_wq) {
2852                 ti->error = "Cannot allocate workqueue";
2853                 r = -ENOMEM;
2854                 goto bad;
2855         }
2856
2857         ic->commit_wq = alloc_workqueue("dm-integrity-commit", WQ_MEM_RECLAIM, 1);
2858         if (!ic->commit_wq) {
2859                 ti->error = "Cannot allocate workqueue";
2860                 r = -ENOMEM;
2861                 goto bad;
2862         }
2863         INIT_WORK(&ic->commit_work, integrity_commit);
2864
2865         if (ic->mode == 'J') {
2866                 ic->writer_wq = alloc_workqueue("dm-integrity-writer", WQ_MEM_RECLAIM, 1);
2867                 if (!ic->writer_wq) {
2868                         ti->error = "Cannot allocate workqueue";
2869                         r = -ENOMEM;
2870                         goto bad;
2871                 }
2872                 INIT_WORK(&ic->writer_work, integrity_writer);
2873         }
2874
2875         ic->sb = alloc_pages_exact(SB_SECTORS << SECTOR_SHIFT, GFP_KERNEL);
2876         if (!ic->sb) {
2877                 r = -ENOMEM;
2878                 ti->error = "Cannot allocate superblock area";
2879                 goto bad;
2880         }
2881
2882         r = sync_rw_sb(ic, REQ_OP_READ, 0);
2883         if (r) {
2884                 ti->error = "Error reading superblock";
2885                 goto bad;
2886         }
2887         should_write_sb = false;
2888         if (memcmp(ic->sb->magic, SB_MAGIC, 8)) {
2889                 if (ic->mode != 'R') {
2890                         if (memchr_inv(ic->sb, 0, SB_SECTORS << SECTOR_SHIFT)) {
2891                                 r = -EINVAL;
2892                                 ti->error = "The device is not initialized";
2893                                 goto bad;
2894                         }
2895                 }
2896
2897                 r = initialize_superblock(ic, journal_sectors, interleave_sectors);
2898                 if (r) {
2899                         ti->error = "Could not initialize superblock";
2900                         goto bad;
2901                 }
2902                 if (ic->mode != 'R')
2903                         should_write_sb = true;
2904         }
2905
2906         if (ic->sb->version != SB_VERSION) {
2907                 r = -EINVAL;
2908                 ti->error = "Unknown version";
2909                 goto bad;
2910         }
2911         if (le16_to_cpu(ic->sb->integrity_tag_size) != ic->tag_size) {
2912                 r = -EINVAL;
2913                 ti->error = "Invalid tag size";
2914                 goto bad;
2915         }
2916         /* make sure that ti->max_io_len doesn't overflow */
2917         if (ic->sb->log2_interleave_sectors < MIN_LOG2_INTERLEAVE_SECTORS ||
2918             ic->sb->log2_interleave_sectors > MAX_LOG2_INTERLEAVE_SECTORS) {
2919                 r = -EINVAL;
2920                 ti->error = "Invalid interleave_sectors in the superblock";
2921                 goto bad;
2922         }
2923         ic->provided_data_sectors = le64_to_cpu(ic->sb->provided_data_sectors);
2924         if (ic->provided_data_sectors != le64_to_cpu(ic->sb->provided_data_sectors)) {
2925                 /* test for overflow */
2926                 r = -EINVAL;
2927                 ti->error = "The superblock has 64-bit device size, but the kernel was compiled with 32-bit sectors";
2928                 goto bad;
2929         }
2930         if (!!(ic->sb->flags & cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC)) != !!ic->journal_mac_alg.alg_string) {
2931                 r = -EINVAL;
2932                 ti->error = "Journal mac mismatch";
2933                 goto bad;
2934         }
2935         r = calculate_device_limits(ic);
2936         if (r) {
2937                 ti->error = "The device is too small";
2938                 goto bad;
2939         }
2940
2941         if (!buffer_sectors)
2942                 buffer_sectors = 1;
2943         ic->log2_buffer_sectors = min3((int)__fls(buffer_sectors), (int)__ffs(ic->metadata_run), 31 - SECTOR_SHIFT);
2944
2945         threshold = (__u64)ic->journal_entries * (100 - journal_watermark);
2946         threshold += 50;
2947         do_div(threshold, 100);
2948         ic->free_sectors_threshold = threshold;
2949
2950         DEBUG_print("initialized:\n");
2951         DEBUG_print("   integrity_tag_size %u\n", le16_to_cpu(ic->sb->integrity_tag_size));
2952         DEBUG_print("   journal_entry_size %u\n", ic->journal_entry_size);
2953         DEBUG_print("   journal_entries_per_sector %u\n", ic->journal_entries_per_sector);
2954         DEBUG_print("   journal_section_entries %u\n", ic->journal_section_entries);
2955         DEBUG_print("   journal_section_sectors %u\n", ic->journal_section_sectors);
2956         DEBUG_print("   journal_sections %u\n", (unsigned)le32_to_cpu(ic->sb->journal_sections));
2957         DEBUG_print("   journal_entries %u\n", ic->journal_entries);
2958         DEBUG_print("   log2_interleave_sectors %d\n", ic->sb->log2_interleave_sectors);
2959         DEBUG_print("   device_sectors 0x%llx\n", (unsigned long long)ic->device_sectors);
2960         DEBUG_print("   initial_sectors 0x%x\n", ic->initial_sectors);
2961         DEBUG_print("   metadata_run 0x%x\n", ic->metadata_run);
2962         DEBUG_print("   log2_metadata_run %d\n", ic->log2_metadata_run);
2963         DEBUG_print("   provided_data_sectors 0x%llx (%llu)\n", (unsigned long long)ic->provided_data_sectors,
2964                     (unsigned long long)ic->provided_data_sectors);
2965         DEBUG_print("   log2_buffer_sectors %u\n", ic->log2_buffer_sectors);
2966
2967         ic->bufio = dm_bufio_client_create(ic->dev->bdev, 1U << (SECTOR_SHIFT + ic->log2_buffer_sectors),
2968                                            1, 0, NULL, NULL);
2969         if (IS_ERR(ic->bufio)) {
2970                 r = PTR_ERR(ic->bufio);
2971                 ti->error = "Cannot initialize dm-bufio";
2972                 ic->bufio = NULL;
2973                 goto bad;
2974         }
2975         dm_bufio_set_sector_offset(ic->bufio, ic->start + ic->initial_sectors);
2976
2977         if (ic->mode != 'R') {
2978                 r = create_journal(ic, &ti->error);
2979                 if (r)
2980                         goto bad;
2981         }
2982
2983         if (should_write_sb) {
2984                 int r;
2985
2986                 init_journal(ic, 0, ic->journal_sections, 0);
2987                 r = dm_integrity_failed(ic);
2988                 if (unlikely(r)) {
2989                         ti->error = "Error initializing journal";
2990                         goto bad;
2991                 }
2992                 r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
2993                 if (r) {
2994                         ti->error = "Error initializing superblock";
2995                         goto bad;
2996                 }
2997                 ic->just_formatted = true;
2998         }
2999
3000         r = dm_set_target_max_io_len(ti, 1U << ic->sb->log2_interleave_sectors);
3001         if (r)
3002                 goto bad;
3003
3004         if (!ic->internal_hash)
3005                 dm_integrity_set(ti, ic);
3006
3007         ti->num_flush_bios = 1;
3008         ti->flush_supported = true;
3009
3010         return 0;
3011 bad:
3012         dm_integrity_dtr(ti);
3013         return r;
3014 }
3015
3016 static void dm_integrity_dtr(struct dm_target *ti)
3017 {
3018         struct dm_integrity_c *ic = ti->private;
3019
3020         BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
3021
3022         if (ic->metadata_wq)
3023                 destroy_workqueue(ic->metadata_wq);
3024         if (ic->wait_wq)
3025                 destroy_workqueue(ic->wait_wq);
3026         if (ic->commit_wq)
3027                 destroy_workqueue(ic->commit_wq);
3028         if (ic->writer_wq)
3029                 destroy_workqueue(ic->writer_wq);
3030         if (ic->bufio)
3031                 dm_bufio_client_destroy(ic->bufio);
3032         mempool_destroy(ic->journal_io_mempool);
3033         if (ic->io)
3034                 dm_io_client_destroy(ic->io);
3035         if (ic->dev)
3036                 dm_put_device(ti, ic->dev);
3037         dm_integrity_free_page_list(ic, ic->journal);
3038         dm_integrity_free_page_list(ic, ic->journal_io);
3039         dm_integrity_free_page_list(ic, ic->journal_xor);
3040         if (ic->journal_scatterlist)
3041                 dm_integrity_free_journal_scatterlist(ic, ic->journal_scatterlist);
3042         if (ic->journal_io_scatterlist)
3043                 dm_integrity_free_journal_scatterlist(ic, ic->journal_io_scatterlist);
3044         if (ic->sk_requests) {
3045                 unsigned i;
3046
3047                 for (i = 0; i < ic->journal_sections; i++) {
3048                         struct skcipher_request *req = ic->sk_requests[i];
3049                         if (req) {
3050                                 kzfree(req->iv);
3051                                 skcipher_request_free(req);
3052                         }
3053                 }
3054                 kvfree(ic->sk_requests);
3055         }
3056         kvfree(ic->journal_tree);
3057         if (ic->sb)
3058                 free_pages_exact(ic->sb, SB_SECTORS << SECTOR_SHIFT);
3059
3060         if (ic->internal_hash)
3061                 crypto_free_shash(ic->internal_hash);
3062         free_alg(&ic->internal_hash_alg);
3063
3064         if (ic->journal_crypt)
3065                 crypto_free_skcipher(ic->journal_crypt);
3066         free_alg(&ic->journal_crypt_alg);
3067
3068         if (ic->journal_mac)
3069                 crypto_free_shash(ic->journal_mac);
3070         free_alg(&ic->journal_mac_alg);
3071
3072         kfree(ic);
3073 }
3074
3075 static struct target_type integrity_target = {
3076         .name                   = "integrity",
3077         .version                = {1, 0, 0},
3078         .module                 = THIS_MODULE,
3079         .features               = DM_TARGET_SINGLETON | DM_TARGET_INTEGRITY,
3080         .ctr                    = dm_integrity_ctr,
3081         .dtr                    = dm_integrity_dtr,
3082         .map                    = dm_integrity_map,
3083         .postsuspend            = dm_integrity_postsuspend,
3084         .resume                 = dm_integrity_resume,
3085         .status                 = dm_integrity_status,
3086         .iterate_devices        = dm_integrity_iterate_devices,
3087 };
3088
3089 int __init dm_integrity_init(void)
3090 {
3091         int r;
3092
3093         journal_io_cache = kmem_cache_create("integrity_journal_io",
3094                                              sizeof(struct journal_io), 0, 0, NULL);
3095         if (!journal_io_cache) {
3096                 DMERR("can't allocate journal io cache");
3097                 return -ENOMEM;
3098         }
3099
3100         r = dm_register_target(&integrity_target);
3101
3102         if (r < 0)
3103                 DMERR("register failed %d", r);
3104
3105         return r;
3106 }
3107
3108 void dm_integrity_exit(void)
3109 {
3110         dm_unregister_target(&integrity_target);
3111         kmem_cache_destroy(journal_io_cache);
3112 }
3113
3114 module_init(dm_integrity_init);
3115 module_exit(dm_integrity_exit);
3116
3117 MODULE_AUTHOR("Milan Broz");
3118 MODULE_AUTHOR("Mikulas Patocka");
3119 MODULE_DESCRIPTION(DM_NAME " target for integrity tags extension");
3120 MODULE_LICENSE("GPL");