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 <linux/log2.h>
  18 #include <crypto/hash.h>
  19 #include <crypto/skcipher.h>
  20 #include <linux/async_tx.h>
  21 #include "dm-bufio.h"
  22
  23 #define DM_MSG_PREFIX "integrity"
  24
  25 #define DEFAULT_INTERLEAVE_SECTORS      32768
  26 #define DEFAULT_JOURNAL_SIZE_FACTOR     7
  27 #define DEFAULT_BUFFER_SECTORS          128
  28 #define DEFAULT_JOURNAL_WATERMARK       50
  29 #define DEFAULT_SYNC_MSEC               10000
  30 #define DEFAULT_MAX_JOURNAL_SECTORS     131072
  31 #define MIN_LOG2_INTERLEAVE_SECTORS     3
  32 #define MAX_LOG2_INTERLEAVE_SECTORS     31
  33 #define METADATA_WORKQUEUE_MAX_ACTIVE   16
  34
  35 /*
  36  * Warning - DEBUG_PRINT prints security-sensitive data to the log,
  37  * so it should not be enabled in the official kernel
  38  */
  39 //#define DEBUG_PRINT
  40 //#define INTERNAL_VERIFY
  41
  42 /*
  43  * On disk structures
  44  */
  45
  46 #define SB_MAGIC                        "integrt"
  47 #define SB_VERSION                      1
  48 #define SB_SECTORS                      8
  49 #define MAX_SECTORS_PER_BLOCK           8
  50
  51 struct superblock {
  52         __u8 magic[8];
  53         __u8 version;
  54         __u8 log2_interleave_sectors;
  55         __u16 integrity_tag_size;
  56         __u32 journal_sections;
  57         __u64 provided_data_sectors;    /* userspace uses this value */
  58         __u32 flags;
  59         __u8 log2_sectors_per_block;
  60 };
  61
  62 #define SB_FLAG_HAVE_JOURNAL_MAC        0x1
  63
  64 #define JOURNAL_ENTRY_ROUNDUP           8
  65
  66 typedef __u64 commit_id_t;
  67 #define JOURNAL_MAC_PER_SECTOR          8
  68
  69 struct journal_entry {
  70         union {
  71                 struct {
  72                         __u32 sector_lo;
  73                         __u32 sector_hi;
  74                 } s;
  75                 __u64 sector;
  76         } u;
  77         commit_id_t last_bytes[0];
  78         /* __u8 tag[0]; */
  79 };
  80
  81 #define journal_entry_tag(ic, je)               ((__u8 *)&(je)->last_bytes[(ic)->sectors_per_block])
  82
  83 #if BITS_PER_LONG == 64
  84 #define journal_entry_set_sector(je, x)         do { smp_wmb(); ACCESS_ONCE((je)->u.sector) = cpu_to_le64(x); } while (0)
  85 #define journal_entry_get_sector(je)            le64_to_cpu((je)->u.sector)
  86 #elif defined(CONFIG_LBDAF)
  87 #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)
  88 #define journal_entry_get_sector(je)            le64_to_cpu((je)->u.sector)
  89 #else
  90 #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)
  91 #define journal_entry_get_sector(je)            le32_to_cpu((je)->u.s.sector_lo)
  92 #endif
  93 #define journal_entry_is_unused(je)             ((je)->u.s.sector_hi == cpu_to_le32(-1))
  94 #define journal_entry_set_unused(je)            do { ((je)->u.s.sector_hi = cpu_to_le32(-1)); } while (0)
  95 #define journal_entry_is_inprogress(je)         ((je)->u.s.sector_hi == cpu_to_le32(-2))
  96 #define journal_entry_set_inprogress(je)        do { ((je)->u.s.sector_hi = cpu_to_le32(-2)); } while (0)
  97
  98 #define JOURNAL_BLOCK_SECTORS           8
  99 #define JOURNAL_SECTOR_DATA             ((1 << SECTOR_SHIFT) - sizeof(commit_id_t))
 100 #define JOURNAL_MAC_SIZE                (JOURNAL_MAC_PER_SECTOR * JOURNAL_BLOCK_SECTORS)
 101
 102 struct journal_sector {
 103         __u8 entries[JOURNAL_SECTOR_DATA - JOURNAL_MAC_PER_SECTOR];
 104         __u8 mac[JOURNAL_MAC_PER_SECTOR];
 105         commit_id_t commit_id;
 106 };
 107
 108 #define MAX_TAG_SIZE                    (JOURNAL_SECTOR_DATA - JOURNAL_MAC_PER_SECTOR - offsetof(struct journal_entry, last_bytes[MAX_SECTORS_PER_BLOCK]))
 109
 110 #define METADATA_PADDING_SECTORS        8
 111
 112 #define N_COMMIT_IDS                    4
 113
 114 static unsigned char prev_commit_seq(unsigned char seq)
 115 {
 116         return (seq + N_COMMIT_IDS - 1) % N_COMMIT_IDS;
 117 }
 118
 119 static unsigned char next_commit_seq(unsigned char seq)
 120 {
 121         return (seq + 1) % N_COMMIT_IDS;
 122 }
 123
 124 /*
 125  * In-memory structures
 126  */
 127
 128 struct journal_node {
 129         struct rb_node node;
 130         sector_t sector;
 131 };
 132
 133 struct alg_spec {
 134         char *alg_string;
 135         char *key_string;
 136         __u8 *key;
 137         unsigned key_size;
 138 };
 139
 140 struct dm_integrity_c {
 141         struct dm_dev *dev;
 142         unsigned tag_size;
 143         __s8 log2_tag_size;
 144         sector_t start;
 145         mempool_t *journal_io_mempool;
 146         struct dm_io_client *io;
 147         struct dm_bufio_client *bufio;
 148         struct workqueue_struct *metadata_wq;
 149         struct superblock *sb;
 150         unsigned journal_pages;
 151         struct page_list *journal;
 152         struct page_list *journal_io;
 153         struct page_list *journal_xor;
 154
 155         struct crypto_skcipher *journal_crypt;
 156         struct scatterlist **journal_scatterlist;
 157         struct scatterlist **journal_io_scatterlist;
 158         struct skcipher_request **sk_requests;
 159
 160         struct crypto_shash *journal_mac;
 161
 162         struct journal_node *journal_tree;
 163         struct rb_root journal_tree_root;
 164
 165         sector_t provided_data_sectors;
 166
 167         unsigned short journal_entry_size;
 168         unsigned char journal_entries_per_sector;
 169         unsigned char journal_section_entries;
 170         unsigned short journal_section_sectors;
 171         unsigned journal_sections;
 172         unsigned journal_entries;
 173         sector_t device_sectors;
 174         unsigned initial_sectors;
 175         unsigned metadata_run;
 176         __s8 log2_metadata_run;
 177         __u8 log2_buffer_sectors;
 178         __u8 sectors_per_block;
 179
 180         unsigned char mode;
 181         bool suspending;
 182
 183         int failed;
 184
 185         struct crypto_shash *internal_hash;
 186
 187         /* these variables are locked with endio_wait.lock */
 188         struct rb_root in_progress;
 189         wait_queue_head_t endio_wait;
 190         struct workqueue_struct *wait_wq;
 191
 192         unsigned char commit_seq;
 193         commit_id_t commit_ids[N_COMMIT_IDS];
 194
 195         unsigned committed_section;
 196         unsigned n_committed_sections;
 197
 198         unsigned uncommitted_section;
 199         unsigned n_uncommitted_sections;
 200
 201         unsigned free_section;
 202         unsigned char free_section_entry;
 203         unsigned free_sectors;
 204
 205         unsigned free_sectors_threshold;
 206
 207         struct workqueue_struct *commit_wq;
 208         struct work_struct commit_work;
 209
 210         struct workqueue_struct *writer_wq;
 211         struct work_struct writer_work;
 212
 213         struct bio_list flush_bio_list;
 214
 215         unsigned long autocommit_jiffies;
 216         struct timer_list autocommit_timer;
 217         unsigned autocommit_msec;
 218
 219         wait_queue_head_t copy_to_journal_wait;
 220
 221         struct completion crypto_backoff;
 222
 223         bool journal_uptodate;
 224         bool just_formatted;
 225
 226         struct alg_spec internal_hash_alg;
 227         struct alg_spec journal_crypt_alg;
 228         struct alg_spec journal_mac_alg;
 229 };
 230
 231 struct dm_integrity_range {
 232         sector_t logical_sector;
 233         unsigned n_sectors;
 234         struct rb_node node;
 235 };
 236
 237 struct dm_integrity_io {
 238         struct work_struct work;
 239
 240         struct dm_integrity_c *ic;
 241         bool write;
 242         bool fua;
 243
 244         struct dm_integrity_range range;
 245
 246         sector_t metadata_block;
 247         unsigned metadata_offset;
 248
 249         atomic_t in_flight;
 250         int bi_error;
 251
 252         struct completion *completion;
 253
 254         struct block_device *orig_bi_bdev;
 255         bio_end_io_t *orig_bi_end_io;
 256         struct bio_integrity_payload *orig_bi_integrity;
 257         struct bvec_iter orig_bi_iter;
 258 };
 259
 260 struct journal_completion {
 261         struct dm_integrity_c *ic;
 262         atomic_t in_flight;
 263         struct completion comp;
 264 };
 265
 266 struct journal_io {
 267         struct dm_integrity_range range;
 268         struct journal_completion *comp;
 269 };
 270
 271 static struct kmem_cache *journal_io_cache;
 272
 273 #define JOURNAL_IO_MEMPOOL      32
 274
 275 #ifdef DEBUG_PRINT
 276 #define DEBUG_print(x, ...)     printk(KERN_DEBUG x, ##__VA_ARGS__)
 277 static void __DEBUG_bytes(__u8 *bytes, size_t len, const char *msg, ...)
 278 {
 279         va_list args;
 280         va_start(args, msg);
 281         vprintk(msg, args);
 282         va_end(args);
 283         if (len)
 284                 pr_cont(":");
 285         while (len) {
 286                 pr_cont(" %02x", *bytes);
 287                 bytes++;
 288                 len--;
 289         }
 290         pr_cont("\n");
 291 }
 292 #define DEBUG_bytes(bytes, len, msg, ...)       __DEBUG_bytes(bytes, len, KERN_DEBUG msg, ##__VA_ARGS__)
 293 #else
 294 #define DEBUG_print(x, ...)                     do { } while (0)
 295 #define DEBUG_bytes(bytes, len, msg, ...)       do { } while (0)
 296 #endif
 297
 298 /*
 299  * DM Integrity profile, protection is performed layer above (dm-crypt)
 300  */
 301 static struct blk_integrity_profile dm_integrity_profile = {
 302         .name                   = "DM-DIF-EXT-TAG",
 303         .generate_fn            = NULL,
 304         .verify_fn              = NULL,
 305 };
 306
 307 static void dm_integrity_map_continue(struct dm_integrity_io *dio, bool from_map);
 308 static void integrity_bio_wait(struct work_struct *w);
 309 static void dm_integrity_dtr(struct dm_target *ti);
 310
 311 static void dm_integrity_io_error(struct dm_integrity_c *ic, const char *msg, int err)
 312 {
 313         if (!cmpxchg(&ic->failed, 0, err))
 314                 DMERR("Error on %s: %d", msg, err);
 315 }
 316
 317 static int dm_integrity_failed(struct dm_integrity_c *ic)
 318 {
 319         return ACCESS_ONCE(ic->failed);
 320 }
 321
 322 static commit_id_t dm_integrity_commit_id(struct dm_integrity_c *ic, unsigned i,
 323                                           unsigned j, unsigned char seq)
 324 {
 325         /*
 326          * Xor the number with section and sector, so that if a piece of
 327          * journal is written at wrong place, it is detected.
 328          */
 329         return ic->commit_ids[seq] ^ cpu_to_le64(((__u64)i << 32) ^ j);
 330 }
 331
 332 static void get_area_and_offset(struct dm_integrity_c *ic, sector_t data_sector,
 333                                 sector_t *area, sector_t *offset)
 334 {
 335         __u8 log2_interleave_sectors = ic->sb->log2_interleave_sectors;
 336
 337         *area = data_sector >> log2_interleave_sectors;
 338         *offset = (unsigned)data_sector & ((1U << log2_interleave_sectors) - 1);
 339 }
 340
 341 #define sector_to_block(ic, n)                                          \
 342 do {                                                                    \
 343         BUG_ON((n) & (unsigned)((ic)->sectors_per_block - 1));          \
 344         (n) >>= (ic)->sb->log2_sectors_per_block;                       \
 345 } while (0)
 346
 347 static __u64 get_metadata_sector_and_offset(struct dm_integrity_c *ic, sector_t area,
 348                                             sector_t offset, unsigned *metadata_offset)
 349 {
 350         __u64 ms;
 351         unsigned mo;
 352
 353         ms = area << ic->sb->log2_interleave_sectors;
 354         if (likely(ic->log2_metadata_run >= 0))
 355                 ms += area << ic->log2_metadata_run;
 356         else
 357                 ms += area * ic->metadata_run;
 358         ms >>= ic->log2_buffer_sectors;
 359
 360         sector_to_block(ic, offset);
 361
 362         if (likely(ic->log2_tag_size >= 0)) {
 363                 ms += offset >> (SECTOR_SHIFT + ic->log2_buffer_sectors - ic->log2_tag_size);
 364                 mo = (offset << ic->log2_tag_size) & ((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - 1);
 365         } else {
 366                 ms += (__u64)offset * ic->tag_size >> (SECTOR_SHIFT + ic->log2_buffer_sectors);
 367                 mo = (offset * ic->tag_size) & ((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - 1);
 368         }
 369         *metadata_offset = mo;
 370         return ms;
 371 }
 372
 373 static sector_t get_data_sector(struct dm_integrity_c *ic, sector_t area, sector_t offset)
 374 {
 375         sector_t result;
 376
 377         result = area << ic->sb->log2_interleave_sectors;
 378         if (likely(ic->log2_metadata_run >= 0))
 379                 result += (area + 1) << ic->log2_metadata_run;
 380         else
 381                 result += (area + 1) * ic->metadata_run;
 382
 383         result += (sector_t)ic->initial_sectors + offset;
 384         return result;
 385 }
 386
 387 static void wraparound_section(struct dm_integrity_c *ic, unsigned *sec_ptr)
 388 {
 389         if (unlikely(*sec_ptr >= ic->journal_sections))
 390                 *sec_ptr -= ic->journal_sections;
 391 }
 392
 393 static int sync_rw_sb(struct dm_integrity_c *ic, int op, int op_flags)
 394 {
 395         struct dm_io_request io_req;
 396         struct dm_io_region io_loc;
 397
 398         io_req.bi_op = op;
 399         io_req.bi_op_flags = op_flags;
 400         io_req.mem.type = DM_IO_KMEM;
 401         io_req.mem.ptr.addr = ic->sb;
 402         io_req.notify.fn = NULL;
 403         io_req.client = ic->io;
 404         io_loc.bdev = ic->dev->bdev;
 405         io_loc.sector = ic->start;
 406         io_loc.count = SB_SECTORS;
 407
 408         return dm_io(&io_req, 1, &io_loc, NULL);
 409 }
 410
 411 static void access_journal_check(struct dm_integrity_c *ic, unsigned section, unsigned offset,
 412                                  bool e, const char *function)
 413 {
 414 #if defined(CONFIG_DM_DEBUG) || defined(INTERNAL_VERIFY)
 415         unsigned limit = e ? ic->journal_section_entries : ic->journal_section_sectors;
 416
 417         if (unlikely(section >= ic->journal_sections) ||
 418             unlikely(offset >= limit)) {
 419                 printk(KERN_CRIT "%s: invalid access at (%u,%u), limit (%u,%u)\n",
 420                         function, section, offset, ic->journal_sections, limit);
 421                 BUG();
 422         }
 423 #endif
 424 }
 425
 426 static void page_list_location(struct dm_integrity_c *ic, unsigned section, unsigned offset,
 427                                unsigned *pl_index, unsigned *pl_offset)
 428 {
 429         unsigned sector;
 430
 431         access_journal_check(ic, section, offset, false, "page_list_location");
 432
 433         sector = section * ic->journal_section_sectors + offset;
 434
 435         *pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
 436         *pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
 437 }
 438
 439 static struct journal_sector *access_page_list(struct dm_integrity_c *ic, struct page_list *pl,
 440                                                unsigned section, unsigned offset, unsigned *n_sectors)
 441 {
 442         unsigned pl_index, pl_offset;
 443         char *va;
 444
 445         page_list_location(ic, section, offset, &pl_index, &pl_offset);
 446
 447         if (n_sectors)
 448                 *n_sectors = (PAGE_SIZE - pl_offset) >> SECTOR_SHIFT;
 449
 450         va = lowmem_page_address(pl[pl_index].page);
 451
 452         return (struct journal_sector *)(va + pl_offset);
 453 }
 454
 455 static struct journal_sector *access_journal(struct dm_integrity_c *ic, unsigned section, unsigned offset)
 456 {
 457         return access_page_list(ic, ic->journal, section, offset, NULL);
 458 }
 459
 460 static struct journal_entry *access_journal_entry(struct dm_integrity_c *ic, unsigned section, unsigned n)
 461 {
 462         unsigned rel_sector, offset;
 463         struct journal_sector *js;
 464
 465         access_journal_check(ic, section, n, true, "access_journal_entry");
 466
 467         rel_sector = n % JOURNAL_BLOCK_SECTORS;
 468         offset = n / JOURNAL_BLOCK_SECTORS;
 469
 470         js = access_journal(ic, section, rel_sector);
 471         return (struct journal_entry *)((char *)js + offset * ic->journal_entry_size);
 472 }
 473
 474 static struct journal_sector *access_journal_data(struct dm_integrity_c *ic, unsigned section, unsigned n)
 475 {
 476         n <<= ic->sb->log2_sectors_per_block;
 477
 478         n += JOURNAL_BLOCK_SECTORS;
 479
 480         access_journal_check(ic, section, n, false, "access_journal_data");
 481
 482         return access_journal(ic, section, n);
 483 }
 484
 485 static void section_mac(struct dm_integrity_c *ic, unsigned section, __u8 result[JOURNAL_MAC_SIZE])
 486 {
 487         SHASH_DESC_ON_STACK(desc, ic->journal_mac);
 488         int r;
 489         unsigned j, size;
 490
 491         desc->tfm = ic->journal_mac;
 492         desc->flags = CRYPTO_TFM_REQ_MAY_SLEEP;
 493
 494         r = crypto_shash_init(desc);
 495         if (unlikely(r)) {
 496                 dm_integrity_io_error(ic, "crypto_shash_init", r);
 497                 goto err;
 498         }
 499
 500         for (j = 0; j < ic->journal_section_entries; j++) {
 501                 struct journal_entry *je = access_journal_entry(ic, section, j);
 502                 r = crypto_shash_update(desc, (__u8 *)&je->u.sector, sizeof je->u.sector);
 503                 if (unlikely(r)) {
 504                         dm_integrity_io_error(ic, "crypto_shash_update", r);
 505                         goto err;
 506                 }
 507         }
 508
 509         size = crypto_shash_digestsize(ic->journal_mac);
 510
 511         if (likely(size <= JOURNAL_MAC_SIZE)) {
 512                 r = crypto_shash_final(desc, result);
 513                 if (unlikely(r)) {
 514                         dm_integrity_io_error(ic, "crypto_shash_final", r);
 515                         goto err;
 516                 }
 517                 memset(result + size, 0, JOURNAL_MAC_SIZE - size);
 518         } else {
 519                 __u8 digest[size];
 520                 r = crypto_shash_final(desc, digest);
 521                 if (unlikely(r)) {
 522                         dm_integrity_io_error(ic, "crypto_shash_final", r);
 523                         goto err;
 524                 }
 525                 memcpy(result, digest, JOURNAL_MAC_SIZE);
 526         }
 527
 528         return;
 529 err:
 530         memset(result, 0, JOURNAL_MAC_SIZE);
 531 }
 532
 533 static void rw_section_mac(struct dm_integrity_c *ic, unsigned section, bool wr)
 534 {
 535         __u8 result[JOURNAL_MAC_SIZE];
 536         unsigned j;
 537
 538         if (!ic->journal_mac)
 539                 return;
 540
 541         section_mac(ic, section, result);
 542
 543         for (j = 0; j < JOURNAL_BLOCK_SECTORS; j++) {
 544                 struct journal_sector *js = access_journal(ic, section, j);
 545
 546                 if (likely(wr))
 547                         memcpy(&js->mac, result + (j * JOURNAL_MAC_PER_SECTOR), JOURNAL_MAC_PER_SECTOR);
 548                 else {
 549                         if (memcmp(&js->mac, result + (j * JOURNAL_MAC_PER_SECTOR), JOURNAL_MAC_PER_SECTOR))
 550                                 dm_integrity_io_error(ic, "journal mac", -EILSEQ);
 551                 }
 552         }
 553 }
 554
 555 static void complete_journal_op(void *context)
 556 {
 557         struct journal_completion *comp = context;
 558         BUG_ON(!atomic_read(&comp->in_flight));
 559         if (likely(atomic_dec_and_test(&comp->in_flight)))
 560                 complete(&comp->comp);
 561 }
 562
 563 static void xor_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
 564                         unsigned n_sections, struct journal_completion *comp)
 565 {
 566         struct async_submit_ctl submit;
 567         size_t n_bytes = (size_t)(n_sections * ic->journal_section_sectors) << SECTOR_SHIFT;
 568         unsigned pl_index, pl_offset, section_index;
 569         struct page_list *source_pl, *target_pl;
 570
 571         if (likely(encrypt)) {
 572                 source_pl = ic->journal;
 573                 target_pl = ic->journal_io;
 574         } else {
 575                 source_pl = ic->journal_io;
 576                 target_pl = ic->journal;
 577         }
 578
 579         page_list_location(ic, section, 0, &pl_index, &pl_offset);
 580
 581         atomic_add(roundup(pl_offset + n_bytes, PAGE_SIZE) >> PAGE_SHIFT, &comp->in_flight);
 582
 583         init_async_submit(&submit, ASYNC_TX_XOR_ZERO_DST, NULL, complete_journal_op, comp, NULL);
 584
 585         section_index = pl_index;
 586
 587         do {
 588                 size_t this_step;
 589                 struct page *src_pages[2];
 590                 struct page *dst_page;
 591
 592                 while (unlikely(pl_index == section_index)) {
 593                         unsigned dummy;
 594                         if (likely(encrypt))
 595                                 rw_section_mac(ic, section, true);
 596                         section++;
 597                         n_sections--;
 598                         if (!n_sections)
 599                                 break;
 600                         page_list_location(ic, section, 0, &section_index, &dummy);
 601                 }
 602
 603                 this_step = min(n_bytes, (size_t)PAGE_SIZE - pl_offset);
 604                 dst_page = target_pl[pl_index].page;
 605                 src_pages[0] = source_pl[pl_index].page;
 606                 src_pages[1] = ic->journal_xor[pl_index].page;
 607
 608                 async_xor(dst_page, src_pages, pl_offset, 2, this_step, &submit);
 609
 610                 pl_index++;
 611                 pl_offset = 0;
 612                 n_bytes -= this_step;
 613         } while (n_bytes);
 614
 615         BUG_ON(n_sections);
 616
 617         async_tx_issue_pending_all();
 618 }
 619
 620 static void complete_journal_encrypt(struct crypto_async_request *req, int err)
 621 {
 622         struct journal_completion *comp = req->data;
 623         if (unlikely(err)) {
 624                 if (likely(err == -EINPROGRESS)) {
 625                         complete(&comp->ic->crypto_backoff);
 626                         return;
 627                 }
 628                 dm_integrity_io_error(comp->ic, "asynchronous encrypt", err);
 629         }
 630         complete_journal_op(comp);
 631 }
 632
 633 static bool do_crypt(bool encrypt, struct skcipher_request *req, struct journal_completion *comp)
 634 {
 635         int r;
 636         skcipher_request_set_callback(req, CRYPTO_TFM_REQ_MAY_BACKLOG | CRYPTO_TFM_REQ_MAY_SLEEP,
 637                                       complete_journal_encrypt, comp);
 638         if (likely(encrypt))
 639                 r = crypto_skcipher_encrypt(req);
 640         else
 641                 r = crypto_skcipher_decrypt(req);
 642         if (likely(!r))
 643                 return false;
 644         if (likely(r == -EINPROGRESS))
 645                 return true;
 646         if (likely(r == -EBUSY)) {
 647                 wait_for_completion(&comp->ic->crypto_backoff);
 648                 reinit_completion(&comp->ic->crypto_backoff);
 649                 return true;
 650         }
 651         dm_integrity_io_error(comp->ic, "encrypt", r);
 652         return false;
 653 }
 654
 655 static void crypt_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
 656                           unsigned n_sections, struct journal_completion *comp)
 657 {
 658         struct scatterlist **source_sg;
 659         struct scatterlist **target_sg;
 660
 661         atomic_add(2, &comp->in_flight);
 662
 663         if (likely(encrypt)) {
 664                 source_sg = ic->journal_scatterlist;
 665                 target_sg = ic->journal_io_scatterlist;
 666         } else {
 667                 source_sg = ic->journal_io_scatterlist;
 668                 target_sg = ic->journal_scatterlist;
 669         }
 670
 671         do {
 672                 struct skcipher_request *req;
 673                 unsigned ivsize;
 674                 char *iv;
 675
 676                 if (likely(encrypt))
 677                         rw_section_mac(ic, section, true);
 678
 679                 req = ic->sk_requests[section];
 680                 ivsize = crypto_skcipher_ivsize(ic->journal_crypt);
 681                 iv = req->iv;
 682
 683                 memcpy(iv, iv + ivsize, ivsize);
 684
 685                 req->src = source_sg[section];
 686                 req->dst = target_sg[section];
 687
 688                 if (unlikely(do_crypt(encrypt, req, comp)))
 689                         atomic_inc(&comp->in_flight);
 690
 691                 section++;
 692                 n_sections--;
 693         } while (n_sections);
 694
 695         atomic_dec(&comp->in_flight);
 696         complete_journal_op(comp);
 697 }
 698
 699 static void encrypt_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
 700                             unsigned n_sections, struct journal_completion *comp)
 701 {
 702         if (ic->journal_xor)
 703                 return xor_journal(ic, encrypt, section, n_sections, comp);
 704         else
 705                 return crypt_journal(ic, encrypt, section, n_sections, comp);
 706 }
 707
 708 static void complete_journal_io(unsigned long error, void *context)
 709 {
 710         struct journal_completion *comp = context;
 711         if (unlikely(error != 0))
 712                 dm_integrity_io_error(comp->ic, "writing journal", -EIO);
 713         complete_journal_op(comp);
 714 }
 715
 716 static void rw_journal(struct dm_integrity_c *ic, int op, int op_flags, unsigned section,
 717                        unsigned n_sections, struct journal_completion *comp)
 718 {
 719         struct dm_io_request io_req;
 720         struct dm_io_region io_loc;
 721         unsigned sector, n_sectors, pl_index, pl_offset;
 722         int r;
 723
 724         if (unlikely(dm_integrity_failed(ic))) {
 725                 if (comp)
 726                         complete_journal_io(-1UL, comp);
 727                 return;
 728         }
 729
 730         sector = section * ic->journal_section_sectors;
 731         n_sectors = n_sections * ic->journal_section_sectors;
 732
 733         pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
 734         pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
 735
 736         io_req.bi_op = op;
 737         io_req.bi_op_flags = op_flags;
 738         io_req.mem.type = DM_IO_PAGE_LIST;
 739         if (ic->journal_io)
 740                 io_req.mem.ptr.pl = &ic->journal_io[pl_index];
 741         else
 742                 io_req.mem.ptr.pl = &ic->journal[pl_index];
 743         io_req.mem.offset = pl_offset;
 744         if (likely(comp != NULL)) {
 745                 io_req.notify.fn = complete_journal_io;
 746                 io_req.notify.context = comp;
 747         } else {
 748                 io_req.notify.fn = NULL;
 749         }
 750         io_req.client = ic->io;
 751         io_loc.bdev = ic->dev->bdev;
 752         io_loc.sector = ic->start + SB_SECTORS + sector;
 753         io_loc.count = n_sectors;
 754
 755         r = dm_io(&io_req, 1, &io_loc, NULL);
 756         if (unlikely(r)) {
 757                 dm_integrity_io_error(ic, op == REQ_OP_READ ? "reading journal" : "writing journal", r);
 758                 if (comp) {
 759                         WARN_ONCE(1, "asynchronous dm_io failed: %d", r);
 760                         complete_journal_io(-1UL, comp);
 761                 }
 762         }
 763 }
 764
 765 static void write_journal(struct dm_integrity_c *ic, unsigned commit_start, unsigned commit_sections)
 766 {
 767         struct journal_completion io_comp;
 768         struct journal_completion crypt_comp_1;
 769         struct journal_completion crypt_comp_2;
 770         unsigned i;
 771
 772         io_comp.ic = ic;
 773         io_comp.comp = COMPLETION_INITIALIZER_ONSTACK(io_comp.comp);
 774
 775         if (commit_start + commit_sections <= ic->journal_sections) {
 776                 io_comp.in_flight = (atomic_t)ATOMIC_INIT(1);
 777                 if (ic->journal_io) {
 778                         crypt_comp_1.ic = ic;
 779                         crypt_comp_1.comp = COMPLETION_INITIALIZER_ONSTACK(crypt_comp_1.comp);
 780                         crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
 781                         encrypt_journal(ic, true, commit_start, commit_sections, &crypt_comp_1);
 782                         wait_for_completion_io(&crypt_comp_1.comp);
 783                 } else {
 784                         for (i = 0; i < commit_sections; i++)
 785                                 rw_section_mac(ic, commit_start + i, true);
 786                 }
 787                 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, commit_sections, &io_comp);
 788         } else {
 789                 unsigned to_end;
 790                 io_comp.in_flight = (atomic_t)ATOMIC_INIT(2);
 791                 to_end = ic->journal_sections - commit_start;
 792                 if (ic->journal_io) {
 793                         crypt_comp_1.ic = ic;
 794                         crypt_comp_1.comp = COMPLETION_INITIALIZER_ONSTACK(crypt_comp_1.comp);
 795                         crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
 796                         encrypt_journal(ic, true, commit_start, to_end, &crypt_comp_1);
 797                         if (try_wait_for_completion(&crypt_comp_1.comp)) {
 798                                 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
 799                                 crypt_comp_1.comp = COMPLETION_INITIALIZER_ONSTACK(crypt_comp_1.comp);
 800                                 crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
 801                                 encrypt_journal(ic, true, 0, commit_sections - to_end, &crypt_comp_1);
 802                                 wait_for_completion_io(&crypt_comp_1.comp);
 803                         } else {
 804                                 crypt_comp_2.ic = ic;
 805                                 crypt_comp_2.comp = COMPLETION_INITIALIZER_ONSTACK(crypt_comp_2.comp);
 806                                 crypt_comp_2.in_flight = (atomic_t)ATOMIC_INIT(0);
 807                                 encrypt_journal(ic, true, 0, commit_sections - to_end, &crypt_comp_2);
 808                                 wait_for_completion_io(&crypt_comp_1.comp);
 809                                 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
 810                                 wait_for_completion_io(&crypt_comp_2.comp);
 811                         }
 812                 } else {
 813                         for (i = 0; i < to_end; i++)
 814                                 rw_section_mac(ic, commit_start + i, true);
 815                         rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
 816                         for (i = 0; i < commit_sections - to_end; i++)
 817                                 rw_section_mac(ic, i, true);
 818                 }
 819                 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, 0, commit_sections - to_end, &io_comp);
 820         }
 821
 822         wait_for_completion_io(&io_comp.comp);
 823 }
 824
 825 static void copy_from_journal(struct dm_integrity_c *ic, unsigned section, unsigned offset,
 826                               unsigned n_sectors, sector_t target, io_notify_fn fn, void *data)
 827 {
 828         struct dm_io_request io_req;
 829         struct dm_io_region io_loc;
 830         int r;
 831         unsigned sector, pl_index, pl_offset;
 832
 833         BUG_ON((target | n_sectors | offset) & (unsigned)(ic->sectors_per_block - 1));
 834
 835         if (unlikely(dm_integrity_failed(ic))) {
 836                 fn(-1UL, data);
 837                 return;
 838         }
 839
 840         sector = section * ic->journal_section_sectors + JOURNAL_BLOCK_SECTORS + offset;
 841
 842         pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
 843         pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
 844
 845         io_req.bi_op = REQ_OP_WRITE;
 846         io_req.bi_op_flags = 0;
 847         io_req.mem.type = DM_IO_PAGE_LIST;
 848         io_req.mem.ptr.pl = &ic->journal[pl_index];
 849         io_req.mem.offset = pl_offset;
 850         io_req.notify.fn = fn;
 851         io_req.notify.context = data;
 852         io_req.client = ic->io;
 853         io_loc.bdev = ic->dev->bdev;
 854         io_loc.sector = ic->start + target;
 855         io_loc.count = n_sectors;
 856
 857         r = dm_io(&io_req, 1, &io_loc, NULL);
 858         if (unlikely(r)) {
 859                 WARN_ONCE(1, "asynchronous dm_io failed: %d", r);
 860                 fn(-1UL, data);
 861         }
 862 }
 863
 864 static bool add_new_range(struct dm_integrity_c *ic, struct dm_integrity_range *new_range)
 865 {
 866         struct rb_node **n = &ic->in_progress.rb_node;
 867         struct rb_node *parent;
 868
 869         BUG_ON((new_range->logical_sector | new_range->n_sectors) & (unsigned)(ic->sectors_per_block - 1));
 870
 871         parent = NULL;
 872
 873         while (*n) {
 874                 struct dm_integrity_range *range = container_of(*n, struct dm_integrity_range, node);
 875
 876                 parent = *n;
 877                 if (new_range->logical_sector + new_range->n_sectors <= range->logical_sector) {
 878                         n = &range->node.rb_left;
 879                 } else if (new_range->logical_sector >= range->logical_sector + range->n_sectors) {
 880                         n = &range->node.rb_right;
 881                 } else {
 882                         return false;
 883                 }
 884         }
 885
 886         rb_link_node(&new_range->node, parent, n);
 887         rb_insert_color(&new_range->node, &ic->in_progress);
 888
 889         return true;
 890 }
 891
 892 static void remove_range_unlocked(struct dm_integrity_c *ic, struct dm_integrity_range *range)
 893 {
 894         rb_erase(&range->node, &ic->in_progress);
 895         wake_up_locked(&ic->endio_wait);
 896 }
 897
 898 static void remove_range(struct dm_integrity_c *ic, struct dm_integrity_range *range)
 899 {
 900         unsigned long flags;
 901
 902         spin_lock_irqsave(&ic->endio_wait.lock, flags);
 903         remove_range_unlocked(ic, range);
 904         spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
 905 }
 906
 907 static void init_journal_node(struct journal_node *node)
 908 {
 909         RB_CLEAR_NODE(&node->node);
 910         node->sector = (sector_t)-1;
 911 }
 912
 913 static void add_journal_node(struct dm_integrity_c *ic, struct journal_node *node, sector_t sector)
 914 {
 915         struct rb_node **link;
 916         struct rb_node *parent;
 917
 918         node->sector = sector;
 919         BUG_ON(!RB_EMPTY_NODE(&node->node));
 920
 921         link = &ic->journal_tree_root.rb_node;
 922         parent = NULL;
 923
 924         while (*link) {
 925                 struct journal_node *j;
 926                 parent = *link;
 927                 j = container_of(parent, struct journal_node, node);
 928                 if (sector < j->sector)
 929                         link = &j->node.rb_left;
 930                 else
 931                         link = &j->node.rb_right;
 932         }
 933
 934         rb_link_node(&node->node, parent, link);
 935         rb_insert_color(&node->node, &ic->journal_tree_root);
 936 }
 937
 938 static void remove_journal_node(struct dm_integrity_c *ic, struct journal_node *node)
 939 {
 940         BUG_ON(RB_EMPTY_NODE(&node->node));
 941         rb_erase(&node->node, &ic->journal_tree_root);
 942         init_journal_node(node);
 943 }
 944
 945 #define NOT_FOUND       (-1U)
 946
 947 static unsigned find_journal_node(struct dm_integrity_c *ic, sector_t sector, sector_t *next_sector)
 948 {
 949         struct rb_node *n = ic->journal_tree_root.rb_node;
 950         unsigned found = NOT_FOUND;
 951         *next_sector = (sector_t)-1;
 952         while (n) {
 953                 struct journal_node *j = container_of(n, struct journal_node, node);
 954                 if (sector == j->sector) {
 955                         found = j - ic->journal_tree;
 956                 }
 957                 if (sector < j->sector) {
 958                         *next_sector = j->sector;
 959                         n = j->node.rb_left;
 960                 } else {
 961                         n = j->node.rb_right;
 962                 }
 963         }
 964
 965         return found;
 966 }
 967
 968 static bool test_journal_node(struct dm_integrity_c *ic, unsigned pos, sector_t sector)
 969 {
 970         struct journal_node *node, *next_node;
 971         struct rb_node *next;
 972
 973         if (unlikely(pos >= ic->journal_entries))
 974                 return false;
 975         node = &ic->journal_tree[pos];
 976         if (unlikely(RB_EMPTY_NODE(&node->node)))
 977                 return false;
 978         if (unlikely(node->sector != sector))
 979                 return false;
 980
 981         next = rb_next(&node->node);
 982         if (unlikely(!next))
 983                 return true;
 984
 985         next_node = container_of(next, struct journal_node, node);
 986         return next_node->sector != sector;
 987 }
 988
 989 static bool find_newer_committed_node(struct dm_integrity_c *ic, struct journal_node *node)
 990 {
 991         struct rb_node *next;
 992         struct journal_node *next_node;
 993         unsigned next_section;
 994
 995         BUG_ON(RB_EMPTY_NODE(&node->node));
 996
 997         next = rb_next(&node->node);
 998         if (unlikely(!next))
 999                 return false;
1000
1001         next_node = container_of(next, struct journal_node, node);
1002
1003         if (next_node->sector != node->sector)
1004                 return false;
1005
1006         next_section = (unsigned)(next_node - ic->journal_tree) / ic->journal_section_entries;
1007         if (next_section >= ic->committed_section &&
1008             next_section < ic->committed_section + ic->n_committed_sections)
1009                 return true;
1010         if (next_section + ic->journal_sections < ic->committed_section + ic->n_committed_sections)
1011                 return true;
1012
1013         return false;
1014 }
1015
1016 #define TAG_READ        0
1017 #define TAG_WRITE       1
1018 #define TAG_CMP         2
1019
1020 static int dm_integrity_rw_tag(struct dm_integrity_c *ic, unsigned char *tag, sector_t *metadata_block,
1021                                unsigned *metadata_offset, unsigned total_size, int op)
1022 {
1023         do {
1024                 unsigned char *data, *dp;
1025                 struct dm_buffer *b;
1026                 unsigned to_copy;
1027                 int r;
1028
1029                 r = dm_integrity_failed(ic);
1030                 if (unlikely(r))
1031                         return r;
1032
1033                 data = dm_bufio_read(ic->bufio, *metadata_block, &b);
1034                 if (unlikely(IS_ERR(data)))
1035                         return PTR_ERR(data);
1036
1037                 to_copy = min((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - *metadata_offset, total_size);
1038                 dp = data + *metadata_offset;
1039                 if (op == TAG_READ) {
1040                         memcpy(tag, dp, to_copy);
1041                 } else if (op == TAG_WRITE) {
1042                         memcpy(dp, tag, to_copy);
1043                         dm_bufio_mark_buffer_dirty(b);
1044                 } else  {
1045                         /* e.g.: op == TAG_CMP */
1046                         if (unlikely(memcmp(dp, tag, to_copy))) {
1047                                 unsigned i;
1048
1049                                 for (i = 0; i < to_copy; i++) {
1050                                         if (dp[i] != tag[i])
1051                                                 break;
1052                                         total_size--;
1053                                 }
1054                                 dm_bufio_release(b);
1055                                 return total_size;
1056                         }
1057                 }
1058                 dm_bufio_release(b);
1059
1060                 tag += to_copy;
1061                 *metadata_offset += to_copy;
1062                 if (unlikely(*metadata_offset == 1U << SECTOR_SHIFT << ic->log2_buffer_sectors)) {
1063                         (*metadata_block)++;
1064                         *metadata_offset = 0;
1065                 }
1066                 total_size -= to_copy;
1067         } while (unlikely(total_size));
1068
1069         return 0;
1070 }
1071
1072 static void dm_integrity_flush_buffers(struct dm_integrity_c *ic)
1073 {
1074         int r;
1075         r = dm_bufio_write_dirty_buffers(ic->bufio);
1076         if (unlikely(r))
1077                 dm_integrity_io_error(ic, "writing tags", r);
1078 }
1079
1080 static void sleep_on_endio_wait(struct dm_integrity_c *ic)
1081 {
1082         DECLARE_WAITQUEUE(wait, current);
1083         __add_wait_queue(&ic->endio_wait, &wait);
1084         __set_current_state(TASK_UNINTERRUPTIBLE);
1085         spin_unlock_irq(&ic->endio_wait.lock);
1086         io_schedule();
1087         spin_lock_irq(&ic->endio_wait.lock);
1088         __remove_wait_queue(&ic->endio_wait, &wait);
1089 }
1090
1091 static void autocommit_fn(unsigned long data)
1092 {
1093         struct dm_integrity_c *ic = (struct dm_integrity_c *)data;
1094
1095         if (likely(!dm_integrity_failed(ic)))
1096                 queue_work(ic->commit_wq, &ic->commit_work);
1097 }
1098
1099 static void schedule_autocommit(struct dm_integrity_c *ic)
1100 {
1101         if (!timer_pending(&ic->autocommit_timer))
1102                 mod_timer(&ic->autocommit_timer, jiffies + ic->autocommit_jiffies);
1103 }
1104
1105 static void submit_flush_bio(struct dm_integrity_c *ic, struct dm_integrity_io *dio)
1106 {
1107         struct bio *bio;
1108         spin_lock_irq(&ic->endio_wait.lock);
1109         bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1110         bio_list_add(&ic->flush_bio_list, bio);
1111         spin_unlock_irq(&ic->endio_wait.lock);
1112         queue_work(ic->commit_wq, &ic->commit_work);
1113 }
1114
1115 static void do_endio(struct dm_integrity_c *ic, struct bio *bio)
1116 {
1117         int r = dm_integrity_failed(ic);
1118         if (unlikely(r) && !bio->bi_error)
1119                 bio->bi_error = r;
1120         bio_endio(bio);
1121 }
1122
1123 static void do_endio_flush(struct dm_integrity_c *ic, struct dm_integrity_io *dio)
1124 {
1125         struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1126
1127         if (unlikely(dio->fua) && likely(!bio->bi_error) && likely(!dm_integrity_failed(ic)))
1128                 submit_flush_bio(ic, dio);
1129         else
1130                 do_endio(ic, bio);
1131 }
1132
1133 static void dec_in_flight(struct dm_integrity_io *dio)
1134 {
1135         if (atomic_dec_and_test(&dio->in_flight)) {
1136                 struct dm_integrity_c *ic = dio->ic;
1137                 struct bio *bio;
1138
1139                 remove_range(ic, &dio->range);
1140
1141                 if (unlikely(dio->write))
1142                         schedule_autocommit(ic);
1143
1144                 bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1145
1146                 if (unlikely(dio->bi_error) && !bio->bi_error)
1147                         bio->bi_error = dio->bi_error;
1148                 if (likely(!bio->bi_error) && unlikely(bio_sectors(bio) != dio->range.n_sectors)) {
1149                         dio->range.logical_sector += dio->range.n_sectors;
1150                         bio_advance(bio, dio->range.n_sectors << SECTOR_SHIFT);
1151                         INIT_WORK(&dio->work, integrity_bio_wait);
1152                         queue_work(ic->wait_wq, &dio->work);
1153                         return;
1154                 }
1155                 do_endio_flush(ic, dio);
1156         }
1157 }
1158
1159 static void integrity_end_io(struct bio *bio)
1160 {
1161         struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
1162
1163         bio->bi_iter = dio->orig_bi_iter;
1164         bio->bi_bdev = dio->orig_bi_bdev;
1165         if (dio->orig_bi_integrity) {
1166                 bio->bi_integrity = dio->orig_bi_integrity;
1167                 bio->bi_opf |= REQ_INTEGRITY;
1168         }
1169         bio->bi_end_io = dio->orig_bi_end_io;
1170
1171         if (dio->completion)
1172                 complete(dio->completion);
1173
1174         dec_in_flight(dio);
1175 }
1176
1177 static void integrity_sector_checksum(struct dm_integrity_c *ic, sector_t sector,
1178                                       const char *data, char *result)
1179 {
1180         __u64 sector_le = cpu_to_le64(sector);
1181         SHASH_DESC_ON_STACK(req, ic->internal_hash);
1182         int r;
1183         unsigned digest_size;
1184
1185         req->tfm = ic->internal_hash;
1186         req->flags = 0;
1187
1188         r = crypto_shash_init(req);
1189         if (unlikely(r < 0)) {
1190                 dm_integrity_io_error(ic, "crypto_shash_init", r);
1191                 goto failed;
1192         }
1193
1194         r = crypto_shash_update(req, (const __u8 *)&sector_le, sizeof sector_le);
1195         if (unlikely(r < 0)) {
1196                 dm_integrity_io_error(ic, "crypto_shash_update", r);
1197                 goto failed;
1198         }
1199
1200         r = crypto_shash_update(req, data, ic->sectors_per_block << SECTOR_SHIFT);
1201         if (unlikely(r < 0)) {
1202                 dm_integrity_io_error(ic, "crypto_shash_update", r);
1203                 goto failed;
1204         }
1205
1206         r = crypto_shash_final(req, result);
1207         if (unlikely(r < 0)) {
1208                 dm_integrity_io_error(ic, "crypto_shash_final", r);
1209                 goto failed;
1210         }
1211
1212         digest_size = crypto_shash_digestsize(ic->internal_hash);
1213         if (unlikely(digest_size < ic->tag_size))
1214                 memset(result + digest_size, 0, ic->tag_size - digest_size);
1215
1216         return;
1217
1218 failed:
1219         /* this shouldn't happen anyway, the hash functions have no reason to fail */
1220         get_random_bytes(result, ic->tag_size);
1221 }
1222
1223 static void integrity_metadata(struct work_struct *w)
1224 {
1225         struct dm_integrity_io *dio = container_of(w, struct dm_integrity_io, work);
1226         struct dm_integrity_c *ic = dio->ic;
1227
1228         int r;
1229
1230         if (ic->internal_hash) {
1231                 struct bvec_iter iter;
1232                 struct bio_vec bv;
1233                 unsigned digest_size = crypto_shash_digestsize(ic->internal_hash);
1234                 struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1235                 char *checksums;
1236                 unsigned extra_space = unlikely(digest_size > ic->tag_size) ? digest_size - ic->tag_size : 0;
1237                 char checksums_onstack[ic->tag_size + extra_space];
1238                 unsigned sectors_to_process = dio->range.n_sectors;
1239                 sector_t sector = dio->range.logical_sector;
1240
1241                 if (unlikely(ic->mode == 'R'))
1242                         goto skip_io;
1243
1244                 checksums = kmalloc((PAGE_SIZE >> SECTOR_SHIFT >> ic->sb->log2_sectors_per_block) * ic->tag_size + extra_space,
1245                                     GFP_NOIO | __GFP_NORETRY | __GFP_NOWARN);
1246                 if (!checksums)
1247                         checksums = checksums_onstack;
1248
1249                 __bio_for_each_segment(bv, bio, iter, dio->orig_bi_iter) {
1250                         unsigned pos;
1251                         char *mem, *checksums_ptr;
1252
1253 again:
1254                         mem = (char *)kmap_atomic(bv.bv_page) + bv.bv_offset;
1255                         pos = 0;
1256                         checksums_ptr = checksums;
1257                         do {
1258                                 integrity_sector_checksum(ic, sector, mem + pos, checksums_ptr);
1259                                 checksums_ptr += ic->tag_size;
1260                                 sectors_to_process -= ic->sectors_per_block;
1261                                 pos += ic->sectors_per_block << SECTOR_SHIFT;
1262                                 sector += ic->sectors_per_block;
1263                         } while (pos < bv.bv_len && sectors_to_process && checksums != checksums_onstack);
1264                         kunmap_atomic(mem);
1265
1266                         r = dm_integrity_rw_tag(ic, checksums, &dio->metadata_block, &dio->metadata_offset,
1267                                                 checksums_ptr - checksums, !dio->write ? TAG_CMP : TAG_WRITE);
1268                         if (unlikely(r)) {
1269                                 if (r > 0) {
1270                                         DMERR("Checksum failed at sector 0x%llx",
1271                                               (unsigned long long)(sector - ((r + ic->tag_size - 1) / ic->tag_size)));
1272                                         r = -EILSEQ;
1273                                 }
1274                                 if (likely(checksums != checksums_onstack))
1275                                         kfree(checksums);
1276                                 goto error;
1277                         }
1278
1279                         if (!sectors_to_process)
1280                                 break;
1281
1282                         if (unlikely(pos < bv.bv_len)) {
1283                                 bv.bv_offset += pos;
1284                                 bv.bv_len -= pos;
1285                                 goto again;
1286                         }
1287                 }
1288
1289                 if (likely(checksums != checksums_onstack))
1290                         kfree(checksums);
1291         } else {
1292                 struct bio_integrity_payload *bip = dio->orig_bi_integrity;
1293
1294                 if (bip) {
1295                         struct bio_vec biv;
1296                         struct bvec_iter iter;
1297                         unsigned data_to_process = dio->range.n_sectors;
1298                         sector_to_block(ic, data_to_process);
1299                         data_to_process *= ic->tag_size;
1300
1301                         bip_for_each_vec(biv, bip, iter) {
1302                                 unsigned char *tag;
1303                                 unsigned this_len;
1304
1305                                 BUG_ON(PageHighMem(biv.bv_page));
1306                                 tag = lowmem_page_address(biv.bv_page) + biv.bv_offset;
1307                                 this_len = min(biv.bv_len, data_to_process);
1308                                 r = dm_integrity_rw_tag(ic, tag, &dio->metadata_block, &dio->metadata_offset,
1309                                                         this_len, !dio->write ? TAG_READ : TAG_WRITE);
1310                                 if (unlikely(r))
1311                                         goto error;
1312                                 data_to_process -= this_len;
1313                                 if (!data_to_process)
1314                                         break;
1315                         }
1316                 }
1317         }
1318 skip_io:
1319         dec_in_flight(dio);
1320         return;
1321 error:
1322         dio->bi_error = r;
1323         dec_in_flight(dio);
1324 }
1325
1326 static int dm_integrity_map(struct dm_target *ti, struct bio *bio)
1327 {
1328         struct dm_integrity_c *ic = ti->private;
1329         struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
1330         struct bio_integrity_payload *bip;
1331
1332         sector_t area, offset;
1333
1334         dio->ic = ic;
1335         dio->bi_error = 0;
1336
1337         if (unlikely(bio->bi_opf & REQ_PREFLUSH)) {
1338                 submit_flush_bio(ic, dio);
1339                 return DM_MAPIO_SUBMITTED;
1340         }
1341
1342         dio->range.logical_sector = dm_target_offset(ti, bio->bi_iter.bi_sector);
1343         dio->write = bio_op(bio) == REQ_OP_WRITE;
1344         dio->fua = dio->write && bio->bi_opf & REQ_FUA;
1345         if (unlikely(dio->fua)) {
1346                 /*
1347                  * Don't pass down the FUA flag because we have to flush
1348                  * disk cache anyway.
1349                  */
1350                 bio->bi_opf &= ~REQ_FUA;
1351         }
1352         if (unlikely(dio->range.logical_sector + bio_sectors(bio) > ic->provided_data_sectors)) {
1353                 DMERR("Too big sector number: 0x%llx + 0x%x > 0x%llx",
1354                       (unsigned long long)dio->range.logical_sector, bio_sectors(bio),
1355                       (unsigned long long)ic->provided_data_sectors);
1356                 return -EIO;
1357         }
1358         if (unlikely((dio->range.logical_sector | bio_sectors(bio)) & (unsigned)(ic->sectors_per_block - 1))) {
1359                 DMERR("Bio not aligned on %u sectors: 0x%llx, 0x%x",
1360                       ic->sectors_per_block,
1361                       (unsigned long long)dio->range.logical_sector, bio_sectors(bio));
1362                 return -EIO;
1363         }
1364
1365         if (ic->sectors_per_block > 1) {
1366                 struct bvec_iter iter;
1367                 struct bio_vec bv;
1368                 bio_for_each_segment(bv, bio, iter) {
1369                         if (unlikely((bv.bv_offset | bv.bv_len) & ((ic->sectors_per_block << SECTOR_SHIFT) - 1))) {
1370                                 DMERR("Bio vector (%u,%u) is not aligned on %u-sector boundary",
1371                                         bv.bv_offset, bv.bv_len, ic->sectors_per_block);
1372                                 return -EIO;
1373                         }
1374                 }
1375         }
1376
1377         bip = bio_integrity(bio);
1378         if (!ic->internal_hash) {
1379                 if (bip) {
1380                         unsigned wanted_tag_size = bio_sectors(bio) >> ic->sb->log2_sectors_per_block;
1381                         if (ic->log2_tag_size >= 0)
1382                                 wanted_tag_size <<= ic->log2_tag_size;
1383                         else
1384                                 wanted_tag_size *= ic->tag_size;
1385                         if (unlikely(wanted_tag_size != bip->bip_iter.bi_size)) {
1386                                 DMERR("Invalid integrity data size %u, expected %u", bip->bip_iter.bi_size, wanted_tag_size);
1387                                 return -EIO;
1388                         }
1389                 }
1390         } else {
1391                 if (unlikely(bip != NULL)) {
1392                         DMERR("Unexpected integrity data when using internal hash");
1393                         return -EIO;
1394                 }
1395         }
1396
1397         if (unlikely(ic->mode == 'R') && unlikely(dio->write))
1398                 return -EIO;
1399
1400         get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
1401         dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset, &dio->metadata_offset);
1402         bio->bi_iter.bi_sector = get_data_sector(ic, area, offset);
1403
1404         dm_integrity_map_continue(dio, true);
1405         return DM_MAPIO_SUBMITTED;
1406 }
1407
1408 static bool __journal_read_write(struct dm_integrity_io *dio, struct bio *bio,
1409                                  unsigned journal_section, unsigned journal_entry)
1410 {
1411         struct dm_integrity_c *ic = dio->ic;
1412         sector_t logical_sector;
1413         unsigned n_sectors;
1414
1415         logical_sector = dio->range.logical_sector;
1416         n_sectors = dio->range.n_sectors;
1417         do {
1418                 struct bio_vec bv = bio_iovec(bio);
1419                 char *mem;
1420
1421                 if (unlikely(bv.bv_len >> SECTOR_SHIFT > n_sectors))
1422                         bv.bv_len = n_sectors << SECTOR_SHIFT;
1423                 n_sectors -= bv.bv_len >> SECTOR_SHIFT;
1424                 bio_advance_iter(bio, &bio->bi_iter, bv.bv_len);
1425 retry_kmap:
1426                 mem = kmap_atomic(bv.bv_page);
1427                 if (likely(dio->write))
1428                         flush_dcache_page(bv.bv_page);
1429
1430                 do {
1431                         struct journal_entry *je = access_journal_entry(ic, journal_section, journal_entry);
1432
1433                         if (unlikely(!dio->write)) {
1434                                 struct journal_sector *js;
1435                                 char *mem_ptr;
1436                                 unsigned s;
1437
1438                                 if (unlikely(journal_entry_is_inprogress(je))) {
1439                                         flush_dcache_page(bv.bv_page);
1440                                         kunmap_atomic(mem);
1441
1442                                         __io_wait_event(ic->copy_to_journal_wait, !journal_entry_is_inprogress(je));
1443                                         goto retry_kmap;
1444                                 }
1445                                 smp_rmb();
1446                                 BUG_ON(journal_entry_get_sector(je) != logical_sector);
1447                                 js = access_journal_data(ic, journal_section, journal_entry);
1448                                 mem_ptr = mem + bv.bv_offset;
1449                                 s = 0;
1450                                 do {
1451                                         memcpy(mem_ptr, js, JOURNAL_SECTOR_DATA);
1452                                         *(commit_id_t *)(mem_ptr + JOURNAL_SECTOR_DATA) = je->last_bytes[s];
1453                                         js++;
1454                                         mem_ptr += 1 << SECTOR_SHIFT;
1455                                 } while (++s < ic->sectors_per_block);
1456 #ifdef INTERNAL_VERIFY
1457                                 if (ic->internal_hash) {
1458                                         char checksums_onstack[max(crypto_shash_digestsize(ic->internal_hash), ic->tag_size)];
1459
1460                                         integrity_sector_checksum(ic, logical_sector, mem + bv.bv_offset, checksums_onstack);
1461                                         if (unlikely(memcmp(checksums_onstack, journal_entry_tag(ic, je), ic->tag_size))) {
1462                                                 DMERR("Checksum failed when reading from journal, at sector 0x%llx",
1463                                                       (unsigned long long)logical_sector);
1464                                         }
1465                                 }
1466 #endif
1467                         }
1468
1469                         if (!ic->internal_hash) {
1470                                 struct bio_integrity_payload *bip = bio_integrity(bio);
1471                                 unsigned tag_todo = ic->tag_size;
1472                                 char *tag_ptr = journal_entry_tag(ic, je);
1473
1474                                 if (bip) do {
1475                                         struct bio_vec biv = bvec_iter_bvec(bip->bip_vec, bip->bip_iter);
1476                                         unsigned tag_now = min(biv.bv_len, tag_todo);
1477                                         char *tag_addr;
1478                                         BUG_ON(PageHighMem(biv.bv_page));
1479                                         tag_addr = lowmem_page_address(biv.bv_page) + biv.bv_offset;
1480                                         if (likely(dio->write))
1481                                                 memcpy(tag_ptr, tag_addr, tag_now);
1482                                         else
1483                                                 memcpy(tag_addr, tag_ptr, tag_now);
1484                                         bvec_iter_advance(bip->bip_vec, &bip->bip_iter, tag_now);
1485                                         tag_ptr += tag_now;
1486                                         tag_todo -= tag_now;
1487                                 } while (unlikely(tag_todo)); else {
1488                                         if (likely(dio->write))
1489                                                 memset(tag_ptr, 0, tag_todo);
1490                                 }
1491                         }
1492
1493                         if (likely(dio->write)) {
1494                                 struct journal_sector *js;
1495                                 unsigned s;
1496
1497                                 js = access_journal_data(ic, journal_section, journal_entry);
1498                                 memcpy(js, mem + bv.bv_offset, ic->sectors_per_block << SECTOR_SHIFT);
1499
1500                                 s = 0;
1501                                 do {
1502                                         je->last_bytes[s] = js[s].commit_id;
1503                                 } while (++s < ic->sectors_per_block);
1504
1505                                 if (ic->internal_hash) {
1506                                         unsigned digest_size = crypto_shash_digestsize(ic->internal_hash);
1507                                         if (unlikely(digest_size > ic->tag_size)) {
1508                                                 char checksums_onstack[digest_size];
1509                                                 integrity_sector_checksum(ic, logical_sector, (char *)js, checksums_onstack);
1510                                                 memcpy(journal_entry_tag(ic, je), checksums_onstack, ic->tag_size);
1511                                         } else
1512                                                 integrity_sector_checksum(ic, logical_sector, (char *)js, journal_entry_tag(ic, je));
1513                                 }
1514
1515                                 journal_entry_set_sector(je, logical_sector);
1516                         }
1517                         logical_sector += ic->sectors_per_block;
1518
1519                         journal_entry++;
1520                         if (unlikely(journal_entry == ic->journal_section_entries)) {
1521                                 journal_entry = 0;
1522                                 journal_section++;
1523                                 wraparound_section(ic, &journal_section);
1524                         }
1525
1526                         bv.bv_offset += ic->sectors_per_block << SECTOR_SHIFT;
1527                 } while (bv.bv_len -= ic->sectors_per_block << SECTOR_SHIFT);
1528
1529                 if (unlikely(!dio->write))
1530                         flush_dcache_page(bv.bv_page);
1531                 kunmap_atomic(mem);
1532         } while (n_sectors);
1533
1534         if (likely(dio->write)) {
1535                 smp_mb();
1536                 if (unlikely(waitqueue_active(&ic->copy_to_journal_wait)))
1537                         wake_up(&ic->copy_to_journal_wait);
1538                 if (ACCESS_ONCE(ic->free_sectors) <= ic->free_sectors_threshold) {
1539                         queue_work(ic->commit_wq, &ic->commit_work);
1540                 } else {
1541                         schedule_autocommit(ic);
1542                 }
1543         } else {
1544                 remove_range(ic, &dio->range);
1545         }
1546
1547         if (unlikely(bio->bi_iter.bi_size)) {
1548                 sector_t area, offset;
1549
1550                 dio->range.logical_sector = logical_sector;
1551                 get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
1552                 dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset, &dio->metadata_offset);
1553                 return true;
1554         }
1555
1556         return false;
1557 }
1558
1559 static void dm_integrity_map_continue(struct dm_integrity_io *dio, bool from_map)
1560 {
1561         struct dm_integrity_c *ic = dio->ic;
1562         struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1563         unsigned journal_section, journal_entry;
1564         unsigned journal_read_pos;
1565         struct completion read_comp;
1566         bool need_sync_io = ic->internal_hash && !dio->write;
1567
1568         if (need_sync_io && from_map) {
1569                 INIT_WORK(&dio->work, integrity_bio_wait);
1570                 queue_work(ic->metadata_wq, &dio->work);
1571                 return;
1572         }
1573
1574 lock_retry:
1575         spin_lock_irq(&ic->endio_wait.lock);
1576 retry:
1577         if (unlikely(dm_integrity_failed(ic))) {
1578                 spin_unlock_irq(&ic->endio_wait.lock);
1579                 do_endio(ic, bio);
1580                 return;
1581         }
1582         dio->range.n_sectors = bio_sectors(bio);
1583         journal_read_pos = NOT_FOUND;
1584         if (likely(ic->mode == 'J')) {
1585                 if (dio->write) {
1586                         unsigned next_entry, i, pos;
1587                         unsigned ws, we;
1588
1589                         dio->range.n_sectors = min(dio->range.n_sectors, ic->free_sectors);
1590                         if (unlikely(!dio->range.n_sectors))
1591                                 goto sleep;
1592                         ic->free_sectors -= dio->range.n_sectors;
1593                         journal_section = ic->free_section;
1594                         journal_entry = ic->free_section_entry;
1595
1596                         next_entry = ic->free_section_entry + dio->range.n_sectors;
1597                         ic->free_section_entry = next_entry % ic->journal_section_entries;
1598                         ic->free_section += next_entry / ic->journal_section_entries;
1599                         ic->n_uncommitted_sections += next_entry / ic->journal_section_entries;
1600                         wraparound_section(ic, &ic->free_section);
1601
1602                         pos = journal_section * ic->journal_section_entries + journal_entry;
1603                         ws = journal_section;
1604                         we = journal_entry;
1605                         i = 0;
1606                         do {
1607                                 struct journal_entry *je;
1608
1609                                 add_journal_node(ic, &ic->journal_tree[pos], dio->range.logical_sector + i);
1610                                 pos++;
1611                                 if (unlikely(pos >= ic->journal_entries))
1612                                         pos = 0;
1613
1614                                 je = access_journal_entry(ic, ws, we);
1615                                 BUG_ON(!journal_entry_is_unused(je));
1616                                 journal_entry_set_inprogress(je);
1617                                 we++;
1618                                 if (unlikely(we == ic->journal_section_entries)) {
1619                                         we = 0;
1620                                         ws++;
1621                                         wraparound_section(ic, &ws);
1622                                 }
1623                         } while ((i += ic->sectors_per_block) < dio->range.n_sectors);
1624
1625                         spin_unlock_irq(&ic->endio_wait.lock);
1626                         goto journal_read_write;
1627                 } else {
1628                         sector_t next_sector;
1629                         journal_read_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
1630                         if (likely(journal_read_pos == NOT_FOUND)) {
1631                                 if (unlikely(dio->range.n_sectors > next_sector - dio->range.logical_sector))
1632                                         dio->range.n_sectors = next_sector - dio->range.logical_sector;
1633                         } else {
1634                                 unsigned i;
1635                                 unsigned jp = journal_read_pos + 1;
1636                                 for (i = ic->sectors_per_block; i < dio->range.n_sectors; i += ic->sectors_per_block, jp++) {
1637                                         if (!test_journal_node(ic, jp, dio->range.logical_sector + i))
1638                                                 break;
1639                                 }
1640                                 dio->range.n_sectors = i;
1641                         }
1642                 }
1643         }
1644         if (unlikely(!add_new_range(ic, &dio->range))) {
1645                 /*
1646                  * We must not sleep in the request routine because it could
1647                  * stall bios on current->bio_list.
1648                  * So, we offload the bio to a workqueue if we have to sleep.
1649                  */
1650 sleep:
1651                 if (from_map) {
1652                         spin_unlock_irq(&ic->endio_wait.lock);
1653                         INIT_WORK(&dio->work, integrity_bio_wait);
1654                         queue_work(ic->wait_wq, &dio->work);
1655                         return;
1656                 } else {
1657                         sleep_on_endio_wait(ic);
1658                         goto retry;
1659                 }
1660         }
1661         spin_unlock_irq(&ic->endio_wait.lock);
1662
1663         if (unlikely(journal_read_pos != NOT_FOUND)) {
1664                 journal_section = journal_read_pos / ic->journal_section_entries;
1665                 journal_entry = journal_read_pos % ic->journal_section_entries;
1666                 goto journal_read_write;
1667         }
1668
1669         dio->in_flight = (atomic_t)ATOMIC_INIT(2);
1670
1671         if (need_sync_io) {
1672                 read_comp = COMPLETION_INITIALIZER_ONSTACK(read_comp);
1673                 dio->completion = &read_comp;
1674         } else
1675                 dio->completion = NULL;
1676
1677         dio->orig_bi_iter = bio->bi_iter;
1678
1679         dio->orig_bi_bdev = bio->bi_bdev;
1680         bio->bi_bdev = ic->dev->bdev;
1681
1682         dio->orig_bi_integrity = bio_integrity(bio);
1683         bio->bi_integrity = NULL;
1684         bio->bi_opf &= ~REQ_INTEGRITY;
1685
1686         dio->orig_bi_end_io = bio->bi_end_io;
1687         bio->bi_end_io = integrity_end_io;
1688
1689         bio->bi_iter.bi_size = dio->range.n_sectors << SECTOR_SHIFT;
1690         bio->bi_iter.bi_sector += ic->start;
1691         generic_make_request(bio);
1692
1693         if (need_sync_io) {
1694                 wait_for_completion_io(&read_comp);
1695                 integrity_metadata(&dio->work);
1696         } else {
1697                 INIT_WORK(&dio->work, integrity_metadata);
1698                 queue_work(ic->metadata_wq, &dio->work);
1699         }
1700
1701         return;
1702
1703 journal_read_write:
1704         if (unlikely(__journal_read_write(dio, bio, journal_section, journal_entry)))
1705                 goto lock_retry;
1706
1707         do_endio_flush(ic, dio);
1708 }
1709
1710
1711 static void integrity_bio_wait(struct work_struct *w)
1712 {
1713         struct dm_integrity_io *dio = container_of(w, struct dm_integrity_io, work);
1714
1715         dm_integrity_map_continue(dio, false);
1716 }
1717
1718 static void pad_uncommitted(struct dm_integrity_c *ic)
1719 {
1720         if (ic->free_section_entry) {
1721                 ic->free_sectors -= ic->journal_section_entries - ic->free_section_entry;
1722                 ic->free_section_entry = 0;
1723                 ic->free_section++;
1724                 wraparound_section(ic, &ic->free_section);
1725                 ic->n_uncommitted_sections++;
1726         }
1727 }
1728
1729 static void integrity_commit(struct work_struct *w)
1730 {
1731         struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, commit_work);
1732         unsigned commit_start, commit_sections;
1733         unsigned i, j, n;
1734         struct bio *flushes;
1735
1736         del_timer(&ic->autocommit_timer);
1737
1738         spin_lock_irq(&ic->endio_wait.lock);
1739         flushes = bio_list_get(&ic->flush_bio_list);
1740         if (unlikely(ic->mode != 'J')) {
1741                 spin_unlock_irq(&ic->endio_wait.lock);
1742                 dm_integrity_flush_buffers(ic);
1743                 goto release_flush_bios;
1744         }
1745
1746         pad_uncommitted(ic);
1747         commit_start = ic->uncommitted_section;
1748         commit_sections = ic->n_uncommitted_sections;
1749         spin_unlock_irq(&ic->endio_wait.lock);
1750
1751         if (!commit_sections)
1752                 goto release_flush_bios;
1753
1754         i = commit_start;
1755         for (n = 0; n < commit_sections; n++) {
1756                 for (j = 0; j < ic->journal_section_entries; j++) {
1757                         struct journal_entry *je;
1758                         je = access_journal_entry(ic, i, j);
1759                         io_wait_event(ic->copy_to_journal_wait, !journal_entry_is_inprogress(je));
1760                 }
1761                 for (j = 0; j < ic->journal_section_sectors; j++) {
1762                         struct journal_sector *js;
1763                         js = access_journal(ic, i, j);
1764                         js->commit_id = dm_integrity_commit_id(ic, i, j, ic->commit_seq);
1765                 }
1766                 i++;
1767                 if (unlikely(i >= ic->journal_sections))
1768                         ic->commit_seq = next_commit_seq(ic->commit_seq);
1769                 wraparound_section(ic, &i);
1770         }
1771         smp_rmb();
1772
1773         write_journal(ic, commit_start, commit_sections);
1774
1775         spin_lock_irq(&ic->endio_wait.lock);
1776         ic->uncommitted_section += commit_sections;
1777         wraparound_section(ic, &ic->uncommitted_section);
1778         ic->n_uncommitted_sections -= commit_sections;
1779         ic->n_committed_sections += commit_sections;
1780         spin_unlock_irq(&ic->endio_wait.lock);
1781
1782         if (ACCESS_ONCE(ic->free_sectors) <= ic->free_sectors_threshold)
1783                 queue_work(ic->writer_wq, &ic->writer_work);
1784
1785 release_flush_bios:
1786         while (flushes) {
1787                 struct bio *next = flushes->bi_next;
1788                 flushes->bi_next = NULL;
1789                 do_endio(ic, flushes);
1790                 flushes = next;
1791         }
1792 }
1793
1794 static void complete_copy_from_journal(unsigned long error, void *context)
1795 {
1796         struct journal_io *io = context;
1797         struct journal_completion *comp = io->comp;
1798         struct dm_integrity_c *ic = comp->ic;
1799         remove_range(ic, &io->range);
1800         mempool_free(io, ic->journal_io_mempool);
1801         if (unlikely(error != 0))
1802                 dm_integrity_io_error(ic, "copying from journal", -EIO);
1803         complete_journal_op(comp);
1804 }
1805
1806 static void restore_last_bytes(struct dm_integrity_c *ic, struct journal_sector *js,
1807                                struct journal_entry *je)
1808 {
1809         unsigned s = 0;
1810         do {
1811                 js->commit_id = je->last_bytes[s];
1812                 js++;
1813         } while (++s < ic->sectors_per_block);
1814 }
1815
1816 static void do_journal_write(struct dm_integrity_c *ic, unsigned write_start,
1817                              unsigned write_sections, bool from_replay)
1818 {
1819         unsigned i, j, n;
1820         struct journal_completion comp;
1821
1822         comp.ic = ic;
1823         comp.in_flight = (atomic_t)ATOMIC_INIT(1);
1824         comp.comp = COMPLETION_INITIALIZER_ONSTACK(comp.comp);
1825
1826         i = write_start;
1827         for (n = 0; n < write_sections; n++, i++, wraparound_section(ic, &i)) {
1828 #ifndef INTERNAL_VERIFY
1829                 if (unlikely(from_replay))
1830 #endif
1831                         rw_section_mac(ic, i, false);
1832                 for (j = 0; j < ic->journal_section_entries; j++) {
1833                         struct journal_entry *je = access_journal_entry(ic, i, j);
1834                         sector_t sec, area, offset;
1835                         unsigned k, l, next_loop;
1836                         sector_t metadata_block;
1837                         unsigned metadata_offset;
1838                         struct journal_io *io;
1839
1840                         if (journal_entry_is_unused(je))
1841                                 continue;
1842                         BUG_ON(unlikely(journal_entry_is_inprogress(je)) && !from_replay);
1843                         sec = journal_entry_get_sector(je);
1844                         if (unlikely(from_replay)) {
1845                                 if (unlikely(sec & (unsigned)(ic->sectors_per_block - 1))) {
1846                                         dm_integrity_io_error(ic, "invalid sector in journal", -EIO);
1847                                         sec &= ~(sector_t)(ic->sectors_per_block - 1);
1848                                 }
1849                         }
1850                         get_area_and_offset(ic, sec, &area, &offset);
1851                         restore_last_bytes(ic, access_journal_data(ic, i, j), je);
1852                         for (k = j + 1; k < ic->journal_section_entries; k++) {
1853                                 struct journal_entry *je2 = access_journal_entry(ic, i, k);
1854                                 sector_t sec2, area2, offset2;
1855                                 if (journal_entry_is_unused(je2))
1856                                         break;
1857                                 BUG_ON(unlikely(journal_entry_is_inprogress(je2)) && !from_replay);
1858                                 sec2 = journal_entry_get_sector(je2);
1859                                 get_area_and_offset(ic, sec2, &area2, &offset2);
1860                                 if (area2 != area || offset2 != offset + ((k - j) << ic->sb->log2_sectors_per_block))
1861                                         break;
1862                                 restore_last_bytes(ic, access_journal_data(ic, i, k), je2);
1863                         }
1864                         next_loop = k - 1;
1865
1866                         io = mempool_alloc(ic->journal_io_mempool, GFP_NOIO);
1867                         io->comp = &comp;
1868                         io->range.logical_sector = sec;
1869                         io->range.n_sectors = (k - j) << ic->sb->log2_sectors_per_block;
1870
1871                         spin_lock_irq(&ic->endio_wait.lock);
1872                         while (unlikely(!add_new_range(ic, &io->range)))
1873                                 sleep_on_endio_wait(ic);
1874
1875                         if (likely(!from_replay)) {
1876                                 struct journal_node *section_node = &ic->journal_tree[i * ic->journal_section_entries];
1877
1878                                 /* don't write if there is newer committed sector */
1879                                 while (j < k && find_newer_committed_node(ic, &section_node[j])) {
1880                                         struct journal_entry *je2 = access_journal_entry(ic, i, j);
1881
1882                                         journal_entry_set_unused(je2);
1883                                         remove_journal_node(ic, &section_node[j]);
1884                                         j++;
1885                                         sec += ic->sectors_per_block;
1886                                         offset += ic->sectors_per_block;
1887                                 }
1888                                 while (j < k && find_newer_committed_node(ic, &section_node[k - 1])) {
1889                                         struct journal_entry *je2 = access_journal_entry(ic, i, k - 1);
1890
1891                                         journal_entry_set_unused(je2);
1892                                         remove_journal_node(ic, &section_node[k - 1]);
1893                                         k--;
1894                                 }
1895                                 if (j == k) {
1896                                         remove_range_unlocked(ic, &io->range);
1897                                         spin_unlock_irq(&ic->endio_wait.lock);
1898                                         mempool_free(io, ic->journal_io_mempool);
1899                                         goto skip_io;
1900                                 }
1901                                 for (l = j; l < k; l++) {
1902                                         remove_journal_node(ic, &section_node[l]);
1903                                 }
1904                         }
1905                         spin_unlock_irq(&ic->endio_wait.lock);
1906
1907                         metadata_block = get_metadata_sector_and_offset(ic, area, offset, &metadata_offset);
1908                         for (l = j; l < k; l++) {
1909                                 int r;
1910                                 struct journal_entry *je2 = access_journal_entry(ic, i, l);
1911
1912                                 if (
1913 #ifndef INTERNAL_VERIFY
1914                                     unlikely(from_replay) &&
1915 #endif
1916                                     ic->internal_hash) {
1917                                         char test_tag[max(crypto_shash_digestsize(ic->internal_hash), ic->tag_size)];
1918
1919                                         integrity_sector_checksum(ic, sec + ((l - j) << ic->sb->log2_sectors_per_block),
1920                                                                   (char *)access_journal_data(ic, i, l), test_tag);
1921                                         if (unlikely(memcmp(test_tag, journal_entry_tag(ic, je2), ic->tag_size)))
1922                                                 dm_integrity_io_error(ic, "tag mismatch when replaying journal", -EILSEQ);
1923                                 }
1924
1925                                 journal_entry_set_unused(je2);
1926                                 r = dm_integrity_rw_tag(ic, journal_entry_tag(ic, je2), &metadata_block, &metadata_offset,
1927                                                         ic->tag_size, TAG_WRITE);
1928                                 if (unlikely(r)) {
1929                                         dm_integrity_io_error(ic, "reading tags", r);
1930                                 }
1931                         }
1932
1933                         atomic_inc(&comp.in_flight);
1934                         copy_from_journal(ic, i, j << ic->sb->log2_sectors_per_block,
1935                                           (k - j) << ic->sb->log2_sectors_per_block,
1936                                           get_data_sector(ic, area, offset),
1937                                           complete_copy_from_journal, io);
1938 skip_io:
1939                         j = next_loop;
1940                 }
1941         }
1942
1943         dm_bufio_write_dirty_buffers_async(ic->bufio);
1944
1945         complete_journal_op(&comp);
1946         wait_for_completion_io(&comp.comp);
1947
1948         dm_integrity_flush_buffers(ic);
1949 }
1950
1951 static void integrity_writer(struct work_struct *w)
1952 {
1953         struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, writer_work);
1954         unsigned write_start, write_sections;
1955
1956         unsigned prev_free_sectors;
1957
1958         /* the following test is not needed, but it tests the replay code */
1959         if (ACCESS_ONCE(ic->suspending))
1960                 return;
1961
1962         spin_lock_irq(&ic->endio_wait.lock);
1963         write_start = ic->committed_section;
1964         write_sections = ic->n_committed_sections;
1965         spin_unlock_irq(&ic->endio_wait.lock);
1966
1967         if (!write_sections)
1968                 return;
1969
1970         do_journal_write(ic, write_start, write_sections, false);
1971
1972         spin_lock_irq(&ic->endio_wait.lock);
1973
1974         ic->committed_section += write_sections;
1975         wraparound_section(ic, &ic->committed_section);
1976         ic->n_committed_sections -= write_sections;
1977
1978         prev_free_sectors = ic->free_sectors;
1979         ic->free_sectors += write_sections * ic->journal_section_entries;
1980         if (unlikely(!prev_free_sectors))
1981                 wake_up_locked(&ic->endio_wait);
1982
1983         spin_unlock_irq(&ic->endio_wait.lock);
1984 }
1985
1986 static void init_journal(struct dm_integrity_c *ic, unsigned start_section,
1987                          unsigned n_sections, unsigned char commit_seq)
1988 {
1989         unsigned i, j, n;
1990
1991         if (!n_sections)
1992                 return;
1993
1994         for (n = 0; n < n_sections; n++) {
1995                 i = start_section + n;
1996                 wraparound_section(ic, &i);
1997                 for (j = 0; j < ic->journal_section_sectors; j++) {
1998                         struct journal_sector *js = access_journal(ic, i, j);
1999                         memset(&js->entries, 0, JOURNAL_SECTOR_DATA);
2000                         js->commit_id = dm_integrity_commit_id(ic, i, j, commit_seq);
2001                 }
2002                 for (j = 0; j < ic->journal_section_entries; j++) {
2003                         struct journal_entry *je = access_journal_entry(ic, i, j);
2004                         journal_entry_set_unused(je);
2005                 }
2006         }
2007
2008         write_journal(ic, start_section, n_sections);
2009 }
2010
2011 static int find_commit_seq(struct dm_integrity_c *ic, unsigned i, unsigned j, commit_id_t id)
2012 {
2013         unsigned char k;
2014         for (k = 0; k < N_COMMIT_IDS; k++) {
2015                 if (dm_integrity_commit_id(ic, i, j, k) == id)
2016                         return k;
2017         }
2018         dm_integrity_io_error(ic, "journal commit id", -EIO);
2019         return -EIO;
2020 }
2021
2022 static void replay_journal(struct dm_integrity_c *ic)
2023 {
2024         unsigned i, j;
2025         bool used_commit_ids[N_COMMIT_IDS];
2026         unsigned max_commit_id_sections[N_COMMIT_IDS];
2027         unsigned write_start, write_sections;
2028         unsigned continue_section;
2029         bool journal_empty;
2030         unsigned char unused, last_used, want_commit_seq;
2031
2032         if (ic->mode == 'R')
2033                 return;
2034
2035         if (ic->journal_uptodate)
2036                 return;
2037
2038         last_used = 0;
2039         write_start = 0;
2040
2041         if (!ic->just_formatted) {
2042                 DEBUG_print("reading journal\n");
2043                 rw_journal(ic, REQ_OP_READ, 0, 0, ic->journal_sections, NULL);
2044                 if (ic->journal_io)
2045                         DEBUG_bytes(lowmem_page_address(ic->journal_io[0].page), 64, "read journal");
2046                 if (ic->journal_io) {
2047                         struct journal_completion crypt_comp;
2048                         crypt_comp.ic = ic;
2049                         crypt_comp.comp = COMPLETION_INITIALIZER_ONSTACK(crypt_comp.comp);
2050                         crypt_comp.in_flight = (atomic_t)ATOMIC_INIT(0);
2051                         encrypt_journal(ic, false, 0, ic->journal_sections, &crypt_comp);
2052                         wait_for_completion(&crypt_comp.comp);
2053                 }
2054                 DEBUG_bytes(lowmem_page_address(ic->journal[0].page), 64, "decrypted journal");
2055         }
2056
2057         if (dm_integrity_failed(ic))
2058                 goto clear_journal;
2059
2060         journal_empty = true;
2061         memset(used_commit_ids, 0, sizeof used_commit_ids);
2062         memset(max_commit_id_sections, 0, sizeof max_commit_id_sections);
2063         for (i = 0; i < ic->journal_sections; i++) {
2064                 for (j = 0; j < ic->journal_section_sectors; j++) {
2065                         int k;
2066                         struct journal_sector *js = access_journal(ic, i, j);
2067                         k = find_commit_seq(ic, i, j, js->commit_id);
2068                         if (k < 0)
2069                                 goto clear_journal;
2070                         used_commit_ids[k] = true;
2071                         max_commit_id_sections[k] = i;
2072                 }
2073                 if (journal_empty) {
2074                         for (j = 0; j < ic->journal_section_entries; j++) {
2075                                 struct journal_entry *je = access_journal_entry(ic, i, j);
2076                                 if (!journal_entry_is_unused(je)) {
2077                                         journal_empty = false;
2078                                         break;
2079                                 }
2080                         }
2081                 }
2082         }
2083
2084         if (!used_commit_ids[N_COMMIT_IDS - 1]) {
2085                 unused = N_COMMIT_IDS - 1;
2086                 while (unused && !used_commit_ids[unused - 1])
2087                         unused--;
2088         } else {
2089                 for (unused = 0; unused < N_COMMIT_IDS; unused++)
2090                         if (!used_commit_ids[unused])
2091                                 break;
2092                 if (unused == N_COMMIT_IDS) {
2093                         dm_integrity_io_error(ic, "journal commit ids", -EIO);
2094                         goto clear_journal;
2095                 }
2096         }
2097         DEBUG_print("first unused commit seq %d [%d,%d,%d,%d]\n",
2098                     unused, used_commit_ids[0], used_commit_ids[1],
2099                     used_commit_ids[2], used_commit_ids[3]);
2100
2101         last_used = prev_commit_seq(unused);
2102         want_commit_seq = prev_commit_seq(last_used);
2103
2104         if (!used_commit_ids[want_commit_seq] && used_commit_ids[prev_commit_seq(want_commit_seq)])
2105                 journal_empty = true;
2106
2107         write_start = max_commit_id_sections[last_used] + 1;
2108         if (unlikely(write_start >= ic->journal_sections))
2109                 want_commit_seq = next_commit_seq(want_commit_seq);
2110         wraparound_section(ic, &write_start);
2111
2112         i = write_start;
2113         for (write_sections = 0; write_sections < ic->journal_sections; write_sections++) {
2114                 for (j = 0; j < ic->journal_section_sectors; j++) {
2115                         struct journal_sector *js = access_journal(ic, i, j);
2116
2117                         if (js->commit_id != dm_integrity_commit_id(ic, i, j, want_commit_seq)) {
2118                                 /*
2119                                  * This could be caused by crash during writing.
2120                                  * We won't replay the inconsistent part of the
2121                                  * journal.
2122                                  */
2123                                 DEBUG_print("commit id mismatch at position (%u, %u): %d != %d\n",
2124                                             i, j, find_commit_seq(ic, i, j, js->commit_id), want_commit_seq);
2125                                 goto brk;
2126                         }
2127                 }
2128                 i++;
2129                 if (unlikely(i >= ic->journal_sections))
2130                         want_commit_seq = next_commit_seq(want_commit_seq);
2131                 wraparound_section(ic, &i);
2132         }
2133 brk:
2134
2135         if (!journal_empty) {
2136                 DEBUG_print("replaying %u sections, starting at %u, commit seq %d\n",
2137                             write_sections, write_start, want_commit_seq);
2138                 do_journal_write(ic, write_start, write_sections, true);
2139         }
2140
2141         if (write_sections == ic->journal_sections && (ic->mode == 'J' || journal_empty)) {
2142                 continue_section = write_start;
2143                 ic->commit_seq = want_commit_seq;
2144                 DEBUG_print("continuing from section %u, commit seq %d\n", write_start, ic->commit_seq);
2145         } else {
2146                 unsigned s;
2147                 unsigned char erase_seq;
2148 clear_journal:
2149                 DEBUG_print("clearing journal\n");
2150
2151                 erase_seq = prev_commit_seq(prev_commit_seq(last_used));
2152                 s = write_start;
2153                 init_journal(ic, s, 1, erase_seq);
2154                 s++;
2155                 wraparound_section(ic, &s);
2156                 if (ic->journal_sections >= 2) {
2157                         init_journal(ic, s, ic->journal_sections - 2, erase_seq);
2158                         s += ic->journal_sections - 2;
2159                         wraparound_section(ic, &s);
2160                         init_journal(ic, s, 1, erase_seq);
2161                 }
2162
2163                 continue_section = 0;
2164                 ic->commit_seq = next_commit_seq(erase_seq);
2165         }
2166
2167         ic->committed_section = continue_section;
2168         ic->n_committed_sections = 0;
2169
2170         ic->uncommitted_section = continue_section;
2171         ic->n_uncommitted_sections = 0;
2172
2173         ic->free_section = continue_section;
2174         ic->free_section_entry = 0;
2175         ic->free_sectors = ic->journal_entries;
2176
2177         ic->journal_tree_root = RB_ROOT;
2178         for (i = 0; i < ic->journal_entries; i++)
2179                 init_journal_node(&ic->journal_tree[i]);
2180 }
2181
2182 static void dm_integrity_postsuspend(struct dm_target *ti)
2183 {
2184         struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
2185
2186         del_timer_sync(&ic->autocommit_timer);
2187
2188         ic->suspending = true;
2189
2190         queue_work(ic->commit_wq, &ic->commit_work);
2191         drain_workqueue(ic->commit_wq);
2192
2193         if (ic->mode == 'J') {
2194                 drain_workqueue(ic->writer_wq);
2195                 dm_integrity_flush_buffers(ic);
2196         }
2197
2198         ic->suspending = false;
2199
2200         BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
2201
2202         ic->journal_uptodate = true;
2203 }
2204
2205 static void dm_integrity_resume(struct dm_target *ti)
2206 {
2207         struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
2208
2209         replay_journal(ic);
2210 }
2211
2212 static void dm_integrity_status(struct dm_target *ti, status_type_t type,
2213                                 unsigned status_flags, char *result, unsigned maxlen)
2214 {
2215         struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
2216         unsigned arg_count;
2217         size_t sz = 0;
2218
2219         switch (type) {
2220         case STATUSTYPE_INFO:
2221                 result[0] = '\0';
2222                 break;
2223
2224         case STATUSTYPE_TABLE: {
2225                 __u64 watermark_percentage = (__u64)(ic->journal_entries - ic->free_sectors_threshold) * 100;
2226                 watermark_percentage += ic->journal_entries / 2;
2227                 do_div(watermark_percentage, ic->journal_entries);
2228                 arg_count = 5;
2229                 arg_count += ic->sectors_per_block != 1;
2230                 arg_count += !!ic->internal_hash_alg.alg_string;
2231                 arg_count += !!ic->journal_crypt_alg.alg_string;
2232                 arg_count += !!ic->journal_mac_alg.alg_string;
2233                 DMEMIT("%s %llu %u %c %u", ic->dev->name, (unsigned long long)ic->start,
2234                        ic->tag_size, ic->mode, arg_count);
2235                 DMEMIT(" journal_sectors:%u", ic->initial_sectors - SB_SECTORS);
2236                 DMEMIT(" interleave_sectors:%u", 1U << ic->sb->log2_interleave_sectors);
2237                 DMEMIT(" buffer_sectors:%u", 1U << ic->log2_buffer_sectors);
2238                 DMEMIT(" journal_watermark:%u", (unsigned)watermark_percentage);
2239                 DMEMIT(" commit_time:%u", ic->autocommit_msec);
2240                 if (ic->sectors_per_block != 1)
2241                         DMEMIT(" block_size:%u", ic->sectors_per_block << SECTOR_SHIFT);
2242
2243 #define EMIT_ALG(a, n)                                                  \
2244                 do {                                                    \
2245                         if (ic->a.alg_string) {                         \
2246                                 DMEMIT(" %s:%s", n, ic->a.alg_string);  \
2247                                 if (ic->a.key_string)                   \
2248                                         DMEMIT(":%s", ic->a.key_string);\
2249                         }                                               \
2250                 } while (0)
2251                 EMIT_ALG(internal_hash_alg, "internal_hash");
2252                 EMIT_ALG(journal_crypt_alg, "journal_crypt");
2253                 EMIT_ALG(journal_mac_alg, "journal_mac");
2254                 break;
2255         }
2256         }
2257 }
2258
2259 static int dm_integrity_iterate_devices(struct dm_target *ti,
2260                                         iterate_devices_callout_fn fn, void *data)
2261 {
2262         struct dm_integrity_c *ic = ti->private;
2263
2264         return fn(ti, ic->dev, ic->start + ic->initial_sectors + ic->metadata_run, ti->len, data);
2265 }
2266
2267 static void dm_integrity_io_hints(struct dm_target *ti, struct queue_limits *limits)
2268 {
2269         struct dm_integrity_c *ic = ti->private;
2270
2271         if (ic->sectors_per_block > 1) {
2272                 limits->logical_block_size = ic->sectors_per_block << SECTOR_SHIFT;
2273                 limits->physical_block_size = ic->sectors_per_block << SECTOR_SHIFT;
2274                 blk_limits_io_min(limits, ic->sectors_per_block << SECTOR_SHIFT);
2275         }
2276 }
2277
2278 static void calculate_journal_section_size(struct dm_integrity_c *ic)
2279 {
2280         unsigned sector_space = JOURNAL_SECTOR_DATA;
2281
2282         ic->journal_sections = le32_to_cpu(ic->sb->journal_sections);
2283         ic->journal_entry_size = roundup(offsetof(struct journal_entry, last_bytes[ic->sectors_per_block]) + ic->tag_size,
2284                                          JOURNAL_ENTRY_ROUNDUP);
2285
2286         if (ic->sb->flags & cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC))
2287                 sector_space -= JOURNAL_MAC_PER_SECTOR;
2288         ic->journal_entries_per_sector = sector_space / ic->journal_entry_size;
2289         ic->journal_section_entries = ic->journal_entries_per_sector * JOURNAL_BLOCK_SECTORS;
2290         ic->journal_section_sectors = (ic->journal_section_entries << ic->sb->log2_sectors_per_block) + JOURNAL_BLOCK_SECTORS;
2291         ic->journal_entries = ic->journal_section_entries * ic->journal_sections;
2292 }
2293
2294 static int calculate_device_limits(struct dm_integrity_c *ic)
2295 {
2296         __u64 initial_sectors;
2297         sector_t last_sector, last_area, last_offset;
2298
2299         calculate_journal_section_size(ic);
2300         initial_sectors = SB_SECTORS + (__u64)ic->journal_section_sectors * ic->journal_sections;
2301         if (initial_sectors + METADATA_PADDING_SECTORS >= ic->device_sectors || initial_sectors > UINT_MAX)
2302                 return -EINVAL;
2303         ic->initial_sectors = initial_sectors;
2304
2305         ic->metadata_run = roundup((__u64)ic->tag_size << (ic->sb->log2_interleave_sectors - ic->sb->log2_sectors_per_block),
2306                                    (__u64)(1 << SECTOR_SHIFT << METADATA_PADDING_SECTORS)) >> SECTOR_SHIFT;
2307         if (!(ic->metadata_run & (ic->metadata_run - 1)))
2308                 ic->log2_metadata_run = __ffs(ic->metadata_run);
2309         else
2310                 ic->log2_metadata_run = -1;
2311
2312         get_area_and_offset(ic, ic->provided_data_sectors - 1, &last_area, &last_offset);
2313         last_sector = get_data_sector(ic, last_area, last_offset);
2314
2315         if (ic->start + last_sector < last_sector || ic->start + last_sector >= ic->device_sectors)
2316                 return -EINVAL;
2317
2318         return 0;
2319 }
2320
2321 static int initialize_superblock(struct dm_integrity_c *ic, unsigned journal_sectors, unsigned interleave_sectors)
2322 {
2323         unsigned journal_sections;
2324         int test_bit;
2325
2326         memset(ic->sb, 0, SB_SECTORS << SECTOR_SHIFT);
2327         memcpy(ic->sb->magic, SB_MAGIC, 8);
2328         ic->sb->version = SB_VERSION;
2329         ic->sb->integrity_tag_size = cpu_to_le16(ic->tag_size);
2330         ic->sb->log2_sectors_per_block = __ffs(ic->sectors_per_block);
2331         if (ic->journal_mac_alg.alg_string)
2332                 ic->sb->flags |= cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC);
2333
2334         calculate_journal_section_size(ic);
2335         journal_sections = journal_sectors / ic->journal_section_sectors;
2336         if (!journal_sections)
2337                 journal_sections = 1;
2338         ic->sb->journal_sections = cpu_to_le32(journal_sections);
2339
2340         if (!interleave_sectors)
2341                 interleave_sectors = DEFAULT_INTERLEAVE_SECTORS;
2342         ic->sb->log2_interleave_sectors = __fls(interleave_sectors);
2343         ic->sb->log2_interleave_sectors = max((__u8)MIN_LOG2_INTERLEAVE_SECTORS, ic->sb->log2_interleave_sectors);
2344         ic->sb->log2_interleave_sectors = min((__u8)MAX_LOG2_INTERLEAVE_SECTORS, ic->sb->log2_interleave_sectors);
2345
2346         ic->provided_data_sectors = 0;
2347         for (test_bit = fls64(ic->device_sectors) - 1; test_bit >= 3; test_bit--) {
2348                 __u64 prev_data_sectors = ic->provided_data_sectors;
2349
2350                 ic->provided_data_sectors |= (sector_t)1 << test_bit;
2351                 if (calculate_device_limits(ic))
2352                         ic->provided_data_sectors = prev_data_sectors;
2353         }
2354
2355         if (!ic->provided_data_sectors)
2356                 return -EINVAL;
2357
2358         ic->sb->provided_data_sectors = cpu_to_le64(ic->provided_data_sectors);
2359
2360         return 0;
2361 }
2362
2363 static void dm_integrity_set(struct dm_target *ti, struct dm_integrity_c *ic)
2364 {
2365         struct gendisk *disk = dm_disk(dm_table_get_md(ti->table));
2366         struct blk_integrity bi;
2367
2368         memset(&bi, 0, sizeof(bi));
2369         bi.profile = &dm_integrity_profile;
2370         bi.tuple_size = ic->tag_size;
2371         bi.tag_size = bi.tuple_size;
2372         bi.interval_exp = ilog2(ic->sectors_per_block << SECTOR_SHIFT);
2373
2374         blk_integrity_register(disk, &bi);
2375         blk_queue_max_integrity_segments(disk->queue, UINT_MAX);
2376 }
2377
2378 /* FIXME: use new kvmalloc */
2379 static void *dm_integrity_kvmalloc(size_t size, gfp_t gfp)
2380 {
2381         void *ptr = NULL;
2382
2383         if (size <= PAGE_SIZE)
2384                 ptr = kmalloc(size, GFP_KERNEL | gfp);
2385         if (!ptr && size <= KMALLOC_MAX_SIZE)
2386                 ptr = kmalloc(size, GFP_KERNEL | __GFP_NOWARN | __GFP_NORETRY | gfp);
2387         if (!ptr)
2388                 ptr = __vmalloc(size, GFP_KERNEL | gfp, PAGE_KERNEL);
2389
2390         return ptr;
2391 }
2392
2393 static void dm_integrity_free_page_list(struct dm_integrity_c *ic, struct page_list *pl)
2394 {
2395         unsigned i;
2396
2397         if (!pl)
2398                 return;
2399         for (i = 0; i < ic->journal_pages; i++)
2400                 if (pl[i].page)
2401                         __free_page(pl[i].page);
2402         kvfree(pl);
2403 }
2404
2405 static struct page_list *dm_integrity_alloc_page_list(struct dm_integrity_c *ic)
2406 {
2407         size_t page_list_desc_size = ic->journal_pages * sizeof(struct page_list);
2408         struct page_list *pl;
2409         unsigned i;
2410
2411         pl = dm_integrity_kvmalloc(page_list_desc_size, __GFP_ZERO);
2412         if (!pl)
2413                 return NULL;
2414
2415         for (i = 0; i < ic->journal_pages; i++) {
2416                 pl[i].page = alloc_page(GFP_KERNEL);
2417                 if (!pl[i].page) {
2418                         dm_integrity_free_page_list(ic, pl);
2419                         return NULL;
2420                 }
2421                 if (i)
2422                         pl[i - 1].next = &pl[i];
2423         }
2424
2425         return pl;
2426 }
2427
2428 static void dm_integrity_free_journal_scatterlist(struct dm_integrity_c *ic, struct scatterlist **sl)
2429 {
2430         unsigned i;
2431         for (i = 0; i < ic->journal_sections; i++)
2432                 kvfree(sl[i]);
2433         kfree(sl);
2434 }
2435
2436 static struct scatterlist **dm_integrity_alloc_journal_scatterlist(struct dm_integrity_c *ic, struct page_list *pl)
2437 {
2438         struct scatterlist **sl;
2439         unsigned i;
2440
2441         sl = dm_integrity_kvmalloc(ic->journal_sections * sizeof(struct scatterlist *), __GFP_ZERO);
2442         if (!sl)
2443                 return NULL;
2444
2445         for (i = 0; i < ic->journal_sections; i++) {
2446                 struct scatterlist *s;
2447                 unsigned start_index, start_offset;
2448                 unsigned end_index, end_offset;
2449                 unsigned n_pages;
2450                 unsigned idx;
2451
2452                 page_list_location(ic, i, 0, &start_index, &start_offset);
2453                 page_list_location(ic, i, ic->journal_section_sectors - 1, &end_index, &end_offset);
2454
2455                 n_pages = (end_index - start_index + 1);
2456
2457                 s = dm_integrity_kvmalloc(n_pages * sizeof(struct scatterlist), 0);
2458                 if (!s) {
2459                         dm_integrity_free_journal_scatterlist(ic, sl);
2460                         return NULL;
2461                 }
2462
2463                 sg_init_table(s, n_pages);
2464                 for (idx = start_index; idx <= end_index; idx++) {
2465                         char *va = lowmem_page_address(pl[idx].page);
2466                         unsigned start = 0, end = PAGE_SIZE;
2467                         if (idx == start_index)
2468                                 start = start_offset;
2469                         if (idx == end_index)
2470                                 end = end_offset + (1 << SECTOR_SHIFT);
2471                         sg_set_buf(&s[idx - start_index], va + start, end - start);
2472                 }
2473
2474                 sl[i] = s;
2475         }
2476
2477         return sl;
2478 }
2479
2480 static void free_alg(struct alg_spec *a)
2481 {
2482         kzfree(a->alg_string);
2483         kzfree(a->key);
2484         memset(a, 0, sizeof *a);
2485 }
2486
2487 static int get_alg_and_key(const char *arg, struct alg_spec *a, char **error, char *error_inval)
2488 {
2489         char *k;
2490
2491         free_alg(a);
2492
2493         a->alg_string = kstrdup(strchr(arg, ':') + 1, GFP_KERNEL);
2494         if (!a->alg_string)
2495                 goto nomem;
2496
2497         k = strchr(a->alg_string, ':');
2498         if (k) {
2499                 unsigned i;
2500
2501                 *k = 0;
2502                 a->key_string = k + 1;
2503                 if (strlen(a->key_string) & 1)
2504                         goto inval;
2505
2506                 a->key_size = strlen(a->key_string) / 2;
2507                 a->key = kmalloc(a->key_size, GFP_KERNEL);
2508                 if (!a->key)
2509                         goto nomem;
2510                 for (i = 0; i < a->key_size; i++) {
2511                         char digit[3];
2512                         digit[0] = a->key_string[i * 2];
2513                         digit[1] = a->key_string[i * 2 + 1];
2514                         digit[2] = 0;
2515                         if (strspn(digit, "0123456789abcdefABCDEF") != 2)
2516                                 goto inval;
2517                         if (kstrtou8(digit, 16, &a->key[i]))
2518                                 goto inval;
2519                 }
2520         }
2521
2522         return 0;
2523 inval:
2524         *error = error_inval;
2525         return -EINVAL;
2526 nomem:
2527         *error = "Out of memory for an argument";
2528         return -ENOMEM;
2529 }
2530
2531 static int get_mac(struct crypto_shash **hash, struct alg_spec *a, char **error,
2532                    char *error_alg, char *error_key)
2533 {
2534         int r;
2535
2536         if (a->alg_string) {
2537                 *hash = crypto_alloc_shash(a->alg_string, 0, CRYPTO_ALG_ASYNC);
2538                 if (IS_ERR(*hash)) {
2539                         *error = error_alg;
2540                         r = PTR_ERR(*hash);
2541                         *hash = NULL;
2542                         return r;
2543                 }
2544
2545                 if (a->key) {
2546                         r = crypto_shash_setkey(*hash, a->key, a->key_size);
2547                         if (r) {
2548                                 *error = error_key;
2549                                 return r;
2550                         }
2551                 }
2552         }
2553
2554         return 0;
2555 }
2556
2557 static int create_journal(struct dm_integrity_c *ic, char **error)
2558 {
2559         int r = 0;
2560         unsigned i;
2561         __u64 journal_pages, journal_desc_size, journal_tree_size;
2562         unsigned char *crypt_data = NULL;
2563
2564         ic->commit_ids[0] = cpu_to_le64(0x1111111111111111ULL);
2565         ic->commit_ids[1] = cpu_to_le64(0x2222222222222222ULL);
2566         ic->commit_ids[2] = cpu_to_le64(0x3333333333333333ULL);
2567         ic->commit_ids[3] = cpu_to_le64(0x4444444444444444ULL);
2568
2569         journal_pages = roundup((__u64)ic->journal_sections * ic->journal_section_sectors,
2570                                 PAGE_SIZE >> SECTOR_SHIFT) >> (PAGE_SHIFT - SECTOR_SHIFT);
2571         journal_desc_size = journal_pages * sizeof(struct page_list);
2572         if (journal_pages >= totalram_pages - totalhigh_pages || journal_desc_size > ULONG_MAX) {
2573                 *error = "Journal doesn't fit into memory";
2574                 r = -ENOMEM;
2575                 goto bad;
2576         }
2577         ic->journal_pages = journal_pages;
2578
2579         ic->journal = dm_integrity_alloc_page_list(ic);
2580         if (!ic->journal) {
2581                 *error = "Could not allocate memory for journal";
2582                 r = -ENOMEM;
2583                 goto bad;
2584         }
2585         if (ic->journal_crypt_alg.alg_string) {
2586                 unsigned ivsize, blocksize;
2587                 struct journal_completion comp;
2588
2589                 comp.ic = ic;
2590                 ic->journal_crypt = crypto_alloc_skcipher(ic->journal_crypt_alg.alg_string, 0, 0);
2591                 if (IS_ERR(ic->journal_crypt)) {
2592                         *error = "Invalid journal cipher";
2593                         r = PTR_ERR(ic->journal_crypt);
2594                         ic->journal_crypt = NULL;
2595                         goto bad;
2596                 }
2597                 ivsize = crypto_skcipher_ivsize(ic->journal_crypt);
2598                 blocksize = crypto_skcipher_blocksize(ic->journal_crypt);
2599
2600                 if (ic->journal_crypt_alg.key) {
2601                         r = crypto_skcipher_setkey(ic->journal_crypt, ic->journal_crypt_alg.key,
2602                                                    ic->journal_crypt_alg.key_size);
2603                         if (r) {
2604                                 *error = "Error setting encryption key";
2605                                 goto bad;
2606                         }
2607                 }
2608                 DEBUG_print("cipher %s, block size %u iv size %u\n",
2609                             ic->journal_crypt_alg.alg_string, blocksize, ivsize);
2610
2611                 ic->journal_io = dm_integrity_alloc_page_list(ic);
2612                 if (!ic->journal_io) {
2613                         *error = "Could not allocate memory for journal io";
2614                         r = -ENOMEM;
2615                         goto bad;
2616                 }
2617
2618                 if (blocksize == 1) {
2619                         struct scatterlist *sg;
2620                         SKCIPHER_REQUEST_ON_STACK(req, ic->journal_crypt);
2621                         unsigned char iv[ivsize];
2622                         skcipher_request_set_tfm(req, ic->journal_crypt);
2623
2624                         ic->journal_xor = dm_integrity_alloc_page_list(ic);
2625                         if (!ic->journal_xor) {
2626                                 *error = "Could not allocate memory for journal xor";
2627                                 r = -ENOMEM;
2628                                 goto bad;
2629                         }
2630
2631                         sg = dm_integrity_kvmalloc((ic->journal_pages + 1) * sizeof(struct scatterlist), 0);
2632                         if (!sg) {
2633                                 *error = "Unable to allocate sg list";
2634                                 r = -ENOMEM;
2635                                 goto bad;
2636                         }
2637                         sg_init_table(sg, ic->journal_pages + 1);
2638                         for (i = 0; i < ic->journal_pages; i++) {
2639                                 char *va = lowmem_page_address(ic->journal_xor[i].page);
2640                                 clear_page(va);
2641                                 sg_set_buf(&sg[i], va, PAGE_SIZE);
2642                         }
2643                         sg_set_buf(&sg[i], &ic->commit_ids, sizeof ic->commit_ids);
2644                         memset(iv, 0x00, ivsize);
2645
2646                         skcipher_request_set_crypt(req, sg, sg, PAGE_SIZE * ic->journal_pages + sizeof ic->commit_ids, iv);
2647                         comp.comp = COMPLETION_INITIALIZER_ONSTACK(comp.comp);
2648                         comp.in_flight = (atomic_t)ATOMIC_INIT(1);
2649                         if (do_crypt(true, req, &comp))
2650                                 wait_for_completion(&comp.comp);
2651                         kvfree(sg);
2652                         r = dm_integrity_failed(ic);
2653                         if (r) {
2654                                 *error = "Unable to encrypt journal";
2655                                 goto bad;
2656                         }
2657                         DEBUG_bytes(lowmem_page_address(ic->journal_xor[0].page), 64, "xor data");
2658
2659                         crypto_free_skcipher(ic->journal_crypt);
2660                         ic->journal_crypt = NULL;
2661                 } else {
2662                         SKCIPHER_REQUEST_ON_STACK(req, ic->journal_crypt);
2663                         unsigned char iv[ivsize];
2664                         unsigned crypt_len = roundup(ivsize, blocksize);
2665
2666                         crypt_data = kmalloc(crypt_len, GFP_KERNEL);
2667                         if (!crypt_data) {
2668                                 *error = "Unable to allocate crypt data";
2669                                 r = -ENOMEM;
2670                                 goto bad;
2671                         }
2672
2673                         skcipher_request_set_tfm(req, ic->journal_crypt);
2674
2675                         ic->journal_scatterlist = dm_integrity_alloc_journal_scatterlist(ic, ic->journal);
2676                         if (!ic->journal_scatterlist) {
2677                                 *error = "Unable to allocate sg list";
2678                                 r = -ENOMEM;
2679                                 goto bad;
2680                         }
2681                         ic->journal_io_scatterlist = dm_integrity_alloc_journal_scatterlist(ic, ic->journal_io);
2682                         if (!ic->journal_io_scatterlist) {
2683                                 *error = "Unable to allocate sg list";
2684                                 r = -ENOMEM;
2685                                 goto bad;
2686                         }
2687                         ic->sk_requests = dm_integrity_kvmalloc(ic->journal_sections * sizeof(struct skcipher_request *), __GFP_ZERO);
2688                         if (!ic->sk_requests) {
2689                                 *error = "Unable to allocate sk requests";
2690                                 r = -ENOMEM;
2691                                 goto bad;
2692                         }
2693                         for (i = 0; i < ic->journal_sections; i++) {
2694                                 struct scatterlist sg;
2695                                 struct skcipher_request *section_req;
2696                                 __u32 section_le = cpu_to_le32(i);
2697
2698                                 memset(iv, 0x00, ivsize);
2699                                 memset(crypt_data, 0x00, crypt_len);
2700                                 memcpy(crypt_data, &section_le, min((size_t)crypt_len, sizeof(section_le)));
2701
2702                                 sg_init_one(&sg, crypt_data, crypt_len);
2703                                 skcipher_request_set_crypt(req, &sg, &sg, crypt_len, iv);
2704                                 comp.comp = COMPLETION_INITIALIZER_ONSTACK(comp.comp);
2705                                 comp.in_flight = (atomic_t)ATOMIC_INIT(1);
2706                                 if (do_crypt(true, req, &comp))
2707                                         wait_for_completion(&comp.comp);
2708
2709                                 r = dm_integrity_failed(ic);
2710                                 if (r) {
2711                                         *error = "Unable to generate iv";
2712                                         goto bad;
2713                                 }
2714
2715                                 section_req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
2716                                 if (!section_req) {
2717                                         *error = "Unable to allocate crypt request";
2718                                         r = -ENOMEM;
2719                                         goto bad;
2720                                 }
2721                                 section_req->iv = kmalloc(ivsize * 2, GFP_KERNEL);
2722                                 if (!section_req->iv) {
2723                                         skcipher_request_free(section_req);
2724                                         *error = "Unable to allocate iv";
2725                                         r = -ENOMEM;
2726                                         goto bad;
2727                                 }
2728                                 memcpy(section_req->iv + ivsize, crypt_data, ivsize);
2729                                 section_req->cryptlen = (size_t)ic->journal_section_sectors << SECTOR_SHIFT;
2730                                 ic->sk_requests[i] = section_req;
2731                                 DEBUG_bytes(crypt_data, ivsize, "iv(%u)", i);
2732                         }
2733                 }
2734         }
2735
2736         for (i = 0; i < N_COMMIT_IDS; i++) {
2737                 unsigned j;
2738 retest_commit_id:
2739                 for (j = 0; j < i; j++) {
2740                         if (ic->commit_ids[j] == ic->commit_ids[i]) {
2741                                 ic->commit_ids[i] = cpu_to_le64(le64_to_cpu(ic->commit_ids[i]) + 1);
2742                                 goto retest_commit_id;
2743                         }
2744                 }
2745                 DEBUG_print("commit id %u: %016llx\n", i, ic->commit_ids[i]);
2746         }
2747
2748         journal_tree_size = (__u64)ic->journal_entries * sizeof(struct journal_node);
2749         if (journal_tree_size > ULONG_MAX) {
2750                 *error = "Journal doesn't fit into memory";
2751                 r = -ENOMEM;
2752                 goto bad;
2753         }
2754         ic->journal_tree = dm_integrity_kvmalloc(journal_tree_size, 0);
2755         if (!ic->journal_tree) {
2756                 *error = "Could not allocate memory for journal tree";
2757                 r = -ENOMEM;
2758         }
2759 bad:
2760         kfree(crypt_data);
2761         return r;
2762 }
2763
2764 /*
2765  * Construct a integrity mapping
2766  *
2767  * Arguments:
2768  *      device
2769  *      offset from the start of the device
2770  *      tag size
2771  *      D - direct writes, J - journal writes, R - recovery mode
2772  *      number of optional arguments
2773  *      optional arguments:
2774  *              journal_sectors
2775  *              interleave_sectors
2776  *              buffer_sectors
2777  *              journal_watermark
2778  *              commit_time
2779  *              internal_hash
2780  *              journal_crypt
2781  *              journal_mac
2782  *              block_size
2783  */
2784 static int dm_integrity_ctr(struct dm_target *ti, unsigned argc, char **argv)
2785 {
2786         struct dm_integrity_c *ic;
2787         char dummy;
2788         int r;
2789         unsigned extra_args;
2790         struct dm_arg_set as;
2791         static struct dm_arg _args[] = {
2792                 {0, 9, "Invalid number of feature args"},
2793         };
2794         unsigned journal_sectors, interleave_sectors, buffer_sectors, journal_watermark, sync_msec;
2795         bool should_write_sb;
2796         __u64 threshold;
2797         unsigned long long start;
2798
2799 #define DIRECT_ARGUMENTS        4
2800
2801         if (argc <= DIRECT_ARGUMENTS) {
2802                 ti->error = "Invalid argument count";
2803                 return -EINVAL;
2804         }
2805
2806         ic = kzalloc(sizeof(struct dm_integrity_c), GFP_KERNEL);
2807         if (!ic) {
2808                 ti->error = "Cannot allocate integrity context";
2809                 return -ENOMEM;
2810         }
2811         ti->private = ic;
2812         ti->per_io_data_size = sizeof(struct dm_integrity_io);
2813
2814         ic->in_progress = RB_ROOT;
2815         init_waitqueue_head(&ic->endio_wait);
2816         bio_list_init(&ic->flush_bio_list);
2817         init_waitqueue_head(&ic->copy_to_journal_wait);
2818         init_completion(&ic->crypto_backoff);
2819
2820         r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &ic->dev);
2821         if (r) {
2822                 ti->error = "Device lookup failed";
2823                 goto bad;
2824         }
2825
2826         if (sscanf(argv[1], "%llu%c", &start, &dummy) != 1 || start != (sector_t)start) {
2827                 ti->error = "Invalid starting offset";
2828                 r = -EINVAL;
2829                 goto bad;
2830         }
2831         ic->start = start;
2832
2833         if (strcmp(argv[2], "-")) {
2834                 if (sscanf(argv[2], "%u%c", &ic->tag_size, &dummy) != 1 || !ic->tag_size) {
2835                         ti->error = "Invalid tag size";
2836                         r = -EINVAL;
2837                         goto bad;
2838                 }
2839         }
2840
2841         if (!strcmp(argv[3], "J") || !strcmp(argv[3], "D") || !strcmp(argv[3], "R"))
2842                 ic->mode = argv[3][0];
2843         else {
2844                 ti->error = "Invalid mode (expecting J, D, R)";
2845                 r = -EINVAL;
2846                 goto bad;
2847         }
2848
2849         ic->device_sectors = i_size_read(ic->dev->bdev->bd_inode) >> SECTOR_SHIFT;
2850         journal_sectors = min((sector_t)DEFAULT_MAX_JOURNAL_SECTORS,
2851                         ic->device_sectors >> DEFAULT_JOURNAL_SIZE_FACTOR);
2852         interleave_sectors = DEFAULT_INTERLEAVE_SECTORS;
2853         buffer_sectors = DEFAULT_BUFFER_SECTORS;
2854         journal_watermark = DEFAULT_JOURNAL_WATERMARK;
2855         sync_msec = DEFAULT_SYNC_MSEC;
2856         ic->sectors_per_block = 1;
2857
2858         as.argc = argc - DIRECT_ARGUMENTS;
2859         as.argv = argv + DIRECT_ARGUMENTS;
2860         r = dm_read_arg_group(_args, &as, &extra_args, &ti->error);
2861         if (r)
2862                 goto bad;
2863
2864         while (extra_args--) {
2865                 const char *opt_string;
2866                 unsigned val;
2867                 opt_string = dm_shift_arg(&as);
2868                 if (!opt_string) {
2869                         r = -EINVAL;
2870                         ti->error = "Not enough feature arguments";
2871                         goto bad;
2872                 }
2873                 if (sscanf(opt_string, "journal_sectors:%u%c", &val, &dummy) == 1)
2874                         journal_sectors = val;
2875                 else if (sscanf(opt_string, "interleave_sectors:%u%c", &val, &dummy) == 1)
2876                         interleave_sectors = val;
2877                 else if (sscanf(opt_string, "buffer_sectors:%u%c", &val, &dummy) == 1)
2878                         buffer_sectors = val;
2879                 else if (sscanf(opt_string, "journal_watermark:%u%c", &val, &dummy) == 1 && val <= 100)
2880                         journal_watermark = val;
2881                 else if (sscanf(opt_string, "commit_time:%u%c", &val, &dummy) == 1)
2882                         sync_msec = val;
2883                 else if (sscanf(opt_string, "block_size:%u%c", &val, &dummy) == 1) {
2884                         if (val < 1 << SECTOR_SHIFT ||
2885                             val > MAX_SECTORS_PER_BLOCK << SECTOR_SHIFT ||
2886                             (val & (val -1))) {
2887                                 r = -EINVAL;
2888                                 ti->error = "Invalid block_size argument";
2889                                 goto bad;
2890                         }
2891                         ic->sectors_per_block = val >> SECTOR_SHIFT;
2892                 } else if (!memcmp(opt_string, "internal_hash:", strlen("internal_hash:"))) {
2893                         r = get_alg_and_key(opt_string, &ic->internal_hash_alg, &ti->error,
2894                                             "Invalid internal_hash argument");
2895                         if (r)
2896                                 goto bad;
2897                 } else if (!memcmp(opt_string, "journal_crypt:", strlen("journal_crypt:"))) {
2898                         r = get_alg_and_key(opt_string, &ic->journal_crypt_alg, &ti->error,
2899                                             "Invalid journal_crypt argument");
2900                         if (r)
2901                                 goto bad;
2902                 } else if (!memcmp(opt_string, "journal_mac:", strlen("journal_mac:"))) {
2903                         r = get_alg_and_key(opt_string, &ic->journal_mac_alg,  &ti->error,
2904                                             "Invalid journal_mac argument");
2905                         if (r)
2906                                 goto bad;
2907                 } else {
2908                         r = -EINVAL;
2909                         ti->error = "Invalid argument";
2910                         goto bad;
2911                 }
2912         }
2913
2914         r = get_mac(&ic->internal_hash, &ic->internal_hash_alg, &ti->error,
2915                     "Invalid internal hash", "Error setting internal hash key");
2916         if (r)
2917                 goto bad;
2918
2919         r = get_mac(&ic->journal_mac, &ic->journal_mac_alg, &ti->error,
2920                     "Invalid journal mac", "Error setting journal mac key");
2921         if (r)
2922                 goto bad;
2923
2924         if (!ic->tag_size) {
2925                 if (!ic->internal_hash) {
2926                         ti->error = "Unknown tag size";
2927                         r = -EINVAL;
2928                         goto bad;
2929                 }
2930                 ic->tag_size = crypto_shash_digestsize(ic->internal_hash);
2931         }
2932         if (ic->tag_size > MAX_TAG_SIZE) {
2933                 ti->error = "Too big tag size";
2934                 r = -EINVAL;
2935                 goto bad;
2936         }
2937         if (!(ic->tag_size & (ic->tag_size - 1)))
2938                 ic->log2_tag_size = __ffs(ic->tag_size);
2939         else
2940                 ic->log2_tag_size = -1;
2941
2942         ic->autocommit_jiffies = msecs_to_jiffies(sync_msec);
2943         ic->autocommit_msec = sync_msec;
2944         setup_timer(&ic->autocommit_timer, autocommit_fn, (unsigned long)ic);
2945
2946         ic->io = dm_io_client_create();
2947         if (IS_ERR(ic->io)) {
2948                 r = PTR_ERR(ic->io);
2949                 ic->io = NULL;
2950                 ti->error = "Cannot allocate dm io";
2951                 goto bad;
2952         }
2953
2954         ic->journal_io_mempool = mempool_create_slab_pool(JOURNAL_IO_MEMPOOL, journal_io_cache);
2955         if (!ic->journal_io_mempool) {
2956                 r = -ENOMEM;
2957                 ti->error = "Cannot allocate mempool";
2958                 goto bad;
2959         }
2960
2961         ic->metadata_wq = alloc_workqueue("dm-integrity-metadata",
2962                                           WQ_MEM_RECLAIM, METADATA_WORKQUEUE_MAX_ACTIVE);
2963         if (!ic->metadata_wq) {
2964                 ti->error = "Cannot allocate workqueue";
2965                 r = -ENOMEM;
2966                 goto bad;
2967         }
2968
2969         /*
2970          * If this workqueue were percpu, it would cause bio reordering
2971          * and reduced performance.
2972          */
2973         ic->wait_wq = alloc_workqueue("dm-integrity-wait", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
2974         if (!ic->wait_wq) {
2975                 ti->error = "Cannot allocate workqueue";
2976                 r = -ENOMEM;
2977                 goto bad;
2978         }
2979
2980         ic->commit_wq = alloc_workqueue("dm-integrity-commit", WQ_MEM_RECLAIM, 1);
2981         if (!ic->commit_wq) {
2982                 ti->error = "Cannot allocate workqueue";
2983                 r = -ENOMEM;
2984                 goto bad;
2985         }
2986         INIT_WORK(&ic->commit_work, integrity_commit);
2987
2988         if (ic->mode == 'J') {
2989                 ic->writer_wq = alloc_workqueue("dm-integrity-writer", WQ_MEM_RECLAIM, 1);
2990                 if (!ic->writer_wq) {
2991                         ti->error = "Cannot allocate workqueue";
2992                         r = -ENOMEM;
2993                         goto bad;
2994                 }
2995                 INIT_WORK(&ic->writer_work, integrity_writer);
2996         }
2997
2998         ic->sb = alloc_pages_exact(SB_SECTORS << SECTOR_SHIFT, GFP_KERNEL);
2999         if (!ic->sb) {
3000                 r = -ENOMEM;
3001                 ti->error = "Cannot allocate superblock area";
3002                 goto bad;
3003         }
3004
3005         r = sync_rw_sb(ic, REQ_OP_READ, 0);
3006         if (r) {
3007                 ti->error = "Error reading superblock";
3008                 goto bad;
3009         }
3010         should_write_sb = false;
3011         if (memcmp(ic->sb->magic, SB_MAGIC, 8)) {
3012                 if (ic->mode != 'R') {
3013                         if (memchr_inv(ic->sb, 0, SB_SECTORS << SECTOR_SHIFT)) {
3014                                 r = -EINVAL;
3015                                 ti->error = "The device is not initialized";
3016                                 goto bad;
3017                         }
3018                 }
3019
3020                 r = initialize_superblock(ic, journal_sectors, interleave_sectors);
3021                 if (r) {
3022                         ti->error = "Could not initialize superblock";
3023                         goto bad;
3024                 }
3025                 if (ic->mode != 'R')
3026                         should_write_sb = true;
3027         }
3028
3029         if (ic->sb->version != SB_VERSION) {
3030                 r = -EINVAL;
3031                 ti->error = "Unknown version";
3032                 goto bad;
3033         }
3034         if (le16_to_cpu(ic->sb->integrity_tag_size) != ic->tag_size) {
3035                 r = -EINVAL;
3036                 ti->error = "Tag size doesn't match the information in superblock";
3037                 goto bad;
3038         }
3039         if (ic->sb->log2_sectors_per_block != __ffs(ic->sectors_per_block)) {
3040                 r = -EINVAL;
3041                 ti->error = "Block size doesn't match the information in superblock";
3042                 goto bad;
3043         }
3044         /* make sure that ti->max_io_len doesn't overflow */
3045         if (ic->sb->log2_interleave_sectors < MIN_LOG2_INTERLEAVE_SECTORS ||
3046             ic->sb->log2_interleave_sectors > MAX_LOG2_INTERLEAVE_SECTORS) {
3047                 r = -EINVAL;
3048                 ti->error = "Invalid interleave_sectors in the superblock";
3049                 goto bad;
3050         }
3051         ic->provided_data_sectors = le64_to_cpu(ic->sb->provided_data_sectors);
3052         if (ic->provided_data_sectors != le64_to_cpu(ic->sb->provided_data_sectors)) {
3053                 /* test for overflow */
3054                 r = -EINVAL;
3055                 ti->error = "The superblock has 64-bit device size, but the kernel was compiled with 32-bit sectors";
3056                 goto bad;
3057         }
3058         if (!!(ic->sb->flags & cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC)) != !!ic->journal_mac_alg.alg_string) {
3059                 r = -EINVAL;
3060                 ti->error = "Journal mac mismatch";
3061                 goto bad;
3062         }
3063         r = calculate_device_limits(ic);
3064         if (r) {
3065                 ti->error = "The device is too small";
3066                 goto bad;
3067         }
3068
3069         if (!buffer_sectors)
3070                 buffer_sectors = 1;
3071         ic->log2_buffer_sectors = min3((int)__fls(buffer_sectors), (int)__ffs(ic->metadata_run), 31 - SECTOR_SHIFT);
3072
3073         threshold = (__u64)ic->journal_entries * (100 - journal_watermark);
3074         threshold += 50;
3075         do_div(threshold, 100);
3076         ic->free_sectors_threshold = threshold;
3077
3078         DEBUG_print("initialized:\n");
3079         DEBUG_print("   integrity_tag_size %u\n", le16_to_cpu(ic->sb->integrity_tag_size));
3080         DEBUG_print("   journal_entry_size %u\n", ic->journal_entry_size);
3081         DEBUG_print("   journal_entries_per_sector %u\n", ic->journal_entries_per_sector);
3082         DEBUG_print("   journal_section_entries %u\n", ic->journal_section_entries);
3083         DEBUG_print("   journal_section_sectors %u\n", ic->journal_section_sectors);
3084         DEBUG_print("   journal_sections %u\n", (unsigned)le32_to_cpu(ic->sb->journal_sections));
3085         DEBUG_print("   journal_entries %u\n", ic->journal_entries);
3086         DEBUG_print("   log2_interleave_sectors %d\n", ic->sb->log2_interleave_sectors);
3087         DEBUG_print("   device_sectors 0x%llx\n", (unsigned long long)ic->device_sectors);
3088         DEBUG_print("   initial_sectors 0x%x\n", ic->initial_sectors);
3089         DEBUG_print("   metadata_run 0x%x\n", ic->metadata_run);
3090         DEBUG_print("   log2_metadata_run %d\n", ic->log2_metadata_run);
3091         DEBUG_print("   provided_data_sectors 0x%llx (%llu)\n", (unsigned long long)ic->provided_data_sectors,
3092                     (unsigned long long)ic->provided_data_sectors);
3093         DEBUG_print("   log2_buffer_sectors %u\n", ic->log2_buffer_sectors);
3094
3095         ic->bufio = dm_bufio_client_create(ic->dev->bdev, 1U << (SECTOR_SHIFT + ic->log2_buffer_sectors),
3096                                            1, 0, NULL, NULL);
3097         if (IS_ERR(ic->bufio)) {
3098                 r = PTR_ERR(ic->bufio);
3099                 ti->error = "Cannot initialize dm-bufio";
3100                 ic->bufio = NULL;
3101                 goto bad;
3102         }
3103         dm_bufio_set_sector_offset(ic->bufio, ic->start + ic->initial_sectors);
3104
3105         if (ic->mode != 'R') {
3106                 r = create_journal(ic, &ti->error);
3107                 if (r)
3108                         goto bad;
3109         }
3110
3111         if (should_write_sb) {
3112                 int r;
3113
3114                 init_journal(ic, 0, ic->journal_sections, 0);
3115                 r = dm_integrity_failed(ic);
3116                 if (unlikely(r)) {
3117                         ti->error = "Error initializing journal";
3118                         goto bad;
3119                 }
3120                 r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
3121                 if (r) {
3122                         ti->error = "Error initializing superblock";
3123                         goto bad;
3124                 }
3125                 ic->just_formatted = true;
3126         }
3127
3128         r = dm_set_target_max_io_len(ti, 1U << ic->sb->log2_interleave_sectors);
3129         if (r)
3130                 goto bad;
3131
3132         if (!ic->internal_hash)
3133                 dm_integrity_set(ti, ic);
3134
3135         ti->num_flush_bios = 1;
3136         ti->flush_supported = true;
3137
3138         return 0;
3139 bad:
3140         dm_integrity_dtr(ti);
3141         return r;
3142 }
3143
3144 static void dm_integrity_dtr(struct dm_target *ti)
3145 {
3146         struct dm_integrity_c *ic = ti->private;
3147
3148         BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
3149
3150         if (ic->metadata_wq)
3151                 destroy_workqueue(ic->metadata_wq);
3152         if (ic->wait_wq)
3153                 destroy_workqueue(ic->wait_wq);
3154         if (ic->commit_wq)
3155                 destroy_workqueue(ic->commit_wq);
3156         if (ic->writer_wq)
3157                 destroy_workqueue(ic->writer_wq);
3158         if (ic->bufio)
3159                 dm_bufio_client_destroy(ic->bufio);
3160         mempool_destroy(ic->journal_io_mempool);
3161         if (ic->io)
3162                 dm_io_client_destroy(ic->io);
3163         if (ic->dev)
3164                 dm_put_device(ti, ic->dev);
3165         dm_integrity_free_page_list(ic, ic->journal);
3166         dm_integrity_free_page_list(ic, ic->journal_io);
3167         dm_integrity_free_page_list(ic, ic->journal_xor);
3168         if (ic->journal_scatterlist)
3169                 dm_integrity_free_journal_scatterlist(ic, ic->journal_scatterlist);
3170         if (ic->journal_io_scatterlist)
3171                 dm_integrity_free_journal_scatterlist(ic, ic->journal_io_scatterlist);
3172         if (ic->sk_requests) {
3173                 unsigned i;
3174
3175                 for (i = 0; i < ic->journal_sections; i++) {
3176                         struct skcipher_request *req = ic->sk_requests[i];
3177                         if (req) {
3178                                 kzfree(req->iv);
3179                                 skcipher_request_free(req);
3180                         }
3181                 }
3182                 kvfree(ic->sk_requests);
3183         }
3184         kvfree(ic->journal_tree);
3185         if (ic->sb)
3186                 free_pages_exact(ic->sb, SB_SECTORS << SECTOR_SHIFT);
3187
3188         if (ic->internal_hash)
3189                 crypto_free_shash(ic->internal_hash);
3190         free_alg(&ic->internal_hash_alg);
3191
3192         if (ic->journal_crypt)
3193                 crypto_free_skcipher(ic->journal_crypt);
3194         free_alg(&ic->journal_crypt_alg);
3195
3196         if (ic->journal_mac)
3197                 crypto_free_shash(ic->journal_mac);
3198         free_alg(&ic->journal_mac_alg);
3199
3200         kfree(ic);
3201 }
3202
3203 static struct target_type integrity_target = {
3204         .name                   = "integrity",
3205         .version                = {1, 0, 0},
3206         .module                 = THIS_MODULE,
3207         .features               = DM_TARGET_SINGLETON | DM_TARGET_INTEGRITY,
3208         .ctr                    = dm_integrity_ctr,
3209         .dtr                    = dm_integrity_dtr,
3210         .map                    = dm_integrity_map,
3211         .postsuspend            = dm_integrity_postsuspend,
3212         .resume                 = dm_integrity_resume,
3213         .status                 = dm_integrity_status,
3214         .iterate_devices        = dm_integrity_iterate_devices,
3215         .io_hints               = dm_integrity_io_hints,
3216 };
3217
3218 int __init dm_integrity_init(void)
3219 {
3220         int r;
3221
3222         journal_io_cache = kmem_cache_create("integrity_journal_io",
3223                                              sizeof(struct journal_io), 0, 0, NULL);
3224         if (!journal_io_cache) {
3225                 DMERR("can't allocate journal io cache");
3226                 return -ENOMEM;
3227         }
3228
3229         r = dm_register_target(&integrity_target);
3230
3231         if (r < 0)
3232                 DMERR("register failed %d", r);
3233
3234         return r;
3235 }
3236
3237 void dm_integrity_exit(void)
3238 {
3239         dm_unregister_target(&integrity_target);
3240         kmem_cache_destroy(journal_io_cache);
3241 }
3242
3243 module_init(dm_integrity_init);
3244 module_exit(dm_integrity_exit);
3245
3246 MODULE_AUTHOR("Milan Broz");
3247 MODULE_AUTHOR("Mikulas Patocka");
3248 MODULE_DESCRIPTION(DM_NAME " target for integrity tags extension");
3249 MODULE_LICENSE("GPL");