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/compiler.h>
  10 #include <linux/module.h>
  11 #include <linux/device-mapper.h>
  12 #include <linux/dm-io.h>
  13 #include <linux/vmalloc.h>
  14 #include <linux/sort.h>
  15 #include <linux/rbtree.h>
  16 #include <linux/delay.h>
  17 #include <linux/random.h>
  18 #include <linux/reboot.h>
  19 #include <crypto/hash.h>
  20 #include <crypto/skcipher.h>
  21 #include <linux/async_tx.h>
  22 #include <linux/dm-bufio.h>
  23
  24 #define DM_MSG_PREFIX "integrity"
  25
  26 #define DEFAULT_INTERLEAVE_SECTORS      32768
  27 #define DEFAULT_JOURNAL_SIZE_FACTOR     7
  28 #define DEFAULT_SECTORS_PER_BITMAP_BIT  32768
  29 #define DEFAULT_BUFFER_SECTORS          128
  30 #define DEFAULT_JOURNAL_WATERMARK       50
  31 #define DEFAULT_SYNC_MSEC               10000
  32 #define DEFAULT_MAX_JOURNAL_SECTORS     131072
  33 #define MIN_LOG2_INTERLEAVE_SECTORS     3
  34 #define MAX_LOG2_INTERLEAVE_SECTORS     31
  35 #define METADATA_WORKQUEUE_MAX_ACTIVE   16
  36 #define RECALC_SECTORS                  8192
  37 #define RECALC_WRITE_SUPER              16
  38 #define BITMAP_BLOCK_SIZE               4096    /* don't change it */
  39 #define BITMAP_FLUSH_INTERVAL           (10 * HZ)
  40
  41 /*
  42  * Warning - DEBUG_PRINT prints security-sensitive data to the log,
  43  * so it should not be enabled in the official kernel
  44  */
  45 //#define DEBUG_PRINT
  46 //#define INTERNAL_VERIFY
  47
  48 /*
  49  * On disk structures
  50  */
  51
  52 #define SB_MAGIC                        "integrt"
  53 #define SB_VERSION_1                    1
  54 #define SB_VERSION_2                    2
  55 #define SB_VERSION_3                    3
  56 #define SB_SECTORS                      8
  57 #define MAX_SECTORS_PER_BLOCK           8
  58
  59 struct superblock {
  60         __u8 magic[8];
  61         __u8 version;
  62         __u8 log2_interleave_sectors;
  63         __u16 integrity_tag_size;
  64         __u32 journal_sections;
  65         __u64 provided_data_sectors;    /* userspace uses this value */
  66         __u32 flags;
  67         __u8 log2_sectors_per_block;
  68         __u8 log2_blocks_per_bitmap_bit;
  69         __u8 pad[2];
  70         __u64 recalc_sector;
  71 };
  72
  73 #define SB_FLAG_HAVE_JOURNAL_MAC        0x1
  74 #define SB_FLAG_RECALCULATING           0x2
  75 #define SB_FLAG_DIRTY_BITMAP            0x4
  76
  77 #define JOURNAL_ENTRY_ROUNDUP           8
  78
  79 typedef __u64 commit_id_t;
  80 #define JOURNAL_MAC_PER_SECTOR          8
  81
  82 struct journal_entry {
  83         union {
  84                 struct {
  85                         __u32 sector_lo;
  86                         __u32 sector_hi;
  87                 } s;
  88                 __u64 sector;
  89         } u;
  90         commit_id_t last_bytes[0];
  91         /* __u8 tag[0]; */
  92 };
  93
  94 #define journal_entry_tag(ic, je)               ((__u8 *)&(je)->last_bytes[(ic)->sectors_per_block])
  95
  96 #if BITS_PER_LONG == 64
  97 #define journal_entry_set_sector(je, x)         do { smp_wmb(); WRITE_ONCE((je)->u.sector, cpu_to_le64(x)); } while (0)
  98 #define journal_entry_get_sector(je)            le64_to_cpu((je)->u.sector)
  99 #elif defined(CONFIG_LBDAF)
 100 #define journal_entry_set_sector(je, x)         do { (je)->u.s.sector_lo = cpu_to_le32(x); smp_wmb(); WRITE_ONCE((je)->u.s.sector_hi, cpu_to_le32((x) >> 32)); } while (0)
 101 #define journal_entry_get_sector(je)            le64_to_cpu((je)->u.sector)
 102 #else
 103 #define journal_entry_set_sector(je, x)         do { (je)->u.s.sector_lo = cpu_to_le32(x); smp_wmb(); WRITE_ONCE((je)->u.s.sector_hi, cpu_to_le32(0)); } while (0)
 104 #define journal_entry_get_sector(je)            le32_to_cpu((je)->u.s.sector_lo)
 105 #endif
 106 #define journal_entry_is_unused(je)             ((je)->u.s.sector_hi == cpu_to_le32(-1))
 107 #define journal_entry_set_unused(je)            do { ((je)->u.s.sector_hi = cpu_to_le32(-1)); } while (0)
 108 #define journal_entry_is_inprogress(je)         ((je)->u.s.sector_hi == cpu_to_le32(-2))
 109 #define journal_entry_set_inprogress(je)        do { ((je)->u.s.sector_hi = cpu_to_le32(-2)); } while (0)
 110
 111 #define JOURNAL_BLOCK_SECTORS           8
 112 #define JOURNAL_SECTOR_DATA             ((1 << SECTOR_SHIFT) - sizeof(commit_id_t))
 113 #define JOURNAL_MAC_SIZE                (JOURNAL_MAC_PER_SECTOR * JOURNAL_BLOCK_SECTORS)
 114
 115 struct journal_sector {
 116         __u8 entries[JOURNAL_SECTOR_DATA - JOURNAL_MAC_PER_SECTOR];
 117         __u8 mac[JOURNAL_MAC_PER_SECTOR];
 118         commit_id_t commit_id;
 119 };
 120
 121 #define MAX_TAG_SIZE                    (JOURNAL_SECTOR_DATA - JOURNAL_MAC_PER_SECTOR - offsetof(struct journal_entry, last_bytes[MAX_SECTORS_PER_BLOCK]))
 122
 123 #define METADATA_PADDING_SECTORS        8
 124
 125 #define N_COMMIT_IDS                    4
 126
 127 static unsigned char prev_commit_seq(unsigned char seq)
 128 {
 129         return (seq + N_COMMIT_IDS - 1) % N_COMMIT_IDS;
 130 }
 131
 132 static unsigned char next_commit_seq(unsigned char seq)
 133 {
 134         return (seq + 1) % N_COMMIT_IDS;
 135 }
 136
 137 /*
 138  * In-memory structures
 139  */
 140
 141 struct journal_node {
 142         struct rb_node node;
 143         sector_t sector;
 144 };
 145
 146 struct alg_spec {
 147         char *alg_string;
 148         char *key_string;
 149         __u8 *key;
 150         unsigned key_size;
 151 };
 152
 153 struct dm_integrity_c {
 154         struct dm_dev *dev;
 155         struct dm_dev *meta_dev;
 156         unsigned tag_size;
 157         __s8 log2_tag_size;
 158         sector_t start;
 159         mempool_t journal_io_mempool;
 160         struct dm_io_client *io;
 161         struct dm_bufio_client *bufio;
 162         struct workqueue_struct *metadata_wq;
 163         struct superblock *sb;
 164         unsigned journal_pages;
 165         unsigned n_bitmap_blocks;
 166
 167         struct page_list *journal;
 168         struct page_list *journal_io;
 169         struct page_list *journal_xor;
 170         struct page_list *recalc_bitmap;
 171         struct page_list *may_write_bitmap;
 172         struct bitmap_block_status *bbs;
 173         unsigned bitmap_flush_interval;
 174         int synchronous_mode;
 175         struct bio_list synchronous_bios;
 176         struct delayed_work bitmap_flush_work;
 177
 178         struct crypto_skcipher *journal_crypt;
 179         struct scatterlist **journal_scatterlist;
 180         struct scatterlist **journal_io_scatterlist;
 181         struct skcipher_request **sk_requests;
 182
 183         struct crypto_shash *journal_mac;
 184
 185         struct journal_node *journal_tree;
 186         struct rb_root journal_tree_root;
 187
 188         sector_t provided_data_sectors;
 189
 190         unsigned short journal_entry_size;
 191         unsigned char journal_entries_per_sector;
 192         unsigned char journal_section_entries;
 193         unsigned short journal_section_sectors;
 194         unsigned journal_sections;
 195         unsigned journal_entries;
 196         sector_t data_device_sectors;
 197         sector_t meta_device_sectors;
 198         unsigned initial_sectors;
 199         unsigned metadata_run;
 200         __s8 log2_metadata_run;
 201         __u8 log2_buffer_sectors;
 202         __u8 sectors_per_block;
 203         __u8 log2_blocks_per_bitmap_bit;
 204
 205         unsigned char mode;
 206         int suspending;
 207
 208         int failed;
 209
 210         struct crypto_shash *internal_hash;
 211
 212         /* these variables are locked with endio_wait.lock */
 213         struct rb_root in_progress;
 214         struct list_head wait_list;
 215         wait_queue_head_t endio_wait;
 216         struct workqueue_struct *wait_wq;
 217
 218         unsigned char commit_seq;
 219         commit_id_t commit_ids[N_COMMIT_IDS];
 220
 221         unsigned committed_section;
 222         unsigned n_committed_sections;
 223
 224         unsigned uncommitted_section;
 225         unsigned n_uncommitted_sections;
 226
 227         unsigned free_section;
 228         unsigned char free_section_entry;
 229         unsigned free_sectors;
 230
 231         unsigned free_sectors_threshold;
 232
 233         struct workqueue_struct *commit_wq;
 234         struct work_struct commit_work;
 235
 236         struct workqueue_struct *writer_wq;
 237         struct work_struct writer_work;
 238
 239         struct workqueue_struct *recalc_wq;
 240         struct work_struct recalc_work;
 241         u8 *recalc_buffer;
 242         u8 *recalc_tags;
 243
 244         struct bio_list flush_bio_list;
 245
 246         unsigned long autocommit_jiffies;
 247         struct timer_list autocommit_timer;
 248         unsigned autocommit_msec;
 249
 250         wait_queue_head_t copy_to_journal_wait;
 251
 252         struct completion crypto_backoff;
 253
 254         bool journal_uptodate;
 255         bool just_formatted;
 256         bool recalculate_flag;
 257
 258         struct alg_spec internal_hash_alg;
 259         struct alg_spec journal_crypt_alg;
 260         struct alg_spec journal_mac_alg;
 261
 262         atomic64_t number_of_mismatches;
 263
 264         struct notifier_block reboot_notifier;
 265 };
 266
 267 struct dm_integrity_range {
 268         sector_t logical_sector;
 269         sector_t n_sectors;
 270         bool waiting;
 271         union {
 272                 struct rb_node node;
 273                 struct {
 274                         struct task_struct *task;
 275                         struct list_head wait_entry;
 276                 };
 277         };
 278 };
 279
 280 struct dm_integrity_io {
 281         struct work_struct work;
 282
 283         struct dm_integrity_c *ic;
 284         bool write;
 285         bool fua;
 286
 287         struct dm_integrity_range range;
 288
 289         sector_t metadata_block;
 290         unsigned metadata_offset;
 291
 292         atomic_t in_flight;
 293         blk_status_t bi_status;
 294
 295         struct completion *completion;
 296
 297         struct gendisk *orig_bi_disk;
 298         u8 orig_bi_partno;
 299         bio_end_io_t *orig_bi_end_io;
 300         struct bio_integrity_payload *orig_bi_integrity;
 301         struct bvec_iter orig_bi_iter;
 302 };
 303
 304 struct journal_completion {
 305         struct dm_integrity_c *ic;
 306         atomic_t in_flight;
 307         struct completion comp;
 308 };
 309
 310 struct journal_io {
 311         struct dm_integrity_range range;
 312         struct journal_completion *comp;
 313 };
 314
 315 struct bitmap_block_status {
 316         struct work_struct work;
 317         struct dm_integrity_c *ic;
 318         unsigned idx;
 319         unsigned long *bitmap;
 320         struct bio_list bio_queue;
 321         spinlock_t bio_queue_lock;
 322
 323 };
 324
 325 static struct kmem_cache *journal_io_cache;
 326
 327 #define JOURNAL_IO_MEMPOOL      32
 328
 329 #ifdef DEBUG_PRINT
 330 #define DEBUG_print(x, ...)     printk(KERN_DEBUG x, ##__VA_ARGS__)
 331 static void __DEBUG_bytes(__u8 *bytes, size_t len, const char *msg, ...)
 332 {
 333         va_list args;
 334         va_start(args, msg);
 335         vprintk(msg, args);
 336         va_end(args);
 337         if (len)
 338                 pr_cont(":");
 339         while (len) {
 340                 pr_cont(" %02x", *bytes);
 341                 bytes++;
 342                 len--;
 343         }
 344         pr_cont("\n");
 345 }
 346 #define DEBUG_bytes(bytes, len, msg, ...)       __DEBUG_bytes(bytes, len, KERN_DEBUG msg, ##__VA_ARGS__)
 347 #else
 348 #define DEBUG_print(x, ...)                     do { } while (0)
 349 #define DEBUG_bytes(bytes, len, msg, ...)       do { } while (0)
 350 #endif
 351
 352 /*
 353  * DM Integrity profile, protection is performed layer above (dm-crypt)
 354  */
 355 static const struct blk_integrity_profile dm_integrity_profile = {
 356         .name                   = "DM-DIF-EXT-TAG",
 357         .generate_fn            = NULL,
 358         .verify_fn              = NULL,
 359 };
 360
 361 static void dm_integrity_map_continue(struct dm_integrity_io *dio, bool from_map);
 362 static void integrity_bio_wait(struct work_struct *w);
 363 static void dm_integrity_dtr(struct dm_target *ti);
 364
 365 static void dm_integrity_io_error(struct dm_integrity_c *ic, const char *msg, int err)
 366 {
 367         if (err == -EILSEQ)
 368                 atomic64_inc(&ic->number_of_mismatches);
 369         if (!cmpxchg(&ic->failed, 0, err))
 370                 DMERR("Error on %s: %d", msg, err);
 371 }
 372
 373 static int dm_integrity_failed(struct dm_integrity_c *ic)
 374 {
 375         return READ_ONCE(ic->failed);
 376 }
 377
 378 static commit_id_t dm_integrity_commit_id(struct dm_integrity_c *ic, unsigned i,
 379                                           unsigned j, unsigned char seq)
 380 {
 381         /*
 382          * Xor the number with section and sector, so that if a piece of
 383          * journal is written at wrong place, it is detected.
 384          */
 385         return ic->commit_ids[seq] ^ cpu_to_le64(((__u64)i << 32) ^ j);
 386 }
 387
 388 static void get_area_and_offset(struct dm_integrity_c *ic, sector_t data_sector,
 389                                 sector_t *area, sector_t *offset)
 390 {
 391         if (!ic->meta_dev) {
 392                 __u8 log2_interleave_sectors = ic->sb->log2_interleave_sectors;
 393                 *area = data_sector >> log2_interleave_sectors;
 394                 *offset = (unsigned)data_sector & ((1U << log2_interleave_sectors) - 1);
 395         } else {
 396                 *area = 0;
 397                 *offset = data_sector;
 398         }
 399 }
 400
 401 #define sector_to_block(ic, n)                                          \
 402 do {                                                                    \
 403         BUG_ON((n) & (unsigned)((ic)->sectors_per_block - 1));          \
 404         (n) >>= (ic)->sb->log2_sectors_per_block;                       \
 405 } while (0)
 406
 407 static __u64 get_metadata_sector_and_offset(struct dm_integrity_c *ic, sector_t area,
 408                                             sector_t offset, unsigned *metadata_offset)
 409 {
 410         __u64 ms;
 411         unsigned mo;
 412
 413         ms = area << ic->sb->log2_interleave_sectors;
 414         if (likely(ic->log2_metadata_run >= 0))
 415                 ms += area << ic->log2_metadata_run;
 416         else
 417                 ms += area * ic->metadata_run;
 418         ms >>= ic->log2_buffer_sectors;
 419
 420         sector_to_block(ic, offset);
 421
 422         if (likely(ic->log2_tag_size >= 0)) {
 423                 ms += offset >> (SECTOR_SHIFT + ic->log2_buffer_sectors - ic->log2_tag_size);
 424                 mo = (offset << ic->log2_tag_size) & ((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - 1);
 425         } else {
 426                 ms += (__u64)offset * ic->tag_size >> (SECTOR_SHIFT + ic->log2_buffer_sectors);
 427                 mo = (offset * ic->tag_size) & ((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - 1);
 428         }
 429         *metadata_offset = mo;
 430         return ms;
 431 }
 432
 433 static sector_t get_data_sector(struct dm_integrity_c *ic, sector_t area, sector_t offset)
 434 {
 435         sector_t result;
 436
 437         if (ic->meta_dev)
 438                 return offset;
 439
 440         result = area << ic->sb->log2_interleave_sectors;
 441         if (likely(ic->log2_metadata_run >= 0))
 442                 result += (area + 1) << ic->log2_metadata_run;
 443         else
 444                 result += (area + 1) * ic->metadata_run;
 445
 446         result += (sector_t)ic->initial_sectors + offset;
 447         result += ic->start;
 448
 449         return result;
 450 }
 451
 452 static void wraparound_section(struct dm_integrity_c *ic, unsigned *sec_ptr)
 453 {
 454         if (unlikely(*sec_ptr >= ic->journal_sections))
 455                 *sec_ptr -= ic->journal_sections;
 456 }
 457
 458 static void sb_set_version(struct dm_integrity_c *ic)
 459 {
 460         if (ic->mode == 'B' || ic->sb->flags & cpu_to_le32(SB_FLAG_DIRTY_BITMAP))
 461                 ic->sb->version = SB_VERSION_3;
 462         else if (ic->meta_dev || ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
 463                 ic->sb->version = SB_VERSION_2;
 464         else
 465                 ic->sb->version = SB_VERSION_1;
 466 }
 467
 468 static int sync_rw_sb(struct dm_integrity_c *ic, int op, int op_flags)
 469 {
 470         struct dm_io_request io_req;
 471         struct dm_io_region io_loc;
 472
 473         io_req.bi_op = op;
 474         io_req.bi_op_flags = op_flags;
 475         io_req.mem.type = DM_IO_KMEM;
 476         io_req.mem.ptr.addr = ic->sb;
 477         io_req.notify.fn = NULL;
 478         io_req.client = ic->io;
 479         io_loc.bdev = ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev;
 480         io_loc.sector = ic->start;
 481         io_loc.count = SB_SECTORS;
 482
 483         return dm_io(&io_req, 1, &io_loc, NULL);
 484 }
 485
 486 #define BITMAP_OP_TEST_ALL_SET          0
 487 #define BITMAP_OP_TEST_ALL_CLEAR        1
 488 #define BITMAP_OP_SET                   2
 489 #define BITMAP_OP_CLEAR                 3
 490
 491 static bool block_bitmap_op(struct dm_integrity_c *ic, struct page_list *bitmap, sector_t sector, sector_t n_sectors, int mode)
 492 {
 493         unsigned long bit, end_bit, this_end_bit, page, end_page;
 494         unsigned long *data;
 495
 496         if (unlikely(((sector | n_sectors) & ((1 << ic->sb->log2_sectors_per_block) - 1)) != 0)) {
 497                 DMCRIT("invalid bitmap access (%llx,%llx,%d,%d,%d)\n",
 498                         (unsigned long long)sector,
 499                         (unsigned long long)n_sectors,
 500                         ic->sb->log2_sectors_per_block,
 501                         ic->log2_blocks_per_bitmap_bit,
 502                         mode);
 503                 BUG();
 504         }
 505
 506         if (unlikely(!n_sectors))
 507                 return true;
 508
 509         bit = sector >> (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
 510         end_bit = (sector + n_sectors - 1) >> (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
 511
 512         page = bit / (PAGE_SIZE * 8);
 513         bit %= PAGE_SIZE * 8;
 514
 515         end_page = end_bit / (PAGE_SIZE * 8);
 516         end_bit %= PAGE_SIZE * 8;
 517
 518 repeat:
 519         if (page < end_page) {
 520                 this_end_bit = PAGE_SIZE * 8 - 1;
 521         } else {
 522                 this_end_bit = end_bit;
 523         }
 524
 525         data = lowmem_page_address(bitmap[page].page);
 526
 527         if (mode == BITMAP_OP_TEST_ALL_SET) {
 528                 while (bit <= this_end_bit) {
 529                         if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
 530                                 do {
 531                                         if (data[bit / BITS_PER_LONG] != -1)
 532                                                 return false;
 533                                         bit += BITS_PER_LONG;
 534                                 } while (this_end_bit >= bit + BITS_PER_LONG - 1);
 535                                 continue;
 536                         }
 537                         if (!test_bit(bit, data))
 538                                 return false;
 539                         bit++;
 540                 }
 541         } else if (mode == BITMAP_OP_TEST_ALL_CLEAR) {
 542                 while (bit <= this_end_bit) {
 543                         if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
 544                                 do {
 545                                         if (data[bit / BITS_PER_LONG] != 0)
 546                                                 return false;
 547                                         bit += BITS_PER_LONG;
 548                                 } while (this_end_bit >= bit + BITS_PER_LONG - 1);
 549                                 continue;
 550                         }
 551                         if (test_bit(bit, data))
 552                                 return false;
 553                         bit++;
 554                 }
 555         } else if (mode == BITMAP_OP_SET) {
 556                 while (bit <= this_end_bit) {
 557                         if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
 558                                 do {
 559                                         data[bit / BITS_PER_LONG] = -1;
 560                                         bit += BITS_PER_LONG;
 561                                 } while (this_end_bit >= bit + BITS_PER_LONG - 1);
 562                                 continue;
 563                         }
 564                         __set_bit(bit, data);
 565                         bit++;
 566                 }
 567         } else if (mode == BITMAP_OP_CLEAR) {
 568                 if (!bit && this_end_bit == PAGE_SIZE * 8 - 1)
 569                         clear_page(data);
 570                 else while (bit <= this_end_bit) {
 571                         if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
 572                                 do {
 573                                         data[bit / BITS_PER_LONG] = 0;
 574                                         bit += BITS_PER_LONG;
 575                                 } while (this_end_bit >= bit + BITS_PER_LONG - 1);
 576                                 continue;
 577                         }
 578                         __clear_bit(bit, data);
 579                         bit++;
 580                 }
 581         } else {
 582                 BUG();
 583         }
 584
 585         if (unlikely(page < end_page)) {
 586                 bit = 0;
 587                 page++;
 588                 goto repeat;
 589         }
 590
 591         return true;
 592 }
 593
 594 static void block_bitmap_copy(struct dm_integrity_c *ic, struct page_list *dst, struct page_list *src)
 595 {
 596         unsigned n_bitmap_pages = DIV_ROUND_UP(ic->n_bitmap_blocks, PAGE_SIZE / BITMAP_BLOCK_SIZE);
 597         unsigned i;
 598
 599         for (i = 0; i < n_bitmap_pages; i++) {
 600                 unsigned long *dst_data = lowmem_page_address(dst[i].page);
 601                 unsigned long *src_data = lowmem_page_address(src[i].page);
 602                 copy_page(dst_data, src_data);
 603         }
 604 }
 605
 606 static struct bitmap_block_status *sector_to_bitmap_block(struct dm_integrity_c *ic, sector_t sector)
 607 {
 608         unsigned bit = sector >> (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
 609         unsigned bitmap_block = bit / (BITMAP_BLOCK_SIZE * 8);
 610
 611         BUG_ON(bitmap_block >= ic->n_bitmap_blocks);
 612         return &ic->bbs[bitmap_block];
 613 }
 614
 615 static void access_journal_check(struct dm_integrity_c *ic, unsigned section, unsigned offset,
 616                                  bool e, const char *function)
 617 {
 618 #if defined(CONFIG_DM_DEBUG) || defined(INTERNAL_VERIFY)
 619         unsigned limit = e ? ic->journal_section_entries : ic->journal_section_sectors;
 620
 621         if (unlikely(section >= ic->journal_sections) ||
 622             unlikely(offset >= limit)) {
 623                 printk(KERN_CRIT "%s: invalid access at (%u,%u), limit (%u,%u)\n",
 624                         function, section, offset, ic->journal_sections, limit);
 625                 BUG();
 626         }
 627 #endif
 628 }
 629
 630 static void page_list_location(struct dm_integrity_c *ic, unsigned section, unsigned offset,
 631                                unsigned *pl_index, unsigned *pl_offset)
 632 {
 633         unsigned sector;
 634
 635         access_journal_check(ic, section, offset, false, "page_list_location");
 636
 637         sector = section * ic->journal_section_sectors + offset;
 638
 639         *pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
 640         *pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
 641 }
 642
 643 static struct journal_sector *access_page_list(struct dm_integrity_c *ic, struct page_list *pl,
 644                                                unsigned section, unsigned offset, unsigned *n_sectors)
 645 {
 646         unsigned pl_index, pl_offset;
 647         char *va;
 648
 649         page_list_location(ic, section, offset, &pl_index, &pl_offset);
 650
 651         if (n_sectors)
 652                 *n_sectors = (PAGE_SIZE - pl_offset) >> SECTOR_SHIFT;
 653
 654         va = lowmem_page_address(pl[pl_index].page);
 655
 656         return (struct journal_sector *)(va + pl_offset);
 657 }
 658
 659 static struct journal_sector *access_journal(struct dm_integrity_c *ic, unsigned section, unsigned offset)
 660 {
 661         return access_page_list(ic, ic->journal, section, offset, NULL);
 662 }
 663
 664 static struct journal_entry *access_journal_entry(struct dm_integrity_c *ic, unsigned section, unsigned n)
 665 {
 666         unsigned rel_sector, offset;
 667         struct journal_sector *js;
 668
 669         access_journal_check(ic, section, n, true, "access_journal_entry");
 670
 671         rel_sector = n % JOURNAL_BLOCK_SECTORS;
 672         offset = n / JOURNAL_BLOCK_SECTORS;
 673
 674         js = access_journal(ic, section, rel_sector);
 675         return (struct journal_entry *)((char *)js + offset * ic->journal_entry_size);
 676 }
 677
 678 static struct journal_sector *access_journal_data(struct dm_integrity_c *ic, unsigned section, unsigned n)
 679 {
 680         n <<= ic->sb->log2_sectors_per_block;
 681
 682         n += JOURNAL_BLOCK_SECTORS;
 683
 684         access_journal_check(ic, section, n, false, "access_journal_data");
 685
 686         return access_journal(ic, section, n);
 687 }
 688
 689 static void section_mac(struct dm_integrity_c *ic, unsigned section, __u8 result[JOURNAL_MAC_SIZE])
 690 {
 691         SHASH_DESC_ON_STACK(desc, ic->journal_mac);
 692         int r;
 693         unsigned j, size;
 694
 695         desc->tfm = ic->journal_mac;
 696         desc->flags = 0;
 697
 698         r = crypto_shash_init(desc);
 699         if (unlikely(r)) {
 700                 dm_integrity_io_error(ic, "crypto_shash_init", r);
 701                 goto err;
 702         }
 703
 704         for (j = 0; j < ic->journal_section_entries; j++) {
 705                 struct journal_entry *je = access_journal_entry(ic, section, j);
 706                 r = crypto_shash_update(desc, (__u8 *)&je->u.sector, sizeof je->u.sector);
 707                 if (unlikely(r)) {
 708                         dm_integrity_io_error(ic, "crypto_shash_update", r);
 709                         goto err;
 710                 }
 711         }
 712
 713         size = crypto_shash_digestsize(ic->journal_mac);
 714
 715         if (likely(size <= JOURNAL_MAC_SIZE)) {
 716                 r = crypto_shash_final(desc, result);
 717                 if (unlikely(r)) {
 718                         dm_integrity_io_error(ic, "crypto_shash_final", r);
 719                         goto err;
 720                 }
 721                 memset(result + size, 0, JOURNAL_MAC_SIZE - size);
 722         } else {
 723                 __u8 digest[HASH_MAX_DIGESTSIZE];
 724
 725                 if (WARN_ON(size > sizeof(digest))) {
 726                         dm_integrity_io_error(ic, "digest_size", -EINVAL);
 727                         goto err;
 728                 }
 729                 r = crypto_shash_final(desc, digest);
 730                 if (unlikely(r)) {
 731                         dm_integrity_io_error(ic, "crypto_shash_final", r);
 732                         goto err;
 733                 }
 734                 memcpy(result, digest, JOURNAL_MAC_SIZE);
 735         }
 736
 737         return;
 738 err:
 739         memset(result, 0, JOURNAL_MAC_SIZE);
 740 }
 741
 742 static void rw_section_mac(struct dm_integrity_c *ic, unsigned section, bool wr)
 743 {
 744         __u8 result[JOURNAL_MAC_SIZE];
 745         unsigned j;
 746
 747         if (!ic->journal_mac)
 748                 return;
 749
 750         section_mac(ic, section, result);
 751
 752         for (j = 0; j < JOURNAL_BLOCK_SECTORS; j++) {
 753                 struct journal_sector *js = access_journal(ic, section, j);
 754
 755                 if (likely(wr))
 756                         memcpy(&js->mac, result + (j * JOURNAL_MAC_PER_SECTOR), JOURNAL_MAC_PER_SECTOR);
 757                 else {
 758                         if (memcmp(&js->mac, result + (j * JOURNAL_MAC_PER_SECTOR), JOURNAL_MAC_PER_SECTOR))
 759                                 dm_integrity_io_error(ic, "journal mac", -EILSEQ);
 760                 }
 761         }
 762 }
 763
 764 static void complete_journal_op(void *context)
 765 {
 766         struct journal_completion *comp = context;
 767         BUG_ON(!atomic_read(&comp->in_flight));
 768         if (likely(atomic_dec_and_test(&comp->in_flight)))
 769                 complete(&comp->comp);
 770 }
 771
 772 static void xor_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
 773                         unsigned n_sections, struct journal_completion *comp)
 774 {
 775         struct async_submit_ctl submit;
 776         size_t n_bytes = (size_t)(n_sections * ic->journal_section_sectors) << SECTOR_SHIFT;
 777         unsigned pl_index, pl_offset, section_index;
 778         struct page_list *source_pl, *target_pl;
 779
 780         if (likely(encrypt)) {
 781                 source_pl = ic->journal;
 782                 target_pl = ic->journal_io;
 783         } else {
 784                 source_pl = ic->journal_io;
 785                 target_pl = ic->journal;
 786         }
 787
 788         page_list_location(ic, section, 0, &pl_index, &pl_offset);
 789
 790         atomic_add(roundup(pl_offset + n_bytes, PAGE_SIZE) >> PAGE_SHIFT, &comp->in_flight);
 791
 792         init_async_submit(&submit, ASYNC_TX_XOR_ZERO_DST, NULL, complete_journal_op, comp, NULL);
 793
 794         section_index = pl_index;
 795
 796         do {
 797                 size_t this_step;
 798                 struct page *src_pages[2];
 799                 struct page *dst_page;
 800
 801                 while (unlikely(pl_index == section_index)) {
 802                         unsigned dummy;
 803                         if (likely(encrypt))
 804                                 rw_section_mac(ic, section, true);
 805                         section++;
 806                         n_sections--;
 807                         if (!n_sections)
 808                                 break;
 809                         page_list_location(ic, section, 0, &section_index, &dummy);
 810                 }
 811
 812                 this_step = min(n_bytes, (size_t)PAGE_SIZE - pl_offset);
 813                 dst_page = target_pl[pl_index].page;
 814                 src_pages[0] = source_pl[pl_index].page;
 815                 src_pages[1] = ic->journal_xor[pl_index].page;
 816
 817                 async_xor(dst_page, src_pages, pl_offset, 2, this_step, &submit);
 818
 819                 pl_index++;
 820                 pl_offset = 0;
 821                 n_bytes -= this_step;
 822         } while (n_bytes);
 823
 824         BUG_ON(n_sections);
 825
 826         async_tx_issue_pending_all();
 827 }
 828
 829 static void complete_journal_encrypt(struct crypto_async_request *req, int err)
 830 {
 831         struct journal_completion *comp = req->data;
 832         if (unlikely(err)) {
 833                 if (likely(err == -EINPROGRESS)) {
 834                         complete(&comp->ic->crypto_backoff);
 835                         return;
 836                 }
 837                 dm_integrity_io_error(comp->ic, "asynchronous encrypt", err);
 838         }
 839         complete_journal_op(comp);
 840 }
 841
 842 static bool do_crypt(bool encrypt, struct skcipher_request *req, struct journal_completion *comp)
 843 {
 844         int r;
 845         skcipher_request_set_callback(req, CRYPTO_TFM_REQ_MAY_BACKLOG,
 846                                       complete_journal_encrypt, comp);
 847         if (likely(encrypt))
 848                 r = crypto_skcipher_encrypt(req);
 849         else
 850                 r = crypto_skcipher_decrypt(req);
 851         if (likely(!r))
 852                 return false;
 853         if (likely(r == -EINPROGRESS))
 854                 return true;
 855         if (likely(r == -EBUSY)) {
 856                 wait_for_completion(&comp->ic->crypto_backoff);
 857                 reinit_completion(&comp->ic->crypto_backoff);
 858                 return true;
 859         }
 860         dm_integrity_io_error(comp->ic, "encrypt", r);
 861         return false;
 862 }
 863
 864 static void crypt_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
 865                           unsigned n_sections, struct journal_completion *comp)
 866 {
 867         struct scatterlist **source_sg;
 868         struct scatterlist **target_sg;
 869
 870         atomic_add(2, &comp->in_flight);
 871
 872         if (likely(encrypt)) {
 873                 source_sg = ic->journal_scatterlist;
 874                 target_sg = ic->journal_io_scatterlist;
 875         } else {
 876                 source_sg = ic->journal_io_scatterlist;
 877                 target_sg = ic->journal_scatterlist;
 878         }
 879
 880         do {
 881                 struct skcipher_request *req;
 882                 unsigned ivsize;
 883                 char *iv;
 884
 885                 if (likely(encrypt))
 886                         rw_section_mac(ic, section, true);
 887
 888                 req = ic->sk_requests[section];
 889                 ivsize = crypto_skcipher_ivsize(ic->journal_crypt);
 890                 iv = req->iv;
 891
 892                 memcpy(iv, iv + ivsize, ivsize);
 893
 894                 req->src = source_sg[section];
 895                 req->dst = target_sg[section];
 896
 897                 if (unlikely(do_crypt(encrypt, req, comp)))
 898                         atomic_inc(&comp->in_flight);
 899
 900                 section++;
 901                 n_sections--;
 902         } while (n_sections);
 903
 904         atomic_dec(&comp->in_flight);
 905         complete_journal_op(comp);
 906 }
 907
 908 static void encrypt_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
 909                             unsigned n_sections, struct journal_completion *comp)
 910 {
 911         if (ic->journal_xor)
 912                 return xor_journal(ic, encrypt, section, n_sections, comp);
 913         else
 914                 return crypt_journal(ic, encrypt, section, n_sections, comp);
 915 }
 916
 917 static void complete_journal_io(unsigned long error, void *context)
 918 {
 919         struct journal_completion *comp = context;
 920         if (unlikely(error != 0))
 921                 dm_integrity_io_error(comp->ic, "writing journal", -EIO);
 922         complete_journal_op(comp);
 923 }
 924
 925 static void rw_journal_sectors(struct dm_integrity_c *ic, int op, int op_flags,
 926                                unsigned sector, unsigned n_sectors, struct journal_completion *comp)
 927 {
 928         struct dm_io_request io_req;
 929         struct dm_io_region io_loc;
 930         unsigned pl_index, pl_offset;
 931         int r;
 932
 933         if (unlikely(dm_integrity_failed(ic))) {
 934                 if (comp)
 935                         complete_journal_io(-1UL, comp);
 936                 return;
 937         }
 938
 939         pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
 940         pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
 941
 942         io_req.bi_op = op;
 943         io_req.bi_op_flags = op_flags;
 944         io_req.mem.type = DM_IO_PAGE_LIST;
 945         if (ic->journal_io)
 946                 io_req.mem.ptr.pl = &ic->journal_io[pl_index];
 947         else
 948                 io_req.mem.ptr.pl = &ic->journal[pl_index];
 949         io_req.mem.offset = pl_offset;
 950         if (likely(comp != NULL)) {
 951                 io_req.notify.fn = complete_journal_io;
 952                 io_req.notify.context = comp;
 953         } else {
 954                 io_req.notify.fn = NULL;
 955         }
 956         io_req.client = ic->io;
 957         io_loc.bdev = ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev;
 958         io_loc.sector = ic->start + SB_SECTORS + sector;
 959         io_loc.count = n_sectors;
 960
 961         r = dm_io(&io_req, 1, &io_loc, NULL);
 962         if (unlikely(r)) {
 963                 dm_integrity_io_error(ic, op == REQ_OP_READ ? "reading journal" : "writing journal", r);
 964                 if (comp) {
 965                         WARN_ONCE(1, "asynchronous dm_io failed: %d", r);
 966                         complete_journal_io(-1UL, comp);
 967                 }
 968         }
 969 }
 970
 971 static void rw_journal(struct dm_integrity_c *ic, int op, int op_flags, unsigned section,
 972                        unsigned n_sections, struct journal_completion *comp)
 973 {
 974         unsigned sector, n_sectors;
 975
 976         sector = section * ic->journal_section_sectors;
 977         n_sectors = n_sections * ic->journal_section_sectors;
 978
 979         rw_journal_sectors(ic, op, op_flags, sector, n_sectors, comp);
 980 }
 981
 982 static void write_journal(struct dm_integrity_c *ic, unsigned commit_start, unsigned commit_sections)
 983 {
 984         struct journal_completion io_comp;
 985         struct journal_completion crypt_comp_1;
 986         struct journal_completion crypt_comp_2;
 987         unsigned i;
 988
 989         io_comp.ic = ic;
 990         init_completion(&io_comp.comp);
 991
 992         if (commit_start + commit_sections <= ic->journal_sections) {
 993                 io_comp.in_flight = (atomic_t)ATOMIC_INIT(1);
 994                 if (ic->journal_io) {
 995                         crypt_comp_1.ic = ic;
 996                         init_completion(&crypt_comp_1.comp);
 997                         crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
 998                         encrypt_journal(ic, true, commit_start, commit_sections, &crypt_comp_1);
 999                         wait_for_completion_io(&crypt_comp_1.comp);
1000                 } else {
1001                         for (i = 0; i < commit_sections; i++)
1002                                 rw_section_mac(ic, commit_start + i, true);
1003                 }
1004                 rw_journal(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, commit_start,
1005                            commit_sections, &io_comp);
1006         } else {
1007                 unsigned to_end;
1008                 io_comp.in_flight = (atomic_t)ATOMIC_INIT(2);
1009                 to_end = ic->journal_sections - commit_start;
1010                 if (ic->journal_io) {
1011                         crypt_comp_1.ic = ic;
1012                         init_completion(&crypt_comp_1.comp);
1013                         crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
1014                         encrypt_journal(ic, true, commit_start, to_end, &crypt_comp_1);
1015                         if (try_wait_for_completion(&crypt_comp_1.comp)) {
1016                                 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
1017                                 reinit_completion(&crypt_comp_1.comp);
1018                                 crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
1019                                 encrypt_journal(ic, true, 0, commit_sections - to_end, &crypt_comp_1);
1020                                 wait_for_completion_io(&crypt_comp_1.comp);
1021                         } else {
1022                                 crypt_comp_2.ic = ic;
1023                                 init_completion(&crypt_comp_2.comp);
1024                                 crypt_comp_2.in_flight = (atomic_t)ATOMIC_INIT(0);
1025                                 encrypt_journal(ic, true, 0, commit_sections - to_end, &crypt_comp_2);
1026                                 wait_for_completion_io(&crypt_comp_1.comp);
1027                                 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
1028                                 wait_for_completion_io(&crypt_comp_2.comp);
1029                         }
1030                 } else {
1031                         for (i = 0; i < to_end; i++)
1032                                 rw_section_mac(ic, commit_start + i, true);
1033                         rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
1034                         for (i = 0; i < commit_sections - to_end; i++)
1035                                 rw_section_mac(ic, i, true);
1036                 }
1037                 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, 0, commit_sections - to_end, &io_comp);
1038         }
1039
1040         wait_for_completion_io(&io_comp.comp);
1041 }
1042
1043 static void copy_from_journal(struct dm_integrity_c *ic, unsigned section, unsigned offset,
1044                               unsigned n_sectors, sector_t target, io_notify_fn fn, void *data)
1045 {
1046         struct dm_io_request io_req;
1047         struct dm_io_region io_loc;
1048         int r;
1049         unsigned sector, pl_index, pl_offset;
1050
1051         BUG_ON((target | n_sectors | offset) & (unsigned)(ic->sectors_per_block - 1));
1052
1053         if (unlikely(dm_integrity_failed(ic))) {
1054                 fn(-1UL, data);
1055                 return;
1056         }
1057
1058         sector = section * ic->journal_section_sectors + JOURNAL_BLOCK_SECTORS + offset;
1059
1060         pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
1061         pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
1062
1063         io_req.bi_op = REQ_OP_WRITE;
1064         io_req.bi_op_flags = 0;
1065         io_req.mem.type = DM_IO_PAGE_LIST;
1066         io_req.mem.ptr.pl = &ic->journal[pl_index];
1067         io_req.mem.offset = pl_offset;
1068         io_req.notify.fn = fn;
1069         io_req.notify.context = data;
1070         io_req.client = ic->io;
1071         io_loc.bdev = ic->dev->bdev;
1072         io_loc.sector = target;
1073         io_loc.count = n_sectors;
1074
1075         r = dm_io(&io_req, 1, &io_loc, NULL);
1076         if (unlikely(r)) {
1077                 WARN_ONCE(1, "asynchronous dm_io failed: %d", r);
1078                 fn(-1UL, data);
1079         }
1080 }
1081
1082 static bool ranges_overlap(struct dm_integrity_range *range1, struct dm_integrity_range *range2)
1083 {
1084         return range1->logical_sector < range2->logical_sector + range2->n_sectors &&
1085                range1->logical_sector + range1->n_sectors > range2->logical_sector;
1086 }
1087
1088 static bool add_new_range(struct dm_integrity_c *ic, struct dm_integrity_range *new_range, bool check_waiting)
1089 {
1090         struct rb_node **n = &ic->in_progress.rb_node;
1091         struct rb_node *parent;
1092
1093         BUG_ON((new_range->logical_sector | new_range->n_sectors) & (unsigned)(ic->sectors_per_block - 1));
1094
1095         if (likely(check_waiting)) {
1096                 struct dm_integrity_range *range;
1097                 list_for_each_entry(range, &ic->wait_list, wait_entry) {
1098                         if (unlikely(ranges_overlap(range, new_range)))
1099                                 return false;
1100                 }
1101         }
1102
1103         parent = NULL;
1104
1105         while (*n) {
1106                 struct dm_integrity_range *range = container_of(*n, struct dm_integrity_range, node);
1107
1108                 parent = *n;
1109                 if (new_range->logical_sector + new_range->n_sectors <= range->logical_sector) {
1110                         n = &range->node.rb_left;
1111                 } else if (new_range->logical_sector >= range->logical_sector + range->n_sectors) {
1112                         n = &range->node.rb_right;
1113                 } else {
1114                         return false;
1115                 }
1116         }
1117
1118         rb_link_node(&new_range->node, parent, n);
1119         rb_insert_color(&new_range->node, &ic->in_progress);
1120
1121         return true;
1122 }
1123
1124 static void remove_range_unlocked(struct dm_integrity_c *ic, struct dm_integrity_range *range)
1125 {
1126         rb_erase(&range->node, &ic->in_progress);
1127         while (unlikely(!list_empty(&ic->wait_list))) {
1128                 struct dm_integrity_range *last_range =
1129                         list_first_entry(&ic->wait_list, struct dm_integrity_range, wait_entry);
1130                 struct task_struct *last_range_task;
1131                 last_range_task = last_range->task;
1132                 list_del(&last_range->wait_entry);
1133                 if (!add_new_range(ic, last_range, false)) {
1134                         last_range->task = last_range_task;
1135                         list_add(&last_range->wait_entry, &ic->wait_list);
1136                         break;
1137                 }
1138                 last_range->waiting = false;
1139                 wake_up_process(last_range_task);
1140         }
1141 }
1142
1143 static void remove_range(struct dm_integrity_c *ic, struct dm_integrity_range *range)
1144 {
1145         unsigned long flags;
1146
1147         spin_lock_irqsave(&ic->endio_wait.lock, flags);
1148         remove_range_unlocked(ic, range);
1149         spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
1150 }
1151
1152 static void wait_and_add_new_range(struct dm_integrity_c *ic, struct dm_integrity_range *new_range)
1153 {
1154         new_range->waiting = true;
1155         list_add_tail(&new_range->wait_entry, &ic->wait_list);
1156         new_range->task = current;
1157         do {
1158                 __set_current_state(TASK_UNINTERRUPTIBLE);
1159                 spin_unlock_irq(&ic->endio_wait.lock);
1160                 io_schedule();
1161                 spin_lock_irq(&ic->endio_wait.lock);
1162         } while (unlikely(new_range->waiting));
1163 }
1164
1165 static void add_new_range_and_wait(struct dm_integrity_c *ic, struct dm_integrity_range *new_range)
1166 {
1167         if (unlikely(!add_new_range(ic, new_range, true)))
1168                 wait_and_add_new_range(ic, new_range);
1169 }
1170
1171 static void init_journal_node(struct journal_node *node)
1172 {
1173         RB_CLEAR_NODE(&node->node);
1174         node->sector = (sector_t)-1;
1175 }
1176
1177 static void add_journal_node(struct dm_integrity_c *ic, struct journal_node *node, sector_t sector)
1178 {
1179         struct rb_node **link;
1180         struct rb_node *parent;
1181
1182         node->sector = sector;
1183         BUG_ON(!RB_EMPTY_NODE(&node->node));
1184
1185         link = &ic->journal_tree_root.rb_node;
1186         parent = NULL;
1187
1188         while (*link) {
1189                 struct journal_node *j;
1190                 parent = *link;
1191                 j = container_of(parent, struct journal_node, node);
1192                 if (sector < j->sector)
1193                         link = &j->node.rb_left;
1194                 else
1195                         link = &j->node.rb_right;
1196         }
1197
1198         rb_link_node(&node->node, parent, link);
1199         rb_insert_color(&node->node, &ic->journal_tree_root);
1200 }
1201
1202 static void remove_journal_node(struct dm_integrity_c *ic, struct journal_node *node)
1203 {
1204         BUG_ON(RB_EMPTY_NODE(&node->node));
1205         rb_erase(&node->node, &ic->journal_tree_root);
1206         init_journal_node(node);
1207 }
1208
1209 #define NOT_FOUND       (-1U)
1210
1211 static unsigned find_journal_node(struct dm_integrity_c *ic, sector_t sector, sector_t *next_sector)
1212 {
1213         struct rb_node *n = ic->journal_tree_root.rb_node;
1214         unsigned found = NOT_FOUND;
1215         *next_sector = (sector_t)-1;
1216         while (n) {
1217                 struct journal_node *j = container_of(n, struct journal_node, node);
1218                 if (sector == j->sector) {
1219                         found = j - ic->journal_tree;
1220                 }
1221                 if (sector < j->sector) {
1222                         *next_sector = j->sector;
1223                         n = j->node.rb_left;
1224                 } else {
1225                         n = j->node.rb_right;
1226                 }
1227         }
1228
1229         return found;
1230 }
1231
1232 static bool test_journal_node(struct dm_integrity_c *ic, unsigned pos, sector_t sector)
1233 {
1234         struct journal_node *node, *next_node;
1235         struct rb_node *next;
1236
1237         if (unlikely(pos >= ic->journal_entries))
1238                 return false;
1239         node = &ic->journal_tree[pos];
1240         if (unlikely(RB_EMPTY_NODE(&node->node)))
1241                 return false;
1242         if (unlikely(node->sector != sector))
1243                 return false;
1244
1245         next = rb_next(&node->node);
1246         if (unlikely(!next))
1247                 return true;
1248
1249         next_node = container_of(next, struct journal_node, node);
1250         return next_node->sector != sector;
1251 }
1252
1253 static bool find_newer_committed_node(struct dm_integrity_c *ic, struct journal_node *node)
1254 {
1255         struct rb_node *next;
1256         struct journal_node *next_node;
1257         unsigned next_section;
1258
1259         BUG_ON(RB_EMPTY_NODE(&node->node));
1260
1261         next = rb_next(&node->node);
1262         if (unlikely(!next))
1263                 return false;
1264
1265         next_node = container_of(next, struct journal_node, node);
1266
1267         if (next_node->sector != node->sector)
1268                 return false;
1269
1270         next_section = (unsigned)(next_node - ic->journal_tree) / ic->journal_section_entries;
1271         if (next_section >= ic->committed_section &&
1272             next_section < ic->committed_section + ic->n_committed_sections)
1273                 return true;
1274         if (next_section + ic->journal_sections < ic->committed_section + ic->n_committed_sections)
1275                 return true;
1276
1277         return false;
1278 }
1279
1280 #define TAG_READ        0
1281 #define TAG_WRITE       1
1282 #define TAG_CMP         2
1283
1284 static int dm_integrity_rw_tag(struct dm_integrity_c *ic, unsigned char *tag, sector_t *metadata_block,
1285                                unsigned *metadata_offset, unsigned total_size, int op)
1286 {
1287         do {
1288                 unsigned char *data, *dp;
1289                 struct dm_buffer *b;
1290                 unsigned to_copy;
1291                 int r;
1292
1293                 r = dm_integrity_failed(ic);
1294                 if (unlikely(r))
1295                         return r;
1296
1297                 data = dm_bufio_read(ic->bufio, *metadata_block, &b);
1298                 if (IS_ERR(data))
1299                         return PTR_ERR(data);
1300
1301                 to_copy = min((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - *metadata_offset, total_size);
1302                 dp = data + *metadata_offset;
1303                 if (op == TAG_READ) {
1304                         memcpy(tag, dp, to_copy);
1305                 } else if (op == TAG_WRITE) {
1306                         memcpy(dp, tag, to_copy);
1307                         dm_bufio_mark_partial_buffer_dirty(b, *metadata_offset, *metadata_offset + to_copy);
1308                 } else  {
1309                         /* e.g.: op == TAG_CMP */
1310                         if (unlikely(memcmp(dp, tag, to_copy))) {
1311                                 unsigned i;
1312
1313                                 for (i = 0; i < to_copy; i++) {
1314                                         if (dp[i] != tag[i])
1315                                                 break;
1316                                         total_size--;
1317                                 }
1318                                 dm_bufio_release(b);
1319                                 return total_size;
1320                         }
1321                 }
1322                 dm_bufio_release(b);
1323
1324                 tag += to_copy;
1325                 *metadata_offset += to_copy;
1326                 if (unlikely(*metadata_offset == 1U << SECTOR_SHIFT << ic->log2_buffer_sectors)) {
1327                         (*metadata_block)++;
1328                         *metadata_offset = 0;
1329                 }
1330                 total_size -= to_copy;
1331         } while (unlikely(total_size));
1332
1333         return 0;
1334 }
1335
1336 static void dm_integrity_flush_buffers(struct dm_integrity_c *ic)
1337 {
1338         int r;
1339         r = dm_bufio_write_dirty_buffers(ic->bufio);
1340         if (unlikely(r))
1341                 dm_integrity_io_error(ic, "writing tags", r);
1342 }
1343
1344 static void sleep_on_endio_wait(struct dm_integrity_c *ic)
1345 {
1346         DECLARE_WAITQUEUE(wait, current);
1347         __add_wait_queue(&ic->endio_wait, &wait);
1348         __set_current_state(TASK_UNINTERRUPTIBLE);
1349         spin_unlock_irq(&ic->endio_wait.lock);
1350         io_schedule();
1351         spin_lock_irq(&ic->endio_wait.lock);
1352         __remove_wait_queue(&ic->endio_wait, &wait);
1353 }
1354
1355 static void autocommit_fn(struct timer_list *t)
1356 {
1357         struct dm_integrity_c *ic = from_timer(ic, t, autocommit_timer);
1358
1359         if (likely(!dm_integrity_failed(ic)))
1360                 queue_work(ic->commit_wq, &ic->commit_work);
1361 }
1362
1363 static void schedule_autocommit(struct dm_integrity_c *ic)
1364 {
1365         if (!timer_pending(&ic->autocommit_timer))
1366                 mod_timer(&ic->autocommit_timer, jiffies + ic->autocommit_jiffies);
1367 }
1368
1369 static void submit_flush_bio(struct dm_integrity_c *ic, struct dm_integrity_io *dio)
1370 {
1371         struct bio *bio;
1372         unsigned long flags;
1373
1374         spin_lock_irqsave(&ic->endio_wait.lock, flags);
1375         bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1376         bio_list_add(&ic->flush_bio_list, bio);
1377         spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
1378
1379         queue_work(ic->commit_wq, &ic->commit_work);
1380 }
1381
1382 static void do_endio(struct dm_integrity_c *ic, struct bio *bio)
1383 {
1384         int r = dm_integrity_failed(ic);
1385         if (unlikely(r) && !bio->bi_status)
1386                 bio->bi_status = errno_to_blk_status(r);
1387         if (unlikely(ic->synchronous_mode) && bio_op(bio) == REQ_OP_WRITE) {
1388                 unsigned long flags;
1389                 spin_lock_irqsave(&ic->endio_wait.lock, flags);
1390                 bio_list_add(&ic->synchronous_bios, bio);
1391                 queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, 0);
1392                 spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
1393                 return;
1394         }
1395         bio_endio(bio);
1396 }
1397
1398 static void do_endio_flush(struct dm_integrity_c *ic, struct dm_integrity_io *dio)
1399 {
1400         struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1401
1402         if (unlikely(dio->fua) && likely(!bio->bi_status) && likely(!dm_integrity_failed(ic)))
1403                 submit_flush_bio(ic, dio);
1404         else
1405                 do_endio(ic, bio);
1406 }
1407
1408 static void dec_in_flight(struct dm_integrity_io *dio)
1409 {
1410         if (atomic_dec_and_test(&dio->in_flight)) {
1411                 struct dm_integrity_c *ic = dio->ic;
1412                 struct bio *bio;
1413
1414                 remove_range(ic, &dio->range);
1415
1416                 if (unlikely(dio->write))
1417                         schedule_autocommit(ic);
1418
1419                 bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1420
1421                 if (unlikely(dio->bi_status) && !bio->bi_status)
1422                         bio->bi_status = dio->bi_status;
1423                 if (likely(!bio->bi_status) && unlikely(bio_sectors(bio) != dio->range.n_sectors)) {
1424                         dio->range.logical_sector += dio->range.n_sectors;
1425                         bio_advance(bio, dio->range.n_sectors << SECTOR_SHIFT);
1426                         INIT_WORK(&dio->work, integrity_bio_wait);
1427                         queue_work(ic->wait_wq, &dio->work);
1428                         return;
1429                 }
1430                 do_endio_flush(ic, dio);
1431         }
1432 }
1433
1434 static void integrity_end_io(struct bio *bio)
1435 {
1436         struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
1437
1438         bio->bi_iter = dio->orig_bi_iter;
1439         bio->bi_disk = dio->orig_bi_disk;
1440         bio->bi_partno = dio->orig_bi_partno;
1441         if (dio->orig_bi_integrity) {
1442                 bio->bi_integrity = dio->orig_bi_integrity;
1443                 bio->bi_opf |= REQ_INTEGRITY;
1444         }
1445         bio->bi_end_io = dio->orig_bi_end_io;
1446
1447         if (dio->completion)
1448                 complete(dio->completion);
1449
1450         dec_in_flight(dio);
1451 }
1452
1453 static void integrity_sector_checksum(struct dm_integrity_c *ic, sector_t sector,
1454                                       const char *data, char *result)
1455 {
1456         __u64 sector_le = cpu_to_le64(sector);
1457         SHASH_DESC_ON_STACK(req, ic->internal_hash);
1458         int r;
1459         unsigned digest_size;
1460
1461         req->tfm = ic->internal_hash;
1462         req->flags = 0;
1463
1464         r = crypto_shash_init(req);
1465         if (unlikely(r < 0)) {
1466                 dm_integrity_io_error(ic, "crypto_shash_init", r);
1467                 goto failed;
1468         }
1469
1470         r = crypto_shash_update(req, (const __u8 *)&sector_le, sizeof sector_le);
1471         if (unlikely(r < 0)) {
1472                 dm_integrity_io_error(ic, "crypto_shash_update", r);
1473                 goto failed;
1474         }
1475
1476         r = crypto_shash_update(req, data, ic->sectors_per_block << SECTOR_SHIFT);
1477         if (unlikely(r < 0)) {
1478                 dm_integrity_io_error(ic, "crypto_shash_update", r);
1479                 goto failed;
1480         }
1481
1482         r = crypto_shash_final(req, result);
1483         if (unlikely(r < 0)) {
1484                 dm_integrity_io_error(ic, "crypto_shash_final", r);
1485                 goto failed;
1486         }
1487
1488         digest_size = crypto_shash_digestsize(ic->internal_hash);
1489         if (unlikely(digest_size < ic->tag_size))
1490                 memset(result + digest_size, 0, ic->tag_size - digest_size);
1491
1492         return;
1493
1494 failed:
1495         /* this shouldn't happen anyway, the hash functions have no reason to fail */
1496         get_random_bytes(result, ic->tag_size);
1497 }
1498
1499 static void integrity_metadata(struct work_struct *w)
1500 {
1501         struct dm_integrity_io *dio = container_of(w, struct dm_integrity_io, work);
1502         struct dm_integrity_c *ic = dio->ic;
1503
1504         int r;
1505
1506         if (ic->internal_hash) {
1507                 struct bvec_iter iter;
1508                 struct bio_vec bv;
1509                 unsigned digest_size = crypto_shash_digestsize(ic->internal_hash);
1510                 struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1511                 char *checksums;
1512                 unsigned extra_space = unlikely(digest_size > ic->tag_size) ? digest_size - ic->tag_size : 0;
1513                 char checksums_onstack[HASH_MAX_DIGESTSIZE];
1514                 unsigned sectors_to_process = dio->range.n_sectors;
1515                 sector_t sector = dio->range.logical_sector;
1516
1517                 if (unlikely(ic->mode == 'R'))
1518                         goto skip_io;
1519
1520                 checksums = kmalloc((PAGE_SIZE >> SECTOR_SHIFT >> ic->sb->log2_sectors_per_block) * ic->tag_size + extra_space,
1521                                     GFP_NOIO | __GFP_NORETRY | __GFP_NOWARN);
1522                 if (!checksums) {
1523                         checksums = checksums_onstack;
1524                         if (WARN_ON(extra_space &&
1525                                     digest_size > sizeof(checksums_onstack))) {
1526                                 r = -EINVAL;
1527                                 goto error;
1528                         }
1529                 }
1530
1531                 __bio_for_each_segment(bv, bio, iter, dio->orig_bi_iter) {
1532                         unsigned pos;
1533                         char *mem, *checksums_ptr;
1534
1535 again:
1536                         mem = (char *)kmap_atomic(bv.bv_page) + bv.bv_offset;
1537                         pos = 0;
1538                         checksums_ptr = checksums;
1539                         do {
1540                                 integrity_sector_checksum(ic, sector, mem + pos, checksums_ptr);
1541                                 checksums_ptr += ic->tag_size;
1542                                 sectors_to_process -= ic->sectors_per_block;
1543                                 pos += ic->sectors_per_block << SECTOR_SHIFT;
1544                                 sector += ic->sectors_per_block;
1545                         } while (pos < bv.bv_len && sectors_to_process && checksums != checksums_onstack);
1546                         kunmap_atomic(mem);
1547
1548                         r = dm_integrity_rw_tag(ic, checksums, &dio->metadata_block, &dio->metadata_offset,
1549                                                 checksums_ptr - checksums, !dio->write ? TAG_CMP : TAG_WRITE);
1550                         if (unlikely(r)) {
1551                                 if (r > 0) {
1552                                         DMERR_LIMIT("Checksum failed at sector 0x%llx",
1553                                                     (unsigned long long)(sector - ((r + ic->tag_size - 1) / ic->tag_size)));
1554                                         r = -EILSEQ;
1555                                         atomic64_inc(&ic->number_of_mismatches);
1556                                 }
1557                                 if (likely(checksums != checksums_onstack))
1558                                         kfree(checksums);
1559                                 goto error;
1560                         }
1561
1562                         if (!sectors_to_process)
1563                                 break;
1564
1565                         if (unlikely(pos < bv.bv_len)) {
1566                                 bv.bv_offset += pos;
1567                                 bv.bv_len -= pos;
1568                                 goto again;
1569                         }
1570                 }
1571
1572                 if (likely(checksums != checksums_onstack))
1573                         kfree(checksums);
1574         } else {
1575                 struct bio_integrity_payload *bip = dio->orig_bi_integrity;
1576
1577                 if (bip) {
1578                         struct bio_vec biv;
1579                         struct bvec_iter iter;
1580                         unsigned data_to_process = dio->range.n_sectors;
1581                         sector_to_block(ic, data_to_process);
1582                         data_to_process *= ic->tag_size;
1583
1584                         bip_for_each_vec(biv, bip, iter) {
1585                                 unsigned char *tag;
1586                                 unsigned this_len;
1587
1588                                 BUG_ON(PageHighMem(biv.bv_page));
1589                                 tag = lowmem_page_address(biv.bv_page) + biv.bv_offset;
1590                                 this_len = min(biv.bv_len, data_to_process);
1591                                 r = dm_integrity_rw_tag(ic, tag, &dio->metadata_block, &dio->metadata_offset,
1592                                                         this_len, !dio->write ? TAG_READ : TAG_WRITE);
1593                                 if (unlikely(r))
1594                                         goto error;
1595                                 data_to_process -= this_len;
1596                                 if (!data_to_process)
1597                                         break;
1598                         }
1599                 }
1600         }
1601 skip_io:
1602         dec_in_flight(dio);
1603         return;
1604 error:
1605         dio->bi_status = errno_to_blk_status(r);
1606         dec_in_flight(dio);
1607 }
1608
1609 static int dm_integrity_map(struct dm_target *ti, struct bio *bio)
1610 {
1611         struct dm_integrity_c *ic = ti->private;
1612         struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
1613         struct bio_integrity_payload *bip;
1614
1615         sector_t area, offset;
1616
1617         dio->ic = ic;
1618         dio->bi_status = 0;
1619
1620         if (unlikely(bio->bi_opf & REQ_PREFLUSH)) {
1621                 submit_flush_bio(ic, dio);
1622                 return DM_MAPIO_SUBMITTED;
1623         }
1624
1625         dio->range.logical_sector = dm_target_offset(ti, bio->bi_iter.bi_sector);
1626         dio->write = bio_op(bio) == REQ_OP_WRITE;
1627         dio->fua = dio->write && bio->bi_opf & REQ_FUA;
1628         if (unlikely(dio->fua)) {
1629                 /*
1630                  * Don't pass down the FUA flag because we have to flush
1631                  * disk cache anyway.
1632                  */
1633                 bio->bi_opf &= ~REQ_FUA;
1634         }
1635         if (unlikely(dio->range.logical_sector + bio_sectors(bio) > ic->provided_data_sectors)) {
1636                 DMERR("Too big sector number: 0x%llx + 0x%x > 0x%llx",
1637                       (unsigned long long)dio->range.logical_sector, bio_sectors(bio),
1638                       (unsigned long long)ic->provided_data_sectors);
1639                 return DM_MAPIO_KILL;
1640         }
1641         if (unlikely((dio->range.logical_sector | bio_sectors(bio)) & (unsigned)(ic->sectors_per_block - 1))) {
1642                 DMERR("Bio not aligned on %u sectors: 0x%llx, 0x%x",
1643                       ic->sectors_per_block,
1644                       (unsigned long long)dio->range.logical_sector, bio_sectors(bio));
1645                 return DM_MAPIO_KILL;
1646         }
1647
1648         if (ic->sectors_per_block > 1) {
1649                 struct bvec_iter iter;
1650                 struct bio_vec bv;
1651                 bio_for_each_segment(bv, bio, iter) {
1652                         if (unlikely(bv.bv_len & ((ic->sectors_per_block << SECTOR_SHIFT) - 1))) {
1653                                 DMERR("Bio vector (%u,%u) is not aligned on %u-sector boundary",
1654                                         bv.bv_offset, bv.bv_len, ic->sectors_per_block);
1655                                 return DM_MAPIO_KILL;
1656                         }
1657                 }
1658         }
1659
1660         bip = bio_integrity(bio);
1661         if (!ic->internal_hash) {
1662                 if (bip) {
1663                         unsigned wanted_tag_size = bio_sectors(bio) >> ic->sb->log2_sectors_per_block;
1664                         if (ic->log2_tag_size >= 0)
1665                                 wanted_tag_size <<= ic->log2_tag_size;
1666                         else
1667                                 wanted_tag_size *= ic->tag_size;
1668                         if (unlikely(wanted_tag_size != bip->bip_iter.bi_size)) {
1669                                 DMERR("Invalid integrity data size %u, expected %u", bip->bip_iter.bi_size, wanted_tag_size);
1670                                 return DM_MAPIO_KILL;
1671                         }
1672                 }
1673         } else {
1674                 if (unlikely(bip != NULL)) {
1675                         DMERR("Unexpected integrity data when using internal hash");
1676                         return DM_MAPIO_KILL;
1677                 }
1678         }
1679
1680         if (unlikely(ic->mode == 'R') && unlikely(dio->write))
1681                 return DM_MAPIO_KILL;
1682
1683         get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
1684         dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset, &dio->metadata_offset);
1685         bio->bi_iter.bi_sector = get_data_sector(ic, area, offset);
1686
1687         dm_integrity_map_continue(dio, true);
1688         return DM_MAPIO_SUBMITTED;
1689 }
1690
1691 static bool __journal_read_write(struct dm_integrity_io *dio, struct bio *bio,
1692                                  unsigned journal_section, unsigned journal_entry)
1693 {
1694         struct dm_integrity_c *ic = dio->ic;
1695         sector_t logical_sector;
1696         unsigned n_sectors;
1697
1698         logical_sector = dio->range.logical_sector;
1699         n_sectors = dio->range.n_sectors;
1700         do {
1701                 struct bio_vec bv = bio_iovec(bio);
1702                 char *mem;
1703
1704                 if (unlikely(bv.bv_len >> SECTOR_SHIFT > n_sectors))
1705                         bv.bv_len = n_sectors << SECTOR_SHIFT;
1706                 n_sectors -= bv.bv_len >> SECTOR_SHIFT;
1707                 bio_advance_iter(bio, &bio->bi_iter, bv.bv_len);
1708 retry_kmap:
1709                 mem = kmap_atomic(bv.bv_page);
1710                 if (likely(dio->write))
1711                         flush_dcache_page(bv.bv_page);
1712
1713                 do {
1714                         struct journal_entry *je = access_journal_entry(ic, journal_section, journal_entry);
1715
1716                         if (unlikely(!dio->write)) {
1717                                 struct journal_sector *js;
1718                                 char *mem_ptr;
1719                                 unsigned s;
1720
1721                                 if (unlikely(journal_entry_is_inprogress(je))) {
1722                                         flush_dcache_page(bv.bv_page);
1723                                         kunmap_atomic(mem);
1724
1725                                         __io_wait_event(ic->copy_to_journal_wait, !journal_entry_is_inprogress(je));
1726                                         goto retry_kmap;
1727                                 }
1728                                 smp_rmb();
1729                                 BUG_ON(journal_entry_get_sector(je) != logical_sector);
1730                                 js = access_journal_data(ic, journal_section, journal_entry);
1731                                 mem_ptr = mem + bv.bv_offset;
1732                                 s = 0;
1733                                 do {
1734                                         memcpy(mem_ptr, js, JOURNAL_SECTOR_DATA);
1735                                         *(commit_id_t *)(mem_ptr + JOURNAL_SECTOR_DATA) = je->last_bytes[s];
1736                                         js++;
1737                                         mem_ptr += 1 << SECTOR_SHIFT;
1738                                 } while (++s < ic->sectors_per_block);
1739 #ifdef INTERNAL_VERIFY
1740                                 if (ic->internal_hash) {
1741                                         char checksums_onstack[max(HASH_MAX_DIGESTSIZE, MAX_TAG_SIZE)];
1742
1743                                         integrity_sector_checksum(ic, logical_sector, mem + bv.bv_offset, checksums_onstack);
1744                                         if (unlikely(memcmp(checksums_onstack, journal_entry_tag(ic, je), ic->tag_size))) {
1745                                                 DMERR_LIMIT("Checksum failed when reading from journal, at sector 0x%llx",
1746                                                             (unsigned long long)logical_sector);
1747                                         }
1748                                 }
1749 #endif
1750                         }
1751
1752                         if (!ic->internal_hash) {
1753                                 struct bio_integrity_payload *bip = bio_integrity(bio);
1754                                 unsigned tag_todo = ic->tag_size;
1755                                 char *tag_ptr = journal_entry_tag(ic, je);
1756
1757                                 if (bip) do {
1758                                         struct bio_vec biv = bvec_iter_bvec(bip->bip_vec, bip->bip_iter);
1759                                         unsigned tag_now = min(biv.bv_len, tag_todo);
1760                                         char *tag_addr;
1761                                         BUG_ON(PageHighMem(biv.bv_page));
1762                                         tag_addr = lowmem_page_address(biv.bv_page) + biv.bv_offset;
1763                                         if (likely(dio->write))
1764                                                 memcpy(tag_ptr, tag_addr, tag_now);
1765                                         else
1766                                                 memcpy(tag_addr, tag_ptr, tag_now);
1767                                         bvec_iter_advance(bip->bip_vec, &bip->bip_iter, tag_now);
1768                                         tag_ptr += tag_now;
1769                                         tag_todo -= tag_now;
1770                                 } while (unlikely(tag_todo)); else {
1771                                         if (likely(dio->write))
1772                                                 memset(tag_ptr, 0, tag_todo);
1773                                 }
1774                         }
1775
1776                         if (likely(dio->write)) {
1777                                 struct journal_sector *js;
1778                                 unsigned s;
1779
1780                                 js = access_journal_data(ic, journal_section, journal_entry);
1781                                 memcpy(js, mem + bv.bv_offset, ic->sectors_per_block << SECTOR_SHIFT);
1782
1783                                 s = 0;
1784                                 do {
1785                                         je->last_bytes[s] = js[s].commit_id;
1786                                 } while (++s < ic->sectors_per_block);
1787
1788                                 if (ic->internal_hash) {
1789                                         unsigned digest_size = crypto_shash_digestsize(ic->internal_hash);
1790                                         if (unlikely(digest_size > ic->tag_size)) {
1791                                                 char checksums_onstack[HASH_MAX_DIGESTSIZE];
1792                                                 integrity_sector_checksum(ic, logical_sector, (char *)js, checksums_onstack);
1793                                                 memcpy(journal_entry_tag(ic, je), checksums_onstack, ic->tag_size);
1794                                         } else
1795                                                 integrity_sector_checksum(ic, logical_sector, (char *)js, journal_entry_tag(ic, je));
1796                                 }
1797
1798                                 journal_entry_set_sector(je, logical_sector);
1799                         }
1800                         logical_sector += ic->sectors_per_block;
1801
1802                         journal_entry++;
1803                         if (unlikely(journal_entry == ic->journal_section_entries)) {
1804                                 journal_entry = 0;
1805                                 journal_section++;
1806                                 wraparound_section(ic, &journal_section);
1807                         }
1808
1809                         bv.bv_offset += ic->sectors_per_block << SECTOR_SHIFT;
1810                 } while (bv.bv_len -= ic->sectors_per_block << SECTOR_SHIFT);
1811
1812                 if (unlikely(!dio->write))
1813                         flush_dcache_page(bv.bv_page);
1814                 kunmap_atomic(mem);
1815         } while (n_sectors);
1816
1817         if (likely(dio->write)) {
1818                 smp_mb();
1819                 if (unlikely(waitqueue_active(&ic->copy_to_journal_wait)))
1820                         wake_up(&ic->copy_to_journal_wait);
1821                 if (READ_ONCE(ic->free_sectors) <= ic->free_sectors_threshold) {
1822                         queue_work(ic->commit_wq, &ic->commit_work);
1823                 } else {
1824                         schedule_autocommit(ic);
1825                 }
1826         } else {
1827                 remove_range(ic, &dio->range);
1828         }
1829
1830         if (unlikely(bio->bi_iter.bi_size)) {
1831                 sector_t area, offset;
1832
1833                 dio->range.logical_sector = logical_sector;
1834                 get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
1835                 dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset, &dio->metadata_offset);
1836                 return true;
1837         }
1838
1839         return false;
1840 }
1841
1842 static void dm_integrity_map_continue(struct dm_integrity_io *dio, bool from_map)
1843 {
1844         struct dm_integrity_c *ic = dio->ic;
1845         struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1846         unsigned journal_section, journal_entry;
1847         unsigned journal_read_pos;
1848         struct completion read_comp;
1849         bool need_sync_io = ic->internal_hash && !dio->write;
1850
1851         if (need_sync_io && from_map) {
1852                 INIT_WORK(&dio->work, integrity_bio_wait);
1853                 queue_work(ic->metadata_wq, &dio->work);
1854                 return;
1855         }
1856
1857 lock_retry:
1858         spin_lock_irq(&ic->endio_wait.lock);
1859 retry:
1860         if (unlikely(dm_integrity_failed(ic))) {
1861                 spin_unlock_irq(&ic->endio_wait.lock);
1862                 do_endio(ic, bio);
1863                 return;
1864         }
1865         dio->range.n_sectors = bio_sectors(bio);
1866         journal_read_pos = NOT_FOUND;
1867         if (likely(ic->mode == 'J')) {
1868                 if (dio->write) {
1869                         unsigned next_entry, i, pos;
1870                         unsigned ws, we, range_sectors;
1871
1872                         dio->range.n_sectors = min(dio->range.n_sectors,
1873                                                    (sector_t)ic->free_sectors << ic->sb->log2_sectors_per_block);
1874                         if (unlikely(!dio->range.n_sectors)) {
1875                                 if (from_map)
1876                                         goto offload_to_thread;
1877                                 sleep_on_endio_wait(ic);
1878                                 goto retry;
1879                         }
1880                         range_sectors = dio->range.n_sectors >> ic->sb->log2_sectors_per_block;
1881                         ic->free_sectors -= range_sectors;
1882                         journal_section = ic->free_section;
1883                         journal_entry = ic->free_section_entry;
1884
1885                         next_entry = ic->free_section_entry + range_sectors;
1886                         ic->free_section_entry = next_entry % ic->journal_section_entries;
1887                         ic->free_section += next_entry / ic->journal_section_entries;
1888                         ic->n_uncommitted_sections += next_entry / ic->journal_section_entries;
1889                         wraparound_section(ic, &ic->free_section);
1890
1891                         pos = journal_section * ic->journal_section_entries + journal_entry;
1892                         ws = journal_section;
1893                         we = journal_entry;
1894                         i = 0;
1895                         do {
1896                                 struct journal_entry *je;
1897
1898                                 add_journal_node(ic, &ic->journal_tree[pos], dio->range.logical_sector + i);
1899                                 pos++;
1900                                 if (unlikely(pos >= ic->journal_entries))
1901                                         pos = 0;
1902
1903                                 je = access_journal_entry(ic, ws, we);
1904                                 BUG_ON(!journal_entry_is_unused(je));
1905                                 journal_entry_set_inprogress(je);
1906                                 we++;
1907                                 if (unlikely(we == ic->journal_section_entries)) {
1908                                         we = 0;
1909                                         ws++;
1910                                         wraparound_section(ic, &ws);
1911                                 }
1912                         } while ((i += ic->sectors_per_block) < dio->range.n_sectors);
1913
1914                         spin_unlock_irq(&ic->endio_wait.lock);
1915                         goto journal_read_write;
1916                 } else {
1917                         sector_t next_sector;
1918                         journal_read_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
1919                         if (likely(journal_read_pos == NOT_FOUND)) {
1920                                 if (unlikely(dio->range.n_sectors > next_sector - dio->range.logical_sector))
1921                                         dio->range.n_sectors = next_sector - dio->range.logical_sector;
1922                         } else {
1923                                 unsigned i;
1924                                 unsigned jp = journal_read_pos + 1;
1925                                 for (i = ic->sectors_per_block; i < dio->range.n_sectors; i += ic->sectors_per_block, jp++) {
1926                                         if (!test_journal_node(ic, jp, dio->range.logical_sector + i))
1927                                                 break;
1928                                 }
1929                                 dio->range.n_sectors = i;
1930                         }
1931                 }
1932         }
1933         if (unlikely(!add_new_range(ic, &dio->range, true))) {
1934                 /*
1935                  * We must not sleep in the request routine because it could
1936                  * stall bios on current->bio_list.
1937                  * So, we offload the bio to a workqueue if we have to sleep.
1938                  */
1939                 if (from_map) {
1940 offload_to_thread:
1941                         spin_unlock_irq(&ic->endio_wait.lock);
1942                         INIT_WORK(&dio->work, integrity_bio_wait);
1943                         queue_work(ic->wait_wq, &dio->work);
1944                         return;
1945                 }
1946                 wait_and_add_new_range(ic, &dio->range);
1947         }
1948         spin_unlock_irq(&ic->endio_wait.lock);
1949
1950         if (unlikely(journal_read_pos != NOT_FOUND)) {
1951                 journal_section = journal_read_pos / ic->journal_section_entries;
1952                 journal_entry = journal_read_pos % ic->journal_section_entries;
1953                 goto journal_read_write;
1954         }
1955
1956         if (ic->mode == 'B' && dio->write) {
1957                 if (!block_bitmap_op(ic, ic->may_write_bitmap, dio->range.logical_sector, dio->range.n_sectors, BITMAP_OP_TEST_ALL_SET)) {
1958                         struct bitmap_block_status *bbs = sector_to_bitmap_block(ic, dio->range.logical_sector);
1959
1960                         spin_lock(&bbs->bio_queue_lock);
1961                         bio_list_add(&bbs->bio_queue, bio);
1962                         spin_unlock(&bbs->bio_queue_lock);
1963
1964                         queue_work(ic->writer_wq, &bbs->work);
1965
1966                         return;
1967                 }
1968         }
1969
1970         dio->in_flight = (atomic_t)ATOMIC_INIT(2);
1971
1972         if (need_sync_io) {
1973                 init_completion(&read_comp);
1974                 dio->completion = &read_comp;
1975         } else
1976                 dio->completion = NULL;
1977
1978         dio->orig_bi_iter = bio->bi_iter;
1979
1980         dio->orig_bi_disk = bio->bi_disk;
1981         dio->orig_bi_partno = bio->bi_partno;
1982         bio_set_dev(bio, ic->dev->bdev);
1983
1984         dio->orig_bi_integrity = bio_integrity(bio);
1985         bio->bi_integrity = NULL;
1986         bio->bi_opf &= ~REQ_INTEGRITY;
1987
1988         dio->orig_bi_end_io = bio->bi_end_io;
1989         bio->bi_end_io = integrity_end_io;
1990
1991         bio->bi_iter.bi_size = dio->range.n_sectors << SECTOR_SHIFT;
1992         generic_make_request(bio);
1993
1994         if (need_sync_io) {
1995                 wait_for_completion_io(&read_comp);
1996                 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING) &&
1997                     dio->range.logical_sector + dio->range.n_sectors > le64_to_cpu(ic->sb->recalc_sector))
1998                         goto skip_check;
1999                 if (ic->mode == 'B') {
2000                         if (!block_bitmap_op(ic, ic->recalc_bitmap, dio->range.logical_sector, dio->range.n_sectors, BITMAP_OP_TEST_ALL_CLEAR))
2001                                 goto skip_check;
2002                 }
2003
2004                 if (likely(!bio->bi_status))
2005                         integrity_metadata(&dio->work);
2006                 else
2007 skip_check:
2008                         dec_in_flight(dio);
2009
2010         } else {
2011                 INIT_WORK(&dio->work, integrity_metadata);
2012                 queue_work(ic->metadata_wq, &dio->work);
2013         }
2014
2015         return;
2016
2017 journal_read_write:
2018         if (unlikely(__journal_read_write(dio, bio, journal_section, journal_entry)))
2019                 goto lock_retry;
2020
2021         do_endio_flush(ic, dio);
2022 }
2023
2024
2025 static void integrity_bio_wait(struct work_struct *w)
2026 {
2027         struct dm_integrity_io *dio = container_of(w, struct dm_integrity_io, work);
2028
2029         dm_integrity_map_continue(dio, false);
2030 }
2031
2032 static void pad_uncommitted(struct dm_integrity_c *ic)
2033 {
2034         if (ic->free_section_entry) {
2035                 ic->free_sectors -= ic->journal_section_entries - ic->free_section_entry;
2036                 ic->free_section_entry = 0;
2037                 ic->free_section++;
2038                 wraparound_section(ic, &ic->free_section);
2039                 ic->n_uncommitted_sections++;
2040         }
2041         if (WARN_ON(ic->journal_sections * ic->journal_section_entries !=
2042                 (ic->n_uncommitted_sections + ic->n_committed_sections) * ic->journal_section_entries + ic->free_sectors)) {
2043                 printk(KERN_CRIT "dm-integrity: "
2044                         "journal_sections %u, "
2045                         "journal_section_entries %u, "
2046                         "n_uncommitted_sections %u, "
2047                         "n_committed_sections %u, "
2048                         "journal_section_entries %u, "
2049                         "free_sectors %u\n",
2050                         ic->journal_sections,
2051                         ic->journal_section_entries,
2052                         ic->n_uncommitted_sections,
2053                         ic->n_committed_sections,
2054                         ic->journal_section_entries,
2055                         ic->free_sectors);
2056         }
2057 }
2058
2059 static void integrity_commit(struct work_struct *w)
2060 {
2061         struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, commit_work);
2062         unsigned commit_start, commit_sections;
2063         unsigned i, j, n;
2064         struct bio *flushes;
2065
2066         del_timer(&ic->autocommit_timer);
2067
2068         spin_lock_irq(&ic->endio_wait.lock);
2069         flushes = bio_list_get(&ic->flush_bio_list);
2070         if (unlikely(ic->mode != 'J')) {
2071                 spin_unlock_irq(&ic->endio_wait.lock);
2072                 dm_integrity_flush_buffers(ic);
2073                 goto release_flush_bios;
2074         }
2075
2076         pad_uncommitted(ic);
2077         commit_start = ic->uncommitted_section;
2078         commit_sections = ic->n_uncommitted_sections;
2079         spin_unlock_irq(&ic->endio_wait.lock);
2080
2081         if (!commit_sections)
2082                 goto release_flush_bios;
2083
2084         i = commit_start;
2085         for (n = 0; n < commit_sections; n++) {
2086                 for (j = 0; j < ic->journal_section_entries; j++) {
2087                         struct journal_entry *je;
2088                         je = access_journal_entry(ic, i, j);
2089                         io_wait_event(ic->copy_to_journal_wait, !journal_entry_is_inprogress(je));
2090                 }
2091                 for (j = 0; j < ic->journal_section_sectors; j++) {
2092                         struct journal_sector *js;
2093                         js = access_journal(ic, i, j);
2094                         js->commit_id = dm_integrity_commit_id(ic, i, j, ic->commit_seq);
2095                 }
2096                 i++;
2097                 if (unlikely(i >= ic->journal_sections))
2098                         ic->commit_seq = next_commit_seq(ic->commit_seq);
2099                 wraparound_section(ic, &i);
2100         }
2101         smp_rmb();
2102
2103         write_journal(ic, commit_start, commit_sections);
2104
2105         spin_lock_irq(&ic->endio_wait.lock);
2106         ic->uncommitted_section += commit_sections;
2107         wraparound_section(ic, &ic->uncommitted_section);
2108         ic->n_uncommitted_sections -= commit_sections;
2109         ic->n_committed_sections += commit_sections;
2110         spin_unlock_irq(&ic->endio_wait.lock);
2111
2112         if (READ_ONCE(ic->free_sectors) <= ic->free_sectors_threshold)
2113                 queue_work(ic->writer_wq, &ic->writer_work);
2114
2115 release_flush_bios:
2116         while (flushes) {
2117                 struct bio *next = flushes->bi_next;
2118                 flushes->bi_next = NULL;
2119                 do_endio(ic, flushes);
2120                 flushes = next;
2121         }
2122 }
2123
2124 static void complete_copy_from_journal(unsigned long error, void *context)
2125 {
2126         struct journal_io *io = context;
2127         struct journal_completion *comp = io->comp;
2128         struct dm_integrity_c *ic = comp->ic;
2129         remove_range(ic, &io->range);
2130         mempool_free(io, &ic->journal_io_mempool);
2131         if (unlikely(error != 0))
2132                 dm_integrity_io_error(ic, "copying from journal", -EIO);
2133         complete_journal_op(comp);
2134 }
2135
2136 static void restore_last_bytes(struct dm_integrity_c *ic, struct journal_sector *js,
2137                                struct journal_entry *je)
2138 {
2139         unsigned s = 0;
2140         do {
2141                 js->commit_id = je->last_bytes[s];
2142                 js++;
2143         } while (++s < ic->sectors_per_block);
2144 }
2145
2146 static void do_journal_write(struct dm_integrity_c *ic, unsigned write_start,
2147                              unsigned write_sections, bool from_replay)
2148 {
2149         unsigned i, j, n;
2150         struct journal_completion comp;
2151         struct blk_plug plug;
2152
2153         blk_start_plug(&plug);
2154
2155         comp.ic = ic;
2156         comp.in_flight = (atomic_t)ATOMIC_INIT(1);
2157         init_completion(&comp.comp);
2158
2159         i = write_start;
2160         for (n = 0; n < write_sections; n++, i++, wraparound_section(ic, &i)) {
2161 #ifndef INTERNAL_VERIFY
2162                 if (unlikely(from_replay))
2163 #endif
2164                         rw_section_mac(ic, i, false);
2165                 for (j = 0; j < ic->journal_section_entries; j++) {
2166                         struct journal_entry *je = access_journal_entry(ic, i, j);
2167                         sector_t sec, area, offset;
2168                         unsigned k, l, next_loop;
2169                         sector_t metadata_block;
2170                         unsigned metadata_offset;
2171                         struct journal_io *io;
2172
2173                         if (journal_entry_is_unused(je))
2174                                 continue;
2175                         BUG_ON(unlikely(journal_entry_is_inprogress(je)) && !from_replay);
2176                         sec = journal_entry_get_sector(je);
2177                         if (unlikely(from_replay)) {
2178                                 if (unlikely(sec & (unsigned)(ic->sectors_per_block - 1))) {
2179                                         dm_integrity_io_error(ic, "invalid sector in journal", -EIO);
2180                                         sec &= ~(sector_t)(ic->sectors_per_block - 1);
2181                                 }
2182                         }
2183                         get_area_and_offset(ic, sec, &area, &offset);
2184                         restore_last_bytes(ic, access_journal_data(ic, i, j), je);
2185                         for (k = j + 1; k < ic->journal_section_entries; k++) {
2186                                 struct journal_entry *je2 = access_journal_entry(ic, i, k);
2187                                 sector_t sec2, area2, offset2;
2188                                 if (journal_entry_is_unused(je2))
2189                                         break;
2190                                 BUG_ON(unlikely(journal_entry_is_inprogress(je2)) && !from_replay);
2191                                 sec2 = journal_entry_get_sector(je2);
2192                                 get_area_and_offset(ic, sec2, &area2, &offset2);
2193                                 if (area2 != area || offset2 != offset + ((k - j) << ic->sb->log2_sectors_per_block))
2194                                         break;
2195                                 restore_last_bytes(ic, access_journal_data(ic, i, k), je2);
2196                         }
2197                         next_loop = k - 1;
2198
2199                         io = mempool_alloc(&ic->journal_io_mempool, GFP_NOIO);
2200                         io->comp = &comp;
2201                         io->range.logical_sector = sec;
2202                         io->range.n_sectors = (k - j) << ic->sb->log2_sectors_per_block;
2203
2204                         spin_lock_irq(&ic->endio_wait.lock);
2205                         add_new_range_and_wait(ic, &io->range);
2206
2207                         if (likely(!from_replay)) {
2208                                 struct journal_node *section_node = &ic->journal_tree[i * ic->journal_section_entries];
2209
2210                                 /* don't write if there is newer committed sector */
2211                                 while (j < k && find_newer_committed_node(ic, &section_node[j])) {
2212                                         struct journal_entry *je2 = access_journal_entry(ic, i, j);
2213
2214                                         journal_entry_set_unused(je2);
2215                                         remove_journal_node(ic, &section_node[j]);
2216                                         j++;
2217                                         sec += ic->sectors_per_block;
2218                                         offset += ic->sectors_per_block;
2219                                 }
2220                                 while (j < k && find_newer_committed_node(ic, &section_node[k - 1])) {
2221                                         struct journal_entry *je2 = access_journal_entry(ic, i, k - 1);
2222
2223                                         journal_entry_set_unused(je2);
2224                                         remove_journal_node(ic, &section_node[k - 1]);
2225                                         k--;
2226                                 }
2227                                 if (j == k) {
2228                                         remove_range_unlocked(ic, &io->range);
2229                                         spin_unlock_irq(&ic->endio_wait.lock);
2230                                         mempool_free(io, &ic->journal_io_mempool);
2231                                         goto skip_io;
2232                                 }
2233                                 for (l = j; l < k; l++) {
2234                                         remove_journal_node(ic, &section_node[l]);
2235                                 }
2236                         }
2237                         spin_unlock_irq(&ic->endio_wait.lock);
2238
2239                         metadata_block = get_metadata_sector_and_offset(ic, area, offset, &metadata_offset);
2240                         for (l = j; l < k; l++) {
2241                                 int r;
2242                                 struct journal_entry *je2 = access_journal_entry(ic, i, l);
2243
2244                                 if (
2245 #ifndef INTERNAL_VERIFY
2246                                     unlikely(from_replay) &&
2247 #endif
2248                                     ic->internal_hash) {
2249                                         char test_tag[max_t(size_t, HASH_MAX_DIGESTSIZE, MAX_TAG_SIZE)];
2250
2251                                         integrity_sector_checksum(ic, sec + ((l - j) << ic->sb->log2_sectors_per_block),
2252                                                                   (char *)access_journal_data(ic, i, l), test_tag);
2253                                         if (unlikely(memcmp(test_tag, journal_entry_tag(ic, je2), ic->tag_size)))
2254                                                 dm_integrity_io_error(ic, "tag mismatch when replaying journal", -EILSEQ);
2255                                 }
2256
2257                                 journal_entry_set_unused(je2);
2258                                 r = dm_integrity_rw_tag(ic, journal_entry_tag(ic, je2), &metadata_block, &metadata_offset,
2259                                                         ic->tag_size, TAG_WRITE);
2260                                 if (unlikely(r)) {
2261                                         dm_integrity_io_error(ic, "reading tags", r);
2262                                 }
2263                         }
2264
2265                         atomic_inc(&comp.in_flight);
2266                         copy_from_journal(ic, i, j << ic->sb->log2_sectors_per_block,
2267                                           (k - j) << ic->sb->log2_sectors_per_block,
2268                                           get_data_sector(ic, area, offset),
2269                                           complete_copy_from_journal, io);
2270 skip_io:
2271                         j = next_loop;
2272                 }
2273         }
2274
2275         dm_bufio_write_dirty_buffers_async(ic->bufio);
2276
2277         blk_finish_plug(&plug);
2278
2279         complete_journal_op(&comp);
2280         wait_for_completion_io(&comp.comp);
2281
2282         dm_integrity_flush_buffers(ic);
2283 }
2284
2285 static void integrity_writer(struct work_struct *w)
2286 {
2287         struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, writer_work);
2288         unsigned write_start, write_sections;
2289
2290         unsigned prev_free_sectors;
2291
2292         /* the following test is not needed, but it tests the replay code */
2293         if (READ_ONCE(ic->suspending) && !ic->meta_dev)
2294                 return;
2295
2296         spin_lock_irq(&ic->endio_wait.lock);
2297         write_start = ic->committed_section;
2298         write_sections = ic->n_committed_sections;
2299         spin_unlock_irq(&ic->endio_wait.lock);
2300
2301         if (!write_sections)
2302                 return;
2303
2304         do_journal_write(ic, write_start, write_sections, false);
2305
2306         spin_lock_irq(&ic->endio_wait.lock);
2307
2308         ic->committed_section += write_sections;
2309         wraparound_section(ic, &ic->committed_section);
2310         ic->n_committed_sections -= write_sections;
2311
2312         prev_free_sectors = ic->free_sectors;
2313         ic->free_sectors += write_sections * ic->journal_section_entries;
2314         if (unlikely(!prev_free_sectors))
2315                 wake_up_locked(&ic->endio_wait);
2316
2317         spin_unlock_irq(&ic->endio_wait.lock);
2318 }
2319
2320 static void recalc_write_super(struct dm_integrity_c *ic)
2321 {
2322         int r;
2323
2324         dm_integrity_flush_buffers(ic);
2325         if (dm_integrity_failed(ic))
2326                 return;
2327
2328         sb_set_version(ic);
2329         r = sync_rw_sb(ic, REQ_OP_WRITE, 0);
2330         if (unlikely(r))
2331                 dm_integrity_io_error(ic, "writing superblock", r);
2332 }
2333
2334 static void integrity_recalc(struct work_struct *w)
2335 {
2336         struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, recalc_work);
2337         struct dm_integrity_range range;
2338         struct dm_io_request io_req;
2339         struct dm_io_region io_loc;
2340         sector_t area, offset;
2341         sector_t metadata_block;
2342         unsigned metadata_offset;
2343         sector_t logical_sector, n_sectors;
2344         __u8 *t;
2345         unsigned i;
2346         int r;
2347         unsigned super_counter = 0;
2348
2349         DEBUG_print("start recalculation... (position %llx)\n", le64_to_cpu(ic->sb->recalc_sector));
2350
2351         spin_lock_irq(&ic->endio_wait.lock);
2352
2353 next_chunk:
2354
2355         if (unlikely(READ_ONCE(ic->suspending)))
2356                 goto unlock_ret;
2357
2358         range.logical_sector = le64_to_cpu(ic->sb->recalc_sector);
2359         if (unlikely(range.logical_sector >= ic->provided_data_sectors)) {
2360                 if (ic->mode == 'B') {
2361                         DEBUG_print("queue_delayed_work: bitmap_flush_work\n");
2362                         queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, 0);
2363                 }
2364                 goto unlock_ret;
2365         }
2366
2367         get_area_and_offset(ic, range.logical_sector, &area, &offset);
2368         range.n_sectors = min((sector_t)RECALC_SECTORS, ic->provided_data_sectors - range.logical_sector);
2369         if (!ic->meta_dev)
2370                 range.n_sectors = min(range.n_sectors, ((sector_t)1U << ic->sb->log2_interleave_sectors) - (unsigned)offset);
2371
2372         add_new_range_and_wait(ic, &range);
2373         spin_unlock_irq(&ic->endio_wait.lock);
2374         logical_sector = range.logical_sector;
2375         n_sectors = range.n_sectors;
2376
2377         if (ic->mode == 'B') {
2378                 if (block_bitmap_op(ic, ic->recalc_bitmap, logical_sector, n_sectors, BITMAP_OP_TEST_ALL_CLEAR)) {
2379                         goto advance_and_next;
2380                 }
2381                 while (block_bitmap_op(ic, ic->recalc_bitmap, logical_sector, ic->sectors_per_block, BITMAP_OP_TEST_ALL_CLEAR)) {
2382                         logical_sector += ic->sectors_per_block;
2383                         n_sectors -= ic->sectors_per_block;
2384                         cond_resched();
2385                 }
2386                 while (block_bitmap_op(ic, ic->recalc_bitmap, logical_sector + n_sectors - ic->sectors_per_block, ic->sectors_per_block, BITMAP_OP_TEST_ALL_CLEAR)) {
2387                         n_sectors -= ic->sectors_per_block;
2388                         cond_resched();
2389                 }
2390                 get_area_and_offset(ic, logical_sector, &area, &offset);
2391         }
2392
2393         DEBUG_print("recalculating: %lx, %lx\n", logical_sector, n_sectors);
2394
2395         if (unlikely(++super_counter == RECALC_WRITE_SUPER)) {
2396                 recalc_write_super(ic);
2397                 if (ic->mode == 'B') {
2398                         queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, ic->bitmap_flush_interval);
2399                 }
2400                 super_counter = 0;
2401         }
2402
2403         if (unlikely(dm_integrity_failed(ic)))
2404                 goto err;
2405
2406         io_req.bi_op = REQ_OP_READ;
2407         io_req.bi_op_flags = 0;
2408         io_req.mem.type = DM_IO_VMA;
2409         io_req.mem.ptr.addr = ic->recalc_buffer;
2410         io_req.notify.fn = NULL;
2411         io_req.client = ic->io;
2412         io_loc.bdev = ic->dev->bdev;
2413         io_loc.sector = get_data_sector(ic, area, offset);
2414         io_loc.count = n_sectors;
2415
2416         r = dm_io(&io_req, 1, &io_loc, NULL);
2417         if (unlikely(r)) {
2418                 dm_integrity_io_error(ic, "reading data", r);
2419                 goto err;
2420         }
2421
2422         t = ic->recalc_tags;
2423         for (i = 0; i < n_sectors; i += ic->sectors_per_block) {
2424                 integrity_sector_checksum(ic, logical_sector + i, ic->recalc_buffer + (i << SECTOR_SHIFT), t);
2425                 t += ic->tag_size;
2426         }
2427
2428         metadata_block = get_metadata_sector_and_offset(ic, area, offset, &metadata_offset);
2429
2430         r = dm_integrity_rw_tag(ic, ic->recalc_tags, &metadata_block, &metadata_offset, t - ic->recalc_tags, TAG_WRITE);
2431         if (unlikely(r)) {
2432                 dm_integrity_io_error(ic, "writing tags", r);
2433                 goto err;
2434         }
2435
2436 advance_and_next:
2437         cond_resched();
2438
2439         spin_lock_irq(&ic->endio_wait.lock);
2440         remove_range_unlocked(ic, &range);
2441         ic->sb->recalc_sector = cpu_to_le64(range.logical_sector + range.n_sectors);
2442         goto next_chunk;
2443
2444 err:
2445         remove_range(ic, &range);
2446         return;
2447
2448 unlock_ret:
2449         spin_unlock_irq(&ic->endio_wait.lock);
2450
2451         recalc_write_super(ic);
2452 }
2453
2454 static void bitmap_block_work(struct work_struct *w)
2455 {
2456         struct bitmap_block_status *bbs = container_of(w, struct bitmap_block_status, work);
2457         struct dm_integrity_c *ic = bbs->ic;
2458         struct bio *bio;
2459         struct bio_list bio_queue;
2460         struct bio_list waiting;
2461
2462         bio_list_init(&waiting);
2463
2464         spin_lock(&bbs->bio_queue_lock);
2465         bio_queue = bbs->bio_queue;
2466         bio_list_init(&bbs->bio_queue);
2467         spin_unlock(&bbs->bio_queue_lock);
2468
2469         while ((bio = bio_list_pop(&bio_queue))) {
2470                 struct dm_integrity_io *dio;
2471
2472                 dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
2473
2474                 if (block_bitmap_op(ic, ic->may_write_bitmap, dio->range.logical_sector, dio->range.n_sectors, BITMAP_OP_TEST_ALL_SET)) {
2475                         remove_range(ic, &dio->range);
2476                         INIT_WORK(&dio->work, integrity_bio_wait);
2477                         queue_work(ic->wait_wq, &dio->work);
2478                 } else {
2479                         block_bitmap_op(ic, ic->journal, dio->range.logical_sector, dio->range.n_sectors, BITMAP_OP_SET);
2480                         bio_list_add(&waiting, bio);
2481                 }
2482         }
2483
2484         if (bio_list_empty(&waiting))
2485                 return;
2486
2487         rw_journal_sectors(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, bbs->idx * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), BITMAP_BLOCK_SIZE >> SECTOR_SHIFT, NULL);
2488
2489         while ((bio = bio_list_pop(&waiting))) {
2490                 struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
2491
2492                 block_bitmap_op(ic, ic->may_write_bitmap, dio->range.logical_sector, dio->range.n_sectors, BITMAP_OP_SET);
2493
2494                 remove_range(ic, &dio->range);
2495                 INIT_WORK(&dio->work, integrity_bio_wait);
2496                 queue_work(ic->wait_wq, &dio->work);
2497         }
2498
2499         queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, ic->bitmap_flush_interval);
2500 }
2501
2502 static void bitmap_flush_work(struct work_struct *work)
2503 {
2504         struct dm_integrity_c *ic = container_of(work, struct dm_integrity_c, bitmap_flush_work.work);
2505         struct dm_integrity_range range;
2506         unsigned long limit;
2507         struct bio *bio;
2508
2509         dm_integrity_flush_buffers(ic);
2510
2511         range.logical_sector = 0;
2512         range.n_sectors = ic->provided_data_sectors;
2513
2514         spin_lock_irq(&ic->endio_wait.lock);
2515         add_new_range_and_wait(ic, &range);
2516         spin_unlock_irq(&ic->endio_wait.lock);
2517
2518         dm_integrity_flush_buffers(ic);
2519         if (ic->meta_dev)
2520                 blkdev_issue_flush(ic->dev->bdev, GFP_NOIO, NULL);
2521
2522         limit = ic->provided_data_sectors;
2523         if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
2524                 limit = le64_to_cpu(ic->sb->recalc_sector)
2525                         >> (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit)
2526                         << (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
2527         }
2528         /*DEBUG_print("zeroing journal\n");*/
2529         block_bitmap_op(ic, ic->journal, 0, limit, BITMAP_OP_CLEAR);
2530         block_bitmap_op(ic, ic->may_write_bitmap, 0, limit, BITMAP_OP_CLEAR);
2531
2532         rw_journal_sectors(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, 0, ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
2533
2534         spin_lock_irq(&ic->endio_wait.lock);
2535         remove_range_unlocked(ic, &range);
2536         while (unlikely((bio = bio_list_pop(&ic->synchronous_bios)) != NULL)) {
2537                 bio_endio(bio);
2538                 spin_unlock_irq(&ic->endio_wait.lock);
2539                 spin_lock_irq(&ic->endio_wait.lock);
2540         }
2541         spin_unlock_irq(&ic->endio_wait.lock);
2542 }
2543
2544
2545 static void init_journal(struct dm_integrity_c *ic, unsigned start_section,
2546                          unsigned n_sections, unsigned char commit_seq)
2547 {
2548         unsigned i, j, n;
2549
2550         if (!n_sections)
2551                 return;
2552
2553         for (n = 0; n < n_sections; n++) {
2554                 i = start_section + n;
2555                 wraparound_section(ic, &i);
2556                 for (j = 0; j < ic->journal_section_sectors; j++) {
2557                         struct journal_sector *js = access_journal(ic, i, j);
2558                         memset(&js->entries, 0, JOURNAL_SECTOR_DATA);
2559                         js->commit_id = dm_integrity_commit_id(ic, i, j, commit_seq);
2560                 }
2561                 for (j = 0; j < ic->journal_section_entries; j++) {
2562                         struct journal_entry *je = access_journal_entry(ic, i, j);
2563                         journal_entry_set_unused(je);
2564                 }
2565         }
2566
2567         write_journal(ic, start_section, n_sections);
2568 }
2569
2570 static int find_commit_seq(struct dm_integrity_c *ic, unsigned i, unsigned j, commit_id_t id)
2571 {
2572         unsigned char k;
2573         for (k = 0; k < N_COMMIT_IDS; k++) {
2574                 if (dm_integrity_commit_id(ic, i, j, k) == id)
2575                         return k;
2576         }
2577         dm_integrity_io_error(ic, "journal commit id", -EIO);
2578         return -EIO;
2579 }
2580
2581 static void replay_journal(struct dm_integrity_c *ic)
2582 {
2583         unsigned i, j;
2584         bool used_commit_ids[N_COMMIT_IDS];
2585         unsigned max_commit_id_sections[N_COMMIT_IDS];
2586         unsigned write_start, write_sections;
2587         unsigned continue_section;
2588         bool journal_empty;
2589         unsigned char unused, last_used, want_commit_seq;
2590
2591         if (ic->mode == 'R')
2592                 return;
2593
2594         if (ic->journal_uptodate)
2595                 return;
2596
2597         last_used = 0;
2598         write_start = 0;
2599
2600         if (!ic->just_formatted) {
2601                 DEBUG_print("reading journal\n");
2602                 rw_journal(ic, REQ_OP_READ, 0, 0, ic->journal_sections, NULL);
2603                 if (ic->journal_io)
2604                         DEBUG_bytes(lowmem_page_address(ic->journal_io[0].page), 64, "read journal");
2605                 if (ic->journal_io) {
2606                         struct journal_completion crypt_comp;
2607                         crypt_comp.ic = ic;
2608                         init_completion(&crypt_comp.comp);
2609                         crypt_comp.in_flight = (atomic_t)ATOMIC_INIT(0);
2610                         encrypt_journal(ic, false, 0, ic->journal_sections, &crypt_comp);
2611                         wait_for_completion(&crypt_comp.comp);
2612                 }
2613                 DEBUG_bytes(lowmem_page_address(ic->journal[0].page), 64, "decrypted journal");
2614         }
2615
2616         if (dm_integrity_failed(ic))
2617                 goto clear_journal;
2618
2619         journal_empty = true;
2620         memset(used_commit_ids, 0, sizeof used_commit_ids);
2621         memset(max_commit_id_sections, 0, sizeof max_commit_id_sections);
2622         for (i = 0; i < ic->journal_sections; i++) {
2623                 for (j = 0; j < ic->journal_section_sectors; j++) {
2624                         int k;
2625                         struct journal_sector *js = access_journal(ic, i, j);
2626                         k = find_commit_seq(ic, i, j, js->commit_id);
2627                         if (k < 0)
2628                                 goto clear_journal;
2629                         used_commit_ids[k] = true;
2630                         max_commit_id_sections[k] = i;
2631                 }
2632                 if (journal_empty) {
2633                         for (j = 0; j < ic->journal_section_entries; j++) {
2634                                 struct journal_entry *je = access_journal_entry(ic, i, j);
2635                                 if (!journal_entry_is_unused(je)) {
2636                                         journal_empty = false;
2637                                         break;
2638                                 }
2639                         }
2640                 }
2641         }
2642
2643         if (!used_commit_ids[N_COMMIT_IDS - 1]) {
2644                 unused = N_COMMIT_IDS - 1;
2645                 while (unused && !used_commit_ids[unused - 1])
2646                         unused--;
2647         } else {
2648                 for (unused = 0; unused < N_COMMIT_IDS; unused++)
2649                         if (!used_commit_ids[unused])
2650                                 break;
2651                 if (unused == N_COMMIT_IDS) {
2652                         dm_integrity_io_error(ic, "journal commit ids", -EIO);
2653                         goto clear_journal;
2654                 }
2655         }
2656         DEBUG_print("first unused commit seq %d [%d,%d,%d,%d]\n",
2657                     unused, used_commit_ids[0], used_commit_ids[1],
2658                     used_commit_ids[2], used_commit_ids[3]);
2659
2660         last_used = prev_commit_seq(unused);
2661         want_commit_seq = prev_commit_seq(last_used);
2662
2663         if (!used_commit_ids[want_commit_seq] && used_commit_ids[prev_commit_seq(want_commit_seq)])
2664                 journal_empty = true;
2665
2666         write_start = max_commit_id_sections[last_used] + 1;
2667         if (unlikely(write_start >= ic->journal_sections))
2668                 want_commit_seq = next_commit_seq(want_commit_seq);
2669         wraparound_section(ic, &write_start);
2670
2671         i = write_start;
2672         for (write_sections = 0; write_sections < ic->journal_sections; write_sections++) {
2673                 for (j = 0; j < ic->journal_section_sectors; j++) {
2674                         struct journal_sector *js = access_journal(ic, i, j);
2675
2676                         if (js->commit_id != dm_integrity_commit_id(ic, i, j, want_commit_seq)) {
2677                                 /*
2678                                  * This could be caused by crash during writing.
2679                                  * We won't replay the inconsistent part of the
2680                                  * journal.
2681                                  */
2682                                 DEBUG_print("commit id mismatch at position (%u, %u): %d != %d\n",
2683                                             i, j, find_commit_seq(ic, i, j, js->commit_id), want_commit_seq);
2684                                 goto brk;
2685                         }
2686                 }
2687                 i++;
2688                 if (unlikely(i >= ic->journal_sections))
2689                         want_commit_seq = next_commit_seq(want_commit_seq);
2690                 wraparound_section(ic, &i);
2691         }
2692 brk:
2693
2694         if (!journal_empty) {
2695                 DEBUG_print("replaying %u sections, starting at %u, commit seq %d\n",
2696                             write_sections, write_start, want_commit_seq);
2697                 do_journal_write(ic, write_start, write_sections, true);
2698         }
2699
2700         if (write_sections == ic->journal_sections && (ic->mode == 'J' || journal_empty)) {
2701                 continue_section = write_start;
2702                 ic->commit_seq = want_commit_seq;
2703                 DEBUG_print("continuing from section %u, commit seq %d\n", write_start, ic->commit_seq);
2704         } else {
2705                 unsigned s;
2706                 unsigned char erase_seq;
2707 clear_journal:
2708                 DEBUG_print("clearing journal\n");
2709
2710                 erase_seq = prev_commit_seq(prev_commit_seq(last_used));
2711                 s = write_start;
2712                 init_journal(ic, s, 1, erase_seq);
2713                 s++;
2714                 wraparound_section(ic, &s);
2715                 if (ic->journal_sections >= 2) {
2716                         init_journal(ic, s, ic->journal_sections - 2, erase_seq);
2717                         s += ic->journal_sections - 2;
2718                         wraparound_section(ic, &s);
2719                         init_journal(ic, s, 1, erase_seq);
2720                 }
2721
2722                 continue_section = 0;
2723                 ic->commit_seq = next_commit_seq(erase_seq);
2724         }
2725
2726         ic->committed_section = continue_section;
2727         ic->n_committed_sections = 0;
2728
2729         ic->uncommitted_section = continue_section;
2730         ic->n_uncommitted_sections = 0;
2731
2732         ic->free_section = continue_section;
2733         ic->free_section_entry = 0;
2734         ic->free_sectors = ic->journal_entries;
2735
2736         ic->journal_tree_root = RB_ROOT;
2737         for (i = 0; i < ic->journal_entries; i++)
2738                 init_journal_node(&ic->journal_tree[i]);
2739 }
2740
2741 static void dm_integrity_enter_synchronous_mode(struct dm_integrity_c *ic)
2742 {
2743         DEBUG_print("dm_integrity_enter_synchronous_mode\n");
2744
2745         if (ic->mode == 'B') {
2746                 ic->bitmap_flush_interval = msecs_to_jiffies(10) + 1;
2747                 ic->synchronous_mode = 1;
2748
2749                 cancel_delayed_work_sync(&ic->bitmap_flush_work);
2750                 queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, 0);
2751                 flush_workqueue(ic->commit_wq);
2752         }
2753 }
2754
2755 static int dm_integrity_reboot(struct notifier_block *n, unsigned long code, void *x)
2756 {
2757         struct dm_integrity_c *ic = container_of(n, struct dm_integrity_c, reboot_notifier);
2758
2759         DEBUG_print("dm_integrity_reboot\n");
2760
2761         dm_integrity_enter_synchronous_mode(ic);
2762
2763         return NOTIFY_DONE;
2764 }
2765
2766 static void dm_integrity_postsuspend(struct dm_target *ti)
2767 {
2768         struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
2769         int r;
2770
2771         WARN_ON(unregister_reboot_notifier(&ic->reboot_notifier));
2772
2773         del_timer_sync(&ic->autocommit_timer);
2774
2775         WRITE_ONCE(ic->suspending, 1);
2776
2777         if (ic->recalc_wq)
2778                 drain_workqueue(ic->recalc_wq);
2779
2780         if (ic->mode == 'B')
2781                 cancel_delayed_work_sync(&ic->bitmap_flush_work);
2782
2783         queue_work(ic->commit_wq, &ic->commit_work);
2784         drain_workqueue(ic->commit_wq);
2785
2786         if (ic->mode == 'J') {
2787                 if (ic->meta_dev)
2788                         queue_work(ic->writer_wq, &ic->writer_work);
2789                 drain_workqueue(ic->writer_wq);
2790                 dm_integrity_flush_buffers(ic);
2791         }
2792
2793         if (ic->mode == 'B') {
2794                 dm_integrity_flush_buffers(ic);
2795 #if 1
2796                 init_journal(ic, 0, ic->journal_sections, 0);
2797                 ic->sb->flags &= ~cpu_to_le32(SB_FLAG_DIRTY_BITMAP);
2798                 r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
2799                 if (unlikely(r))
2800                         dm_integrity_io_error(ic, "writing superblock", r);
2801 #endif
2802         }
2803
2804         WRITE_ONCE(ic->suspending, 0);
2805
2806         BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
2807
2808         ic->journal_uptodate = true;
2809 }
2810
2811 static void dm_integrity_resume(struct dm_target *ti)
2812 {
2813         struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
2814         int r;
2815         DEBUG_print("resume\n");
2816
2817         if (ic->sb->flags & cpu_to_le32(SB_FLAG_DIRTY_BITMAP)) {
2818                 DEBUG_print("resume dirty_bitmap\n");
2819                 rw_journal_sectors(ic, REQ_OP_READ, 0, 0, ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
2820                 if (ic->mode == 'B') {
2821                         if (ic->sb->log2_blocks_per_bitmap_bit == ic->log2_blocks_per_bitmap_bit) {
2822                                 block_bitmap_copy(ic, ic->recalc_bitmap, ic->journal);
2823                                 block_bitmap_copy(ic, ic->may_write_bitmap, ic->journal);
2824                                 if (!block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors, BITMAP_OP_TEST_ALL_CLEAR)) {
2825                                         ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
2826                                         ic->sb->recalc_sector = cpu_to_le64(0);
2827                                 }
2828                         } else {
2829                                 DEBUG_print("non-matching blocks_per_bitmap_bit: %u, %u\n", ic->sb->log2_blocks_per_bitmap_bit, ic->log2_blocks_per_bitmap_bit);
2830                                 ic->sb->log2_blocks_per_bitmap_bit = ic->log2_blocks_per_bitmap_bit;
2831                                 block_bitmap_op(ic, ic->recalc_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_SET);
2832                                 block_bitmap_op(ic, ic->may_write_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_SET);
2833                                 block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors, BITMAP_OP_SET);
2834                                 rw_journal_sectors(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, 0, ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
2835                                 ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
2836                                 ic->sb->recalc_sector = cpu_to_le64(0);
2837                         }
2838                 } else {
2839                         if (!(ic->sb->log2_blocks_per_bitmap_bit == ic->log2_blocks_per_bitmap_bit &&
2840                               block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors, BITMAP_OP_TEST_ALL_CLEAR))) {
2841                                 ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
2842                                 ic->sb->recalc_sector = cpu_to_le64(0);
2843                         }
2844                         init_journal(ic, 0, ic->journal_sections, 0);
2845                         replay_journal(ic);
2846                         ic->sb->flags &= ~cpu_to_le32(SB_FLAG_DIRTY_BITMAP);
2847                 }
2848                 r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
2849                 if (unlikely(r))
2850                         dm_integrity_io_error(ic, "writing superblock", r);
2851         } else {
2852                 replay_journal(ic);
2853                 if (ic->mode == 'B') {
2854                         int mode;
2855                         ic->sb->flags |= cpu_to_le32(SB_FLAG_DIRTY_BITMAP);
2856                         ic->sb->log2_blocks_per_bitmap_bit = ic->log2_blocks_per_bitmap_bit;
2857                         r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
2858                         if (unlikely(r))
2859                                 dm_integrity_io_error(ic, "writing superblock", r);
2860
2861                         mode = ic->recalculate_flag ? BITMAP_OP_SET : BITMAP_OP_CLEAR;
2862                         block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors, mode);
2863                         block_bitmap_op(ic, ic->recalc_bitmap, 0, ic->provided_data_sectors, mode);
2864                         block_bitmap_op(ic, ic->may_write_bitmap, 0, ic->provided_data_sectors, mode);
2865                         rw_journal_sectors(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, 0, ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
2866                 }
2867         }
2868
2869         DEBUG_print("testing recalc: %x\n", ic->sb->flags);
2870         if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
2871                 __u64 recalc_pos = le64_to_cpu(ic->sb->recalc_sector);
2872                 DEBUG_print("recalc pos: %lx / %lx\n", (long)recalc_pos, ic->provided_data_sectors);
2873                 if (recalc_pos < ic->provided_data_sectors) {
2874                         queue_work(ic->recalc_wq, &ic->recalc_work);
2875                 } else if (recalc_pos > ic->provided_data_sectors) {
2876                         ic->sb->recalc_sector = cpu_to_le64(ic->provided_data_sectors);
2877                         recalc_write_super(ic);
2878                 }
2879         }
2880
2881         ic->reboot_notifier.notifier_call = dm_integrity_reboot;
2882         ic->reboot_notifier.next = NULL;
2883         ic->reboot_notifier.priority = INT_MAX - 1;     /* be notified after md and before hardware drivers */
2884         WARN_ON(register_reboot_notifier(&ic->reboot_notifier));
2885
2886 #if 0
2887         dm_integrity_enter_synchronous_mode(ic);
2888 #endif
2889 }
2890
2891 static void dm_integrity_status(struct dm_target *ti, status_type_t type,
2892                                 unsigned status_flags, char *result, unsigned maxlen)
2893 {
2894         struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
2895         unsigned arg_count;
2896         size_t sz = 0;
2897
2898         switch (type) {
2899         case STATUSTYPE_INFO:
2900                 DMEMIT("%llu %llu",
2901                         (unsigned long long)atomic64_read(&ic->number_of_mismatches),
2902                         (unsigned long long)ic->provided_data_sectors);
2903                 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
2904                         DMEMIT(" %llu", (unsigned long long)le64_to_cpu(ic->sb->recalc_sector));
2905                 else
2906                         DMEMIT(" -");
2907                 break;
2908
2909         case STATUSTYPE_TABLE: {
2910                 __u64 watermark_percentage = (__u64)(ic->journal_entries - ic->free_sectors_threshold) * 100;
2911                 watermark_percentage += ic->journal_entries / 2;
2912                 do_div(watermark_percentage, ic->journal_entries);
2913                 arg_count = 3;
2914                 arg_count += !!ic->meta_dev;
2915                 arg_count += ic->sectors_per_block != 1;
2916                 arg_count += !!(ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING));
2917                 arg_count += ic->mode == 'J';
2918                 arg_count += ic->mode == 'J';
2919                 arg_count += ic->mode == 'B';
2920                 arg_count += ic->mode == 'B';
2921                 arg_count += !!ic->internal_hash_alg.alg_string;
2922                 arg_count += !!ic->journal_crypt_alg.alg_string;
2923                 arg_count += !!ic->journal_mac_alg.alg_string;
2924                 DMEMIT("%s %llu %u %c %u", ic->dev->name, (unsigned long long)ic->start,
2925                        ic->tag_size, ic->mode, arg_count);
2926                 if (ic->meta_dev)
2927                         DMEMIT(" meta_device:%s", ic->meta_dev->name);
2928                 if (ic->sectors_per_block != 1)
2929                         DMEMIT(" block_size:%u", ic->sectors_per_block << SECTOR_SHIFT);
2930                 if (ic->recalculate_flag)
2931                         DMEMIT(" recalculate");
2932                 DMEMIT(" journal_sectors:%u", ic->initial_sectors - SB_SECTORS);
2933                 DMEMIT(" interleave_sectors:%u", 1U << ic->sb->log2_interleave_sectors);
2934                 DMEMIT(" buffer_sectors:%u", 1U << ic->log2_buffer_sectors);
2935                 if (ic->mode == 'J') {
2936                         DMEMIT(" journal_watermark:%u", (unsigned)watermark_percentage);
2937                         DMEMIT(" commit_time:%u", ic->autocommit_msec);
2938                 }
2939                 if (ic->mode == 'B') {
2940                         DMEMIT(" sectors_per_bit:%llu", (unsigned long long)ic->sectors_per_block << ic->log2_blocks_per_bitmap_bit);
2941                         DMEMIT(" bitmap_flush_interval:%u", jiffies_to_msecs(ic->bitmap_flush_interval));
2942                 }
2943
2944 #define EMIT_ALG(a, n)                                                  \
2945                 do {                                                    \
2946                         if (ic->a.alg_string) {                         \
2947                                 DMEMIT(" %s:%s", n, ic->a.alg_string);  \
2948                                 if (ic->a.key_string)                   \
2949                                         DMEMIT(":%s", ic->a.key_string);\
2950                         }                                               \
2951                 } while (0)
2952                 EMIT_ALG(internal_hash_alg, "internal_hash");
2953                 EMIT_ALG(journal_crypt_alg, "journal_crypt");
2954                 EMIT_ALG(journal_mac_alg, "journal_mac");
2955                 break;
2956         }
2957         }
2958 }
2959
2960 static int dm_integrity_iterate_devices(struct dm_target *ti,
2961                                         iterate_devices_callout_fn fn, void *data)
2962 {
2963         struct dm_integrity_c *ic = ti->private;
2964
2965         if (!ic->meta_dev)
2966                 return fn(ti, ic->dev, ic->start + ic->initial_sectors + ic->metadata_run, ti->len, data);
2967         else
2968                 return fn(ti, ic->dev, 0, ti->len, data);
2969 }
2970
2971 static void dm_integrity_io_hints(struct dm_target *ti, struct queue_limits *limits)
2972 {
2973         struct dm_integrity_c *ic = ti->private;
2974
2975         if (ic->sectors_per_block > 1) {
2976                 limits->logical_block_size = ic->sectors_per_block << SECTOR_SHIFT;
2977                 limits->physical_block_size = ic->sectors_per_block << SECTOR_SHIFT;
2978                 blk_limits_io_min(limits, ic->sectors_per_block << SECTOR_SHIFT);
2979         }
2980 }
2981
2982 static void calculate_journal_section_size(struct dm_integrity_c *ic)
2983 {
2984         unsigned sector_space = JOURNAL_SECTOR_DATA;
2985
2986         ic->journal_sections = le32_to_cpu(ic->sb->journal_sections);
2987         ic->journal_entry_size = roundup(offsetof(struct journal_entry, last_bytes[ic->sectors_per_block]) + ic->tag_size,
2988                                          JOURNAL_ENTRY_ROUNDUP);
2989
2990         if (ic->sb->flags & cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC))
2991                 sector_space -= JOURNAL_MAC_PER_SECTOR;
2992         ic->journal_entries_per_sector = sector_space / ic->journal_entry_size;
2993         ic->journal_section_entries = ic->journal_entries_per_sector * JOURNAL_BLOCK_SECTORS;
2994         ic->journal_section_sectors = (ic->journal_section_entries << ic->sb->log2_sectors_per_block) + JOURNAL_BLOCK_SECTORS;
2995         ic->journal_entries = ic->journal_section_entries * ic->journal_sections;
2996 }
2997
2998 static int calculate_device_limits(struct dm_integrity_c *ic)
2999 {
3000         __u64 initial_sectors;
3001
3002         calculate_journal_section_size(ic);
3003         initial_sectors = SB_SECTORS + (__u64)ic->journal_section_sectors * ic->journal_sections;
3004         if (initial_sectors + METADATA_PADDING_SECTORS >= ic->meta_device_sectors || initial_sectors > UINT_MAX)
3005                 return -EINVAL;
3006         ic->initial_sectors = initial_sectors;
3007
3008         if (!ic->meta_dev) {
3009                 sector_t last_sector, last_area, last_offset;
3010
3011                 ic->metadata_run = roundup((__u64)ic->tag_size << (ic->sb->log2_interleave_sectors - ic->sb->log2_sectors_per_block),
3012                                            (__u64)(1 << SECTOR_SHIFT << METADATA_PADDING_SECTORS)) >> SECTOR_SHIFT;
3013                 if (!(ic->metadata_run & (ic->metadata_run - 1)))
3014                         ic->log2_metadata_run = __ffs(ic->metadata_run);
3015                 else
3016                         ic->log2_metadata_run = -1;
3017
3018                 get_area_and_offset(ic, ic->provided_data_sectors - 1, &last_area, &last_offset);
3019                 last_sector = get_data_sector(ic, last_area, last_offset);
3020                 if (last_sector < ic->start || last_sector >= ic->meta_device_sectors)
3021                         return -EINVAL;
3022         } else {
3023                 __u64 meta_size = (ic->provided_data_sectors >> ic->sb->log2_sectors_per_block) * ic->tag_size;
3024                 meta_size = (meta_size + ((1U << (ic->log2_buffer_sectors + SECTOR_SHIFT)) - 1))
3025                                 >> (ic->log2_buffer_sectors + SECTOR_SHIFT);
3026                 meta_size <<= ic->log2_buffer_sectors;
3027                 if (ic->initial_sectors + meta_size < ic->initial_sectors ||
3028                     ic->initial_sectors + meta_size > ic->meta_device_sectors)
3029                         return -EINVAL;
3030                 ic->metadata_run = 1;
3031                 ic->log2_metadata_run = 0;
3032         }
3033
3034         return 0;
3035 }
3036
3037 static int initialize_superblock(struct dm_integrity_c *ic, unsigned journal_sectors, unsigned interleave_sectors)
3038 {
3039         unsigned journal_sections;
3040         int test_bit;
3041
3042         memset(ic->sb, 0, SB_SECTORS << SECTOR_SHIFT);
3043         memcpy(ic->sb->magic, SB_MAGIC, 8);
3044         ic->sb->integrity_tag_size = cpu_to_le16(ic->tag_size);
3045         ic->sb->log2_sectors_per_block = __ffs(ic->sectors_per_block);
3046         if (ic->journal_mac_alg.alg_string)
3047                 ic->sb->flags |= cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC);
3048
3049         calculate_journal_section_size(ic);
3050         journal_sections = journal_sectors / ic->journal_section_sectors;
3051         if (!journal_sections)
3052                 journal_sections = 1;
3053
3054         if (!ic->meta_dev) {
3055                 ic->sb->journal_sections = cpu_to_le32(journal_sections);
3056                 if (!interleave_sectors)
3057                         interleave_sectors = DEFAULT_INTERLEAVE_SECTORS;
3058                 ic->sb->log2_interleave_sectors = __fls(interleave_sectors);
3059                 ic->sb->log2_interleave_sectors = max((__u8)MIN_LOG2_INTERLEAVE_SECTORS, ic->sb->log2_interleave_sectors);
3060                 ic->sb->log2_interleave_sectors = min((__u8)MAX_LOG2_INTERLEAVE_SECTORS, ic->sb->log2_interleave_sectors);
3061
3062                 ic->provided_data_sectors = 0;
3063                 for (test_bit = fls64(ic->meta_device_sectors) - 1; test_bit >= 3; test_bit--) {
3064                         __u64 prev_data_sectors = ic->provided_data_sectors;
3065
3066                         ic->provided_data_sectors |= (sector_t)1 << test_bit;
3067                         if (calculate_device_limits(ic))
3068                                 ic->provided_data_sectors = prev_data_sectors;
3069                 }
3070                 if (!ic->provided_data_sectors)
3071                         return -EINVAL;
3072         } else {
3073                 ic->sb->log2_interleave_sectors = 0;
3074                 ic->provided_data_sectors = ic->data_device_sectors;
3075                 ic->provided_data_sectors &= ~(sector_t)(ic->sectors_per_block - 1);
3076
3077 try_smaller_buffer:
3078                 ic->sb->journal_sections = cpu_to_le32(0);
3079                 for (test_bit = fls(journal_sections) - 1; test_bit >= 0; test_bit--) {
3080                         __u32 prev_journal_sections = le32_to_cpu(ic->sb->journal_sections);
3081                         __u32 test_journal_sections = prev_journal_sections | (1U << test_bit);
3082                         if (test_journal_sections > journal_sections)
3083                                 continue;
3084                         ic->sb->journal_sections = cpu_to_le32(test_journal_sections);
3085                         if (calculate_device_limits(ic))
3086                                 ic->sb->journal_sections = cpu_to_le32(prev_journal_sections);
3087
3088                 }
3089                 if (!le32_to_cpu(ic->sb->journal_sections)) {
3090                         if (ic->log2_buffer_sectors > 3) {
3091                                 ic->log2_buffer_sectors--;
3092                                 goto try_smaller_buffer;
3093                         }
3094                         return -EINVAL;
3095                 }
3096         }
3097
3098         ic->sb->provided_data_sectors = cpu_to_le64(ic->provided_data_sectors);
3099
3100         sb_set_version(ic);
3101
3102         return 0;
3103 }
3104
3105 static void dm_integrity_set(struct dm_target *ti, struct dm_integrity_c *ic)
3106 {
3107         struct gendisk *disk = dm_disk(dm_table_get_md(ti->table));
3108         struct blk_integrity bi;
3109
3110         memset(&bi, 0, sizeof(bi));
3111         bi.profile = &dm_integrity_profile;
3112         bi.tuple_size = ic->tag_size;
3113         bi.tag_size = bi.tuple_size;
3114         bi.interval_exp = ic->sb->log2_sectors_per_block + SECTOR_SHIFT;
3115
3116         blk_integrity_register(disk, &bi);
3117         blk_queue_max_integrity_segments(disk->queue, UINT_MAX);
3118 }
3119
3120 static void dm_integrity_free_page_list(struct page_list *pl)
3121 {
3122         unsigned i;
3123
3124         if (!pl)
3125                 return;
3126         for (i = 0; pl[i].page; i++)
3127                 __free_page(pl[i].page);
3128         kvfree(pl);
3129 }
3130
3131 static struct page_list *dm_integrity_alloc_page_list(unsigned n_pages)
3132 {
3133         struct page_list *pl;
3134         unsigned i;
3135
3136         pl = kvmalloc_array(n_pages + 1, sizeof(struct page_list), GFP_KERNEL | __GFP_ZERO);
3137         if (!pl)
3138                 return NULL;
3139
3140         for (i = 0; i < n_pages; i++) {
3141                 pl[i].page = alloc_page(GFP_KERNEL);
3142                 if (!pl[i].page) {
3143                         dm_integrity_free_page_list(pl);
3144                         return NULL;
3145                 }
3146                 if (i)
3147                         pl[i - 1].next = &pl[i];
3148         }
3149         pl[i].page = NULL;
3150         pl[i].next = NULL;
3151
3152         return pl;
3153 }
3154
3155 static void dm_integrity_free_journal_scatterlist(struct dm_integrity_c *ic, struct scatterlist **sl)
3156 {
3157         unsigned i;
3158         for (i = 0; i < ic->journal_sections; i++)
3159                 kvfree(sl[i]);
3160         kvfree(sl);
3161 }
3162
3163 static struct scatterlist **dm_integrity_alloc_journal_scatterlist(struct dm_integrity_c *ic, struct page_list *pl)
3164 {
3165         struct scatterlist **sl;
3166         unsigned i;
3167
3168         sl = kvmalloc_array(ic->journal_sections,
3169                             sizeof(struct scatterlist *),
3170                             GFP_KERNEL | __GFP_ZERO);
3171         if (!sl)
3172                 return NULL;
3173
3174         for (i = 0; i < ic->journal_sections; i++) {
3175                 struct scatterlist *s;
3176                 unsigned start_index, start_offset;
3177                 unsigned end_index, end_offset;
3178                 unsigned n_pages;
3179                 unsigned idx;
3180
3181                 page_list_location(ic, i, 0, &start_index, &start_offset);
3182                 page_list_location(ic, i, ic->journal_section_sectors - 1, &end_index, &end_offset);
3183
3184                 n_pages = (end_index - start_index + 1);
3185
3186                 s = kvmalloc_array(n_pages, sizeof(struct scatterlist),
3187                                    GFP_KERNEL);
3188                 if (!s) {
3189                         dm_integrity_free_journal_scatterlist(ic, sl);
3190                         return NULL;
3191                 }
3192
3193                 sg_init_table(s, n_pages);
3194                 for (idx = start_index; idx <= end_index; idx++) {
3195                         char *va = lowmem_page_address(pl[idx].page);
3196                         unsigned start = 0, end = PAGE_SIZE;
3197                         if (idx == start_index)
3198                                 start = start_offset;
3199                         if (idx == end_index)
3200                                 end = end_offset + (1 << SECTOR_SHIFT);
3201                         sg_set_buf(&s[idx - start_index], va + start, end - start);
3202                 }
3203
3204                 sl[i] = s;
3205         }
3206
3207         return sl;
3208 }
3209
3210 static void free_alg(struct alg_spec *a)
3211 {
3212         kzfree(a->alg_string);
3213         kzfree(a->key);
3214         memset(a, 0, sizeof *a);
3215 }
3216
3217 static int get_alg_and_key(const char *arg, struct alg_spec *a, char **error, char *error_inval)
3218 {
3219         char *k;
3220
3221         free_alg(a);
3222
3223         a->alg_string = kstrdup(strchr(arg, ':') + 1, GFP_KERNEL);
3224         if (!a->alg_string)
3225                 goto nomem;
3226
3227         k = strchr(a->alg_string, ':');
3228         if (k) {
3229                 *k = 0;
3230                 a->key_string = k + 1;
3231                 if (strlen(a->key_string) & 1)
3232                         goto inval;
3233
3234                 a->key_size = strlen(a->key_string) / 2;
3235                 a->key = kmalloc(a->key_size, GFP_KERNEL);
3236                 if (!a->key)
3237                         goto nomem;
3238                 if (hex2bin(a->key, a->key_string, a->key_size))
3239                         goto inval;
3240         }
3241
3242         return 0;
3243 inval:
3244         *error = error_inval;
3245         return -EINVAL;
3246 nomem:
3247         *error = "Out of memory for an argument";
3248         return -ENOMEM;
3249 }
3250
3251 static int get_mac(struct crypto_shash **hash, struct alg_spec *a, char **error,
3252                    char *error_alg, char *error_key)
3253 {
3254         int r;
3255
3256         if (a->alg_string) {
3257                 *hash = crypto_alloc_shash(a->alg_string, 0, 0);
3258                 if (IS_ERR(*hash)) {
3259                         *error = error_alg;
3260                         r = PTR_ERR(*hash);
3261                         *hash = NULL;
3262                         return r;
3263                 }
3264
3265                 if (a->key) {
3266                         r = crypto_shash_setkey(*hash, a->key, a->key_size);
3267                         if (r) {
3268                                 *error = error_key;
3269                                 return r;
3270                         }
3271                 } else if (crypto_shash_get_flags(*hash) & CRYPTO_TFM_NEED_KEY) {
3272                         *error = error_key;
3273                         return -ENOKEY;
3274                 }
3275         }
3276
3277         return 0;
3278 }
3279
3280 static int create_journal(struct dm_integrity_c *ic, char **error)
3281 {
3282         int r = 0;
3283         unsigned i;
3284         __u64 journal_pages, journal_desc_size, journal_tree_size;
3285         unsigned char *crypt_data = NULL, *crypt_iv = NULL;
3286         struct skcipher_request *req = NULL;
3287
3288         ic->commit_ids[0] = cpu_to_le64(0x1111111111111111ULL);
3289         ic->commit_ids[1] = cpu_to_le64(0x2222222222222222ULL);
3290         ic->commit_ids[2] = cpu_to_le64(0x3333333333333333ULL);
3291         ic->commit_ids[3] = cpu_to_le64(0x4444444444444444ULL);
3292
3293         journal_pages = roundup((__u64)ic->journal_sections * ic->journal_section_sectors,
3294                                 PAGE_SIZE >> SECTOR_SHIFT) >> (PAGE_SHIFT - SECTOR_SHIFT);
3295         journal_desc_size = journal_pages * sizeof(struct page_list);
3296         if (journal_pages >= totalram_pages() - totalhigh_pages() || journal_desc_size > ULONG_MAX) {
3297                 *error = "Journal doesn't fit into memory";
3298                 r = -ENOMEM;
3299                 goto bad;
3300         }
3301         ic->journal_pages = journal_pages;
3302
3303         ic->journal = dm_integrity_alloc_page_list(ic->journal_pages);
3304         if (!ic->journal) {
3305                 *error = "Could not allocate memory for journal";
3306                 r = -ENOMEM;
3307                 goto bad;
3308         }
3309         if (ic->journal_crypt_alg.alg_string) {
3310                 unsigned ivsize, blocksize;
3311                 struct journal_completion comp;
3312
3313                 comp.ic = ic;
3314                 ic->journal_crypt = crypto_alloc_skcipher(ic->journal_crypt_alg.alg_string, 0, 0);
3315                 if (IS_ERR(ic->journal_crypt)) {
3316                         *error = "Invalid journal cipher";
3317                         r = PTR_ERR(ic->journal_crypt);
3318                         ic->journal_crypt = NULL;
3319                         goto bad;
3320                 }
3321                 ivsize = crypto_skcipher_ivsize(ic->journal_crypt);
3322                 blocksize = crypto_skcipher_blocksize(ic->journal_crypt);
3323
3324                 if (ic->journal_crypt_alg.key) {
3325                         r = crypto_skcipher_setkey(ic->journal_crypt, ic->journal_crypt_alg.key,
3326                                                    ic->journal_crypt_alg.key_size);
3327                         if (r) {
3328                                 *error = "Error setting encryption key";
3329                                 goto bad;
3330                         }
3331                 }
3332                 DEBUG_print("cipher %s, block size %u iv size %u\n",
3333                             ic->journal_crypt_alg.alg_string, blocksize, ivsize);
3334
3335                 ic->journal_io = dm_integrity_alloc_page_list(ic->journal_pages);
3336                 if (!ic->journal_io) {
3337                         *error = "Could not allocate memory for journal io";
3338                         r = -ENOMEM;
3339                         goto bad;
3340                 }
3341
3342                 if (blocksize == 1) {
3343                         struct scatterlist *sg;
3344
3345                         req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
3346                         if (!req) {
3347                                 *error = "Could not allocate crypt request";
3348                                 r = -ENOMEM;
3349                                 goto bad;
3350                         }
3351
3352                         crypt_iv = kmalloc(ivsize, GFP_KERNEL);
3353                         if (!crypt_iv) {
3354                                 *error = "Could not allocate iv";
3355                                 r = -ENOMEM;
3356                                 goto bad;
3357                         }
3358
3359                         ic->journal_xor = dm_integrity_alloc_page_list(ic->journal_pages);
3360                         if (!ic->journal_xor) {
3361                                 *error = "Could not allocate memory for journal xor";
3362                                 r = -ENOMEM;
3363                                 goto bad;
3364                         }
3365
3366                         sg = kvmalloc_array(ic->journal_pages + 1,
3367                                             sizeof(struct scatterlist),
3368                                             GFP_KERNEL);
3369                         if (!sg) {
3370                                 *error = "Unable to allocate sg list";
3371                                 r = -ENOMEM;
3372                                 goto bad;
3373                         }
3374                         sg_init_table(sg, ic->journal_pages + 1);
3375                         for (i = 0; i < ic->journal_pages; i++) {
3376                                 char *va = lowmem_page_address(ic->journal_xor[i].page);
3377                                 clear_page(va);
3378                                 sg_set_buf(&sg[i], va, PAGE_SIZE);
3379                         }
3380                         sg_set_buf(&sg[i], &ic->commit_ids, sizeof ic->commit_ids);
3381                         memset(crypt_iv, 0x00, ivsize);
3382
3383                         skcipher_request_set_crypt(req, sg, sg, PAGE_SIZE * ic->journal_pages + sizeof ic->commit_ids, crypt_iv);
3384                         init_completion(&comp.comp);
3385                         comp.in_flight = (atomic_t)ATOMIC_INIT(1);
3386                         if (do_crypt(true, req, &comp))
3387                                 wait_for_completion(&comp.comp);
3388                         kvfree(sg);
3389                         r = dm_integrity_failed(ic);
3390                         if (r) {
3391                                 *error = "Unable to encrypt journal";
3392                                 goto bad;
3393                         }
3394                         DEBUG_bytes(lowmem_page_address(ic->journal_xor[0].page), 64, "xor data");
3395
3396                         crypto_free_skcipher(ic->journal_crypt);
3397                         ic->journal_crypt = NULL;
3398                 } else {
3399                         unsigned crypt_len = roundup(ivsize, blocksize);
3400
3401                         req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
3402                         if (!req) {
3403                                 *error = "Could not allocate crypt request";
3404                                 r = -ENOMEM;
3405                                 goto bad;
3406                         }
3407
3408                         crypt_iv = kmalloc(ivsize, GFP_KERNEL);
3409                         if (!crypt_iv) {
3410                                 *error = "Could not allocate iv";
3411                                 r = -ENOMEM;
3412                                 goto bad;
3413                         }
3414
3415                         crypt_data = kmalloc(crypt_len, GFP_KERNEL);
3416                         if (!crypt_data) {
3417                                 *error = "Unable to allocate crypt data";
3418                                 r = -ENOMEM;
3419                                 goto bad;
3420                         }
3421
3422                         ic->journal_scatterlist = dm_integrity_alloc_journal_scatterlist(ic, ic->journal);
3423                         if (!ic->journal_scatterlist) {
3424                                 *error = "Unable to allocate sg list";
3425                                 r = -ENOMEM;
3426                                 goto bad;
3427                         }
3428                         ic->journal_io_scatterlist = dm_integrity_alloc_journal_scatterlist(ic, ic->journal_io);
3429                         if (!ic->journal_io_scatterlist) {
3430                                 *error = "Unable to allocate sg list";
3431                                 r = -ENOMEM;
3432                                 goto bad;
3433                         }
3434                         ic->sk_requests = kvmalloc_array(ic->journal_sections,
3435                                                          sizeof(struct skcipher_request *),
3436                                                          GFP_KERNEL | __GFP_ZERO);
3437                         if (!ic->sk_requests) {
3438                                 *error = "Unable to allocate sk requests";
3439                                 r = -ENOMEM;
3440                                 goto bad;
3441                         }
3442                         for (i = 0; i < ic->journal_sections; i++) {
3443                                 struct scatterlist sg;
3444                                 struct skcipher_request *section_req;
3445                                 __u32 section_le = cpu_to_le32(i);
3446
3447                                 memset(crypt_iv, 0x00, ivsize);
3448                                 memset(crypt_data, 0x00, crypt_len);
3449                                 memcpy(crypt_data, &section_le, min((size_t)crypt_len, sizeof(section_le)));
3450
3451                                 sg_init_one(&sg, crypt_data, crypt_len);
3452                                 skcipher_request_set_crypt(req, &sg, &sg, crypt_len, crypt_iv);
3453                                 init_completion(&comp.comp);
3454                                 comp.in_flight = (atomic_t)ATOMIC_INIT(1);
3455                                 if (do_crypt(true, req, &comp))
3456                                         wait_for_completion(&comp.comp);
3457
3458                                 r = dm_integrity_failed(ic);
3459                                 if (r) {
3460                                         *error = "Unable to generate iv";
3461                                         goto bad;
3462                                 }
3463
3464                                 section_req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
3465                                 if (!section_req) {
3466                                         *error = "Unable to allocate crypt request";
3467                                         r = -ENOMEM;
3468                                         goto bad;
3469                                 }
3470                                 section_req->iv = kmalloc_array(ivsize, 2,
3471                                                                 GFP_KERNEL);
3472                                 if (!section_req->iv) {
3473                                         skcipher_request_free(section_req);
3474                                         *error = "Unable to allocate iv";
3475                                         r = -ENOMEM;
3476                                         goto bad;
3477                                 }
3478                                 memcpy(section_req->iv + ivsize, crypt_data, ivsize);
3479                                 section_req->cryptlen = (size_t)ic->journal_section_sectors << SECTOR_SHIFT;
3480                                 ic->sk_requests[i] = section_req;
3481                                 DEBUG_bytes(crypt_data, ivsize, "iv(%u)", i);
3482                         }
3483                 }
3484         }
3485
3486         for (i = 0; i < N_COMMIT_IDS; i++) {
3487                 unsigned j;
3488 retest_commit_id:
3489                 for (j = 0; j < i; j++) {
3490                         if (ic->commit_ids[j] == ic->commit_ids[i]) {
3491                                 ic->commit_ids[i] = cpu_to_le64(le64_to_cpu(ic->commit_ids[i]) + 1);
3492                                 goto retest_commit_id;
3493                         }
3494                 }
3495                 DEBUG_print("commit id %u: %016llx\n", i, ic->commit_ids[i]);
3496         }
3497
3498         journal_tree_size = (__u64)ic->journal_entries * sizeof(struct journal_node);
3499         if (journal_tree_size > ULONG_MAX) {
3500                 *error = "Journal doesn't fit into memory";
3501                 r = -ENOMEM;
3502                 goto bad;
3503         }
3504         ic->journal_tree = kvmalloc(journal_tree_size, GFP_KERNEL);
3505         if (!ic->journal_tree) {
3506                 *error = "Could not allocate memory for journal tree";
3507                 r = -ENOMEM;
3508         }
3509 bad:
3510         kfree(crypt_data);
3511         kfree(crypt_iv);
3512         skcipher_request_free(req);
3513
3514         return r;
3515 }
3516
3517 /*
3518  * Construct a integrity mapping
3519  *
3520  * Arguments:
3521  *      device
3522  *      offset from the start of the device
3523  *      tag size
3524  *      D - direct writes, J - journal writes, B - bitmap mode, R - recovery mode
3525  *      number of optional arguments
3526  *      optional arguments:
3527  *              journal_sectors
3528  *              interleave_sectors
3529  *              buffer_sectors
3530  *              journal_watermark
3531  *              commit_time
3532  *              meta_device
3533  *              block_size
3534  *              sectors_per_bit
3535  *              bitmap_flush_interval
3536  *              internal_hash
3537  *              journal_crypt
3538  *              journal_mac
3539  *              recalculate
3540  */
3541 static int dm_integrity_ctr(struct dm_target *ti, unsigned argc, char **argv)
3542 {
3543         struct dm_integrity_c *ic;
3544         char dummy;
3545         int r;
3546         unsigned extra_args;
3547         struct dm_arg_set as;
3548         static const struct dm_arg _args[] = {
3549                 {0, 9, "Invalid number of feature args"},
3550         };
3551         unsigned journal_sectors, interleave_sectors, buffer_sectors, journal_watermark, sync_msec;
3552         bool should_write_sb;
3553         __u64 threshold;
3554         unsigned long long start;
3555         __s8 log2_sectors_per_bitmap_bit = -1;
3556         __s8 log2_blocks_per_bitmap_bit;
3557         __u64 bits_in_journal;
3558         __u64 n_bitmap_bits;
3559
3560 #define DIRECT_ARGUMENTS        4
3561
3562         if (argc <= DIRECT_ARGUMENTS) {
3563                 ti->error = "Invalid argument count";
3564                 return -EINVAL;
3565         }
3566
3567         ic = kzalloc(sizeof(struct dm_integrity_c), GFP_KERNEL);
3568         if (!ic) {
3569                 ti->error = "Cannot allocate integrity context";
3570                 return -ENOMEM;
3571         }
3572         ti->private = ic;
3573         ti->per_io_data_size = sizeof(struct dm_integrity_io);
3574
3575         ic->in_progress = RB_ROOT;
3576         INIT_LIST_HEAD(&ic->wait_list);
3577         init_waitqueue_head(&ic->endio_wait);
3578         bio_list_init(&ic->flush_bio_list);
3579         init_waitqueue_head(&ic->copy_to_journal_wait);
3580         init_completion(&ic->crypto_backoff);
3581         atomic64_set(&ic->number_of_mismatches, 0);
3582         ic->bitmap_flush_interval = BITMAP_FLUSH_INTERVAL;
3583
3584         r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &ic->dev);
3585         if (r) {
3586                 ti->error = "Device lookup failed";
3587                 goto bad;
3588         }
3589
3590         if (sscanf(argv[1], "%llu%c", &start, &dummy) != 1 || start != (sector_t)start) {
3591                 ti->error = "Invalid starting offset";
3592                 r = -EINVAL;
3593                 goto bad;
3594         }
3595         ic->start = start;
3596
3597         if (strcmp(argv[2], "-")) {
3598                 if (sscanf(argv[2], "%u%c", &ic->tag_size, &dummy) != 1 || !ic->tag_size) {
3599                         ti->error = "Invalid tag size";
3600                         r = -EINVAL;
3601                         goto bad;
3602                 }
3603         }
3604
3605         if (!strcmp(argv[3], "J") || !strcmp(argv[3], "B") || !strcmp(argv[3], "D") || !strcmp(argv[3], "R")) {
3606                 ic->mode = argv[3][0];
3607         } else {
3608                 ti->error = "Invalid mode (expecting J, B, D, R)";
3609                 r = -EINVAL;
3610                 goto bad;
3611         }
3612
3613         journal_sectors = 0;
3614         interleave_sectors = DEFAULT_INTERLEAVE_SECTORS;
3615         buffer_sectors = DEFAULT_BUFFER_SECTORS;
3616         journal_watermark = DEFAULT_JOURNAL_WATERMARK;
3617         sync_msec = DEFAULT_SYNC_MSEC;
3618         ic->sectors_per_block = 1;
3619
3620         as.argc = argc - DIRECT_ARGUMENTS;
3621         as.argv = argv + DIRECT_ARGUMENTS;
3622         r = dm_read_arg_group(_args, &as, &extra_args, &ti->error);
3623         if (r)
3624                 goto bad;
3625
3626         while (extra_args--) {
3627                 const char *opt_string;
3628                 unsigned val;
3629                 unsigned long long llval;
3630                 opt_string = dm_shift_arg(&as);
3631                 if (!opt_string) {
3632                         r = -EINVAL;
3633                         ti->error = "Not enough feature arguments";
3634                         goto bad;
3635                 }
3636                 if (sscanf(opt_string, "journal_sectors:%u%c", &val, &dummy) == 1)
3637                         journal_sectors = val ? val : 1;
3638                 else if (sscanf(opt_string, "interleave_sectors:%u%c", &val, &dummy) == 1)
3639                         interleave_sectors = val;
3640                 else if (sscanf(opt_string, "buffer_sectors:%u%c", &val, &dummy) == 1)
3641                         buffer_sectors = val;
3642                 else if (sscanf(opt_string, "journal_watermark:%u%c", &val, &dummy) == 1 && val <= 100)
3643                         journal_watermark = val;
3644                 else if (sscanf(opt_string, "commit_time:%u%c", &val, &dummy) == 1)
3645                         sync_msec = val;
3646                 else if (!strncmp(opt_string, "meta_device:", strlen("meta_device:"))) {
3647                         if (ic->meta_dev) {
3648                                 dm_put_device(ti, ic->meta_dev);
3649                                 ic->meta_dev = NULL;
3650                         }
3651                         r = dm_get_device(ti, strchr(opt_string, ':') + 1, dm_table_get_mode(ti->table), &ic->meta_dev);
3652                         if (r) {
3653                                 ti->error = "Device lookup failed";
3654                                 goto bad;
3655                         }
3656                 } else if (sscanf(opt_string, "block_size:%u%c", &val, &dummy) == 1) {
3657                         if (val < 1 << SECTOR_SHIFT ||
3658                             val > MAX_SECTORS_PER_BLOCK << SECTOR_SHIFT ||
3659                             (val & (val -1))) {
3660                                 r = -EINVAL;
3661                                 ti->error = "Invalid block_size argument";
3662                                 goto bad;
3663                         }
3664                         ic->sectors_per_block = val >> SECTOR_SHIFT;
3665                 } else if (sscanf(opt_string, "sectors_per_bit:%llu%c", &llval, &dummy) == 1) {
3666                         log2_sectors_per_bitmap_bit = !llval ? 0 : __ilog2_u64(llval);
3667                 } else if (sscanf(opt_string, "bitmap_flush_interval:%u%c", &val, &dummy) == 1) {
3668                         if (val >= (uint64_t)UINT_MAX * 1000 / HZ) {
3669                                 r = -EINVAL;
3670                                 ti->error = "Invalid bitmap_flush_interval argument";
3671                         }
3672                         ic->bitmap_flush_interval = msecs_to_jiffies(val);
3673                 } else if (!strncmp(opt_string, "internal_hash:", strlen("internal_hash:"))) {
3674                         r = get_alg_and_key(opt_string, &ic->internal_hash_alg, &ti->error,
3675                                             "Invalid internal_hash argument");
3676                         if (r)
3677                                 goto bad;
3678                 } else if (!strncmp(opt_string, "journal_crypt:", strlen("journal_crypt:"))) {
3679                         r = get_alg_and_key(opt_string, &ic->journal_crypt_alg, &ti->error,
3680                                             "Invalid journal_crypt argument");
3681                         if (r)
3682                                 goto bad;
3683                 } else if (!strncmp(opt_string, "journal_mac:", strlen("journal_mac:"))) {
3684                         r = get_alg_and_key(opt_string, &ic->journal_mac_alg,  &ti->error,
3685                                             "Invalid journal_mac argument");
3686                         if (r)
3687                                 goto bad;
3688                 } else if (!strcmp(opt_string, "recalculate")) {
3689                         ic->recalculate_flag = true;
3690                 } else {
3691                         r = -EINVAL;
3692                         ti->error = "Invalid argument";
3693                         goto bad;
3694                 }
3695         }
3696
3697         ic->data_device_sectors = i_size_read(ic->dev->bdev->bd_inode) >> SECTOR_SHIFT;
3698         if (!ic->meta_dev)
3699                 ic->meta_device_sectors = ic->data_device_sectors;
3700         else
3701                 ic->meta_device_sectors = i_size_read(ic->meta_dev->bdev->bd_inode) >> SECTOR_SHIFT;
3702
3703         if (!journal_sectors) {
3704                 journal_sectors = min((sector_t)DEFAULT_MAX_JOURNAL_SECTORS,
3705                         ic->data_device_sectors >> DEFAULT_JOURNAL_SIZE_FACTOR);
3706         }
3707
3708         if (!buffer_sectors)
3709                 buffer_sectors = 1;
3710         ic->log2_buffer_sectors = min((int)__fls(buffer_sectors), 31 - SECTOR_SHIFT);
3711
3712         r = get_mac(&ic->internal_hash, &ic->internal_hash_alg, &ti->error,
3713                     "Invalid internal hash", "Error setting internal hash key");
3714         if (r)
3715                 goto bad;
3716
3717         r = get_mac(&ic->journal_mac, &ic->journal_mac_alg, &ti->error,
3718                     "Invalid journal mac", "Error setting journal mac key");
3719         if (r)
3720                 goto bad;
3721
3722         if (!ic->tag_size) {
3723                 if (!ic->internal_hash) {
3724                         ti->error = "Unknown tag size";
3725                         r = -EINVAL;
3726                         goto bad;
3727                 }
3728                 ic->tag_size = crypto_shash_digestsize(ic->internal_hash);
3729         }
3730         if (ic->tag_size > MAX_TAG_SIZE) {
3731                 ti->error = "Too big tag size";
3732                 r = -EINVAL;
3733                 goto bad;
3734         }
3735         if (!(ic->tag_size & (ic->tag_size - 1)))
3736                 ic->log2_tag_size = __ffs(ic->tag_size);
3737         else
3738                 ic->log2_tag_size = -1;
3739
3740         if (ic->mode == 'B' && !ic->internal_hash) {
3741                 r = -EINVAL;
3742                 ti->error = "Bitmap mode can be only used with internal hash";
3743                 goto bad;
3744         }
3745
3746         ic->autocommit_jiffies = msecs_to_jiffies(sync_msec);
3747         ic->autocommit_msec = sync_msec;
3748         timer_setup(&ic->autocommit_timer, autocommit_fn, 0);
3749
3750         ic->io = dm_io_client_create();
3751         if (IS_ERR(ic->io)) {
3752                 r = PTR_ERR(ic->io);
3753                 ic->io = NULL;
3754                 ti->error = "Cannot allocate dm io";
3755                 goto bad;
3756         }
3757
3758         r = mempool_init_slab_pool(&ic->journal_io_mempool, JOURNAL_IO_MEMPOOL, journal_io_cache);
3759         if (r) {
3760                 ti->error = "Cannot allocate mempool";
3761                 goto bad;
3762         }
3763
3764         ic->metadata_wq = alloc_workqueue("dm-integrity-metadata",
3765                                           WQ_MEM_RECLAIM, METADATA_WORKQUEUE_MAX_ACTIVE);
3766         if (!ic->metadata_wq) {
3767                 ti->error = "Cannot allocate workqueue";
3768                 r = -ENOMEM;
3769                 goto bad;
3770         }
3771
3772         /*
3773          * If this workqueue were percpu, it would cause bio reordering
3774          * and reduced performance.
3775          */
3776         ic->wait_wq = alloc_workqueue("dm-integrity-wait", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3777         if (!ic->wait_wq) {
3778                 ti->error = "Cannot allocate workqueue";
3779                 r = -ENOMEM;
3780                 goto bad;
3781         }
3782
3783         ic->commit_wq = alloc_workqueue("dm-integrity-commit", WQ_MEM_RECLAIM, 1);
3784         if (!ic->commit_wq) {
3785                 ti->error = "Cannot allocate workqueue";
3786                 r = -ENOMEM;
3787                 goto bad;
3788         }
3789         INIT_WORK(&ic->commit_work, integrity_commit);
3790
3791         if (ic->mode == 'J' || ic->mode == 'B') {
3792                 ic->writer_wq = alloc_workqueue("dm-integrity-writer", WQ_MEM_RECLAIM, 1);
3793                 if (!ic->writer_wq) {
3794                         ti->error = "Cannot allocate workqueue";
3795                         r = -ENOMEM;
3796                         goto bad;
3797                 }
3798                 INIT_WORK(&ic->writer_work, integrity_writer);
3799         }
3800
3801         ic->sb = alloc_pages_exact(SB_SECTORS << SECTOR_SHIFT, GFP_KERNEL);
3802         if (!ic->sb) {
3803                 r = -ENOMEM;
3804                 ti->error = "Cannot allocate superblock area";
3805                 goto bad;
3806         }
3807
3808         r = sync_rw_sb(ic, REQ_OP_READ, 0);
3809         if (r) {
3810                 ti->error = "Error reading superblock";
3811                 goto bad;
3812         }
3813         should_write_sb = false;
3814         if (memcmp(ic->sb->magic, SB_MAGIC, 8)) {
3815                 if (ic->mode != 'R') {
3816                         if (memchr_inv(ic->sb, 0, SB_SECTORS << SECTOR_SHIFT)) {
3817                                 r = -EINVAL;
3818                                 ti->error = "The device is not initialized";
3819                                 goto bad;
3820                         }
3821                 }
3822
3823                 r = initialize_superblock(ic, journal_sectors, interleave_sectors);
3824                 if (r) {
3825                         ti->error = "Could not initialize superblock";
3826                         goto bad;
3827                 }
3828                 if (ic->mode != 'R')
3829                         should_write_sb = true;
3830         }
3831
3832         if (!ic->sb->version || ic->sb->version > SB_VERSION_3) {
3833                 r = -EINVAL;
3834                 ti->error = "Unknown version";
3835                 goto bad;
3836         }
3837         if (le16_to_cpu(ic->sb->integrity_tag_size) != ic->tag_size) {
3838                 r = -EINVAL;
3839                 ti->error = "Tag size doesn't match the information in superblock";
3840                 goto bad;
3841         }
3842         if (ic->sb->log2_sectors_per_block != __ffs(ic->sectors_per_block)) {
3843                 r = -EINVAL;
3844                 ti->error = "Block size doesn't match the information in superblock";
3845                 goto bad;
3846         }
3847         if (!le32_to_cpu(ic->sb->journal_sections)) {
3848                 r = -EINVAL;
3849                 ti->error = "Corrupted superblock, journal_sections is 0";
3850                 goto bad;
3851         }
3852         /* make sure that ti->max_io_len doesn't overflow */
3853         if (!ic->meta_dev) {
3854                 if (ic->sb->log2_interleave_sectors < MIN_LOG2_INTERLEAVE_SECTORS ||
3855                     ic->sb->log2_interleave_sectors > MAX_LOG2_INTERLEAVE_SECTORS) {
3856                         r = -EINVAL;
3857                         ti->error = "Invalid interleave_sectors in the superblock";
3858                         goto bad;
3859                 }
3860         } else {
3861                 if (ic->sb->log2_interleave_sectors) {
3862                         r = -EINVAL;
3863                         ti->error = "Invalid interleave_sectors in the superblock";
3864                         goto bad;
3865                 }
3866         }
3867         ic->provided_data_sectors = le64_to_cpu(ic->sb->provided_data_sectors);
3868         if (ic->provided_data_sectors != le64_to_cpu(ic->sb->provided_data_sectors)) {
3869                 /* test for overflow */
3870                 r = -EINVAL;
3871                 ti->error = "The superblock has 64-bit device size, but the kernel was compiled with 32-bit sectors";
3872                 goto bad;
3873         }
3874         if (!!(ic->sb->flags & cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC)) != !!ic->journal_mac_alg.alg_string) {
3875                 r = -EINVAL;
3876                 ti->error = "Journal mac mismatch";
3877                 goto bad;
3878         }
3879
3880 try_smaller_buffer:
3881         r = calculate_device_limits(ic);
3882         if (r) {
3883                 if (ic->meta_dev) {
3884                         if (ic->log2_buffer_sectors > 3) {
3885                                 ic->log2_buffer_sectors--;
3886                                 goto try_smaller_buffer;
3887                         }
3888                 }
3889                 ti->error = "The device is too small";
3890                 goto bad;
3891         }
3892
3893         if (log2_sectors_per_bitmap_bit < 0)
3894                 log2_sectors_per_bitmap_bit = __fls(DEFAULT_SECTORS_PER_BITMAP_BIT);
3895         if (log2_sectors_per_bitmap_bit < ic->sb->log2_sectors_per_block)
3896                 log2_sectors_per_bitmap_bit = ic->sb->log2_sectors_per_block;
3897
3898         bits_in_journal = ((__u64)ic->journal_section_sectors * ic->journal_sections) << (SECTOR_SHIFT + 3);
3899         if (bits_in_journal > UINT_MAX)
3900                 bits_in_journal = UINT_MAX;
3901         while (bits_in_journal < (ic->provided_data_sectors + ((sector_t)1 << log2_sectors_per_bitmap_bit) - 1) >> log2_sectors_per_bitmap_bit)
3902                 log2_sectors_per_bitmap_bit++;
3903
3904         log2_blocks_per_bitmap_bit = log2_sectors_per_bitmap_bit - ic->sb->log2_sectors_per_block;
3905         ic->log2_blocks_per_bitmap_bit = log2_blocks_per_bitmap_bit;
3906         if (should_write_sb) {
3907                 ic->sb->log2_blocks_per_bitmap_bit = log2_blocks_per_bitmap_bit;
3908         }
3909         n_bitmap_bits = ((ic->provided_data_sectors >> ic->sb->log2_sectors_per_block)
3910                                 + (((sector_t)1 << log2_blocks_per_bitmap_bit) - 1)) >> log2_blocks_per_bitmap_bit;
3911         ic->n_bitmap_blocks = DIV_ROUND_UP(n_bitmap_bits, BITMAP_BLOCK_SIZE * 8);
3912
3913         if (!ic->meta_dev)
3914                 ic->log2_buffer_sectors = min(ic->log2_buffer_sectors, (__u8)__ffs(ic->metadata_run));
3915
3916         if (ti->len > ic->provided_data_sectors) {
3917                 r = -EINVAL;
3918                 ti->error = "Not enough provided sectors for requested mapping size";
3919                 goto bad;
3920         }
3921
3922
3923         threshold = (__u64)ic->journal_entries * (100 - journal_watermark);
3924         threshold += 50;
3925         do_div(threshold, 100);
3926         ic->free_sectors_threshold = threshold;
3927
3928         DEBUG_print("initialized:\n");
3929         DEBUG_print("   integrity_tag_size %u\n", le16_to_cpu(ic->sb->integrity_tag_size));
3930         DEBUG_print("   journal_entry_size %u\n", ic->journal_entry_size);
3931         DEBUG_print("   journal_entries_per_sector %u\n", ic->journal_entries_per_sector);
3932         DEBUG_print("   journal_section_entries %u\n", ic->journal_section_entries);
3933         DEBUG_print("   journal_section_sectors %u\n", ic->journal_section_sectors);
3934         DEBUG_print("   journal_sections %u\n", (unsigned)le32_to_cpu(ic->sb->journal_sections));
3935         DEBUG_print("   journal_entries %u\n", ic->journal_entries);
3936         DEBUG_print("   log2_interleave_sectors %d\n", ic->sb->log2_interleave_sectors);
3937         DEBUG_print("   data_device_sectors 0x%llx\n", i_size_read(ic->dev->bdev->bd_inode) >> SECTOR_SHIFT);
3938         DEBUG_print("   initial_sectors 0x%x\n", ic->initial_sectors);
3939         DEBUG_print("   metadata_run 0x%x\n", ic->metadata_run);
3940         DEBUG_print("   log2_metadata_run %d\n", ic->log2_metadata_run);
3941         DEBUG_print("   provided_data_sectors 0x%llx (%llu)\n", (unsigned long long)ic->provided_data_sectors,
3942                     (unsigned long long)ic->provided_data_sectors);
3943         DEBUG_print("   log2_buffer_sectors %u\n", ic->log2_buffer_sectors);
3944         DEBUG_print("   bits_in_journal %llu\n", (unsigned long long)bits_in_journal);
3945
3946         if (ic->recalculate_flag && !(ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))) {
3947                 ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
3948                 ic->sb->recalc_sector = cpu_to_le64(0);
3949         }
3950
3951         if (ic->internal_hash) {
3952                 ic->recalc_wq = alloc_workqueue("dm-integrity-recalc", WQ_MEM_RECLAIM, 1);
3953                 if (!ic->recalc_wq ) {
3954                         ti->error = "Cannot allocate workqueue";
3955                         r = -ENOMEM;
3956                         goto bad;
3957                 }
3958                 INIT_WORK(&ic->recalc_work, integrity_recalc);
3959                 ic->recalc_buffer = vmalloc(RECALC_SECTORS << SECTOR_SHIFT);
3960                 if (!ic->recalc_buffer) {
3961                         ti->error = "Cannot allocate buffer for recalculating";
3962                         r = -ENOMEM;
3963                         goto bad;
3964                 }
3965                 ic->recalc_tags = kvmalloc_array(RECALC_SECTORS >> ic->sb->log2_sectors_per_block,
3966                                                  ic->tag_size, GFP_KERNEL);
3967                 if (!ic->recalc_tags) {
3968                         ti->error = "Cannot allocate tags for recalculating";
3969                         r = -ENOMEM;
3970                         goto bad;
3971                 }
3972         }
3973
3974         ic->bufio = dm_bufio_client_create(ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev,
3975                         1U << (SECTOR_SHIFT + ic->log2_buffer_sectors), 1, 0, NULL, NULL);
3976         if (IS_ERR(ic->bufio)) {
3977                 r = PTR_ERR(ic->bufio);
3978                 ti->error = "Cannot initialize dm-bufio";
3979                 ic->bufio = NULL;
3980                 goto bad;
3981         }
3982         dm_bufio_set_sector_offset(ic->bufio, ic->start + ic->initial_sectors);
3983
3984         if (ic->mode != 'R') {
3985                 r = create_journal(ic, &ti->error);
3986                 if (r)
3987                         goto bad;
3988
3989         }
3990
3991         if (ic->mode == 'B') {
3992                 unsigned i;
3993                 unsigned n_bitmap_pages = DIV_ROUND_UP(ic->n_bitmap_blocks, PAGE_SIZE / BITMAP_BLOCK_SIZE);
3994
3995                 ic->recalc_bitmap = dm_integrity_alloc_page_list(n_bitmap_pages);
3996                 if (!ic->recalc_bitmap) {
3997                         r = -ENOMEM;
3998                         goto bad;
3999                 }
4000                 ic->may_write_bitmap = dm_integrity_alloc_page_list(n_bitmap_pages);
4001                 if (!ic->may_write_bitmap) {
4002                         r = -ENOMEM;
4003                         goto bad;
4004                 }
4005                 ic->bbs = kvmalloc_array(ic->n_bitmap_blocks, sizeof(struct bitmap_block_status), GFP_KERNEL);
4006                 if (!ic->bbs) {
4007                         r = -ENOMEM;
4008                         goto bad;
4009                 }
4010                 INIT_DELAYED_WORK(&ic->bitmap_flush_work, bitmap_flush_work);
4011                 for (i = 0; i < ic->n_bitmap_blocks; i++) {
4012                         struct bitmap_block_status *bbs = &ic->bbs[i];
4013                         unsigned sector, pl_index, pl_offset;
4014
4015                         INIT_WORK(&bbs->work, bitmap_block_work);
4016                         bbs->ic = ic;
4017                         bbs->idx = i;
4018                         bio_list_init(&bbs->bio_queue);
4019                         spin_lock_init(&bbs->bio_queue_lock);
4020
4021                         sector = i * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT);
4022                         pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
4023                         pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
4024
4025                         bbs->bitmap = lowmem_page_address(ic->journal[pl_index].page) + pl_offset;
4026                 }
4027         }
4028
4029         if (should_write_sb) {
4030                 int r;
4031
4032                 init_journal(ic, 0, ic->journal_sections, 0);
4033                 r = dm_integrity_failed(ic);
4034                 if (unlikely(r)) {
4035                         ti->error = "Error initializing journal";
4036                         goto bad;
4037                 }
4038                 r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
4039                 if (r) {
4040                         ti->error = "Error initializing superblock";
4041                         goto bad;
4042                 }
4043                 ic->just_formatted = true;
4044         }
4045
4046         if (!ic->meta_dev) {
4047                 r = dm_set_target_max_io_len(ti, 1U << ic->sb->log2_interleave_sectors);
4048                 if (r)
4049                         goto bad;
4050         }
4051         if (ic->mode == 'B') {
4052                 unsigned max_io_len = ((sector_t)ic->sectors_per_block << ic->log2_blocks_per_bitmap_bit) * (BITMAP_BLOCK_SIZE * 8);
4053                 if (!max_io_len)
4054                         max_io_len = 1U << 31;
4055                 DEBUG_print("max_io_len: old %u, new %u\n", ti->max_io_len, max_io_len);
4056                 if (!ti->max_io_len || ti->max_io_len > max_io_len) {
4057                         r = dm_set_target_max_io_len(ti, max_io_len);
4058                         if (r)
4059                                 goto bad;
4060                 }
4061         }
4062
4063         if (!ic->internal_hash)
4064                 dm_integrity_set(ti, ic);
4065
4066         ti->num_flush_bios = 1;
4067         ti->flush_supported = true;
4068
4069         return 0;
4070
4071 bad:
4072         dm_integrity_dtr(ti);
4073         return r;
4074 }
4075
4076 static void dm_integrity_dtr(struct dm_target *ti)
4077 {
4078         struct dm_integrity_c *ic = ti->private;
4079
4080         BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
4081         BUG_ON(!list_empty(&ic->wait_list));
4082
4083         if (ic->metadata_wq)
4084                 destroy_workqueue(ic->metadata_wq);
4085         if (ic->wait_wq)
4086                 destroy_workqueue(ic->wait_wq);
4087         if (ic->commit_wq)
4088                 destroy_workqueue(ic->commit_wq);
4089         if (ic->writer_wq)
4090                 destroy_workqueue(ic->writer_wq);
4091         if (ic->recalc_wq)
4092                 destroy_workqueue(ic->recalc_wq);
4093         vfree(ic->recalc_buffer);
4094         kvfree(ic->recalc_tags);
4095         kvfree(ic->bbs);
4096         if (ic->bufio)
4097                 dm_bufio_client_destroy(ic->bufio);
4098         mempool_exit(&ic->journal_io_mempool);
4099         if (ic->io)
4100                 dm_io_client_destroy(ic->io);
4101         if (ic->dev)
4102                 dm_put_device(ti, ic->dev);
4103         if (ic->meta_dev)
4104                 dm_put_device(ti, ic->meta_dev);
4105         dm_integrity_free_page_list(ic->journal);
4106         dm_integrity_free_page_list(ic->journal_io);
4107         dm_integrity_free_page_list(ic->journal_xor);
4108         dm_integrity_free_page_list(ic->recalc_bitmap);
4109         dm_integrity_free_page_list(ic->may_write_bitmap);
4110         if (ic->journal_scatterlist)
4111                 dm_integrity_free_journal_scatterlist(ic, ic->journal_scatterlist);
4112         if (ic->journal_io_scatterlist)
4113                 dm_integrity_free_journal_scatterlist(ic, ic->journal_io_scatterlist);
4114         if (ic->sk_requests) {
4115                 unsigned i;
4116
4117                 for (i = 0; i < ic->journal_sections; i++) {
4118                         struct skcipher_request *req = ic->sk_requests[i];
4119                         if (req) {
4120                                 kzfree(req->iv);
4121                                 skcipher_request_free(req);
4122                         }
4123                 }
4124                 kvfree(ic->sk_requests);
4125         }
4126         kvfree(ic->journal_tree);
4127         if (ic->sb)
4128                 free_pages_exact(ic->sb, SB_SECTORS << SECTOR_SHIFT);
4129
4130         if (ic->internal_hash)
4131                 crypto_free_shash(ic->internal_hash);
4132         free_alg(&ic->internal_hash_alg);
4133
4134         if (ic->journal_crypt)
4135                 crypto_free_skcipher(ic->journal_crypt);
4136         free_alg(&ic->journal_crypt_alg);
4137
4138         if (ic->journal_mac)
4139                 crypto_free_shash(ic->journal_mac);
4140         free_alg(&ic->journal_mac_alg);
4141
4142         kfree(ic);
4143 }
4144
4145 static struct target_type integrity_target = {
4146         .name                   = "integrity",
4147         .version                = {1, 3, 0},
4148         .module                 = THIS_MODULE,
4149         .features               = DM_TARGET_SINGLETON | DM_TARGET_INTEGRITY,
4150         .ctr                    = dm_integrity_ctr,
4151         .dtr                    = dm_integrity_dtr,
4152         .map                    = dm_integrity_map,
4153         .postsuspend            = dm_integrity_postsuspend,
4154         .resume                 = dm_integrity_resume,
4155         .status                 = dm_integrity_status,
4156         .iterate_devices        = dm_integrity_iterate_devices,
4157         .io_hints               = dm_integrity_io_hints,
4158 };
4159
4160 static int __init dm_integrity_init(void)
4161 {
4162         int r;
4163
4164         journal_io_cache = kmem_cache_create("integrity_journal_io",
4165                                              sizeof(struct journal_io), 0, 0, NULL);
4166         if (!journal_io_cache) {
4167                 DMERR("can't allocate journal io cache");
4168                 return -ENOMEM;
4169         }
4170
4171         r = dm_register_target(&integrity_target);
4172
4173         if (r < 0)
4174                 DMERR("register failed %d", r);
4175
4176         return r;
4177 }
4178
4179 static void __exit dm_integrity_exit(void)
4180 {
4181         dm_unregister_target(&integrity_target);
4182         kmem_cache_destroy(journal_io_cache);
4183 }
4184
4185 module_init(dm_integrity_init);
4186 module_exit(dm_integrity_exit);
4187
4188 MODULE_AUTHOR("Milan Broz");
4189 MODULE_AUTHOR("Mikulas Patocka");
4190 MODULE_DESCRIPTION(DM_NAME " target for integrity tags extension");
4191 MODULE_LICENSE("GPL");