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 "dm-bio-record.h"
  10
  11 #include <linux/compiler.h>
  12 #include <linux/module.h>
  13 #include <linux/device-mapper.h>
  14 #include <linux/dm-io.h>
  15 #include <linux/vmalloc.h>
  16 #include <linux/sort.h>
  17 #include <linux/rbtree.h>
  18 #include <linux/delay.h>
  19 #include <linux/random.h>
  20 #include <linux/reboot.h>
  21 #include <crypto/hash.h>
  22 #include <crypto/skcipher.h>
  23 #include <linux/async_tx.h>
  24 #include <linux/dm-bufio.h>
  25
  26 #define DM_MSG_PREFIX "integrity"
  27
  28 #define DEFAULT_INTERLEAVE_SECTORS      32768
  29 #define DEFAULT_JOURNAL_SIZE_FACTOR     7
  30 #define DEFAULT_SECTORS_PER_BITMAP_BIT  32768
  31 #define DEFAULT_BUFFER_SECTORS          128
  32 #define DEFAULT_JOURNAL_WATERMARK       50
  33 #define DEFAULT_SYNC_MSEC               10000
  34 #define DEFAULT_MAX_JOURNAL_SECTORS     131072
  35 #define MIN_LOG2_INTERLEAVE_SECTORS     3
  36 #define MAX_LOG2_INTERLEAVE_SECTORS     31
  37 #define METADATA_WORKQUEUE_MAX_ACTIVE   16
  38 #define RECALC_SECTORS                  8192
  39 #define RECALC_WRITE_SUPER              16
  40 #define BITMAP_BLOCK_SIZE               4096    /* don't change it */
  41 #define BITMAP_FLUSH_INTERVAL           (10 * HZ)
  42
  43 /*
  44  * Warning - DEBUG_PRINT prints security-sensitive data to the log,
  45  * so it should not be enabled in the official kernel
  46  */
  47 //#define DEBUG_PRINT
  48 //#define INTERNAL_VERIFY
  49
  50 /*
  51  * On disk structures
  52  */
  53
  54 #define SB_MAGIC                        "integrt"
  55 #define SB_VERSION_1                    1
  56 #define SB_VERSION_2                    2
  57 #define SB_VERSION_3                    3
  58 #define SB_VERSION_4                    4
  59 #define SB_SECTORS                      8
  60 #define MAX_SECTORS_PER_BLOCK           8
  61
  62 struct superblock {
  63         __u8 magic[8];
  64         __u8 version;
  65         __u8 log2_interleave_sectors;
  66         __u16 integrity_tag_size;
  67         __u32 journal_sections;
  68         __u64 provided_data_sectors;    /* userspace uses this value */
  69         __u32 flags;
  70         __u8 log2_sectors_per_block;
  71         __u8 log2_blocks_per_bitmap_bit;
  72         __u8 pad[2];
  73         __u64 recalc_sector;
  74 };
  75
  76 #define SB_FLAG_HAVE_JOURNAL_MAC        0x1
  77 #define SB_FLAG_RECALCULATING           0x2
  78 #define SB_FLAG_DIRTY_BITMAP            0x4
  79 #define SB_FLAG_FIXED_PADDING           0x8
  80
  81 #define JOURNAL_ENTRY_ROUNDUP           8
  82
  83 typedef __u64 commit_id_t;
  84 #define JOURNAL_MAC_PER_SECTOR          8
  85
  86 struct journal_entry {
  87         union {
  88                 struct {
  89                         __u32 sector_lo;
  90                         __u32 sector_hi;
  91                 } s;
  92                 __u64 sector;
  93         } u;
  94         commit_id_t last_bytes[0];
  95         /* __u8 tag[0]; */
  96 };
  97
  98 #define journal_entry_tag(ic, je)               ((__u8 *)&(je)->last_bytes[(ic)->sectors_per_block])
  99
 100 #if BITS_PER_LONG == 64
 101 #define journal_entry_set_sector(je, x)         do { smp_wmb(); WRITE_ONCE((je)->u.sector, cpu_to_le64(x)); } while (0)
 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((x) >> 32)); } while (0)
 104 #endif
 105 #define journal_entry_get_sector(je)            le64_to_cpu((je)->u.sector)
 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
 207         int failed;
 208
 209         struct crypto_shash *internal_hash;
 210
 211         struct dm_target *ti;
 212
 213         /* these variables are locked with endio_wait.lock */
 214         struct rb_root in_progress;
 215         struct list_head wait_list;
 216         wait_queue_head_t endio_wait;
 217         struct workqueue_struct *wait_wq;
 218         struct workqueue_struct *offload_wq;
 219
 220         unsigned char commit_seq;
 221         commit_id_t commit_ids[N_COMMIT_IDS];
 222
 223         unsigned committed_section;
 224         unsigned n_committed_sections;
 225
 226         unsigned uncommitted_section;
 227         unsigned n_uncommitted_sections;
 228
 229         unsigned free_section;
 230         unsigned char free_section_entry;
 231         unsigned free_sectors;
 232
 233         unsigned free_sectors_threshold;
 234
 235         struct workqueue_struct *commit_wq;
 236         struct work_struct commit_work;
 237
 238         struct workqueue_struct *writer_wq;
 239         struct work_struct writer_work;
 240
 241         struct workqueue_struct *recalc_wq;
 242         struct work_struct recalc_work;
 243         u8 *recalc_buffer;
 244         u8 *recalc_tags;
 245
 246         struct bio_list flush_bio_list;
 247
 248         unsigned long autocommit_jiffies;
 249         struct timer_list autocommit_timer;
 250         unsigned autocommit_msec;
 251
 252         wait_queue_head_t copy_to_journal_wait;
 253
 254         struct completion crypto_backoff;
 255
 256         bool journal_uptodate;
 257         bool just_formatted;
 258         bool recalculate_flag;
 259         bool fix_padding;
 260
 261         struct alg_spec internal_hash_alg;
 262         struct alg_spec journal_crypt_alg;
 263         struct alg_spec journal_mac_alg;
 264
 265         atomic64_t number_of_mismatches;
 266
 267         struct notifier_block reboot_notifier;
 268 };
 269
 270 struct dm_integrity_range {
 271         sector_t logical_sector;
 272         sector_t n_sectors;
 273         bool waiting;
 274         union {
 275                 struct rb_node node;
 276                 struct {
 277                         struct task_struct *task;
 278                         struct list_head wait_entry;
 279                 };
 280         };
 281 };
 282
 283 struct dm_integrity_io {
 284         struct work_struct work;
 285
 286         struct dm_integrity_c *ic;
 287         bool write;
 288         bool fua;
 289
 290         struct dm_integrity_range range;
 291
 292         sector_t metadata_block;
 293         unsigned metadata_offset;
 294
 295         atomic_t in_flight;
 296         blk_status_t bi_status;
 297
 298         struct completion *completion;
 299
 300         struct dm_bio_details bio_details;
 301 };
 302
 303 struct journal_completion {
 304         struct dm_integrity_c *ic;
 305         atomic_t in_flight;
 306         struct completion comp;
 307 };
 308
 309 struct journal_io {
 310         struct dm_integrity_range range;
 311         struct journal_completion *comp;
 312 };
 313
 314 struct bitmap_block_status {
 315         struct work_struct work;
 316         struct dm_integrity_c *ic;
 317         unsigned idx;
 318         unsigned long *bitmap;
 319         struct bio_list bio_queue;
 320         spinlock_t bio_queue_lock;
 321
 322 };
 323
 324 static struct kmem_cache *journal_io_cache;
 325
 326 #define JOURNAL_IO_MEMPOOL      32
 327
 328 #ifdef DEBUG_PRINT
 329 #define DEBUG_print(x, ...)     printk(KERN_DEBUG x, ##__VA_ARGS__)
 330 static void __DEBUG_bytes(__u8 *bytes, size_t len, const char *msg, ...)
 331 {
 332         va_list args;
 333         va_start(args, msg);
 334         vprintk(msg, args);
 335         va_end(args);
 336         if (len)
 337                 pr_cont(":");
 338         while (len) {
 339                 pr_cont(" %02x", *bytes);
 340                 bytes++;
 341                 len--;
 342         }
 343         pr_cont("\n");
 344 }
 345 #define DEBUG_bytes(bytes, len, msg, ...)       __DEBUG_bytes(bytes, len, KERN_DEBUG msg, ##__VA_ARGS__)
 346 #else
 347 #define DEBUG_print(x, ...)                     do { } while (0)
 348 #define DEBUG_bytes(bytes, len, msg, ...)       do { } while (0)
 349 #endif
 350
 351 static void dm_integrity_prepare(struct request *rq)
 352 {
 353 }
 354
 355 static void dm_integrity_complete(struct request *rq, unsigned int nr_bytes)
 356 {
 357 }
 358
 359 /*
 360  * DM Integrity profile, protection is performed layer above (dm-crypt)
 361  */
 362 static const struct blk_integrity_profile dm_integrity_profile = {
 363         .name                   = "DM-DIF-EXT-TAG",
 364         .generate_fn            = NULL,
 365         .verify_fn              = NULL,
 366         .prepare_fn             = dm_integrity_prepare,
 367         .complete_fn            = dm_integrity_complete,
 368 };
 369
 370 static void dm_integrity_map_continue(struct dm_integrity_io *dio, bool from_map);
 371 static void integrity_bio_wait(struct work_struct *w);
 372 static void dm_integrity_dtr(struct dm_target *ti);
 373
 374 static void dm_integrity_io_error(struct dm_integrity_c *ic, const char *msg, int err)
 375 {
 376         if (err == -EILSEQ)
 377                 atomic64_inc(&ic->number_of_mismatches);
 378         if (!cmpxchg(&ic->failed, 0, err))
 379                 DMERR("Error on %s: %d", msg, err);
 380 }
 381
 382 static int dm_integrity_failed(struct dm_integrity_c *ic)
 383 {
 384         return READ_ONCE(ic->failed);
 385 }
 386
 387 static commit_id_t dm_integrity_commit_id(struct dm_integrity_c *ic, unsigned i,
 388                                           unsigned j, unsigned char seq)
 389 {
 390         /*
 391          * Xor the number with section and sector, so that if a piece of
 392          * journal is written at wrong place, it is detected.
 393          */
 394         return ic->commit_ids[seq] ^ cpu_to_le64(((__u64)i << 32) ^ j);
 395 }
 396
 397 static void get_area_and_offset(struct dm_integrity_c *ic, sector_t data_sector,
 398                                 sector_t *area, sector_t *offset)
 399 {
 400         if (!ic->meta_dev) {
 401                 __u8 log2_interleave_sectors = ic->sb->log2_interleave_sectors;
 402                 *area = data_sector >> log2_interleave_sectors;
 403                 *offset = (unsigned)data_sector & ((1U << log2_interleave_sectors) - 1);
 404         } else {
 405                 *area = 0;
 406                 *offset = data_sector;
 407         }
 408 }
 409
 410 #define sector_to_block(ic, n)                                          \
 411 do {                                                                    \
 412         BUG_ON((n) & (unsigned)((ic)->sectors_per_block - 1));          \
 413         (n) >>= (ic)->sb->log2_sectors_per_block;                       \
 414 } while (0)
 415
 416 static __u64 get_metadata_sector_and_offset(struct dm_integrity_c *ic, sector_t area,
 417                                             sector_t offset, unsigned *metadata_offset)
 418 {
 419         __u64 ms;
 420         unsigned mo;
 421
 422         ms = area << ic->sb->log2_interleave_sectors;
 423         if (likely(ic->log2_metadata_run >= 0))
 424                 ms += area << ic->log2_metadata_run;
 425         else
 426                 ms += area * ic->metadata_run;
 427         ms >>= ic->log2_buffer_sectors;
 428
 429         sector_to_block(ic, offset);
 430
 431         if (likely(ic->log2_tag_size >= 0)) {
 432                 ms += offset >> (SECTOR_SHIFT + ic->log2_buffer_sectors - ic->log2_tag_size);
 433                 mo = (offset << ic->log2_tag_size) & ((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - 1);
 434         } else {
 435                 ms += (__u64)offset * ic->tag_size >> (SECTOR_SHIFT + ic->log2_buffer_sectors);
 436                 mo = (offset * ic->tag_size) & ((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - 1);
 437         }
 438         *metadata_offset = mo;
 439         return ms;
 440 }
 441
 442 static sector_t get_data_sector(struct dm_integrity_c *ic, sector_t area, sector_t offset)
 443 {
 444         sector_t result;
 445
 446         if (ic->meta_dev)
 447                 return offset;
 448
 449         result = area << ic->sb->log2_interleave_sectors;
 450         if (likely(ic->log2_metadata_run >= 0))
 451                 result += (area + 1) << ic->log2_metadata_run;
 452         else
 453                 result += (area + 1) * ic->metadata_run;
 454
 455         result += (sector_t)ic->initial_sectors + offset;
 456         result += ic->start;
 457
 458         return result;
 459 }
 460
 461 static void wraparound_section(struct dm_integrity_c *ic, unsigned *sec_ptr)
 462 {
 463         if (unlikely(*sec_ptr >= ic->journal_sections))
 464                 *sec_ptr -= ic->journal_sections;
 465 }
 466
 467 static void sb_set_version(struct dm_integrity_c *ic)
 468 {
 469         if (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING))
 470                 ic->sb->version = SB_VERSION_4;
 471         else if (ic->mode == 'B' || ic->sb->flags & cpu_to_le32(SB_FLAG_DIRTY_BITMAP))
 472                 ic->sb->version = SB_VERSION_3;
 473         else if (ic->meta_dev || ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
 474                 ic->sb->version = SB_VERSION_2;
 475         else
 476                 ic->sb->version = SB_VERSION_1;
 477 }
 478
 479 static int sync_rw_sb(struct dm_integrity_c *ic, int op, int op_flags)
 480 {
 481         struct dm_io_request io_req;
 482         struct dm_io_region io_loc;
 483
 484         io_req.bi_op = op;
 485         io_req.bi_op_flags = op_flags;
 486         io_req.mem.type = DM_IO_KMEM;
 487         io_req.mem.ptr.addr = ic->sb;
 488         io_req.notify.fn = NULL;
 489         io_req.client = ic->io;
 490         io_loc.bdev = ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev;
 491         io_loc.sector = ic->start;
 492         io_loc.count = SB_SECTORS;
 493
 494         if (op == REQ_OP_WRITE)
 495                 sb_set_version(ic);
 496
 497         return dm_io(&io_req, 1, &io_loc, NULL);
 498 }
 499
 500 #define BITMAP_OP_TEST_ALL_SET          0
 501 #define BITMAP_OP_TEST_ALL_CLEAR        1
 502 #define BITMAP_OP_SET                   2
 503 #define BITMAP_OP_CLEAR                 3
 504
 505 static bool block_bitmap_op(struct dm_integrity_c *ic, struct page_list *bitmap,
 506                             sector_t sector, sector_t n_sectors, int mode)
 507 {
 508         unsigned long bit, end_bit, this_end_bit, page, end_page;
 509         unsigned long *data;
 510
 511         if (unlikely(((sector | n_sectors) & ((1 << ic->sb->log2_sectors_per_block) - 1)) != 0)) {
 512                 DMCRIT("invalid bitmap access (%llx,%llx,%d,%d,%d)",
 513                         sector,
 514                         n_sectors,
 515                         ic->sb->log2_sectors_per_block,
 516                         ic->log2_blocks_per_bitmap_bit,
 517                         mode);
 518                 BUG();
 519         }
 520
 521         if (unlikely(!n_sectors))
 522                 return true;
 523
 524         bit = sector >> (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
 525         end_bit = (sector + n_sectors - 1) >>
 526                 (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
 527
 528         page = bit / (PAGE_SIZE * 8);
 529         bit %= PAGE_SIZE * 8;
 530
 531         end_page = end_bit / (PAGE_SIZE * 8);
 532         end_bit %= PAGE_SIZE * 8;
 533
 534 repeat:
 535         if (page < end_page) {
 536                 this_end_bit = PAGE_SIZE * 8 - 1;
 537         } else {
 538                 this_end_bit = end_bit;
 539         }
 540
 541         data = lowmem_page_address(bitmap[page].page);
 542
 543         if (mode == BITMAP_OP_TEST_ALL_SET) {
 544                 while (bit <= this_end_bit) {
 545                         if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
 546                                 do {
 547                                         if (data[bit / BITS_PER_LONG] != -1)
 548                                                 return false;
 549                                         bit += BITS_PER_LONG;
 550                                 } while (this_end_bit >= bit + BITS_PER_LONG - 1);
 551                                 continue;
 552                         }
 553                         if (!test_bit(bit, data))
 554                                 return false;
 555                         bit++;
 556                 }
 557         } else if (mode == BITMAP_OP_TEST_ALL_CLEAR) {
 558                 while (bit <= this_end_bit) {
 559                         if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
 560                                 do {
 561                                         if (data[bit / BITS_PER_LONG] != 0)
 562                                                 return false;
 563                                         bit += BITS_PER_LONG;
 564                                 } while (this_end_bit >= bit + BITS_PER_LONG - 1);
 565                                 continue;
 566                         }
 567                         if (test_bit(bit, data))
 568                                 return false;
 569                         bit++;
 570                 }
 571         } else if (mode == BITMAP_OP_SET) {
 572                 while (bit <= this_end_bit) {
 573                         if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
 574                                 do {
 575                                         data[bit / BITS_PER_LONG] = -1;
 576                                         bit += BITS_PER_LONG;
 577                                 } while (this_end_bit >= bit + BITS_PER_LONG - 1);
 578                                 continue;
 579                         }
 580                         __set_bit(bit, data);
 581                         bit++;
 582                 }
 583         } else if (mode == BITMAP_OP_CLEAR) {
 584                 if (!bit && this_end_bit == PAGE_SIZE * 8 - 1)
 585                         clear_page(data);
 586                 else while (bit <= this_end_bit) {
 587                         if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
 588                                 do {
 589                                         data[bit / BITS_PER_LONG] = 0;
 590                                         bit += BITS_PER_LONG;
 591                                 } while (this_end_bit >= bit + BITS_PER_LONG - 1);
 592                                 continue;
 593                         }
 594                         __clear_bit(bit, data);
 595                         bit++;
 596                 }
 597         } else {
 598                 BUG();
 599         }
 600
 601         if (unlikely(page < end_page)) {
 602                 bit = 0;
 603                 page++;
 604                 goto repeat;
 605         }
 606
 607         return true;
 608 }
 609
 610 static void block_bitmap_copy(struct dm_integrity_c *ic, struct page_list *dst, struct page_list *src)
 611 {
 612         unsigned n_bitmap_pages = DIV_ROUND_UP(ic->n_bitmap_blocks, PAGE_SIZE / BITMAP_BLOCK_SIZE);
 613         unsigned i;
 614
 615         for (i = 0; i < n_bitmap_pages; i++) {
 616                 unsigned long *dst_data = lowmem_page_address(dst[i].page);
 617                 unsigned long *src_data = lowmem_page_address(src[i].page);
 618                 copy_page(dst_data, src_data);
 619         }
 620 }
 621
 622 static struct bitmap_block_status *sector_to_bitmap_block(struct dm_integrity_c *ic, sector_t sector)
 623 {
 624         unsigned bit = sector >> (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
 625         unsigned bitmap_block = bit / (BITMAP_BLOCK_SIZE * 8);
 626
 627         BUG_ON(bitmap_block >= ic->n_bitmap_blocks);
 628         return &ic->bbs[bitmap_block];
 629 }
 630
 631 static void access_journal_check(struct dm_integrity_c *ic, unsigned section, unsigned offset,
 632                                  bool e, const char *function)
 633 {
 634 #if defined(CONFIG_DM_DEBUG) || defined(INTERNAL_VERIFY)
 635         unsigned limit = e ? ic->journal_section_entries : ic->journal_section_sectors;
 636
 637         if (unlikely(section >= ic->journal_sections) ||
 638             unlikely(offset >= limit)) {
 639                 DMCRIT("%s: invalid access at (%u,%u), limit (%u,%u)",
 640                        function, section, offset, ic->journal_sections, limit);
 641                 BUG();
 642         }
 643 #endif
 644 }
 645
 646 static void page_list_location(struct dm_integrity_c *ic, unsigned section, unsigned offset,
 647                                unsigned *pl_index, unsigned *pl_offset)
 648 {
 649         unsigned sector;
 650
 651         access_journal_check(ic, section, offset, false, "page_list_location");
 652
 653         sector = section * ic->journal_section_sectors + offset;
 654
 655         *pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
 656         *pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
 657 }
 658
 659 static struct journal_sector *access_page_list(struct dm_integrity_c *ic, struct page_list *pl,
 660                                                unsigned section, unsigned offset, unsigned *n_sectors)
 661 {
 662         unsigned pl_index, pl_offset;
 663         char *va;
 664
 665         page_list_location(ic, section, offset, &pl_index, &pl_offset);
 666
 667         if (n_sectors)
 668                 *n_sectors = (PAGE_SIZE - pl_offset) >> SECTOR_SHIFT;
 669
 670         va = lowmem_page_address(pl[pl_index].page);
 671
 672         return (struct journal_sector *)(va + pl_offset);
 673 }
 674
 675 static struct journal_sector *access_journal(struct dm_integrity_c *ic, unsigned section, unsigned offset)
 676 {
 677         return access_page_list(ic, ic->journal, section, offset, NULL);
 678 }
 679
 680 static struct journal_entry *access_journal_entry(struct dm_integrity_c *ic, unsigned section, unsigned n)
 681 {
 682         unsigned rel_sector, offset;
 683         struct journal_sector *js;
 684
 685         access_journal_check(ic, section, n, true, "access_journal_entry");
 686
 687         rel_sector = n % JOURNAL_BLOCK_SECTORS;
 688         offset = n / JOURNAL_BLOCK_SECTORS;
 689
 690         js = access_journal(ic, section, rel_sector);
 691         return (struct journal_entry *)((char *)js + offset * ic->journal_entry_size);
 692 }
 693
 694 static struct journal_sector *access_journal_data(struct dm_integrity_c *ic, unsigned section, unsigned n)
 695 {
 696         n <<= ic->sb->log2_sectors_per_block;
 697
 698         n += JOURNAL_BLOCK_SECTORS;
 699
 700         access_journal_check(ic, section, n, false, "access_journal_data");
 701
 702         return access_journal(ic, section, n);
 703 }
 704
 705 static void section_mac(struct dm_integrity_c *ic, unsigned section, __u8 result[JOURNAL_MAC_SIZE])
 706 {
 707         SHASH_DESC_ON_STACK(desc, ic->journal_mac);
 708         int r;
 709         unsigned j, size;
 710
 711         desc->tfm = ic->journal_mac;
 712
 713         r = crypto_shash_init(desc);
 714         if (unlikely(r)) {
 715                 dm_integrity_io_error(ic, "crypto_shash_init", r);
 716                 goto err;
 717         }
 718
 719         for (j = 0; j < ic->journal_section_entries; j++) {
 720                 struct journal_entry *je = access_journal_entry(ic, section, j);
 721                 r = crypto_shash_update(desc, (__u8 *)&je->u.sector, sizeof je->u.sector);
 722                 if (unlikely(r)) {
 723                         dm_integrity_io_error(ic, "crypto_shash_update", r);
 724                         goto err;
 725                 }
 726         }
 727
 728         size = crypto_shash_digestsize(ic->journal_mac);
 729
 730         if (likely(size <= JOURNAL_MAC_SIZE)) {
 731                 r = crypto_shash_final(desc, result);
 732                 if (unlikely(r)) {
 733                         dm_integrity_io_error(ic, "crypto_shash_final", r);
 734                         goto err;
 735                 }
 736                 memset(result + size, 0, JOURNAL_MAC_SIZE - size);
 737         } else {
 738                 __u8 digest[HASH_MAX_DIGESTSIZE];
 739
 740                 if (WARN_ON(size > sizeof(digest))) {
 741                         dm_integrity_io_error(ic, "digest_size", -EINVAL);
 742                         goto err;
 743                 }
 744                 r = crypto_shash_final(desc, digest);
 745                 if (unlikely(r)) {
 746                         dm_integrity_io_error(ic, "crypto_shash_final", r);
 747                         goto err;
 748                 }
 749                 memcpy(result, digest, JOURNAL_MAC_SIZE);
 750         }
 751
 752         return;
 753 err:
 754         memset(result, 0, JOURNAL_MAC_SIZE);
 755 }
 756
 757 static void rw_section_mac(struct dm_integrity_c *ic, unsigned section, bool wr)
 758 {
 759         __u8 result[JOURNAL_MAC_SIZE];
 760         unsigned j;
 761
 762         if (!ic->journal_mac)
 763                 return;
 764
 765         section_mac(ic, section, result);
 766
 767         for (j = 0; j < JOURNAL_BLOCK_SECTORS; j++) {
 768                 struct journal_sector *js = access_journal(ic, section, j);
 769
 770                 if (likely(wr))
 771                         memcpy(&js->mac, result + (j * JOURNAL_MAC_PER_SECTOR), JOURNAL_MAC_PER_SECTOR);
 772                 else {
 773                         if (memcmp(&js->mac, result + (j * JOURNAL_MAC_PER_SECTOR), JOURNAL_MAC_PER_SECTOR))
 774                                 dm_integrity_io_error(ic, "journal mac", -EILSEQ);
 775                 }
 776         }
 777 }
 778
 779 static void complete_journal_op(void *context)
 780 {
 781         struct journal_completion *comp = context;
 782         BUG_ON(!atomic_read(&comp->in_flight));
 783         if (likely(atomic_dec_and_test(&comp->in_flight)))
 784                 complete(&comp->comp);
 785 }
 786
 787 static void xor_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
 788                         unsigned n_sections, struct journal_completion *comp)
 789 {
 790         struct async_submit_ctl submit;
 791         size_t n_bytes = (size_t)(n_sections * ic->journal_section_sectors) << SECTOR_SHIFT;
 792         unsigned pl_index, pl_offset, section_index;
 793         struct page_list *source_pl, *target_pl;
 794
 795         if (likely(encrypt)) {
 796                 source_pl = ic->journal;
 797                 target_pl = ic->journal_io;
 798         } else {
 799                 source_pl = ic->journal_io;
 800                 target_pl = ic->journal;
 801         }
 802
 803         page_list_location(ic, section, 0, &pl_index, &pl_offset);
 804
 805         atomic_add(roundup(pl_offset + n_bytes, PAGE_SIZE) >> PAGE_SHIFT, &comp->in_flight);
 806
 807         init_async_submit(&submit, ASYNC_TX_XOR_ZERO_DST, NULL, complete_journal_op, comp, NULL);
 808
 809         section_index = pl_index;
 810
 811         do {
 812                 size_t this_step;
 813                 struct page *src_pages[2];
 814                 struct page *dst_page;
 815
 816                 while (unlikely(pl_index == section_index)) {
 817                         unsigned dummy;
 818                         if (likely(encrypt))
 819                                 rw_section_mac(ic, section, true);
 820                         section++;
 821                         n_sections--;
 822                         if (!n_sections)
 823                                 break;
 824                         page_list_location(ic, section, 0, &section_index, &dummy);
 825                 }
 826
 827                 this_step = min(n_bytes, (size_t)PAGE_SIZE - pl_offset);
 828                 dst_page = target_pl[pl_index].page;
 829                 src_pages[0] = source_pl[pl_index].page;
 830                 src_pages[1] = ic->journal_xor[pl_index].page;
 831
 832                 async_xor(dst_page, src_pages, pl_offset, 2, this_step, &submit);
 833
 834                 pl_index++;
 835                 pl_offset = 0;
 836                 n_bytes -= this_step;
 837         } while (n_bytes);
 838
 839         BUG_ON(n_sections);
 840
 841         async_tx_issue_pending_all();
 842 }
 843
 844 static void complete_journal_encrypt(struct crypto_async_request *req, int err)
 845 {
 846         struct journal_completion *comp = req->data;
 847         if (unlikely(err)) {
 848                 if (likely(err == -EINPROGRESS)) {
 849                         complete(&comp->ic->crypto_backoff);
 850                         return;
 851                 }
 852                 dm_integrity_io_error(comp->ic, "asynchronous encrypt", err);
 853         }
 854         complete_journal_op(comp);
 855 }
 856
 857 static bool do_crypt(bool encrypt, struct skcipher_request *req, struct journal_completion *comp)
 858 {
 859         int r;
 860         skcipher_request_set_callback(req, CRYPTO_TFM_REQ_MAY_BACKLOG,
 861                                       complete_journal_encrypt, comp);
 862         if (likely(encrypt))
 863                 r = crypto_skcipher_encrypt(req);
 864         else
 865                 r = crypto_skcipher_decrypt(req);
 866         if (likely(!r))
 867                 return false;
 868         if (likely(r == -EINPROGRESS))
 869                 return true;
 870         if (likely(r == -EBUSY)) {
 871                 wait_for_completion(&comp->ic->crypto_backoff);
 872                 reinit_completion(&comp->ic->crypto_backoff);
 873                 return true;
 874         }
 875         dm_integrity_io_error(comp->ic, "encrypt", r);
 876         return false;
 877 }
 878
 879 static void crypt_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
 880                           unsigned n_sections, struct journal_completion *comp)
 881 {
 882         struct scatterlist **source_sg;
 883         struct scatterlist **target_sg;
 884
 885         atomic_add(2, &comp->in_flight);
 886
 887         if (likely(encrypt)) {
 888                 source_sg = ic->journal_scatterlist;
 889                 target_sg = ic->journal_io_scatterlist;
 890         } else {
 891                 source_sg = ic->journal_io_scatterlist;
 892                 target_sg = ic->journal_scatterlist;
 893         }
 894
 895         do {
 896                 struct skcipher_request *req;
 897                 unsigned ivsize;
 898                 char *iv;
 899
 900                 if (likely(encrypt))
 901                         rw_section_mac(ic, section, true);
 902
 903                 req = ic->sk_requests[section];
 904                 ivsize = crypto_skcipher_ivsize(ic->journal_crypt);
 905                 iv = req->iv;
 906
 907                 memcpy(iv, iv + ivsize, ivsize);
 908
 909                 req->src = source_sg[section];
 910                 req->dst = target_sg[section];
 911
 912                 if (unlikely(do_crypt(encrypt, req, comp)))
 913                         atomic_inc(&comp->in_flight);
 914
 915                 section++;
 916                 n_sections--;
 917         } while (n_sections);
 918
 919         atomic_dec(&comp->in_flight);
 920         complete_journal_op(comp);
 921 }
 922
 923 static void encrypt_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
 924                             unsigned n_sections, struct journal_completion *comp)
 925 {
 926         if (ic->journal_xor)
 927                 return xor_journal(ic, encrypt, section, n_sections, comp);
 928         else
 929                 return crypt_journal(ic, encrypt, section, n_sections, comp);
 930 }
 931
 932 static void complete_journal_io(unsigned long error, void *context)
 933 {
 934         struct journal_completion *comp = context;
 935         if (unlikely(error != 0))
 936                 dm_integrity_io_error(comp->ic, "writing journal", -EIO);
 937         complete_journal_op(comp);
 938 }
 939
 940 static void rw_journal_sectors(struct dm_integrity_c *ic, int op, int op_flags,
 941                                unsigned sector, unsigned n_sectors, struct journal_completion *comp)
 942 {
 943         struct dm_io_request io_req;
 944         struct dm_io_region io_loc;
 945         unsigned pl_index, pl_offset;
 946         int r;
 947
 948         if (unlikely(dm_integrity_failed(ic))) {
 949                 if (comp)
 950                         complete_journal_io(-1UL, comp);
 951                 return;
 952         }
 953
 954         pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
 955         pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
 956
 957         io_req.bi_op = op;
 958         io_req.bi_op_flags = op_flags;
 959         io_req.mem.type = DM_IO_PAGE_LIST;
 960         if (ic->journal_io)
 961                 io_req.mem.ptr.pl = &ic->journal_io[pl_index];
 962         else
 963                 io_req.mem.ptr.pl = &ic->journal[pl_index];
 964         io_req.mem.offset = pl_offset;
 965         if (likely(comp != NULL)) {
 966                 io_req.notify.fn = complete_journal_io;
 967                 io_req.notify.context = comp;
 968         } else {
 969                 io_req.notify.fn = NULL;
 970         }
 971         io_req.client = ic->io;
 972         io_loc.bdev = ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev;
 973         io_loc.sector = ic->start + SB_SECTORS + sector;
 974         io_loc.count = n_sectors;
 975
 976         r = dm_io(&io_req, 1, &io_loc, NULL);
 977         if (unlikely(r)) {
 978                 dm_integrity_io_error(ic, op == REQ_OP_READ ? "reading journal" : "writing journal", r);
 979                 if (comp) {
 980                         WARN_ONCE(1, "asynchronous dm_io failed: %d", r);
 981                         complete_journal_io(-1UL, comp);
 982                 }
 983         }
 984 }
 985
 986 static void rw_journal(struct dm_integrity_c *ic, int op, int op_flags, unsigned section,
 987                        unsigned n_sections, struct journal_completion *comp)
 988 {
 989         unsigned sector, n_sectors;
 990
 991         sector = section * ic->journal_section_sectors;
 992         n_sectors = n_sections * ic->journal_section_sectors;
 993
 994         rw_journal_sectors(ic, op, op_flags, sector, n_sectors, comp);
 995 }
 996
 997 static void write_journal(struct dm_integrity_c *ic, unsigned commit_start, unsigned commit_sections)
 998 {
 999         struct journal_completion io_comp;
1000         struct journal_completion crypt_comp_1;
1001         struct journal_completion crypt_comp_2;
1002         unsigned i;
1003
1004         io_comp.ic = ic;
1005         init_completion(&io_comp.comp);
1006
1007         if (commit_start + commit_sections <= ic->journal_sections) {
1008                 io_comp.in_flight = (atomic_t)ATOMIC_INIT(1);
1009                 if (ic->journal_io) {
1010                         crypt_comp_1.ic = ic;
1011                         init_completion(&crypt_comp_1.comp);
1012                         crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
1013                         encrypt_journal(ic, true, commit_start, commit_sections, &crypt_comp_1);
1014                         wait_for_completion_io(&crypt_comp_1.comp);
1015                 } else {
1016                         for (i = 0; i < commit_sections; i++)
1017                                 rw_section_mac(ic, commit_start + i, true);
1018                 }
1019                 rw_journal(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, commit_start,
1020                            commit_sections, &io_comp);
1021         } else {
1022                 unsigned to_end;
1023                 io_comp.in_flight = (atomic_t)ATOMIC_INIT(2);
1024                 to_end = ic->journal_sections - commit_start;
1025                 if (ic->journal_io) {
1026                         crypt_comp_1.ic = ic;
1027                         init_completion(&crypt_comp_1.comp);
1028                         crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
1029                         encrypt_journal(ic, true, commit_start, to_end, &crypt_comp_1);
1030                         if (try_wait_for_completion(&crypt_comp_1.comp)) {
1031                                 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
1032                                 reinit_completion(&crypt_comp_1.comp);
1033                                 crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
1034                                 encrypt_journal(ic, true, 0, commit_sections - to_end, &crypt_comp_1);
1035                                 wait_for_completion_io(&crypt_comp_1.comp);
1036                         } else {
1037                                 crypt_comp_2.ic = ic;
1038                                 init_completion(&crypt_comp_2.comp);
1039                                 crypt_comp_2.in_flight = (atomic_t)ATOMIC_INIT(0);
1040                                 encrypt_journal(ic, true, 0, commit_sections - to_end, &crypt_comp_2);
1041                                 wait_for_completion_io(&crypt_comp_1.comp);
1042                                 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
1043                                 wait_for_completion_io(&crypt_comp_2.comp);
1044                         }
1045                 } else {
1046                         for (i = 0; i < to_end; i++)
1047                                 rw_section_mac(ic, commit_start + i, true);
1048                         rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
1049                         for (i = 0; i < commit_sections - to_end; i++)
1050                                 rw_section_mac(ic, i, true);
1051                 }
1052                 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, 0, commit_sections - to_end, &io_comp);
1053         }
1054
1055         wait_for_completion_io(&io_comp.comp);
1056 }
1057
1058 static void copy_from_journal(struct dm_integrity_c *ic, unsigned section, unsigned offset,
1059                               unsigned n_sectors, sector_t target, io_notify_fn fn, void *data)
1060 {
1061         struct dm_io_request io_req;
1062         struct dm_io_region io_loc;
1063         int r;
1064         unsigned sector, pl_index, pl_offset;
1065
1066         BUG_ON((target | n_sectors | offset) & (unsigned)(ic->sectors_per_block - 1));
1067
1068         if (unlikely(dm_integrity_failed(ic))) {
1069                 fn(-1UL, data);
1070                 return;
1071         }
1072
1073         sector = section * ic->journal_section_sectors + JOURNAL_BLOCK_SECTORS + offset;
1074
1075         pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
1076         pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
1077
1078         io_req.bi_op = REQ_OP_WRITE;
1079         io_req.bi_op_flags = 0;
1080         io_req.mem.type = DM_IO_PAGE_LIST;
1081         io_req.mem.ptr.pl = &ic->journal[pl_index];
1082         io_req.mem.offset = pl_offset;
1083         io_req.notify.fn = fn;
1084         io_req.notify.context = data;
1085         io_req.client = ic->io;
1086         io_loc.bdev = ic->dev->bdev;
1087         io_loc.sector = target;
1088         io_loc.count = n_sectors;
1089
1090         r = dm_io(&io_req, 1, &io_loc, NULL);
1091         if (unlikely(r)) {
1092                 WARN_ONCE(1, "asynchronous dm_io failed: %d", r);
1093                 fn(-1UL, data);
1094         }
1095 }
1096
1097 static bool ranges_overlap(struct dm_integrity_range *range1, struct dm_integrity_range *range2)
1098 {
1099         return range1->logical_sector < range2->logical_sector + range2->n_sectors &&
1100                range1->logical_sector + range1->n_sectors > range2->logical_sector;
1101 }
1102
1103 static bool add_new_range(struct dm_integrity_c *ic, struct dm_integrity_range *new_range, bool check_waiting)
1104 {
1105         struct rb_node **n = &ic->in_progress.rb_node;
1106         struct rb_node *parent;
1107
1108         BUG_ON((new_range->logical_sector | new_range->n_sectors) & (unsigned)(ic->sectors_per_block - 1));
1109
1110         if (likely(check_waiting)) {
1111                 struct dm_integrity_range *range;
1112                 list_for_each_entry(range, &ic->wait_list, wait_entry) {
1113                         if (unlikely(ranges_overlap(range, new_range)))
1114                                 return false;
1115                 }
1116         }
1117
1118         parent = NULL;
1119
1120         while (*n) {
1121                 struct dm_integrity_range *range = container_of(*n, struct dm_integrity_range, node);
1122
1123                 parent = *n;
1124                 if (new_range->logical_sector + new_range->n_sectors <= range->logical_sector) {
1125                         n = &range->node.rb_left;
1126                 } else if (new_range->logical_sector >= range->logical_sector + range->n_sectors) {
1127                         n = &range->node.rb_right;
1128                 } else {
1129                         return false;
1130                 }
1131         }
1132
1133         rb_link_node(&new_range->node, parent, n);
1134         rb_insert_color(&new_range->node, &ic->in_progress);
1135
1136         return true;
1137 }
1138
1139 static void remove_range_unlocked(struct dm_integrity_c *ic, struct dm_integrity_range *range)
1140 {
1141         rb_erase(&range->node, &ic->in_progress);
1142         while (unlikely(!list_empty(&ic->wait_list))) {
1143                 struct dm_integrity_range *last_range =
1144                         list_first_entry(&ic->wait_list, struct dm_integrity_range, wait_entry);
1145                 struct task_struct *last_range_task;
1146                 last_range_task = last_range->task;
1147                 list_del(&last_range->wait_entry);
1148                 if (!add_new_range(ic, last_range, false)) {
1149                         last_range->task = last_range_task;
1150                         list_add(&last_range->wait_entry, &ic->wait_list);
1151                         break;
1152                 }
1153                 last_range->waiting = false;
1154                 wake_up_process(last_range_task);
1155         }
1156 }
1157
1158 static void remove_range(struct dm_integrity_c *ic, struct dm_integrity_range *range)
1159 {
1160         unsigned long flags;
1161
1162         spin_lock_irqsave(&ic->endio_wait.lock, flags);
1163         remove_range_unlocked(ic, range);
1164         spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
1165 }
1166
1167 static void wait_and_add_new_range(struct dm_integrity_c *ic, struct dm_integrity_range *new_range)
1168 {
1169         new_range->waiting = true;
1170         list_add_tail(&new_range->wait_entry, &ic->wait_list);
1171         new_range->task = current;
1172         do {
1173                 __set_current_state(TASK_UNINTERRUPTIBLE);
1174                 spin_unlock_irq(&ic->endio_wait.lock);
1175                 io_schedule();
1176                 spin_lock_irq(&ic->endio_wait.lock);
1177         } while (unlikely(new_range->waiting));
1178 }
1179
1180 static void add_new_range_and_wait(struct dm_integrity_c *ic, struct dm_integrity_range *new_range)
1181 {
1182         if (unlikely(!add_new_range(ic, new_range, true)))
1183                 wait_and_add_new_range(ic, new_range);
1184 }
1185
1186 static void init_journal_node(struct journal_node *node)
1187 {
1188         RB_CLEAR_NODE(&node->node);
1189         node->sector = (sector_t)-1;
1190 }
1191
1192 static void add_journal_node(struct dm_integrity_c *ic, struct journal_node *node, sector_t sector)
1193 {
1194         struct rb_node **link;
1195         struct rb_node *parent;
1196
1197         node->sector = sector;
1198         BUG_ON(!RB_EMPTY_NODE(&node->node));
1199
1200         link = &ic->journal_tree_root.rb_node;
1201         parent = NULL;
1202
1203         while (*link) {
1204                 struct journal_node *j;
1205                 parent = *link;
1206                 j = container_of(parent, struct journal_node, node);
1207                 if (sector < j->sector)
1208                         link = &j->node.rb_left;
1209                 else
1210                         link = &j->node.rb_right;
1211         }
1212
1213         rb_link_node(&node->node, parent, link);
1214         rb_insert_color(&node->node, &ic->journal_tree_root);
1215 }
1216
1217 static void remove_journal_node(struct dm_integrity_c *ic, struct journal_node *node)
1218 {
1219         BUG_ON(RB_EMPTY_NODE(&node->node));
1220         rb_erase(&node->node, &ic->journal_tree_root);
1221         init_journal_node(node);
1222 }
1223
1224 #define NOT_FOUND       (-1U)
1225
1226 static unsigned find_journal_node(struct dm_integrity_c *ic, sector_t sector, sector_t *next_sector)
1227 {
1228         struct rb_node *n = ic->journal_tree_root.rb_node;
1229         unsigned found = NOT_FOUND;
1230         *next_sector = (sector_t)-1;
1231         while (n) {
1232                 struct journal_node *j = container_of(n, struct journal_node, node);
1233                 if (sector == j->sector) {
1234                         found = j - ic->journal_tree;
1235                 }
1236                 if (sector < j->sector) {
1237                         *next_sector = j->sector;
1238                         n = j->node.rb_left;
1239                 } else {
1240                         n = j->node.rb_right;
1241                 }
1242         }
1243
1244         return found;
1245 }
1246
1247 static bool test_journal_node(struct dm_integrity_c *ic, unsigned pos, sector_t sector)
1248 {
1249         struct journal_node *node, *next_node;
1250         struct rb_node *next;
1251
1252         if (unlikely(pos >= ic->journal_entries))
1253                 return false;
1254         node = &ic->journal_tree[pos];
1255         if (unlikely(RB_EMPTY_NODE(&node->node)))
1256                 return false;
1257         if (unlikely(node->sector != sector))
1258                 return false;
1259
1260         next = rb_next(&node->node);
1261         if (unlikely(!next))
1262                 return true;
1263
1264         next_node = container_of(next, struct journal_node, node);
1265         return next_node->sector != sector;
1266 }
1267
1268 static bool find_newer_committed_node(struct dm_integrity_c *ic, struct journal_node *node)
1269 {
1270         struct rb_node *next;
1271         struct journal_node *next_node;
1272         unsigned next_section;
1273
1274         BUG_ON(RB_EMPTY_NODE(&node->node));
1275
1276         next = rb_next(&node->node);
1277         if (unlikely(!next))
1278                 return false;
1279
1280         next_node = container_of(next, struct journal_node, node);
1281
1282         if (next_node->sector != node->sector)
1283                 return false;
1284
1285         next_section = (unsigned)(next_node - ic->journal_tree) / ic->journal_section_entries;
1286         if (next_section >= ic->committed_section &&
1287             next_section < ic->committed_section + ic->n_committed_sections)
1288                 return true;
1289         if (next_section + ic->journal_sections < ic->committed_section + ic->n_committed_sections)
1290                 return true;
1291
1292         return false;
1293 }
1294
1295 #define TAG_READ        0
1296 #define TAG_WRITE       1
1297 #define TAG_CMP         2
1298
1299 static int dm_integrity_rw_tag(struct dm_integrity_c *ic, unsigned char *tag, sector_t *metadata_block,
1300                                unsigned *metadata_offset, unsigned total_size, int op)
1301 {
1302         do {
1303                 unsigned char *data, *dp;
1304                 struct dm_buffer *b;
1305                 unsigned to_copy;
1306                 int r;
1307
1308                 r = dm_integrity_failed(ic);
1309                 if (unlikely(r))
1310                         return r;
1311
1312                 data = dm_bufio_read(ic->bufio, *metadata_block, &b);
1313                 if (IS_ERR(data))
1314                         return PTR_ERR(data);
1315
1316                 to_copy = min((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - *metadata_offset, total_size);
1317                 dp = data + *metadata_offset;
1318                 if (op == TAG_READ) {
1319                         memcpy(tag, dp, to_copy);
1320                 } else if (op == TAG_WRITE) {
1321                         memcpy(dp, tag, to_copy);
1322                         dm_bufio_mark_partial_buffer_dirty(b, *metadata_offset, *metadata_offset + to_copy);
1323                 } else  {
1324                         /* e.g.: op == TAG_CMP */
1325                         if (unlikely(memcmp(dp, tag, to_copy))) {
1326                                 unsigned i;
1327
1328                                 for (i = 0; i < to_copy; i++) {
1329                                         if (dp[i] != tag[i])
1330                                                 break;
1331                                         total_size--;
1332                                 }
1333                                 dm_bufio_release(b);
1334                                 return total_size;
1335                         }
1336                 }
1337                 dm_bufio_release(b);
1338
1339                 tag += to_copy;
1340                 *metadata_offset += to_copy;
1341                 if (unlikely(*metadata_offset == 1U << SECTOR_SHIFT << ic->log2_buffer_sectors)) {
1342                         (*metadata_block)++;
1343                         *metadata_offset = 0;
1344                 }
1345                 total_size -= to_copy;
1346         } while (unlikely(total_size));
1347
1348         return 0;
1349 }
1350
1351 static void dm_integrity_flush_buffers(struct dm_integrity_c *ic)
1352 {
1353         int r;
1354         r = dm_bufio_write_dirty_buffers(ic->bufio);
1355         if (unlikely(r))
1356                 dm_integrity_io_error(ic, "writing tags", r);
1357 }
1358
1359 static void sleep_on_endio_wait(struct dm_integrity_c *ic)
1360 {
1361         DECLARE_WAITQUEUE(wait, current);
1362         __add_wait_queue(&ic->endio_wait, &wait);
1363         __set_current_state(TASK_UNINTERRUPTIBLE);
1364         spin_unlock_irq(&ic->endio_wait.lock);
1365         io_schedule();
1366         spin_lock_irq(&ic->endio_wait.lock);
1367         __remove_wait_queue(&ic->endio_wait, &wait);
1368 }
1369
1370 static void autocommit_fn(struct timer_list *t)
1371 {
1372         struct dm_integrity_c *ic = from_timer(ic, t, autocommit_timer);
1373
1374         if (likely(!dm_integrity_failed(ic)))
1375                 queue_work(ic->commit_wq, &ic->commit_work);
1376 }
1377
1378 static void schedule_autocommit(struct dm_integrity_c *ic)
1379 {
1380         if (!timer_pending(&ic->autocommit_timer))
1381                 mod_timer(&ic->autocommit_timer, jiffies + ic->autocommit_jiffies);
1382 }
1383
1384 static void submit_flush_bio(struct dm_integrity_c *ic, struct dm_integrity_io *dio)
1385 {
1386         struct bio *bio;
1387         unsigned long flags;
1388
1389         spin_lock_irqsave(&ic->endio_wait.lock, flags);
1390         bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1391         bio_list_add(&ic->flush_bio_list, bio);
1392         spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
1393
1394         queue_work(ic->commit_wq, &ic->commit_work);
1395 }
1396
1397 static void do_endio(struct dm_integrity_c *ic, struct bio *bio)
1398 {
1399         int r = dm_integrity_failed(ic);
1400         if (unlikely(r) && !bio->bi_status)
1401                 bio->bi_status = errno_to_blk_status(r);
1402         if (unlikely(ic->synchronous_mode) && bio_op(bio) == REQ_OP_WRITE) {
1403                 unsigned long flags;
1404                 spin_lock_irqsave(&ic->endio_wait.lock, flags);
1405                 bio_list_add(&ic->synchronous_bios, bio);
1406                 queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, 0);
1407                 spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
1408                 return;
1409         }
1410         bio_endio(bio);
1411 }
1412
1413 static void do_endio_flush(struct dm_integrity_c *ic, struct dm_integrity_io *dio)
1414 {
1415         struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1416
1417         if (unlikely(dio->fua) && likely(!bio->bi_status) && likely(!dm_integrity_failed(ic)))
1418                 submit_flush_bio(ic, dio);
1419         else
1420                 do_endio(ic, bio);
1421 }
1422
1423 static void dec_in_flight(struct dm_integrity_io *dio)
1424 {
1425         if (atomic_dec_and_test(&dio->in_flight)) {
1426                 struct dm_integrity_c *ic = dio->ic;
1427                 struct bio *bio;
1428
1429                 remove_range(ic, &dio->range);
1430
1431                 if (unlikely(dio->write))
1432                         schedule_autocommit(ic);
1433
1434                 bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1435
1436                 if (unlikely(dio->bi_status) && !bio->bi_status)
1437                         bio->bi_status = dio->bi_status;
1438                 if (likely(!bio->bi_status) && unlikely(bio_sectors(bio) != dio->range.n_sectors)) {
1439                         dio->range.logical_sector += dio->range.n_sectors;
1440                         bio_advance(bio, dio->range.n_sectors << SECTOR_SHIFT);
1441                         INIT_WORK(&dio->work, integrity_bio_wait);
1442                         queue_work(ic->offload_wq, &dio->work);
1443                         return;
1444                 }
1445                 do_endio_flush(ic, dio);
1446         }
1447 }
1448
1449 static void integrity_end_io(struct bio *bio)
1450 {
1451         struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
1452
1453         dm_bio_restore(&dio->bio_details, bio);
1454         if (bio->bi_integrity)
1455                 bio->bi_opf |= REQ_INTEGRITY;
1456
1457         if (dio->completion)
1458                 complete(dio->completion);
1459
1460         dec_in_flight(dio);
1461 }
1462
1463 static void integrity_sector_checksum(struct dm_integrity_c *ic, sector_t sector,
1464                                       const char *data, char *result)
1465 {
1466         __u64 sector_le = cpu_to_le64(sector);
1467         SHASH_DESC_ON_STACK(req, ic->internal_hash);
1468         int r;
1469         unsigned digest_size;
1470
1471         req->tfm = ic->internal_hash;
1472
1473         r = crypto_shash_init(req);
1474         if (unlikely(r < 0)) {
1475                 dm_integrity_io_error(ic, "crypto_shash_init", r);
1476                 goto failed;
1477         }
1478
1479         r = crypto_shash_update(req, (const __u8 *)&sector_le, sizeof sector_le);
1480         if (unlikely(r < 0)) {
1481                 dm_integrity_io_error(ic, "crypto_shash_update", r);
1482                 goto failed;
1483         }
1484
1485         r = crypto_shash_update(req, data, ic->sectors_per_block << SECTOR_SHIFT);
1486         if (unlikely(r < 0)) {
1487                 dm_integrity_io_error(ic, "crypto_shash_update", r);
1488                 goto failed;
1489         }
1490
1491         r = crypto_shash_final(req, result);
1492         if (unlikely(r < 0)) {
1493                 dm_integrity_io_error(ic, "crypto_shash_final", r);
1494                 goto failed;
1495         }
1496
1497         digest_size = crypto_shash_digestsize(ic->internal_hash);
1498         if (unlikely(digest_size < ic->tag_size))
1499                 memset(result + digest_size, 0, ic->tag_size - digest_size);
1500
1501         return;
1502
1503 failed:
1504         /* this shouldn't happen anyway, the hash functions have no reason to fail */
1505         get_random_bytes(result, ic->tag_size);
1506 }
1507
1508 static void integrity_metadata(struct work_struct *w)
1509 {
1510         struct dm_integrity_io *dio = container_of(w, struct dm_integrity_io, work);
1511         struct dm_integrity_c *ic = dio->ic;
1512
1513         int r;
1514
1515         if (ic->internal_hash) {
1516                 struct bvec_iter iter;
1517                 struct bio_vec bv;
1518                 unsigned digest_size = crypto_shash_digestsize(ic->internal_hash);
1519                 struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1520                 char *checksums;
1521                 unsigned extra_space = unlikely(digest_size > ic->tag_size) ? digest_size - ic->tag_size : 0;
1522                 char checksums_onstack[max((size_t)HASH_MAX_DIGESTSIZE, MAX_TAG_SIZE)];
1523                 unsigned sectors_to_process = dio->range.n_sectors;
1524                 sector_t sector = dio->range.logical_sector;
1525
1526                 if (unlikely(ic->mode == 'R'))
1527                         goto skip_io;
1528
1529                 checksums = kmalloc((PAGE_SIZE >> SECTOR_SHIFT >> ic->sb->log2_sectors_per_block) * ic->tag_size + extra_space,
1530                                     GFP_NOIO | __GFP_NORETRY | __GFP_NOWARN);
1531                 if (!checksums) {
1532                         checksums = checksums_onstack;
1533                         if (WARN_ON(extra_space &&
1534                                     digest_size > sizeof(checksums_onstack))) {
1535                                 r = -EINVAL;
1536                                 goto error;
1537                         }
1538                 }
1539
1540                 __bio_for_each_segment(bv, bio, iter, dio->bio_details.bi_iter) {
1541                         unsigned pos;
1542                         char *mem, *checksums_ptr;
1543
1544 again:
1545                         mem = (char *)kmap_atomic(bv.bv_page) + bv.bv_offset;
1546                         pos = 0;
1547                         checksums_ptr = checksums;
1548                         do {
1549                                 integrity_sector_checksum(ic, sector, mem + pos, checksums_ptr);
1550                                 checksums_ptr += ic->tag_size;
1551                                 sectors_to_process -= ic->sectors_per_block;
1552                                 pos += ic->sectors_per_block << SECTOR_SHIFT;
1553                                 sector += ic->sectors_per_block;
1554                         } while (pos < bv.bv_len && sectors_to_process && checksums != checksums_onstack);
1555                         kunmap_atomic(mem);
1556
1557                         r = dm_integrity_rw_tag(ic, checksums, &dio->metadata_block, &dio->metadata_offset,
1558                                                 checksums_ptr - checksums, !dio->write ? TAG_CMP : TAG_WRITE);
1559                         if (unlikely(r)) {
1560                                 if (r > 0) {
1561                                         char b[BDEVNAME_SIZE];
1562                                         DMERR_LIMIT("%s: Checksum failed at sector 0x%llx", bio_devname(bio, b),
1563                                                     (sector - ((r + ic->tag_size - 1) / ic->tag_size)));
1564                                         r = -EILSEQ;
1565                                         atomic64_inc(&ic->number_of_mismatches);
1566                                 }
1567                                 if (likely(checksums != checksums_onstack))
1568                                         kfree(checksums);
1569                                 goto error;
1570                         }
1571
1572                         if (!sectors_to_process)
1573                                 break;
1574
1575                         if (unlikely(pos < bv.bv_len)) {
1576                                 bv.bv_offset += pos;
1577                                 bv.bv_len -= pos;
1578                                 goto again;
1579                         }
1580                 }
1581
1582                 if (likely(checksums != checksums_onstack))
1583                         kfree(checksums);
1584         } else {
1585                 struct bio_integrity_payload *bip = dio->bio_details.bi_integrity;
1586
1587                 if (bip) {
1588                         struct bio_vec biv;
1589                         struct bvec_iter iter;
1590                         unsigned data_to_process = dio->range.n_sectors;
1591                         sector_to_block(ic, data_to_process);
1592                         data_to_process *= ic->tag_size;
1593
1594                         bip_for_each_vec(biv, bip, iter) {
1595                                 unsigned char *tag;
1596                                 unsigned this_len;
1597
1598                                 BUG_ON(PageHighMem(biv.bv_page));
1599                                 tag = lowmem_page_address(biv.bv_page) + biv.bv_offset;
1600                                 this_len = min(biv.bv_len, data_to_process);
1601                                 r = dm_integrity_rw_tag(ic, tag, &dio->metadata_block, &dio->metadata_offset,
1602                                                         this_len, !dio->write ? TAG_READ : TAG_WRITE);
1603                                 if (unlikely(r))
1604                                         goto error;
1605                                 data_to_process -= this_len;
1606                                 if (!data_to_process)
1607                                         break;
1608                         }
1609                 }
1610         }
1611 skip_io:
1612         dec_in_flight(dio);
1613         return;
1614 error:
1615         dio->bi_status = errno_to_blk_status(r);
1616         dec_in_flight(dio);
1617 }
1618
1619 static int dm_integrity_map(struct dm_target *ti, struct bio *bio)
1620 {
1621         struct dm_integrity_c *ic = ti->private;
1622         struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
1623         struct bio_integrity_payload *bip;
1624
1625         sector_t area, offset;
1626
1627         dio->ic = ic;
1628         dio->bi_status = 0;
1629
1630         if (unlikely(bio->bi_opf & REQ_PREFLUSH)) {
1631                 submit_flush_bio(ic, dio);
1632                 return DM_MAPIO_SUBMITTED;
1633         }
1634
1635         dio->range.logical_sector = dm_target_offset(ti, bio->bi_iter.bi_sector);
1636         dio->write = bio_op(bio) == REQ_OP_WRITE;
1637         dio->fua = dio->write && bio->bi_opf & REQ_FUA;
1638         if (unlikely(dio->fua)) {
1639                 /*
1640                  * Don't pass down the FUA flag because we have to flush
1641                  * disk cache anyway.
1642                  */
1643                 bio->bi_opf &= ~REQ_FUA;
1644         }
1645         if (unlikely(dio->range.logical_sector + bio_sectors(bio) > ic->provided_data_sectors)) {
1646                 DMERR("Too big sector number: 0x%llx + 0x%x > 0x%llx",
1647                       dio->range.logical_sector, bio_sectors(bio),
1648                       ic->provided_data_sectors);
1649                 return DM_MAPIO_KILL;
1650         }
1651         if (unlikely((dio->range.logical_sector | bio_sectors(bio)) & (unsigned)(ic->sectors_per_block - 1))) {
1652                 DMERR("Bio not aligned on %u sectors: 0x%llx, 0x%x",
1653                       ic->sectors_per_block,
1654                       dio->range.logical_sector, bio_sectors(bio));
1655                 return DM_MAPIO_KILL;
1656         }
1657
1658         if (ic->sectors_per_block > 1) {
1659                 struct bvec_iter iter;
1660                 struct bio_vec bv;
1661                 bio_for_each_segment(bv, bio, iter) {
1662                         if (unlikely(bv.bv_len & ((ic->sectors_per_block << SECTOR_SHIFT) - 1))) {
1663                                 DMERR("Bio vector (%u,%u) is not aligned on %u-sector boundary",
1664                                         bv.bv_offset, bv.bv_len, ic->sectors_per_block);
1665                                 return DM_MAPIO_KILL;
1666                         }
1667                 }
1668         }
1669
1670         bip = bio_integrity(bio);
1671         if (!ic->internal_hash) {
1672                 if (bip) {
1673                         unsigned wanted_tag_size = bio_sectors(bio) >> ic->sb->log2_sectors_per_block;
1674                         if (ic->log2_tag_size >= 0)
1675                                 wanted_tag_size <<= ic->log2_tag_size;
1676                         else
1677                                 wanted_tag_size *= ic->tag_size;
1678                         if (unlikely(wanted_tag_size != bip->bip_iter.bi_size)) {
1679                                 DMERR("Invalid integrity data size %u, expected %u",
1680                                       bip->bip_iter.bi_size, wanted_tag_size);
1681                                 return DM_MAPIO_KILL;
1682                         }
1683                 }
1684         } else {
1685                 if (unlikely(bip != NULL)) {
1686                         DMERR("Unexpected integrity data when using internal hash");
1687                         return DM_MAPIO_KILL;
1688                 }
1689         }
1690
1691         if (unlikely(ic->mode == 'R') && unlikely(dio->write))
1692                 return DM_MAPIO_KILL;
1693
1694         get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
1695         dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset, &dio->metadata_offset);
1696         bio->bi_iter.bi_sector = get_data_sector(ic, area, offset);
1697
1698         dm_integrity_map_continue(dio, true);
1699         return DM_MAPIO_SUBMITTED;
1700 }
1701
1702 static bool __journal_read_write(struct dm_integrity_io *dio, struct bio *bio,
1703                                  unsigned journal_section, unsigned journal_entry)
1704 {
1705         struct dm_integrity_c *ic = dio->ic;
1706         sector_t logical_sector;
1707         unsigned n_sectors;
1708
1709         logical_sector = dio->range.logical_sector;
1710         n_sectors = dio->range.n_sectors;
1711         do {
1712                 struct bio_vec bv = bio_iovec(bio);
1713                 char *mem;
1714
1715                 if (unlikely(bv.bv_len >> SECTOR_SHIFT > n_sectors))
1716                         bv.bv_len = n_sectors << SECTOR_SHIFT;
1717                 n_sectors -= bv.bv_len >> SECTOR_SHIFT;
1718                 bio_advance_iter(bio, &bio->bi_iter, bv.bv_len);
1719 retry_kmap:
1720                 mem = kmap_atomic(bv.bv_page);
1721                 if (likely(dio->write))
1722                         flush_dcache_page(bv.bv_page);
1723
1724                 do {
1725                         struct journal_entry *je = access_journal_entry(ic, journal_section, journal_entry);
1726
1727                         if (unlikely(!dio->write)) {
1728                                 struct journal_sector *js;
1729                                 char *mem_ptr;
1730                                 unsigned s;
1731
1732                                 if (unlikely(journal_entry_is_inprogress(je))) {
1733                                         flush_dcache_page(bv.bv_page);
1734                                         kunmap_atomic(mem);
1735
1736                                         __io_wait_event(ic->copy_to_journal_wait, !journal_entry_is_inprogress(je));
1737                                         goto retry_kmap;
1738                                 }
1739                                 smp_rmb();
1740                                 BUG_ON(journal_entry_get_sector(je) != logical_sector);
1741                                 js = access_journal_data(ic, journal_section, journal_entry);
1742                                 mem_ptr = mem + bv.bv_offset;
1743                                 s = 0;
1744                                 do {
1745                                         memcpy(mem_ptr, js, JOURNAL_SECTOR_DATA);
1746                                         *(commit_id_t *)(mem_ptr + JOURNAL_SECTOR_DATA) = je->last_bytes[s];
1747                                         js++;
1748                                         mem_ptr += 1 << SECTOR_SHIFT;
1749                                 } while (++s < ic->sectors_per_block);
1750 #ifdef INTERNAL_VERIFY
1751                                 if (ic->internal_hash) {
1752                                         char checksums_onstack[max((size_t)HASH_MAX_DIGESTSIZE, MAX_TAG_SIZE)];
1753
1754                                         integrity_sector_checksum(ic, logical_sector, mem + bv.bv_offset, checksums_onstack);
1755                                         if (unlikely(memcmp(checksums_onstack, journal_entry_tag(ic, je), ic->tag_size))) {
1756                                                 DMERR_LIMIT("Checksum failed when reading from journal, at sector 0x%llx",
1757                                                             logical_sector);
1758                                         }
1759                                 }
1760 #endif
1761                         }
1762
1763                         if (!ic->internal_hash) {
1764                                 struct bio_integrity_payload *bip = bio_integrity(bio);
1765                                 unsigned tag_todo = ic->tag_size;
1766                                 char *tag_ptr = journal_entry_tag(ic, je);
1767
1768                                 if (bip) do {
1769                                         struct bio_vec biv = bvec_iter_bvec(bip->bip_vec, bip->bip_iter);
1770                                         unsigned tag_now = min(biv.bv_len, tag_todo);
1771                                         char *tag_addr;
1772                                         BUG_ON(PageHighMem(biv.bv_page));
1773                                         tag_addr = lowmem_page_address(biv.bv_page) + biv.bv_offset;
1774                                         if (likely(dio->write))
1775                                                 memcpy(tag_ptr, tag_addr, tag_now);
1776                                         else
1777                                                 memcpy(tag_addr, tag_ptr, tag_now);
1778                                         bvec_iter_advance(bip->bip_vec, &bip->bip_iter, tag_now);
1779                                         tag_ptr += tag_now;
1780                                         tag_todo -= tag_now;
1781                                 } while (unlikely(tag_todo)); else {
1782                                         if (likely(dio->write))
1783                                                 memset(tag_ptr, 0, tag_todo);
1784                                 }
1785                         }
1786
1787                         if (likely(dio->write)) {
1788                                 struct journal_sector *js;
1789                                 unsigned s;
1790
1791                                 js = access_journal_data(ic, journal_section, journal_entry);
1792                                 memcpy(js, mem + bv.bv_offset, ic->sectors_per_block << SECTOR_SHIFT);
1793
1794                                 s = 0;
1795                                 do {
1796                                         je->last_bytes[s] = js[s].commit_id;
1797                                 } while (++s < ic->sectors_per_block);
1798
1799                                 if (ic->internal_hash) {
1800                                         unsigned digest_size = crypto_shash_digestsize(ic->internal_hash);
1801                                         if (unlikely(digest_size > ic->tag_size)) {
1802                                                 char checksums_onstack[HASH_MAX_DIGESTSIZE];
1803                                                 integrity_sector_checksum(ic, logical_sector, (char *)js, checksums_onstack);
1804                                                 memcpy(journal_entry_tag(ic, je), checksums_onstack, ic->tag_size);
1805                                         } else
1806                                                 integrity_sector_checksum(ic, logical_sector, (char *)js, journal_entry_tag(ic, je));
1807                                 }
1808
1809                                 journal_entry_set_sector(je, logical_sector);
1810                         }
1811                         logical_sector += ic->sectors_per_block;
1812
1813                         journal_entry++;
1814                         if (unlikely(journal_entry == ic->journal_section_entries)) {
1815                                 journal_entry = 0;
1816                                 journal_section++;
1817                                 wraparound_section(ic, &journal_section);
1818                         }
1819
1820                         bv.bv_offset += ic->sectors_per_block << SECTOR_SHIFT;
1821                 } while (bv.bv_len -= ic->sectors_per_block << SECTOR_SHIFT);
1822
1823                 if (unlikely(!dio->write))
1824                         flush_dcache_page(bv.bv_page);
1825                 kunmap_atomic(mem);
1826         } while (n_sectors);
1827
1828         if (likely(dio->write)) {
1829                 smp_mb();
1830                 if (unlikely(waitqueue_active(&ic->copy_to_journal_wait)))
1831                         wake_up(&ic->copy_to_journal_wait);
1832                 if (READ_ONCE(ic->free_sectors) <= ic->free_sectors_threshold) {
1833                         queue_work(ic->commit_wq, &ic->commit_work);
1834                 } else {
1835                         schedule_autocommit(ic);
1836                 }
1837         } else {
1838                 remove_range(ic, &dio->range);
1839         }
1840
1841         if (unlikely(bio->bi_iter.bi_size)) {
1842                 sector_t area, offset;
1843
1844                 dio->range.logical_sector = logical_sector;
1845                 get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
1846                 dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset, &dio->metadata_offset);
1847                 return true;
1848         }
1849
1850         return false;
1851 }
1852
1853 static void dm_integrity_map_continue(struct dm_integrity_io *dio, bool from_map)
1854 {
1855         struct dm_integrity_c *ic = dio->ic;
1856         struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1857         unsigned journal_section, journal_entry;
1858         unsigned journal_read_pos;
1859         struct completion read_comp;
1860         bool need_sync_io = ic->internal_hash && !dio->write;
1861
1862         if (need_sync_io && from_map) {
1863                 INIT_WORK(&dio->work, integrity_bio_wait);
1864                 queue_work(ic->offload_wq, &dio->work);
1865                 return;
1866         }
1867
1868 lock_retry:
1869         spin_lock_irq(&ic->endio_wait.lock);
1870 retry:
1871         if (unlikely(dm_integrity_failed(ic))) {
1872                 spin_unlock_irq(&ic->endio_wait.lock);
1873                 do_endio(ic, bio);
1874                 return;
1875         }
1876         dio->range.n_sectors = bio_sectors(bio);
1877         journal_read_pos = NOT_FOUND;
1878         if (likely(ic->mode == 'J')) {
1879                 if (dio->write) {
1880                         unsigned next_entry, i, pos;
1881                         unsigned ws, we, range_sectors;
1882
1883                         dio->range.n_sectors = min(dio->range.n_sectors,
1884                                                    (sector_t)ic->free_sectors << ic->sb->log2_sectors_per_block);
1885                         if (unlikely(!dio->range.n_sectors)) {
1886                                 if (from_map)
1887                                         goto offload_to_thread;
1888                                 sleep_on_endio_wait(ic);
1889                                 goto retry;
1890                         }
1891                         range_sectors = dio->range.n_sectors >> ic->sb->log2_sectors_per_block;
1892                         ic->free_sectors -= range_sectors;
1893                         journal_section = ic->free_section;
1894                         journal_entry = ic->free_section_entry;
1895
1896                         next_entry = ic->free_section_entry + range_sectors;
1897                         ic->free_section_entry = next_entry % ic->journal_section_entries;
1898                         ic->free_section += next_entry / ic->journal_section_entries;
1899                         ic->n_uncommitted_sections += next_entry / ic->journal_section_entries;
1900                         wraparound_section(ic, &ic->free_section);
1901
1902                         pos = journal_section * ic->journal_section_entries + journal_entry;
1903                         ws = journal_section;
1904                         we = journal_entry;
1905                         i = 0;
1906                         do {
1907                                 struct journal_entry *je;
1908
1909                                 add_journal_node(ic, &ic->journal_tree[pos], dio->range.logical_sector + i);
1910                                 pos++;
1911                                 if (unlikely(pos >= ic->journal_entries))
1912                                         pos = 0;
1913
1914                                 je = access_journal_entry(ic, ws, we);
1915                                 BUG_ON(!journal_entry_is_unused(je));
1916                                 journal_entry_set_inprogress(je);
1917                                 we++;
1918                                 if (unlikely(we == ic->journal_section_entries)) {
1919                                         we = 0;
1920                                         ws++;
1921                                         wraparound_section(ic, &ws);
1922                                 }
1923                         } while ((i += ic->sectors_per_block) < dio->range.n_sectors);
1924
1925                         spin_unlock_irq(&ic->endio_wait.lock);
1926                         goto journal_read_write;
1927                 } else {
1928                         sector_t next_sector;
1929                         journal_read_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
1930                         if (likely(journal_read_pos == NOT_FOUND)) {
1931                                 if (unlikely(dio->range.n_sectors > next_sector - dio->range.logical_sector))
1932                                         dio->range.n_sectors = next_sector - dio->range.logical_sector;
1933                         } else {
1934                                 unsigned i;
1935                                 unsigned jp = journal_read_pos + 1;
1936                                 for (i = ic->sectors_per_block; i < dio->range.n_sectors; i += ic->sectors_per_block, jp++) {
1937                                         if (!test_journal_node(ic, jp, dio->range.logical_sector + i))
1938                                                 break;
1939                                 }
1940                                 dio->range.n_sectors = i;
1941                         }
1942                 }
1943         }
1944         if (unlikely(!add_new_range(ic, &dio->range, true))) {
1945                 /*
1946                  * We must not sleep in the request routine because it could
1947                  * stall bios on current->bio_list.
1948                  * So, we offload the bio to a workqueue if we have to sleep.
1949                  */
1950                 if (from_map) {
1951 offload_to_thread:
1952                         spin_unlock_irq(&ic->endio_wait.lock);
1953                         INIT_WORK(&dio->work, integrity_bio_wait);
1954                         queue_work(ic->wait_wq, &dio->work);
1955                         return;
1956                 }
1957                 if (journal_read_pos != NOT_FOUND)
1958                         dio->range.n_sectors = ic->sectors_per_block;
1959                 wait_and_add_new_range(ic, &dio->range);
1960                 /*
1961                  * wait_and_add_new_range drops the spinlock, so the journal
1962                  * may have been changed arbitrarily. We need to recheck.
1963                  * To simplify the code, we restrict I/O size to just one block.
1964                  */
1965                 if (journal_read_pos != NOT_FOUND) {
1966                         sector_t next_sector;
1967                         unsigned new_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
1968                         if (unlikely(new_pos != journal_read_pos)) {
1969                                 remove_range_unlocked(ic, &dio->range);
1970                                 goto retry;
1971                         }
1972                 }
1973         }
1974         spin_unlock_irq(&ic->endio_wait.lock);
1975
1976         if (unlikely(journal_read_pos != NOT_FOUND)) {
1977                 journal_section = journal_read_pos / ic->journal_section_entries;
1978                 journal_entry = journal_read_pos % ic->journal_section_entries;
1979                 goto journal_read_write;
1980         }
1981
1982         if (ic->mode == 'B' && dio->write) {
1983                 if (!block_bitmap_op(ic, ic->may_write_bitmap, dio->range.logical_sector,
1984                                      dio->range.n_sectors, BITMAP_OP_TEST_ALL_SET)) {
1985                         struct bitmap_block_status *bbs;
1986
1987                         bbs = sector_to_bitmap_block(ic, dio->range.logical_sector);
1988                         spin_lock(&bbs->bio_queue_lock);
1989                         bio_list_add(&bbs->bio_queue, bio);
1990                         spin_unlock(&bbs->bio_queue_lock);
1991                         queue_work(ic->writer_wq, &bbs->work);
1992                         return;
1993                 }
1994         }
1995
1996         dio->in_flight = (atomic_t)ATOMIC_INIT(2);
1997
1998         if (need_sync_io) {
1999                 init_completion(&read_comp);
2000                 dio->completion = &read_comp;
2001         } else
2002                 dio->completion = NULL;
2003
2004         dm_bio_record(&dio->bio_details, bio);
2005         bio_set_dev(bio, ic->dev->bdev);
2006         bio->bi_integrity = NULL;
2007         bio->bi_opf &= ~REQ_INTEGRITY;
2008         bio->bi_end_io = integrity_end_io;
2009         bio->bi_iter.bi_size = dio->range.n_sectors << SECTOR_SHIFT;
2010
2011         generic_make_request(bio);
2012
2013         if (need_sync_io) {
2014                 wait_for_completion_io(&read_comp);
2015                 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING) &&
2016                     dio->range.logical_sector + dio->range.n_sectors > le64_to_cpu(ic->sb->recalc_sector))
2017                         goto skip_check;
2018                 if (ic->mode == 'B') {
2019                         if (!block_bitmap_op(ic, ic->recalc_bitmap, dio->range.logical_sector,
2020                                              dio->range.n_sectors, BITMAP_OP_TEST_ALL_CLEAR))
2021                                 goto skip_check;
2022                 }
2023
2024                 if (likely(!bio->bi_status))
2025                         integrity_metadata(&dio->work);
2026                 else
2027 skip_check:
2028                         dec_in_flight(dio);
2029
2030         } else {
2031                 INIT_WORK(&dio->work, integrity_metadata);
2032                 queue_work(ic->metadata_wq, &dio->work);
2033         }
2034
2035         return;
2036
2037 journal_read_write:
2038         if (unlikely(__journal_read_write(dio, bio, journal_section, journal_entry)))
2039                 goto lock_retry;
2040
2041         do_endio_flush(ic, dio);
2042 }
2043
2044
2045 static void integrity_bio_wait(struct work_struct *w)
2046 {
2047         struct dm_integrity_io *dio = container_of(w, struct dm_integrity_io, work);
2048
2049         dm_integrity_map_continue(dio, false);
2050 }
2051
2052 static void pad_uncommitted(struct dm_integrity_c *ic)
2053 {
2054         if (ic->free_section_entry) {
2055                 ic->free_sectors -= ic->journal_section_entries - ic->free_section_entry;
2056                 ic->free_section_entry = 0;
2057                 ic->free_section++;
2058                 wraparound_section(ic, &ic->free_section);
2059                 ic->n_uncommitted_sections++;
2060         }
2061         if (WARN_ON(ic->journal_sections * ic->journal_section_entries !=
2062                     (ic->n_uncommitted_sections + ic->n_committed_sections) *
2063                     ic->journal_section_entries + ic->free_sectors)) {
2064                 DMCRIT("journal_sections %u, journal_section_entries %u, "
2065                        "n_uncommitted_sections %u, n_committed_sections %u, "
2066                        "journal_section_entries %u, free_sectors %u",
2067                        ic->journal_sections, ic->journal_section_entries,
2068                        ic->n_uncommitted_sections, ic->n_committed_sections,
2069                        ic->journal_section_entries, ic->free_sectors);
2070         }
2071 }
2072
2073 static void integrity_commit(struct work_struct *w)
2074 {
2075         struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, commit_work);
2076         unsigned commit_start, commit_sections;
2077         unsigned i, j, n;
2078         struct bio *flushes;
2079
2080         del_timer(&ic->autocommit_timer);
2081
2082         spin_lock_irq(&ic->endio_wait.lock);
2083         flushes = bio_list_get(&ic->flush_bio_list);
2084         if (unlikely(ic->mode != 'J')) {
2085                 spin_unlock_irq(&ic->endio_wait.lock);
2086                 dm_integrity_flush_buffers(ic);
2087                 goto release_flush_bios;
2088         }
2089
2090         pad_uncommitted(ic);
2091         commit_start = ic->uncommitted_section;
2092         commit_sections = ic->n_uncommitted_sections;
2093         spin_unlock_irq(&ic->endio_wait.lock);
2094
2095         if (!commit_sections)
2096                 goto release_flush_bios;
2097
2098         i = commit_start;
2099         for (n = 0; n < commit_sections; n++) {
2100                 for (j = 0; j < ic->journal_section_entries; j++) {
2101                         struct journal_entry *je;
2102                         je = access_journal_entry(ic, i, j);
2103                         io_wait_event(ic->copy_to_journal_wait, !journal_entry_is_inprogress(je));
2104                 }
2105                 for (j = 0; j < ic->journal_section_sectors; j++) {
2106                         struct journal_sector *js;
2107                         js = access_journal(ic, i, j);
2108                         js->commit_id = dm_integrity_commit_id(ic, i, j, ic->commit_seq);
2109                 }
2110                 i++;
2111                 if (unlikely(i >= ic->journal_sections))
2112                         ic->commit_seq = next_commit_seq(ic->commit_seq);
2113                 wraparound_section(ic, &i);
2114         }
2115         smp_rmb();
2116
2117         write_journal(ic, commit_start, commit_sections);
2118
2119         spin_lock_irq(&ic->endio_wait.lock);
2120         ic->uncommitted_section += commit_sections;
2121         wraparound_section(ic, &ic->uncommitted_section);
2122         ic->n_uncommitted_sections -= commit_sections;
2123         ic->n_committed_sections += commit_sections;
2124         spin_unlock_irq(&ic->endio_wait.lock);
2125
2126         if (READ_ONCE(ic->free_sectors) <= ic->free_sectors_threshold)
2127                 queue_work(ic->writer_wq, &ic->writer_work);
2128
2129 release_flush_bios:
2130         while (flushes) {
2131                 struct bio *next = flushes->bi_next;
2132                 flushes->bi_next = NULL;
2133                 do_endio(ic, flushes);
2134                 flushes = next;
2135         }
2136 }
2137
2138 static void complete_copy_from_journal(unsigned long error, void *context)
2139 {
2140         struct journal_io *io = context;
2141         struct journal_completion *comp = io->comp;
2142         struct dm_integrity_c *ic = comp->ic;
2143         remove_range(ic, &io->range);
2144         mempool_free(io, &ic->journal_io_mempool);
2145         if (unlikely(error != 0))
2146                 dm_integrity_io_error(ic, "copying from journal", -EIO);
2147         complete_journal_op(comp);
2148 }
2149
2150 static void restore_last_bytes(struct dm_integrity_c *ic, struct journal_sector *js,
2151                                struct journal_entry *je)
2152 {
2153         unsigned s = 0;
2154         do {
2155                 js->commit_id = je->last_bytes[s];
2156                 js++;
2157         } while (++s < ic->sectors_per_block);
2158 }
2159
2160 static void do_journal_write(struct dm_integrity_c *ic, unsigned write_start,
2161                              unsigned write_sections, bool from_replay)
2162 {
2163         unsigned i, j, n;
2164         struct journal_completion comp;
2165         struct blk_plug plug;
2166
2167         blk_start_plug(&plug);
2168
2169         comp.ic = ic;
2170         comp.in_flight = (atomic_t)ATOMIC_INIT(1);
2171         init_completion(&comp.comp);
2172
2173         i = write_start;
2174         for (n = 0; n < write_sections; n++, i++, wraparound_section(ic, &i)) {
2175 #ifndef INTERNAL_VERIFY
2176                 if (unlikely(from_replay))
2177 #endif
2178                         rw_section_mac(ic, i, false);
2179                 for (j = 0; j < ic->journal_section_entries; j++) {
2180                         struct journal_entry *je = access_journal_entry(ic, i, j);
2181                         sector_t sec, area, offset;
2182                         unsigned k, l, next_loop;
2183                         sector_t metadata_block;
2184                         unsigned metadata_offset;
2185                         struct journal_io *io;
2186
2187                         if (journal_entry_is_unused(je))
2188                                 continue;
2189                         BUG_ON(unlikely(journal_entry_is_inprogress(je)) && !from_replay);
2190                         sec = journal_entry_get_sector(je);
2191                         if (unlikely(from_replay)) {
2192                                 if (unlikely(sec & (unsigned)(ic->sectors_per_block - 1))) {
2193                                         dm_integrity_io_error(ic, "invalid sector in journal", -EIO);
2194                                         sec &= ~(sector_t)(ic->sectors_per_block - 1);
2195                                 }
2196                         }
2197                         get_area_and_offset(ic, sec, &area, &offset);
2198                         restore_last_bytes(ic, access_journal_data(ic, i, j), je);
2199                         for (k = j + 1; k < ic->journal_section_entries; k++) {
2200                                 struct journal_entry *je2 = access_journal_entry(ic, i, k);
2201                                 sector_t sec2, area2, offset2;
2202                                 if (journal_entry_is_unused(je2))
2203                                         break;
2204                                 BUG_ON(unlikely(journal_entry_is_inprogress(je2)) && !from_replay);
2205                                 sec2 = journal_entry_get_sector(je2);
2206                                 get_area_and_offset(ic, sec2, &area2, &offset2);
2207                                 if (area2 != area || offset2 != offset + ((k - j) << ic->sb->log2_sectors_per_block))
2208                                         break;
2209                                 restore_last_bytes(ic, access_journal_data(ic, i, k), je2);
2210                         }
2211                         next_loop = k - 1;
2212
2213                         io = mempool_alloc(&ic->journal_io_mempool, GFP_NOIO);
2214                         io->comp = &comp;
2215                         io->range.logical_sector = sec;
2216                         io->range.n_sectors = (k - j) << ic->sb->log2_sectors_per_block;
2217
2218                         spin_lock_irq(&ic->endio_wait.lock);
2219                         add_new_range_and_wait(ic, &io->range);
2220
2221                         if (likely(!from_replay)) {
2222                                 struct journal_node *section_node = &ic->journal_tree[i * ic->journal_section_entries];
2223
2224                                 /* don't write if there is newer committed sector */
2225                                 while (j < k && find_newer_committed_node(ic, &section_node[j])) {
2226                                         struct journal_entry *je2 = access_journal_entry(ic, i, j);
2227
2228                                         journal_entry_set_unused(je2);
2229                                         remove_journal_node(ic, &section_node[j]);
2230                                         j++;
2231                                         sec += ic->sectors_per_block;
2232                                         offset += ic->sectors_per_block;
2233                                 }
2234                                 while (j < k && find_newer_committed_node(ic, &section_node[k - 1])) {
2235                                         struct journal_entry *je2 = access_journal_entry(ic, i, k - 1);
2236
2237                                         journal_entry_set_unused(je2);
2238                                         remove_journal_node(ic, &section_node[k - 1]);
2239                                         k--;
2240                                 }
2241                                 if (j == k) {
2242                                         remove_range_unlocked(ic, &io->range);
2243                                         spin_unlock_irq(&ic->endio_wait.lock);
2244                                         mempool_free(io, &ic->journal_io_mempool);
2245                                         goto skip_io;
2246                                 }
2247                                 for (l = j; l < k; l++) {
2248                                         remove_journal_node(ic, &section_node[l]);
2249                                 }
2250                         }
2251                         spin_unlock_irq(&ic->endio_wait.lock);
2252
2253                         metadata_block = get_metadata_sector_and_offset(ic, area, offset, &metadata_offset);
2254                         for (l = j; l < k; l++) {
2255                                 int r;
2256                                 struct journal_entry *je2 = access_journal_entry(ic, i, l);
2257
2258                                 if (
2259 #ifndef INTERNAL_VERIFY
2260                                     unlikely(from_replay) &&
2261 #endif
2262                                     ic->internal_hash) {
2263                                         char test_tag[max_t(size_t, HASH_MAX_DIGESTSIZE, MAX_TAG_SIZE)];
2264
2265                                         integrity_sector_checksum(ic, sec + ((l - j) << ic->sb->log2_sectors_per_block),
2266                                                                   (char *)access_journal_data(ic, i, l), test_tag);
2267                                         if (unlikely(memcmp(test_tag, journal_entry_tag(ic, je2), ic->tag_size)))
2268                                                 dm_integrity_io_error(ic, "tag mismatch when replaying journal", -EILSEQ);
2269                                 }
2270
2271                                 journal_entry_set_unused(je2);
2272                                 r = dm_integrity_rw_tag(ic, journal_entry_tag(ic, je2), &metadata_block, &metadata_offset,
2273                                                         ic->tag_size, TAG_WRITE);
2274                                 if (unlikely(r)) {
2275                                         dm_integrity_io_error(ic, "reading tags", r);
2276                                 }
2277                         }
2278
2279                         atomic_inc(&comp.in_flight);
2280                         copy_from_journal(ic, i, j << ic->sb->log2_sectors_per_block,
2281                                           (k - j) << ic->sb->log2_sectors_per_block,
2282                                           get_data_sector(ic, area, offset),
2283                                           complete_copy_from_journal, io);
2284 skip_io:
2285                         j = next_loop;
2286                 }
2287         }
2288
2289         dm_bufio_write_dirty_buffers_async(ic->bufio);
2290
2291         blk_finish_plug(&plug);
2292
2293         complete_journal_op(&comp);
2294         wait_for_completion_io(&comp.comp);
2295
2296         dm_integrity_flush_buffers(ic);
2297 }
2298
2299 static void integrity_writer(struct work_struct *w)
2300 {
2301         struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, writer_work);
2302         unsigned write_start, write_sections;
2303
2304         unsigned prev_free_sectors;
2305
2306         /* the following test is not needed, but it tests the replay code */
2307         if (unlikely(dm_suspended(ic->ti)) && !ic->meta_dev)
2308                 return;
2309
2310         spin_lock_irq(&ic->endio_wait.lock);
2311         write_start = ic->committed_section;
2312         write_sections = ic->n_committed_sections;
2313         spin_unlock_irq(&ic->endio_wait.lock);
2314
2315         if (!write_sections)
2316                 return;
2317
2318         do_journal_write(ic, write_start, write_sections, false);
2319
2320         spin_lock_irq(&ic->endio_wait.lock);
2321
2322         ic->committed_section += write_sections;
2323         wraparound_section(ic, &ic->committed_section);
2324         ic->n_committed_sections -= write_sections;
2325
2326         prev_free_sectors = ic->free_sectors;
2327         ic->free_sectors += write_sections * ic->journal_section_entries;
2328         if (unlikely(!prev_free_sectors))
2329                 wake_up_locked(&ic->endio_wait);
2330
2331         spin_unlock_irq(&ic->endio_wait.lock);
2332 }
2333
2334 static void recalc_write_super(struct dm_integrity_c *ic)
2335 {
2336         int r;
2337
2338         dm_integrity_flush_buffers(ic);
2339         if (dm_integrity_failed(ic))
2340                 return;
2341
2342         r = sync_rw_sb(ic, REQ_OP_WRITE, 0);
2343         if (unlikely(r))
2344                 dm_integrity_io_error(ic, "writing superblock", r);
2345 }
2346
2347 static void integrity_recalc(struct work_struct *w)
2348 {
2349         struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, recalc_work);
2350         struct dm_integrity_range range;
2351         struct dm_io_request io_req;
2352         struct dm_io_region io_loc;
2353         sector_t area, offset;
2354         sector_t metadata_block;
2355         unsigned metadata_offset;
2356         sector_t logical_sector, n_sectors;
2357         __u8 *t;
2358         unsigned i;
2359         int r;
2360         unsigned super_counter = 0;
2361
2362         DEBUG_print("start recalculation... (position %llx)\n", le64_to_cpu(ic->sb->recalc_sector));
2363
2364         spin_lock_irq(&ic->endio_wait.lock);
2365
2366 next_chunk:
2367
2368         if (unlikely(dm_suspended(ic->ti)))
2369                 goto unlock_ret;
2370
2371         range.logical_sector = le64_to_cpu(ic->sb->recalc_sector);
2372         if (unlikely(range.logical_sector >= ic->provided_data_sectors)) {
2373                 if (ic->mode == 'B') {
2374                         DEBUG_print("queue_delayed_work: bitmap_flush_work\n");
2375                         queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, 0);
2376                 }
2377                 goto unlock_ret;
2378         }
2379
2380         get_area_and_offset(ic, range.logical_sector, &area, &offset);
2381         range.n_sectors = min((sector_t)RECALC_SECTORS, ic->provided_data_sectors - range.logical_sector);
2382         if (!ic->meta_dev)
2383                 range.n_sectors = min(range.n_sectors, ((sector_t)1U << ic->sb->log2_interleave_sectors) - (unsigned)offset);
2384
2385         add_new_range_and_wait(ic, &range);
2386         spin_unlock_irq(&ic->endio_wait.lock);
2387         logical_sector = range.logical_sector;
2388         n_sectors = range.n_sectors;
2389
2390         if (ic->mode == 'B') {
2391                 if (block_bitmap_op(ic, ic->recalc_bitmap, logical_sector, n_sectors, BITMAP_OP_TEST_ALL_CLEAR)) {
2392                         goto advance_and_next;
2393                 }
2394                 while (block_bitmap_op(ic, ic->recalc_bitmap, logical_sector,
2395                                        ic->sectors_per_block, BITMAP_OP_TEST_ALL_CLEAR)) {
2396                         logical_sector += ic->sectors_per_block;
2397                         n_sectors -= ic->sectors_per_block;
2398                         cond_resched();
2399                 }
2400                 while (block_bitmap_op(ic, ic->recalc_bitmap, logical_sector + n_sectors - ic->sectors_per_block,
2401                                        ic->sectors_per_block, BITMAP_OP_TEST_ALL_CLEAR)) {
2402                         n_sectors -= ic->sectors_per_block;
2403                         cond_resched();
2404                 }
2405                 get_area_and_offset(ic, logical_sector, &area, &offset);
2406         }
2407
2408         DEBUG_print("recalculating: %llx, %llx\n", logical_sector, n_sectors);
2409
2410         if (unlikely(++super_counter == RECALC_WRITE_SUPER)) {
2411                 recalc_write_super(ic);
2412                 if (ic->mode == 'B') {
2413                         queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, ic->bitmap_flush_interval);
2414                 }
2415                 super_counter = 0;
2416         }
2417
2418         if (unlikely(dm_integrity_failed(ic)))
2419                 goto err;
2420
2421         io_req.bi_op = REQ_OP_READ;
2422         io_req.bi_op_flags = 0;
2423         io_req.mem.type = DM_IO_VMA;
2424         io_req.mem.ptr.addr = ic->recalc_buffer;
2425         io_req.notify.fn = NULL;
2426         io_req.client = ic->io;
2427         io_loc.bdev = ic->dev->bdev;
2428         io_loc.sector = get_data_sector(ic, area, offset);
2429         io_loc.count = n_sectors;
2430
2431         r = dm_io(&io_req, 1, &io_loc, NULL);
2432         if (unlikely(r)) {
2433                 dm_integrity_io_error(ic, "reading data", r);
2434                 goto err;
2435         }
2436
2437         t = ic->recalc_tags;
2438         for (i = 0; i < n_sectors; i += ic->sectors_per_block) {
2439                 integrity_sector_checksum(ic, logical_sector + i, ic->recalc_buffer + (i << SECTOR_SHIFT), t);
2440                 t += ic->tag_size;
2441         }
2442
2443         metadata_block = get_metadata_sector_and_offset(ic, area, offset, &metadata_offset);
2444
2445         r = dm_integrity_rw_tag(ic, ic->recalc_tags, &metadata_block, &metadata_offset, t - ic->recalc_tags, TAG_WRITE);
2446         if (unlikely(r)) {
2447                 dm_integrity_io_error(ic, "writing tags", r);
2448                 goto err;
2449         }
2450
2451 advance_and_next:
2452         cond_resched();
2453
2454         spin_lock_irq(&ic->endio_wait.lock);
2455         remove_range_unlocked(ic, &range);
2456         ic->sb->recalc_sector = cpu_to_le64(range.logical_sector + range.n_sectors);
2457         goto next_chunk;
2458
2459 err:
2460         remove_range(ic, &range);
2461         return;
2462
2463 unlock_ret:
2464         spin_unlock_irq(&ic->endio_wait.lock);
2465
2466         recalc_write_super(ic);
2467 }
2468
2469 static void bitmap_block_work(struct work_struct *w)
2470 {
2471         struct bitmap_block_status *bbs = container_of(w, struct bitmap_block_status, work);
2472         struct dm_integrity_c *ic = bbs->ic;
2473         struct bio *bio;
2474         struct bio_list bio_queue;
2475         struct bio_list waiting;
2476
2477         bio_list_init(&waiting);
2478
2479         spin_lock(&bbs->bio_queue_lock);
2480         bio_queue = bbs->bio_queue;
2481         bio_list_init(&bbs->bio_queue);
2482         spin_unlock(&bbs->bio_queue_lock);
2483
2484         while ((bio = bio_list_pop(&bio_queue))) {
2485                 struct dm_integrity_io *dio;
2486
2487                 dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
2488
2489                 if (block_bitmap_op(ic, ic->may_write_bitmap, dio->range.logical_sector,
2490                                     dio->range.n_sectors, BITMAP_OP_TEST_ALL_SET)) {
2491                         remove_range(ic, &dio->range);
2492                         INIT_WORK(&dio->work, integrity_bio_wait);
2493                         queue_work(ic->offload_wq, &dio->work);
2494                 } else {
2495                         block_bitmap_op(ic, ic->journal, dio->range.logical_sector,
2496                                         dio->range.n_sectors, BITMAP_OP_SET);
2497                         bio_list_add(&waiting, bio);
2498                 }
2499         }
2500
2501         if (bio_list_empty(&waiting))
2502                 return;
2503
2504         rw_journal_sectors(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC,
2505                            bbs->idx * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT),
2506                            BITMAP_BLOCK_SIZE >> SECTOR_SHIFT, NULL);
2507
2508         while ((bio = bio_list_pop(&waiting))) {
2509                 struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
2510
2511                 block_bitmap_op(ic, ic->may_write_bitmap, dio->range.logical_sector,
2512                                 dio->range.n_sectors, BITMAP_OP_SET);
2513
2514                 remove_range(ic, &dio->range);
2515                 INIT_WORK(&dio->work, integrity_bio_wait);
2516                 queue_work(ic->offload_wq, &dio->work);
2517         }
2518
2519         queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, ic->bitmap_flush_interval);
2520 }
2521
2522 static void bitmap_flush_work(struct work_struct *work)
2523 {
2524         struct dm_integrity_c *ic = container_of(work, struct dm_integrity_c, bitmap_flush_work.work);
2525         struct dm_integrity_range range;
2526         unsigned long limit;
2527         struct bio *bio;
2528
2529         dm_integrity_flush_buffers(ic);
2530
2531         range.logical_sector = 0;
2532         range.n_sectors = ic->provided_data_sectors;
2533
2534         spin_lock_irq(&ic->endio_wait.lock);
2535         add_new_range_and_wait(ic, &range);
2536         spin_unlock_irq(&ic->endio_wait.lock);
2537
2538         dm_integrity_flush_buffers(ic);
2539         if (ic->meta_dev)
2540                 blkdev_issue_flush(ic->dev->bdev, GFP_NOIO, NULL);
2541
2542         limit = ic->provided_data_sectors;
2543         if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
2544                 limit = le64_to_cpu(ic->sb->recalc_sector)
2545                         >> (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit)
2546                         << (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
2547         }
2548         /*DEBUG_print("zeroing journal\n");*/
2549         block_bitmap_op(ic, ic->journal, 0, limit, BITMAP_OP_CLEAR);
2550         block_bitmap_op(ic, ic->may_write_bitmap, 0, limit, BITMAP_OP_CLEAR);
2551
2552         rw_journal_sectors(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, 0,
2553                            ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
2554
2555         spin_lock_irq(&ic->endio_wait.lock);
2556         remove_range_unlocked(ic, &range);
2557         while (unlikely((bio = bio_list_pop(&ic->synchronous_bios)) != NULL)) {
2558                 bio_endio(bio);
2559                 spin_unlock_irq(&ic->endio_wait.lock);
2560                 spin_lock_irq(&ic->endio_wait.lock);
2561         }
2562         spin_unlock_irq(&ic->endio_wait.lock);
2563 }
2564
2565
2566 static void init_journal(struct dm_integrity_c *ic, unsigned start_section,
2567                          unsigned n_sections, unsigned char commit_seq)
2568 {
2569         unsigned i, j, n;
2570
2571         if (!n_sections)
2572                 return;
2573
2574         for (n = 0; n < n_sections; n++) {
2575                 i = start_section + n;
2576                 wraparound_section(ic, &i);
2577                 for (j = 0; j < ic->journal_section_sectors; j++) {
2578                         struct journal_sector *js = access_journal(ic, i, j);
2579                         memset(&js->entries, 0, JOURNAL_SECTOR_DATA);
2580                         js->commit_id = dm_integrity_commit_id(ic, i, j, commit_seq);
2581                 }
2582                 for (j = 0; j < ic->journal_section_entries; j++) {
2583                         struct journal_entry *je = access_journal_entry(ic, i, j);
2584                         journal_entry_set_unused(je);
2585                 }
2586         }
2587
2588         write_journal(ic, start_section, n_sections);
2589 }
2590
2591 static int find_commit_seq(struct dm_integrity_c *ic, unsigned i, unsigned j, commit_id_t id)
2592 {
2593         unsigned char k;
2594         for (k = 0; k < N_COMMIT_IDS; k++) {
2595                 if (dm_integrity_commit_id(ic, i, j, k) == id)
2596                         return k;
2597         }
2598         dm_integrity_io_error(ic, "journal commit id", -EIO);
2599         return -EIO;
2600 }
2601
2602 static void replay_journal(struct dm_integrity_c *ic)
2603 {
2604         unsigned i, j;
2605         bool used_commit_ids[N_COMMIT_IDS];
2606         unsigned max_commit_id_sections[N_COMMIT_IDS];
2607         unsigned write_start, write_sections;
2608         unsigned continue_section;
2609         bool journal_empty;
2610         unsigned char unused, last_used, want_commit_seq;
2611
2612         if (ic->mode == 'R')
2613                 return;
2614
2615         if (ic->journal_uptodate)
2616                 return;
2617
2618         last_used = 0;
2619         write_start = 0;
2620
2621         if (!ic->just_formatted) {
2622                 DEBUG_print("reading journal\n");
2623                 rw_journal(ic, REQ_OP_READ, 0, 0, ic->journal_sections, NULL);
2624                 if (ic->journal_io)
2625                         DEBUG_bytes(lowmem_page_address(ic->journal_io[0].page), 64, "read journal");
2626                 if (ic->journal_io) {
2627                         struct journal_completion crypt_comp;
2628                         crypt_comp.ic = ic;
2629                         init_completion(&crypt_comp.comp);
2630                         crypt_comp.in_flight = (atomic_t)ATOMIC_INIT(0);
2631                         encrypt_journal(ic, false, 0, ic->journal_sections, &crypt_comp);
2632                         wait_for_completion(&crypt_comp.comp);
2633                 }
2634                 DEBUG_bytes(lowmem_page_address(ic->journal[0].page), 64, "decrypted journal");
2635         }
2636
2637         if (dm_integrity_failed(ic))
2638                 goto clear_journal;
2639
2640         journal_empty = true;
2641         memset(used_commit_ids, 0, sizeof used_commit_ids);
2642         memset(max_commit_id_sections, 0, sizeof max_commit_id_sections);
2643         for (i = 0; i < ic->journal_sections; i++) {
2644                 for (j = 0; j < ic->journal_section_sectors; j++) {
2645                         int k;
2646                         struct journal_sector *js = access_journal(ic, i, j);
2647                         k = find_commit_seq(ic, i, j, js->commit_id);
2648                         if (k < 0)
2649                                 goto clear_journal;
2650                         used_commit_ids[k] = true;
2651                         max_commit_id_sections[k] = i;
2652                 }
2653                 if (journal_empty) {
2654                         for (j = 0; j < ic->journal_section_entries; j++) {
2655                                 struct journal_entry *je = access_journal_entry(ic, i, j);
2656                                 if (!journal_entry_is_unused(je)) {
2657                                         journal_empty = false;
2658                                         break;
2659                                 }
2660                         }
2661                 }
2662         }
2663
2664         if (!used_commit_ids[N_COMMIT_IDS - 1]) {
2665                 unused = N_COMMIT_IDS - 1;
2666                 while (unused && !used_commit_ids[unused - 1])
2667                         unused--;
2668         } else {
2669                 for (unused = 0; unused < N_COMMIT_IDS; unused++)
2670                         if (!used_commit_ids[unused])
2671                                 break;
2672                 if (unused == N_COMMIT_IDS) {
2673                         dm_integrity_io_error(ic, "journal commit ids", -EIO);
2674                         goto clear_journal;
2675                 }
2676         }
2677         DEBUG_print("first unused commit seq %d [%d,%d,%d,%d]\n",
2678                     unused, used_commit_ids[0], used_commit_ids[1],
2679                     used_commit_ids[2], used_commit_ids[3]);
2680
2681         last_used = prev_commit_seq(unused);
2682         want_commit_seq = prev_commit_seq(last_used);
2683
2684         if (!used_commit_ids[want_commit_seq] && used_commit_ids[prev_commit_seq(want_commit_seq)])
2685                 journal_empty = true;
2686
2687         write_start = max_commit_id_sections[last_used] + 1;
2688         if (unlikely(write_start >= ic->journal_sections))
2689                 want_commit_seq = next_commit_seq(want_commit_seq);
2690         wraparound_section(ic, &write_start);
2691
2692         i = write_start;
2693         for (write_sections = 0; write_sections < ic->journal_sections; write_sections++) {
2694                 for (j = 0; j < ic->journal_section_sectors; j++) {
2695                         struct journal_sector *js = access_journal(ic, i, j);
2696
2697                         if (js->commit_id != dm_integrity_commit_id(ic, i, j, want_commit_seq)) {
2698                                 /*
2699                                  * This could be caused by crash during writing.
2700                                  * We won't replay the inconsistent part of the
2701                                  * journal.
2702                                  */
2703                                 DEBUG_print("commit id mismatch at position (%u, %u): %d != %d\n",
2704                                             i, j, find_commit_seq(ic, i, j, js->commit_id), want_commit_seq);
2705                                 goto brk;
2706                         }
2707                 }
2708                 i++;
2709                 if (unlikely(i >= ic->journal_sections))
2710                         want_commit_seq = next_commit_seq(want_commit_seq);
2711                 wraparound_section(ic, &i);
2712         }
2713 brk:
2714
2715         if (!journal_empty) {
2716                 DEBUG_print("replaying %u sections, starting at %u, commit seq %d\n",
2717                             write_sections, write_start, want_commit_seq);
2718                 do_journal_write(ic, write_start, write_sections, true);
2719         }
2720
2721         if (write_sections == ic->journal_sections && (ic->mode == 'J' || journal_empty)) {
2722                 continue_section = write_start;
2723                 ic->commit_seq = want_commit_seq;
2724                 DEBUG_print("continuing from section %u, commit seq %d\n", write_start, ic->commit_seq);
2725         } else {
2726                 unsigned s;
2727                 unsigned char erase_seq;
2728 clear_journal:
2729                 DEBUG_print("clearing journal\n");
2730
2731                 erase_seq = prev_commit_seq(prev_commit_seq(last_used));
2732                 s = write_start;
2733                 init_journal(ic, s, 1, erase_seq);
2734                 s++;
2735                 wraparound_section(ic, &s);
2736                 if (ic->journal_sections >= 2) {
2737                         init_journal(ic, s, ic->journal_sections - 2, erase_seq);
2738                         s += ic->journal_sections - 2;
2739                         wraparound_section(ic, &s);
2740                         init_journal(ic, s, 1, erase_seq);
2741                 }
2742
2743                 continue_section = 0;
2744                 ic->commit_seq = next_commit_seq(erase_seq);
2745         }
2746
2747         ic->committed_section = continue_section;
2748         ic->n_committed_sections = 0;
2749
2750         ic->uncommitted_section = continue_section;
2751         ic->n_uncommitted_sections = 0;
2752
2753         ic->free_section = continue_section;
2754         ic->free_section_entry = 0;
2755         ic->free_sectors = ic->journal_entries;
2756
2757         ic->journal_tree_root = RB_ROOT;
2758         for (i = 0; i < ic->journal_entries; i++)
2759                 init_journal_node(&ic->journal_tree[i]);
2760 }
2761
2762 static void dm_integrity_enter_synchronous_mode(struct dm_integrity_c *ic)
2763 {
2764         DEBUG_print("dm_integrity_enter_synchronous_mode\n");
2765
2766         if (ic->mode == 'B') {
2767                 ic->bitmap_flush_interval = msecs_to_jiffies(10) + 1;
2768                 ic->synchronous_mode = 1;
2769
2770                 cancel_delayed_work_sync(&ic->bitmap_flush_work);
2771                 queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, 0);
2772                 flush_workqueue(ic->commit_wq);
2773         }
2774 }
2775
2776 static int dm_integrity_reboot(struct notifier_block *n, unsigned long code, void *x)
2777 {
2778         struct dm_integrity_c *ic = container_of(n, struct dm_integrity_c, reboot_notifier);
2779
2780         DEBUG_print("dm_integrity_reboot\n");
2781
2782         dm_integrity_enter_synchronous_mode(ic);
2783
2784         return NOTIFY_DONE;
2785 }
2786
2787 static void dm_integrity_postsuspend(struct dm_target *ti)
2788 {
2789         struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
2790         int r;
2791
2792         WARN_ON(unregister_reboot_notifier(&ic->reboot_notifier));
2793
2794         del_timer_sync(&ic->autocommit_timer);
2795
2796         if (ic->recalc_wq)
2797                 drain_workqueue(ic->recalc_wq);
2798
2799         if (ic->mode == 'B')
2800                 cancel_delayed_work_sync(&ic->bitmap_flush_work);
2801
2802         queue_work(ic->commit_wq, &ic->commit_work);
2803         drain_workqueue(ic->commit_wq);
2804
2805         if (ic->mode == 'J') {
2806                 if (ic->meta_dev)
2807                         queue_work(ic->writer_wq, &ic->writer_work);
2808                 drain_workqueue(ic->writer_wq);
2809                 dm_integrity_flush_buffers(ic);
2810         }
2811
2812         if (ic->mode == 'B') {
2813                 dm_integrity_flush_buffers(ic);
2814 #if 1
2815                 /* set to 0 to test bitmap replay code */
2816                 init_journal(ic, 0, ic->journal_sections, 0);
2817                 ic->sb->flags &= ~cpu_to_le32(SB_FLAG_DIRTY_BITMAP);
2818                 r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
2819                 if (unlikely(r))
2820                         dm_integrity_io_error(ic, "writing superblock", r);
2821 #endif
2822         }
2823
2824         BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
2825
2826         ic->journal_uptodate = true;
2827 }
2828
2829 static void dm_integrity_resume(struct dm_target *ti)
2830 {
2831         struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
2832         int r;
2833         DEBUG_print("resume\n");
2834
2835         if (ic->sb->flags & cpu_to_le32(SB_FLAG_DIRTY_BITMAP)) {
2836                 DEBUG_print("resume dirty_bitmap\n");
2837                 rw_journal_sectors(ic, REQ_OP_READ, 0, 0,
2838                                    ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
2839                 if (ic->mode == 'B') {
2840                         if (ic->sb->log2_blocks_per_bitmap_bit == ic->log2_blocks_per_bitmap_bit) {
2841                                 block_bitmap_copy(ic, ic->recalc_bitmap, ic->journal);
2842                                 block_bitmap_copy(ic, ic->may_write_bitmap, ic->journal);
2843                                 if (!block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors,
2844                                                      BITMAP_OP_TEST_ALL_CLEAR)) {
2845                                         ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
2846                                         ic->sb->recalc_sector = cpu_to_le64(0);
2847                                 }
2848                         } else {
2849                                 DEBUG_print("non-matching blocks_per_bitmap_bit: %u, %u\n",
2850                                             ic->sb->log2_blocks_per_bitmap_bit, ic->log2_blocks_per_bitmap_bit);
2851                                 ic->sb->log2_blocks_per_bitmap_bit = ic->log2_blocks_per_bitmap_bit;
2852                                 block_bitmap_op(ic, ic->recalc_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_SET);
2853                                 block_bitmap_op(ic, ic->may_write_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_SET);
2854                                 block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors, BITMAP_OP_SET);
2855                                 rw_journal_sectors(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, 0,
2856                                                    ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
2857                                 ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
2858                                 ic->sb->recalc_sector = cpu_to_le64(0);
2859                         }
2860                 } else {
2861                         if (!(ic->sb->log2_blocks_per_bitmap_bit == ic->log2_blocks_per_bitmap_bit &&
2862                               block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors, BITMAP_OP_TEST_ALL_CLEAR))) {
2863                                 ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
2864                                 ic->sb->recalc_sector = cpu_to_le64(0);
2865                         }
2866                         init_journal(ic, 0, ic->journal_sections, 0);
2867                         replay_journal(ic);
2868                         ic->sb->flags &= ~cpu_to_le32(SB_FLAG_DIRTY_BITMAP);
2869                 }
2870                 r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
2871                 if (unlikely(r))
2872                         dm_integrity_io_error(ic, "writing superblock", r);
2873         } else {
2874                 replay_journal(ic);
2875                 if (ic->mode == 'B') {
2876                         ic->sb->flags |= cpu_to_le32(SB_FLAG_DIRTY_BITMAP);
2877                         ic->sb->log2_blocks_per_bitmap_bit = ic->log2_blocks_per_bitmap_bit;
2878                         r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
2879                         if (unlikely(r))
2880                                 dm_integrity_io_error(ic, "writing superblock", r);
2881
2882                         block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors, BITMAP_OP_CLEAR);
2883                         block_bitmap_op(ic, ic->recalc_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_CLEAR);
2884                         block_bitmap_op(ic, ic->may_write_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_CLEAR);
2885                         if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING) &&
2886                             le64_to_cpu(ic->sb->recalc_sector) < ic->provided_data_sectors) {
2887                                 block_bitmap_op(ic, ic->journal, le64_to_cpu(ic->sb->recalc_sector),
2888                                                 ic->provided_data_sectors - le64_to_cpu(ic->sb->recalc_sector), BITMAP_OP_SET);
2889                                 block_bitmap_op(ic, ic->recalc_bitmap, le64_to_cpu(ic->sb->recalc_sector),
2890                                                 ic->provided_data_sectors - le64_to_cpu(ic->sb->recalc_sector), BITMAP_OP_SET);
2891                                 block_bitmap_op(ic, ic->may_write_bitmap, le64_to_cpu(ic->sb->recalc_sector),
2892                                                 ic->provided_data_sectors - le64_to_cpu(ic->sb->recalc_sector), BITMAP_OP_SET);
2893                         }
2894                         rw_journal_sectors(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, 0,
2895                                            ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
2896                 }
2897         }
2898
2899         DEBUG_print("testing recalc: %x\n", ic->sb->flags);
2900         if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
2901                 __u64 recalc_pos = le64_to_cpu(ic->sb->recalc_sector);
2902                 DEBUG_print("recalc pos: %llx / %llx\n", recalc_pos, ic->provided_data_sectors);
2903                 if (recalc_pos < ic->provided_data_sectors) {
2904                         queue_work(ic->recalc_wq, &ic->recalc_work);
2905                 } else if (recalc_pos > ic->provided_data_sectors) {
2906                         ic->sb->recalc_sector = cpu_to_le64(ic->provided_data_sectors);
2907                         recalc_write_super(ic);
2908                 }
2909         }
2910
2911         ic->reboot_notifier.notifier_call = dm_integrity_reboot;
2912         ic->reboot_notifier.next = NULL;
2913         ic->reboot_notifier.priority = INT_MAX - 1;     /* be notified after md and before hardware drivers */
2914         WARN_ON(register_reboot_notifier(&ic->reboot_notifier));
2915
2916 #if 0
2917         /* set to 1 to stress test synchronous mode */
2918         dm_integrity_enter_synchronous_mode(ic);
2919 #endif
2920 }
2921
2922 static void dm_integrity_status(struct dm_target *ti, status_type_t type,
2923                                 unsigned status_flags, char *result, unsigned maxlen)
2924 {
2925         struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
2926         unsigned arg_count;
2927         size_t sz = 0;
2928
2929         switch (type) {
2930         case STATUSTYPE_INFO:
2931                 DMEMIT("%llu %llu",
2932                         atomic64_read(&ic->number_of_mismatches),
2933                         ic->provided_data_sectors);
2934                 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
2935                         DMEMIT(" %llu", le64_to_cpu(ic->sb->recalc_sector));
2936                 else
2937                         DMEMIT(" -");
2938                 break;
2939
2940         case STATUSTYPE_TABLE: {
2941                 __u64 watermark_percentage = (__u64)(ic->journal_entries - ic->free_sectors_threshold) * 100;
2942                 watermark_percentage += ic->journal_entries / 2;
2943                 do_div(watermark_percentage, ic->journal_entries);
2944                 arg_count = 3;
2945                 arg_count += !!ic->meta_dev;
2946                 arg_count += ic->sectors_per_block != 1;
2947                 arg_count += !!(ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING));
2948                 arg_count += ic->mode == 'J';
2949                 arg_count += ic->mode == 'J';
2950                 arg_count += ic->mode == 'B';
2951                 arg_count += ic->mode == 'B';
2952                 arg_count += !!ic->internal_hash_alg.alg_string;
2953                 arg_count += !!ic->journal_crypt_alg.alg_string;
2954                 arg_count += !!ic->journal_mac_alg.alg_string;
2955                 arg_count += (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING)) != 0;
2956                 DMEMIT("%s %llu %u %c %u", ic->dev->name, ic->start,
2957                        ic->tag_size, ic->mode, arg_count);
2958                 if (ic->meta_dev)
2959                         DMEMIT(" meta_device:%s", ic->meta_dev->name);
2960                 if (ic->sectors_per_block != 1)
2961                         DMEMIT(" block_size:%u", ic->sectors_per_block << SECTOR_SHIFT);
2962                 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
2963                         DMEMIT(" recalculate");
2964                 DMEMIT(" journal_sectors:%u", ic->initial_sectors - SB_SECTORS);
2965                 DMEMIT(" interleave_sectors:%u", 1U << ic->sb->log2_interleave_sectors);
2966                 DMEMIT(" buffer_sectors:%u", 1U << ic->log2_buffer_sectors);
2967                 if (ic->mode == 'J') {
2968                         DMEMIT(" journal_watermark:%u", (unsigned)watermark_percentage);
2969                         DMEMIT(" commit_time:%u", ic->autocommit_msec);
2970                 }
2971                 if (ic->mode == 'B') {
2972                         DMEMIT(" sectors_per_bit:%llu", (sector_t)ic->sectors_per_block << ic->log2_blocks_per_bitmap_bit);
2973                         DMEMIT(" bitmap_flush_interval:%u", jiffies_to_msecs(ic->bitmap_flush_interval));
2974                 }
2975                 if ((ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING)) != 0)
2976                         DMEMIT(" fix_padding");
2977
2978 #define EMIT_ALG(a, n)                                                  \
2979                 do {                                                    \
2980                         if (ic->a.alg_string) {                         \
2981                                 DMEMIT(" %s:%s", n, ic->a.alg_string);  \
2982                                 if (ic->a.key_string)                   \
2983                                         DMEMIT(":%s", ic->a.key_string);\
2984                         }                                               \
2985                 } while (0)
2986                 EMIT_ALG(internal_hash_alg, "internal_hash");
2987                 EMIT_ALG(journal_crypt_alg, "journal_crypt");
2988                 EMIT_ALG(journal_mac_alg, "journal_mac");
2989                 break;
2990         }
2991         }
2992 }
2993
2994 static int dm_integrity_iterate_devices(struct dm_target *ti,
2995                                         iterate_devices_callout_fn fn, void *data)
2996 {
2997         struct dm_integrity_c *ic = ti->private;
2998
2999         if (!ic->meta_dev)
3000                 return fn(ti, ic->dev, ic->start + ic->initial_sectors + ic->metadata_run, ti->len, data);
3001         else
3002                 return fn(ti, ic->dev, 0, ti->len, data);
3003 }
3004
3005 static void dm_integrity_io_hints(struct dm_target *ti, struct queue_limits *limits)
3006 {
3007         struct dm_integrity_c *ic = ti->private;
3008
3009         if (ic->sectors_per_block > 1) {
3010                 limits->logical_block_size = ic->sectors_per_block << SECTOR_SHIFT;
3011                 limits->physical_block_size = ic->sectors_per_block << SECTOR_SHIFT;
3012                 blk_limits_io_min(limits, ic->sectors_per_block << SECTOR_SHIFT);
3013         }
3014 }
3015
3016 static void calculate_journal_section_size(struct dm_integrity_c *ic)
3017 {
3018         unsigned sector_space = JOURNAL_SECTOR_DATA;
3019
3020         ic->journal_sections = le32_to_cpu(ic->sb->journal_sections);
3021         ic->journal_entry_size = roundup(offsetof(struct journal_entry, last_bytes[ic->sectors_per_block]) + ic->tag_size,
3022                                          JOURNAL_ENTRY_ROUNDUP);
3023
3024         if (ic->sb->flags & cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC))
3025                 sector_space -= JOURNAL_MAC_PER_SECTOR;
3026         ic->journal_entries_per_sector = sector_space / ic->journal_entry_size;
3027         ic->journal_section_entries = ic->journal_entries_per_sector * JOURNAL_BLOCK_SECTORS;
3028         ic->journal_section_sectors = (ic->journal_section_entries << ic->sb->log2_sectors_per_block) + JOURNAL_BLOCK_SECTORS;
3029         ic->journal_entries = ic->journal_section_entries * ic->journal_sections;
3030 }
3031
3032 static int calculate_device_limits(struct dm_integrity_c *ic)
3033 {
3034         __u64 initial_sectors;
3035
3036         calculate_journal_section_size(ic);
3037         initial_sectors = SB_SECTORS + (__u64)ic->journal_section_sectors * ic->journal_sections;
3038         if (initial_sectors + METADATA_PADDING_SECTORS >= ic->meta_device_sectors || initial_sectors > UINT_MAX)
3039                 return -EINVAL;
3040         ic->initial_sectors = initial_sectors;
3041
3042         if (!ic->meta_dev) {
3043                 sector_t last_sector, last_area, last_offset;
3044
3045                 /* we have to maintain excessive padding for compatibility with existing volumes */
3046                 __u64 metadata_run_padding =
3047                         ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING) ?
3048                         (__u64)(METADATA_PADDING_SECTORS << SECTOR_SHIFT) :
3049                         (__u64)(1 << SECTOR_SHIFT << METADATA_PADDING_SECTORS);
3050
3051                 ic->metadata_run = round_up((__u64)ic->tag_size << (ic->sb->log2_interleave_sectors - ic->sb->log2_sectors_per_block),
3052                                             metadata_run_padding) >> SECTOR_SHIFT;
3053                 if (!(ic->metadata_run & (ic->metadata_run - 1)))
3054                         ic->log2_metadata_run = __ffs(ic->metadata_run);
3055                 else
3056                         ic->log2_metadata_run = -1;
3057
3058                 get_area_and_offset(ic, ic->provided_data_sectors - 1, &last_area, &last_offset);
3059                 last_sector = get_data_sector(ic, last_area, last_offset);
3060                 if (last_sector < ic->start || last_sector >= ic->meta_device_sectors)
3061                         return -EINVAL;
3062         } else {
3063                 __u64 meta_size = (ic->provided_data_sectors >> ic->sb->log2_sectors_per_block) * ic->tag_size;
3064                 meta_size = (meta_size + ((1U << (ic->log2_buffer_sectors + SECTOR_SHIFT)) - 1))
3065                                 >> (ic->log2_buffer_sectors + SECTOR_SHIFT);
3066                 meta_size <<= ic->log2_buffer_sectors;
3067                 if (ic->initial_sectors + meta_size < ic->initial_sectors ||
3068                     ic->initial_sectors + meta_size > ic->meta_device_sectors)
3069                         return -EINVAL;
3070                 ic->metadata_run = 1;
3071                 ic->log2_metadata_run = 0;
3072         }
3073
3074         return 0;
3075 }
3076
3077 static int initialize_superblock(struct dm_integrity_c *ic, unsigned journal_sectors, unsigned interleave_sectors)
3078 {
3079         unsigned journal_sections;
3080         int test_bit;
3081
3082         memset(ic->sb, 0, SB_SECTORS << SECTOR_SHIFT);
3083         memcpy(ic->sb->magic, SB_MAGIC, 8);
3084         ic->sb->integrity_tag_size = cpu_to_le16(ic->tag_size);
3085         ic->sb->log2_sectors_per_block = __ffs(ic->sectors_per_block);
3086         if (ic->journal_mac_alg.alg_string)
3087                 ic->sb->flags |= cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC);
3088
3089         calculate_journal_section_size(ic);
3090         journal_sections = journal_sectors / ic->journal_section_sectors;
3091         if (!journal_sections)
3092                 journal_sections = 1;
3093
3094         if (!ic->meta_dev) {
3095                 if (ic->fix_padding)
3096                         ic->sb->flags |= cpu_to_le32(SB_FLAG_FIXED_PADDING);
3097                 ic->sb->journal_sections = cpu_to_le32(journal_sections);
3098                 if (!interleave_sectors)
3099                         interleave_sectors = DEFAULT_INTERLEAVE_SECTORS;
3100                 ic->sb->log2_interleave_sectors = __fls(interleave_sectors);
3101                 ic->sb->log2_interleave_sectors = max((__u8)MIN_LOG2_INTERLEAVE_SECTORS, ic->sb->log2_interleave_sectors);
3102                 ic->sb->log2_interleave_sectors = min((__u8)MAX_LOG2_INTERLEAVE_SECTORS, ic->sb->log2_interleave_sectors);
3103
3104                 ic->provided_data_sectors = 0;
3105                 for (test_bit = fls64(ic->meta_device_sectors) - 1; test_bit >= 3; test_bit--) {
3106                         __u64 prev_data_sectors = ic->provided_data_sectors;
3107
3108                         ic->provided_data_sectors |= (sector_t)1 << test_bit;
3109                         if (calculate_device_limits(ic))
3110                                 ic->provided_data_sectors = prev_data_sectors;
3111                 }
3112                 if (!ic->provided_data_sectors)
3113                         return -EINVAL;
3114         } else {
3115                 ic->sb->log2_interleave_sectors = 0;
3116                 ic->provided_data_sectors = ic->data_device_sectors;
3117                 ic->provided_data_sectors &= ~(sector_t)(ic->sectors_per_block - 1);
3118
3119 try_smaller_buffer:
3120                 ic->sb->journal_sections = cpu_to_le32(0);
3121                 for (test_bit = fls(journal_sections) - 1; test_bit >= 0; test_bit--) {
3122                         __u32 prev_journal_sections = le32_to_cpu(ic->sb->journal_sections);
3123                         __u32 test_journal_sections = prev_journal_sections | (1U << test_bit);
3124                         if (test_journal_sections > journal_sections)
3125                                 continue;
3126                         ic->sb->journal_sections = cpu_to_le32(test_journal_sections);
3127                         if (calculate_device_limits(ic))
3128                                 ic->sb->journal_sections = cpu_to_le32(prev_journal_sections);
3129
3130                 }
3131                 if (!le32_to_cpu(ic->sb->journal_sections)) {
3132                         if (ic->log2_buffer_sectors > 3) {
3133                                 ic->log2_buffer_sectors--;
3134                                 goto try_smaller_buffer;
3135                         }
3136                         return -EINVAL;
3137                 }
3138         }
3139
3140         ic->sb->provided_data_sectors = cpu_to_le64(ic->provided_data_sectors);
3141
3142         sb_set_version(ic);
3143
3144         return 0;
3145 }
3146
3147 static void dm_integrity_set(struct dm_target *ti, struct dm_integrity_c *ic)
3148 {
3149         struct gendisk *disk = dm_disk(dm_table_get_md(ti->table));
3150         struct blk_integrity bi;
3151
3152         memset(&bi, 0, sizeof(bi));
3153         bi.profile = &dm_integrity_profile;
3154         bi.tuple_size = ic->tag_size;
3155         bi.tag_size = bi.tuple_size;
3156         bi.interval_exp = ic->sb->log2_sectors_per_block + SECTOR_SHIFT;
3157
3158         blk_integrity_register(disk, &bi);
3159         blk_queue_max_integrity_segments(disk->queue, UINT_MAX);
3160 }
3161
3162 static void dm_integrity_free_page_list(struct page_list *pl)
3163 {
3164         unsigned i;
3165
3166         if (!pl)
3167                 return;
3168         for (i = 0; pl[i].page; i++)
3169                 __free_page(pl[i].page);
3170         kvfree(pl);
3171 }
3172
3173 static struct page_list *dm_integrity_alloc_page_list(unsigned n_pages)
3174 {
3175         struct page_list *pl;
3176         unsigned i;
3177
3178         pl = kvmalloc_array(n_pages + 1, sizeof(struct page_list), GFP_KERNEL | __GFP_ZERO);
3179         if (!pl)
3180                 return NULL;
3181
3182         for (i = 0; i < n_pages; i++) {
3183                 pl[i].page = alloc_page(GFP_KERNEL);
3184                 if (!pl[i].page) {
3185                         dm_integrity_free_page_list(pl);
3186                         return NULL;
3187                 }
3188                 if (i)
3189                         pl[i - 1].next = &pl[i];
3190         }
3191         pl[i].page = NULL;
3192         pl[i].next = NULL;
3193
3194         return pl;
3195 }
3196
3197 static void dm_integrity_free_journal_scatterlist(struct dm_integrity_c *ic, struct scatterlist **sl)
3198 {
3199         unsigned i;
3200         for (i = 0; i < ic->journal_sections; i++)
3201                 kvfree(sl[i]);
3202         kvfree(sl);
3203 }
3204
3205 static struct scatterlist **dm_integrity_alloc_journal_scatterlist(struct dm_integrity_c *ic,
3206                                                                    struct page_list *pl)
3207 {
3208         struct scatterlist **sl;
3209         unsigned i;
3210
3211         sl = kvmalloc_array(ic->journal_sections,
3212                             sizeof(struct scatterlist *),
3213                             GFP_KERNEL | __GFP_ZERO);
3214         if (!sl)
3215                 return NULL;
3216
3217         for (i = 0; i < ic->journal_sections; i++) {
3218                 struct scatterlist *s;
3219                 unsigned start_index, start_offset;
3220                 unsigned end_index, end_offset;
3221                 unsigned n_pages;
3222                 unsigned idx;
3223
3224                 page_list_location(ic, i, 0, &start_index, &start_offset);
3225                 page_list_location(ic, i, ic->journal_section_sectors - 1,
3226                                    &end_index, &end_offset);
3227
3228                 n_pages = (end_index - start_index + 1);
3229
3230                 s = kvmalloc_array(n_pages, sizeof(struct scatterlist),
3231                                    GFP_KERNEL);
3232                 if (!s) {
3233                         dm_integrity_free_journal_scatterlist(ic, sl);
3234                         return NULL;
3235                 }
3236
3237                 sg_init_table(s, n_pages);
3238                 for (idx = start_index; idx <= end_index; idx++) {
3239                         char *va = lowmem_page_address(pl[idx].page);
3240                         unsigned start = 0, end = PAGE_SIZE;
3241                         if (idx == start_index)
3242                                 start = start_offset;
3243                         if (idx == end_index)
3244                                 end = end_offset + (1 << SECTOR_SHIFT);
3245                         sg_set_buf(&s[idx - start_index], va + start, end - start);
3246                 }
3247
3248                 sl[i] = s;
3249         }
3250
3251         return sl;
3252 }
3253
3254 static void free_alg(struct alg_spec *a)
3255 {
3256         kzfree(a->alg_string);
3257         kzfree(a->key);
3258         memset(a, 0, sizeof *a);
3259 }
3260
3261 static int get_alg_and_key(const char *arg, struct alg_spec *a, char **error, char *error_inval)
3262 {
3263         char *k;
3264
3265         free_alg(a);
3266
3267         a->alg_string = kstrdup(strchr(arg, ':') + 1, GFP_KERNEL);
3268         if (!a->alg_string)
3269                 goto nomem;
3270
3271         k = strchr(a->alg_string, ':');
3272         if (k) {
3273                 *k = 0;
3274                 a->key_string = k + 1;
3275                 if (strlen(a->key_string) & 1)
3276                         goto inval;
3277
3278                 a->key_size = strlen(a->key_string) / 2;
3279                 a->key = kmalloc(a->key_size, GFP_KERNEL);
3280                 if (!a->key)
3281                         goto nomem;
3282                 if (hex2bin(a->key, a->key_string, a->key_size))
3283                         goto inval;
3284         }
3285
3286         return 0;
3287 inval:
3288         *error = error_inval;
3289         return -EINVAL;
3290 nomem:
3291         *error = "Out of memory for an argument";
3292         return -ENOMEM;
3293 }
3294
3295 static int get_mac(struct crypto_shash **hash, struct alg_spec *a, char **error,
3296                    char *error_alg, char *error_key)
3297 {
3298         int r;
3299
3300         if (a->alg_string) {
3301                 *hash = crypto_alloc_shash(a->alg_string, 0, 0);
3302                 if (IS_ERR(*hash)) {
3303                         *error = error_alg;
3304                         r = PTR_ERR(*hash);
3305                         *hash = NULL;
3306                         return r;
3307                 }
3308
3309                 if (a->key) {
3310                         r = crypto_shash_setkey(*hash, a->key, a->key_size);
3311                         if (r) {
3312                                 *error = error_key;
3313                                 return r;
3314                         }
3315                 } else if (crypto_shash_get_flags(*hash) & CRYPTO_TFM_NEED_KEY) {
3316                         *error = error_key;
3317                         return -ENOKEY;
3318                 }
3319         }
3320
3321         return 0;
3322 }
3323
3324 static int create_journal(struct dm_integrity_c *ic, char **error)
3325 {
3326         int r = 0;
3327         unsigned i;
3328         __u64 journal_pages, journal_desc_size, journal_tree_size;
3329         unsigned char *crypt_data = NULL, *crypt_iv = NULL;
3330         struct skcipher_request *req = NULL;
3331
3332         ic->commit_ids[0] = cpu_to_le64(0x1111111111111111ULL);
3333         ic->commit_ids[1] = cpu_to_le64(0x2222222222222222ULL);
3334         ic->commit_ids[2] = cpu_to_le64(0x3333333333333333ULL);
3335         ic->commit_ids[3] = cpu_to_le64(0x4444444444444444ULL);
3336
3337         journal_pages = roundup((__u64)ic->journal_sections * ic->journal_section_sectors,
3338                                 PAGE_SIZE >> SECTOR_SHIFT) >> (PAGE_SHIFT - SECTOR_SHIFT);
3339         journal_desc_size = journal_pages * sizeof(struct page_list);
3340         if (journal_pages >= totalram_pages() - totalhigh_pages() || journal_desc_size > ULONG_MAX) {
3341                 *error = "Journal doesn't fit into memory";
3342                 r = -ENOMEM;
3343                 goto bad;
3344         }
3345         ic->journal_pages = journal_pages;
3346
3347         ic->journal = dm_integrity_alloc_page_list(ic->journal_pages);
3348         if (!ic->journal) {
3349                 *error = "Could not allocate memory for journal";
3350                 r = -ENOMEM;
3351                 goto bad;
3352         }
3353         if (ic->journal_crypt_alg.alg_string) {
3354                 unsigned ivsize, blocksize;
3355                 struct journal_completion comp;
3356
3357                 comp.ic = ic;
3358                 ic->journal_crypt = crypto_alloc_skcipher(ic->journal_crypt_alg.alg_string, 0, 0);
3359                 if (IS_ERR(ic->journal_crypt)) {
3360                         *error = "Invalid journal cipher";
3361                         r = PTR_ERR(ic->journal_crypt);
3362                         ic->journal_crypt = NULL;
3363                         goto bad;
3364                 }
3365                 ivsize = crypto_skcipher_ivsize(ic->journal_crypt);
3366                 blocksize = crypto_skcipher_blocksize(ic->journal_crypt);
3367
3368                 if (ic->journal_crypt_alg.key) {
3369                         r = crypto_skcipher_setkey(ic->journal_crypt, ic->journal_crypt_alg.key,
3370                                                    ic->journal_crypt_alg.key_size);
3371                         if (r) {
3372                                 *error = "Error setting encryption key";
3373                                 goto bad;
3374                         }
3375                 }
3376                 DEBUG_print("cipher %s, block size %u iv size %u\n",
3377                             ic->journal_crypt_alg.alg_string, blocksize, ivsize);
3378
3379                 ic->journal_io = dm_integrity_alloc_page_list(ic->journal_pages);
3380                 if (!ic->journal_io) {
3381                         *error = "Could not allocate memory for journal io";
3382                         r = -ENOMEM;
3383                         goto bad;
3384                 }
3385
3386                 if (blocksize == 1) {
3387                         struct scatterlist *sg;
3388
3389                         req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
3390                         if (!req) {
3391                                 *error = "Could not allocate crypt request";
3392                                 r = -ENOMEM;
3393                                 goto bad;
3394                         }
3395
3396                         crypt_iv = kzalloc(ivsize, GFP_KERNEL);
3397                         if (!crypt_iv) {
3398                                 *error = "Could not allocate iv";
3399                                 r = -ENOMEM;
3400                                 goto bad;
3401                         }
3402
3403                         ic->journal_xor = dm_integrity_alloc_page_list(ic->journal_pages);
3404                         if (!ic->journal_xor) {
3405                                 *error = "Could not allocate memory for journal xor";
3406                                 r = -ENOMEM;
3407                                 goto bad;
3408                         }
3409
3410                         sg = kvmalloc_array(ic->journal_pages + 1,
3411                                             sizeof(struct scatterlist),
3412                                             GFP_KERNEL);
3413                         if (!sg) {
3414                                 *error = "Unable to allocate sg list";
3415                                 r = -ENOMEM;
3416                                 goto bad;
3417                         }
3418                         sg_init_table(sg, ic->journal_pages + 1);
3419                         for (i = 0; i < ic->journal_pages; i++) {
3420                                 char *va = lowmem_page_address(ic->journal_xor[i].page);
3421                                 clear_page(va);
3422                                 sg_set_buf(&sg[i], va, PAGE_SIZE);
3423                         }
3424                         sg_set_buf(&sg[i], &ic->commit_ids, sizeof ic->commit_ids);
3425
3426                         skcipher_request_set_crypt(req, sg, sg,
3427                                                    PAGE_SIZE * ic->journal_pages + sizeof ic->commit_ids, crypt_iv);
3428                         init_completion(&comp.comp);
3429                         comp.in_flight = (atomic_t)ATOMIC_INIT(1);
3430                         if (do_crypt(true, req, &comp))
3431                                 wait_for_completion(&comp.comp);
3432                         kvfree(sg);
3433                         r = dm_integrity_failed(ic);
3434                         if (r) {
3435                                 *error = "Unable to encrypt journal";
3436                                 goto bad;
3437                         }
3438                         DEBUG_bytes(lowmem_page_address(ic->journal_xor[0].page), 64, "xor data");
3439
3440                         crypto_free_skcipher(ic->journal_crypt);
3441                         ic->journal_crypt = NULL;
3442                 } else {
3443                         unsigned crypt_len = roundup(ivsize, blocksize);
3444
3445                         req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
3446                         if (!req) {
3447                                 *error = "Could not allocate crypt request";
3448                                 r = -ENOMEM;
3449                                 goto bad;
3450                         }
3451
3452                         crypt_iv = kmalloc(ivsize, GFP_KERNEL);
3453                         if (!crypt_iv) {
3454                                 *error = "Could not allocate iv";
3455                                 r = -ENOMEM;
3456                                 goto bad;
3457                         }
3458
3459                         crypt_data = kmalloc(crypt_len, GFP_KERNEL);
3460                         if (!crypt_data) {
3461                                 *error = "Unable to allocate crypt data";
3462                                 r = -ENOMEM;
3463                                 goto bad;
3464                         }
3465
3466                         ic->journal_scatterlist = dm_integrity_alloc_journal_scatterlist(ic, ic->journal);
3467                         if (!ic->journal_scatterlist) {
3468                                 *error = "Unable to allocate sg list";
3469                                 r = -ENOMEM;
3470                                 goto bad;
3471                         }
3472                         ic->journal_io_scatterlist = dm_integrity_alloc_journal_scatterlist(ic, ic->journal_io);
3473                         if (!ic->journal_io_scatterlist) {
3474                                 *error = "Unable to allocate sg list";
3475                                 r = -ENOMEM;
3476                                 goto bad;
3477                         }
3478                         ic->sk_requests = kvmalloc_array(ic->journal_sections,
3479                                                          sizeof(struct skcipher_request *),
3480                                                          GFP_KERNEL | __GFP_ZERO);
3481                         if (!ic->sk_requests) {
3482                                 *error = "Unable to allocate sk requests";
3483                                 r = -ENOMEM;
3484                                 goto bad;
3485                         }
3486                         for (i = 0; i < ic->journal_sections; i++) {
3487                                 struct scatterlist sg;
3488                                 struct skcipher_request *section_req;
3489                                 __u32 section_le = cpu_to_le32(i);
3490
3491                                 memset(crypt_iv, 0x00, ivsize);
3492                                 memset(crypt_data, 0x00, crypt_len);
3493                                 memcpy(crypt_data, &section_le, min((size_t)crypt_len, sizeof(section_le)));
3494
3495                                 sg_init_one(&sg, crypt_data, crypt_len);
3496                                 skcipher_request_set_crypt(req, &sg, &sg, crypt_len, crypt_iv);
3497                                 init_completion(&comp.comp);
3498                                 comp.in_flight = (atomic_t)ATOMIC_INIT(1);
3499                                 if (do_crypt(true, req, &comp))
3500                                         wait_for_completion(&comp.comp);
3501
3502                                 r = dm_integrity_failed(ic);
3503                                 if (r) {
3504                                         *error = "Unable to generate iv";
3505                                         goto bad;
3506                                 }
3507
3508                                 section_req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
3509                                 if (!section_req) {
3510                                         *error = "Unable to allocate crypt request";
3511                                         r = -ENOMEM;
3512                                         goto bad;
3513                                 }
3514                                 section_req->iv = kmalloc_array(ivsize, 2,
3515                                                                 GFP_KERNEL);
3516                                 if (!section_req->iv) {
3517                                         skcipher_request_free(section_req);
3518                                         *error = "Unable to allocate iv";
3519                                         r = -ENOMEM;
3520                                         goto bad;
3521                                 }
3522                                 memcpy(section_req->iv + ivsize, crypt_data, ivsize);
3523                                 section_req->cryptlen = (size_t)ic->journal_section_sectors << SECTOR_SHIFT;
3524                                 ic->sk_requests[i] = section_req;
3525                                 DEBUG_bytes(crypt_data, ivsize, "iv(%u)", i);
3526                         }
3527                 }
3528         }
3529
3530         for (i = 0; i < N_COMMIT_IDS; i++) {
3531                 unsigned j;
3532 retest_commit_id:
3533                 for (j = 0; j < i; j++) {
3534                         if (ic->commit_ids[j] == ic->commit_ids[i]) {
3535                                 ic->commit_ids[i] = cpu_to_le64(le64_to_cpu(ic->commit_ids[i]) + 1);
3536                                 goto retest_commit_id;
3537                         }
3538                 }
3539                 DEBUG_print("commit id %u: %016llx\n", i, ic->commit_ids[i]);
3540         }
3541
3542         journal_tree_size = (__u64)ic->journal_entries * sizeof(struct journal_node);
3543         if (journal_tree_size > ULONG_MAX) {
3544                 *error = "Journal doesn't fit into memory";
3545                 r = -ENOMEM;
3546                 goto bad;
3547         }
3548         ic->journal_tree = kvmalloc(journal_tree_size, GFP_KERNEL);
3549         if (!ic->journal_tree) {
3550                 *error = "Could not allocate memory for journal tree";
3551                 r = -ENOMEM;
3552         }
3553 bad:
3554         kfree(crypt_data);
3555         kfree(crypt_iv);
3556         skcipher_request_free(req);
3557
3558         return r;
3559 }
3560
3561 /*
3562  * Construct a integrity mapping
3563  *
3564  * Arguments:
3565  *      device
3566  *      offset from the start of the device
3567  *      tag size
3568  *      D - direct writes, J - journal writes, B - bitmap mode, R - recovery mode
3569  *      number of optional arguments
3570  *      optional arguments:
3571  *              journal_sectors
3572  *              interleave_sectors
3573  *              buffer_sectors
3574  *              journal_watermark
3575  *              commit_time
3576  *              meta_device
3577  *              block_size
3578  *              sectors_per_bit
3579  *              bitmap_flush_interval
3580  *              internal_hash
3581  *              journal_crypt
3582  *              journal_mac
3583  *              recalculate
3584  */
3585 static int dm_integrity_ctr(struct dm_target *ti, unsigned argc, char **argv)
3586 {
3587         struct dm_integrity_c *ic;
3588         char dummy;
3589         int r;
3590         unsigned extra_args;
3591         struct dm_arg_set as;
3592         static const struct dm_arg _args[] = {
3593                 {0, 9, "Invalid number of feature args"},
3594         };
3595         unsigned journal_sectors, interleave_sectors, buffer_sectors, journal_watermark, sync_msec;
3596         bool should_write_sb;
3597         __u64 threshold;
3598         unsigned long long start;
3599         __s8 log2_sectors_per_bitmap_bit = -1;
3600         __s8 log2_blocks_per_bitmap_bit;
3601         __u64 bits_in_journal;
3602         __u64 n_bitmap_bits;
3603
3604 #define DIRECT_ARGUMENTS        4
3605
3606         if (argc <= DIRECT_ARGUMENTS) {
3607                 ti->error = "Invalid argument count";
3608                 return -EINVAL;
3609         }
3610
3611         ic = kzalloc(sizeof(struct dm_integrity_c), GFP_KERNEL);
3612         if (!ic) {
3613                 ti->error = "Cannot allocate integrity context";
3614                 return -ENOMEM;
3615         }
3616         ti->private = ic;
3617         ti->per_io_data_size = sizeof(struct dm_integrity_io);
3618         ic->ti = ti;
3619
3620         ic->in_progress = RB_ROOT;
3621         INIT_LIST_HEAD(&ic->wait_list);
3622         init_waitqueue_head(&ic->endio_wait);
3623         bio_list_init(&ic->flush_bio_list);
3624         init_waitqueue_head(&ic->copy_to_journal_wait);
3625         init_completion(&ic->crypto_backoff);
3626         atomic64_set(&ic->number_of_mismatches, 0);
3627         ic->bitmap_flush_interval = BITMAP_FLUSH_INTERVAL;
3628
3629         r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &ic->dev);
3630         if (r) {
3631                 ti->error = "Device lookup failed";
3632                 goto bad;
3633         }
3634
3635         if (sscanf(argv[1], "%llu%c", &start, &dummy) != 1 || start != (sector_t)start) {
3636                 ti->error = "Invalid starting offset";
3637                 r = -EINVAL;
3638                 goto bad;
3639         }
3640         ic->start = start;
3641
3642         if (strcmp(argv[2], "-")) {
3643                 if (sscanf(argv[2], "%u%c", &ic->tag_size, &dummy) != 1 || !ic->tag_size) {
3644                         ti->error = "Invalid tag size";
3645                         r = -EINVAL;
3646                         goto bad;
3647                 }
3648         }
3649
3650         if (!strcmp(argv[3], "J") || !strcmp(argv[3], "B") ||
3651             !strcmp(argv[3], "D") || !strcmp(argv[3], "R")) {
3652                 ic->mode = argv[3][0];
3653         } else {
3654                 ti->error = "Invalid mode (expecting J, B, D, R)";
3655                 r = -EINVAL;
3656                 goto bad;
3657         }
3658
3659         journal_sectors = 0;
3660         interleave_sectors = DEFAULT_INTERLEAVE_SECTORS;
3661         buffer_sectors = DEFAULT_BUFFER_SECTORS;
3662         journal_watermark = DEFAULT_JOURNAL_WATERMARK;
3663         sync_msec = DEFAULT_SYNC_MSEC;
3664         ic->sectors_per_block = 1;
3665
3666         as.argc = argc - DIRECT_ARGUMENTS;
3667         as.argv = argv + DIRECT_ARGUMENTS;
3668         r = dm_read_arg_group(_args, &as, &extra_args, &ti->error);
3669         if (r)
3670                 goto bad;
3671
3672         while (extra_args--) {
3673                 const char *opt_string;
3674                 unsigned val;
3675                 unsigned long long llval;
3676                 opt_string = dm_shift_arg(&as);
3677                 if (!opt_string) {
3678                         r = -EINVAL;
3679                         ti->error = "Not enough feature arguments";
3680                         goto bad;
3681                 }
3682                 if (sscanf(opt_string, "journal_sectors:%u%c", &val, &dummy) == 1)
3683                         journal_sectors = val ? val : 1;
3684                 else if (sscanf(opt_string, "interleave_sectors:%u%c", &val, &dummy) == 1)
3685                         interleave_sectors = val;
3686                 else if (sscanf(opt_string, "buffer_sectors:%u%c", &val, &dummy) == 1)
3687                         buffer_sectors = val;
3688                 else if (sscanf(opt_string, "journal_watermark:%u%c", &val, &dummy) == 1 && val <= 100)
3689                         journal_watermark = val;
3690                 else if (sscanf(opt_string, "commit_time:%u%c", &val, &dummy) == 1)
3691                         sync_msec = val;
3692                 else if (!strncmp(opt_string, "meta_device:", strlen("meta_device:"))) {
3693                         if (ic->meta_dev) {
3694                                 dm_put_device(ti, ic->meta_dev);
3695                                 ic->meta_dev = NULL;
3696                         }
3697                         r = dm_get_device(ti, strchr(opt_string, ':') + 1,
3698                                           dm_table_get_mode(ti->table), &ic->meta_dev);
3699                         if (r) {
3700                                 ti->error = "Device lookup failed";
3701                                 goto bad;
3702                         }
3703                 } else if (sscanf(opt_string, "block_size:%u%c", &val, &dummy) == 1) {
3704                         if (val < 1 << SECTOR_SHIFT ||
3705                             val > MAX_SECTORS_PER_BLOCK << SECTOR_SHIFT ||
3706                             (val & (val -1))) {
3707                                 r = -EINVAL;
3708                                 ti->error = "Invalid block_size argument";
3709                                 goto bad;
3710                         }
3711                         ic->sectors_per_block = val >> SECTOR_SHIFT;
3712                 } else if (sscanf(opt_string, "sectors_per_bit:%llu%c", &llval, &dummy) == 1) {
3713                         log2_sectors_per_bitmap_bit = !llval ? 0 : __ilog2_u64(llval);
3714                 } else if (sscanf(opt_string, "bitmap_flush_interval:%u%c", &val, &dummy) == 1) {
3715                         if (val >= (uint64_t)UINT_MAX * 1000 / HZ) {
3716                                 r = -EINVAL;
3717                                 ti->error = "Invalid bitmap_flush_interval argument";
3718                         }
3719                         ic->bitmap_flush_interval = msecs_to_jiffies(val);
3720                 } else if (!strncmp(opt_string, "internal_hash:", strlen("internal_hash:"))) {
3721                         r = get_alg_and_key(opt_string, &ic->internal_hash_alg, &ti->error,
3722                                             "Invalid internal_hash argument");
3723                         if (r)
3724                                 goto bad;
3725                 } else if (!strncmp(opt_string, "journal_crypt:", strlen("journal_crypt:"))) {
3726                         r = get_alg_and_key(opt_string, &ic->journal_crypt_alg, &ti->error,
3727                                             "Invalid journal_crypt argument");
3728                         if (r)
3729                                 goto bad;
3730                 } else if (!strncmp(opt_string, "journal_mac:", strlen("journal_mac:"))) {
3731                         r = get_alg_and_key(opt_string, &ic->journal_mac_alg,  &ti->error,
3732                                             "Invalid journal_mac argument");
3733                         if (r)
3734                                 goto bad;
3735                 } else if (!strcmp(opt_string, "recalculate")) {
3736                         ic->recalculate_flag = true;
3737                 } else if (!strcmp(opt_string, "fix_padding")) {
3738                         ic->fix_padding = true;
3739                 } else {
3740                         r = -EINVAL;
3741                         ti->error = "Invalid argument";
3742                         goto bad;
3743                 }
3744         }
3745
3746         ic->data_device_sectors = i_size_read(ic->dev->bdev->bd_inode) >> SECTOR_SHIFT;
3747         if (!ic->meta_dev)
3748                 ic->meta_device_sectors = ic->data_device_sectors;
3749         else
3750                 ic->meta_device_sectors = i_size_read(ic->meta_dev->bdev->bd_inode) >> SECTOR_SHIFT;
3751
3752         if (!journal_sectors) {
3753                 journal_sectors = min((sector_t)DEFAULT_MAX_JOURNAL_SECTORS,
3754                                       ic->data_device_sectors >> DEFAULT_JOURNAL_SIZE_FACTOR);
3755         }
3756
3757         if (!buffer_sectors)
3758                 buffer_sectors = 1;
3759         ic->log2_buffer_sectors = min((int)__fls(buffer_sectors), 31 - SECTOR_SHIFT);
3760
3761         r = get_mac(&ic->internal_hash, &ic->internal_hash_alg, &ti->error,
3762                     "Invalid internal hash", "Error setting internal hash key");
3763         if (r)
3764                 goto bad;
3765
3766         r = get_mac(&ic->journal_mac, &ic->journal_mac_alg, &ti->error,
3767                     "Invalid journal mac", "Error setting journal mac key");
3768         if (r)
3769                 goto bad;
3770
3771         if (!ic->tag_size) {
3772                 if (!ic->internal_hash) {
3773                         ti->error = "Unknown tag size";
3774                         r = -EINVAL;
3775                         goto bad;
3776                 }
3777                 ic->tag_size = crypto_shash_digestsize(ic->internal_hash);
3778         }
3779         if (ic->tag_size > MAX_TAG_SIZE) {
3780                 ti->error = "Too big tag size";
3781                 r = -EINVAL;
3782                 goto bad;
3783         }
3784         if (!(ic->tag_size & (ic->tag_size - 1)))
3785                 ic->log2_tag_size = __ffs(ic->tag_size);
3786         else
3787                 ic->log2_tag_size = -1;
3788
3789         if (ic->mode == 'B' && !ic->internal_hash) {
3790                 r = -EINVAL;
3791                 ti->error = "Bitmap mode can be only used with internal hash";
3792                 goto bad;
3793         }
3794
3795         ic->autocommit_jiffies = msecs_to_jiffies(sync_msec);
3796         ic->autocommit_msec = sync_msec;
3797         timer_setup(&ic->autocommit_timer, autocommit_fn, 0);
3798
3799         ic->io = dm_io_client_create();
3800         if (IS_ERR(ic->io)) {
3801                 r = PTR_ERR(ic->io);
3802                 ic->io = NULL;
3803                 ti->error = "Cannot allocate dm io";
3804                 goto bad;
3805         }
3806
3807         r = mempool_init_slab_pool(&ic->journal_io_mempool, JOURNAL_IO_MEMPOOL, journal_io_cache);
3808         if (r) {
3809                 ti->error = "Cannot allocate mempool";
3810                 goto bad;
3811         }
3812
3813         ic->metadata_wq = alloc_workqueue("dm-integrity-metadata",
3814                                           WQ_MEM_RECLAIM, METADATA_WORKQUEUE_MAX_ACTIVE);
3815         if (!ic->metadata_wq) {
3816                 ti->error = "Cannot allocate workqueue";
3817                 r = -ENOMEM;
3818                 goto bad;
3819         }
3820
3821         /*
3822          * If this workqueue were percpu, it would cause bio reordering
3823          * and reduced performance.
3824          */
3825         ic->wait_wq = alloc_workqueue("dm-integrity-wait", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3826         if (!ic->wait_wq) {
3827                 ti->error = "Cannot allocate workqueue";
3828                 r = -ENOMEM;
3829                 goto bad;
3830         }
3831
3832         ic->offload_wq = alloc_workqueue("dm-integrity-offload", WQ_MEM_RECLAIM,
3833                                           METADATA_WORKQUEUE_MAX_ACTIVE);
3834         if (!ic->offload_wq) {
3835                 ti->error = "Cannot allocate workqueue";
3836                 r = -ENOMEM;
3837                 goto bad;
3838         }
3839
3840         ic->commit_wq = alloc_workqueue("dm-integrity-commit", WQ_MEM_RECLAIM, 1);
3841         if (!ic->commit_wq) {
3842                 ti->error = "Cannot allocate workqueue";
3843                 r = -ENOMEM;
3844                 goto bad;
3845         }
3846         INIT_WORK(&ic->commit_work, integrity_commit);
3847
3848         if (ic->mode == 'J' || ic->mode == 'B') {
3849                 ic->writer_wq = alloc_workqueue("dm-integrity-writer", WQ_MEM_RECLAIM, 1);
3850                 if (!ic->writer_wq) {
3851                         ti->error = "Cannot allocate workqueue";
3852                         r = -ENOMEM;
3853                         goto bad;
3854                 }
3855                 INIT_WORK(&ic->writer_work, integrity_writer);
3856         }
3857
3858         ic->sb = alloc_pages_exact(SB_SECTORS << SECTOR_SHIFT, GFP_KERNEL);
3859         if (!ic->sb) {
3860                 r = -ENOMEM;
3861                 ti->error = "Cannot allocate superblock area";
3862                 goto bad;
3863         }
3864
3865         r = sync_rw_sb(ic, REQ_OP_READ, 0);
3866         if (r) {
3867                 ti->error = "Error reading superblock";
3868                 goto bad;
3869         }
3870         should_write_sb = false;
3871         if (memcmp(ic->sb->magic, SB_MAGIC, 8)) {
3872                 if (ic->mode != 'R') {
3873                         if (memchr_inv(ic->sb, 0, SB_SECTORS << SECTOR_SHIFT)) {
3874                                 r = -EINVAL;
3875                                 ti->error = "The device is not initialized";
3876                                 goto bad;
3877                         }
3878                 }
3879
3880                 r = initialize_superblock(ic, journal_sectors, interleave_sectors);
3881                 if (r) {
3882                         ti->error = "Could not initialize superblock";
3883                         goto bad;
3884                 }
3885                 if (ic->mode != 'R')
3886                         should_write_sb = true;
3887         }
3888
3889         if (!ic->sb->version || ic->sb->version > SB_VERSION_4) {
3890                 r = -EINVAL;
3891                 ti->error = "Unknown version";
3892                 goto bad;
3893         }
3894         if (le16_to_cpu(ic->sb->integrity_tag_size) != ic->tag_size) {
3895                 r = -EINVAL;
3896                 ti->error = "Tag size doesn't match the information in superblock";
3897                 goto bad;
3898         }
3899         if (ic->sb->log2_sectors_per_block != __ffs(ic->sectors_per_block)) {
3900                 r = -EINVAL;
3901                 ti->error = "Block size doesn't match the information in superblock";
3902                 goto bad;
3903         }
3904         if (!le32_to_cpu(ic->sb->journal_sections)) {
3905                 r = -EINVAL;
3906                 ti->error = "Corrupted superblock, journal_sections is 0";
3907                 goto bad;
3908         }
3909         /* make sure that ti->max_io_len doesn't overflow */
3910         if (!ic->meta_dev) {
3911                 if (ic->sb->log2_interleave_sectors < MIN_LOG2_INTERLEAVE_SECTORS ||
3912                     ic->sb->log2_interleave_sectors > MAX_LOG2_INTERLEAVE_SECTORS) {
3913                         r = -EINVAL;
3914                         ti->error = "Invalid interleave_sectors in the superblock";
3915                         goto bad;
3916                 }
3917         } else {
3918                 if (ic->sb->log2_interleave_sectors) {
3919                         r = -EINVAL;
3920                         ti->error = "Invalid interleave_sectors in the superblock";
3921                         goto bad;
3922                 }
3923         }
3924         ic->provided_data_sectors = le64_to_cpu(ic->sb->provided_data_sectors);
3925         if (ic->provided_data_sectors != le64_to_cpu(ic->sb->provided_data_sectors)) {
3926                 /* test for overflow */
3927                 r = -EINVAL;
3928                 ti->error = "The superblock has 64-bit device size, but the kernel was compiled with 32-bit sectors";
3929                 goto bad;
3930         }
3931         if (!!(ic->sb->flags & cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC)) != !!ic->journal_mac_alg.alg_string) {
3932                 r = -EINVAL;
3933                 ti->error = "Journal mac mismatch";
3934                 goto bad;
3935         }
3936
3937 try_smaller_buffer:
3938         r = calculate_device_limits(ic);
3939         if (r) {
3940                 if (ic->meta_dev) {
3941                         if (ic->log2_buffer_sectors > 3) {
3942                                 ic->log2_buffer_sectors--;
3943                                 goto try_smaller_buffer;
3944                         }
3945                 }
3946                 ti->error = "The device is too small";
3947                 goto bad;
3948         }
3949
3950         if (log2_sectors_per_bitmap_bit < 0)
3951                 log2_sectors_per_bitmap_bit = __fls(DEFAULT_SECTORS_PER_BITMAP_BIT);
3952         if (log2_sectors_per_bitmap_bit < ic->sb->log2_sectors_per_block)
3953                 log2_sectors_per_bitmap_bit = ic->sb->log2_sectors_per_block;
3954
3955         bits_in_journal = ((__u64)ic->journal_section_sectors * ic->journal_sections) << (SECTOR_SHIFT + 3);
3956         if (bits_in_journal > UINT_MAX)
3957                 bits_in_journal = UINT_MAX;
3958         while (bits_in_journal < (ic->provided_data_sectors + ((sector_t)1 << log2_sectors_per_bitmap_bit) - 1) >> log2_sectors_per_bitmap_bit)
3959                 log2_sectors_per_bitmap_bit++;
3960
3961         log2_blocks_per_bitmap_bit = log2_sectors_per_bitmap_bit - ic->sb->log2_sectors_per_block;
3962         ic->log2_blocks_per_bitmap_bit = log2_blocks_per_bitmap_bit;
3963         if (should_write_sb) {
3964                 ic->sb->log2_blocks_per_bitmap_bit = log2_blocks_per_bitmap_bit;
3965         }
3966         n_bitmap_bits = ((ic->provided_data_sectors >> ic->sb->log2_sectors_per_block)
3967                                 + (((sector_t)1 << log2_blocks_per_bitmap_bit) - 1)) >> log2_blocks_per_bitmap_bit;
3968         ic->n_bitmap_blocks = DIV_ROUND_UP(n_bitmap_bits, BITMAP_BLOCK_SIZE * 8);
3969
3970         if (!ic->meta_dev)
3971                 ic->log2_buffer_sectors = min(ic->log2_buffer_sectors, (__u8)__ffs(ic->metadata_run));
3972
3973         if (ti->len > ic->provided_data_sectors) {
3974                 r = -EINVAL;
3975                 ti->error = "Not enough provided sectors for requested mapping size";
3976                 goto bad;
3977         }
3978
3979
3980         threshold = (__u64)ic->journal_entries * (100 - journal_watermark);
3981         threshold += 50;
3982         do_div(threshold, 100);
3983         ic->free_sectors_threshold = threshold;
3984
3985         DEBUG_print("initialized:\n");
3986         DEBUG_print("   integrity_tag_size %u\n", le16_to_cpu(ic->sb->integrity_tag_size));
3987         DEBUG_print("   journal_entry_size %u\n", ic->journal_entry_size);
3988         DEBUG_print("   journal_entries_per_sector %u\n", ic->journal_entries_per_sector);
3989         DEBUG_print("   journal_section_entries %u\n", ic->journal_section_entries);
3990         DEBUG_print("   journal_section_sectors %u\n", ic->journal_section_sectors);
3991         DEBUG_print("   journal_sections %u\n", (unsigned)le32_to_cpu(ic->sb->journal_sections));
3992         DEBUG_print("   journal_entries %u\n", ic->journal_entries);
3993         DEBUG_print("   log2_interleave_sectors %d\n", ic->sb->log2_interleave_sectors);
3994         DEBUG_print("   data_device_sectors 0x%llx\n", i_size_read(ic->dev->bdev->bd_inode) >> SECTOR_SHIFT);
3995         DEBUG_print("   initial_sectors 0x%x\n", ic->initial_sectors);
3996         DEBUG_print("   metadata_run 0x%x\n", ic->metadata_run);
3997         DEBUG_print("   log2_metadata_run %d\n", ic->log2_metadata_run);
3998         DEBUG_print("   provided_data_sectors 0x%llx (%llu)\n", ic->provided_data_sectors, ic->provided_data_sectors);
3999         DEBUG_print("   log2_buffer_sectors %u\n", ic->log2_buffer_sectors);
4000         DEBUG_print("   bits_in_journal %llu\n", bits_in_journal);
4001
4002         if (ic->recalculate_flag && !(ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))) {
4003                 ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
4004                 ic->sb->recalc_sector = cpu_to_le64(0);
4005         }
4006
4007         if (ic->internal_hash) {
4008                 ic->recalc_wq = alloc_workqueue("dm-integrity-recalc", WQ_MEM_RECLAIM, 1);
4009                 if (!ic->recalc_wq ) {
4010                         ti->error = "Cannot allocate workqueue";
4011                         r = -ENOMEM;
4012                         goto bad;
4013                 }
4014                 INIT_WORK(&ic->recalc_work, integrity_recalc);
4015                 ic->recalc_buffer = vmalloc(RECALC_SECTORS << SECTOR_SHIFT);
4016                 if (!ic->recalc_buffer) {
4017                         ti->error = "Cannot allocate buffer for recalculating";
4018                         r = -ENOMEM;
4019                         goto bad;
4020                 }
4021                 ic->recalc_tags = kvmalloc_array(RECALC_SECTORS >> ic->sb->log2_sectors_per_block,
4022                                                  ic->tag_size, GFP_KERNEL);
4023                 if (!ic->recalc_tags) {
4024                         ti->error = "Cannot allocate tags for recalculating";
4025                         r = -ENOMEM;
4026                         goto bad;
4027                 }
4028         }
4029
4030         ic->bufio = dm_bufio_client_create(ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev,
4031                         1U << (SECTOR_SHIFT + ic->log2_buffer_sectors), 1, 0, NULL, NULL);
4032         if (IS_ERR(ic->bufio)) {
4033                 r = PTR_ERR(ic->bufio);
4034                 ti->error = "Cannot initialize dm-bufio";
4035                 ic->bufio = NULL;
4036                 goto bad;
4037         }
4038         dm_bufio_set_sector_offset(ic->bufio, ic->start + ic->initial_sectors);
4039
4040         if (ic->mode != 'R') {
4041                 r = create_journal(ic, &ti->error);
4042                 if (r)
4043                         goto bad;
4044
4045         }
4046
4047         if (ic->mode == 'B') {
4048                 unsigned i;
4049                 unsigned n_bitmap_pages = DIV_ROUND_UP(ic->n_bitmap_blocks, PAGE_SIZE / BITMAP_BLOCK_SIZE);
4050
4051                 ic->recalc_bitmap = dm_integrity_alloc_page_list(n_bitmap_pages);
4052                 if (!ic->recalc_bitmap) {
4053                         r = -ENOMEM;
4054                         goto bad;
4055                 }
4056                 ic->may_write_bitmap = dm_integrity_alloc_page_list(n_bitmap_pages);
4057                 if (!ic->may_write_bitmap) {
4058                         r = -ENOMEM;
4059                         goto bad;
4060                 }
4061                 ic->bbs = kvmalloc_array(ic->n_bitmap_blocks, sizeof(struct bitmap_block_status), GFP_KERNEL);
4062                 if (!ic->bbs) {
4063                         r = -ENOMEM;
4064                         goto bad;
4065                 }
4066                 INIT_DELAYED_WORK(&ic->bitmap_flush_work, bitmap_flush_work);
4067                 for (i = 0; i < ic->n_bitmap_blocks; i++) {
4068                         struct bitmap_block_status *bbs = &ic->bbs[i];
4069                         unsigned sector, pl_index, pl_offset;
4070
4071                         INIT_WORK(&bbs->work, bitmap_block_work);
4072                         bbs->ic = ic;
4073                         bbs->idx = i;
4074                         bio_list_init(&bbs->bio_queue);
4075                         spin_lock_init(&bbs->bio_queue_lock);
4076
4077                         sector = i * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT);
4078                         pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
4079                         pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
4080
4081                         bbs->bitmap = lowmem_page_address(ic->journal[pl_index].page) + pl_offset;
4082                 }
4083         }
4084
4085         if (should_write_sb) {
4086                 int r;
4087
4088                 init_journal(ic, 0, ic->journal_sections, 0);
4089                 r = dm_integrity_failed(ic);
4090                 if (unlikely(r)) {
4091                         ti->error = "Error initializing journal";
4092                         goto bad;
4093                 }
4094                 r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
4095                 if (r) {
4096                         ti->error = "Error initializing superblock";
4097                         goto bad;
4098                 }
4099                 ic->just_formatted = true;
4100         }
4101
4102         if (!ic->meta_dev) {
4103                 r = dm_set_target_max_io_len(ti, 1U << ic->sb->log2_interleave_sectors);
4104                 if (r)
4105                         goto bad;
4106         }
4107         if (ic->mode == 'B') {
4108                 unsigned max_io_len = ((sector_t)ic->sectors_per_block << ic->log2_blocks_per_bitmap_bit) * (BITMAP_BLOCK_SIZE * 8);
4109                 if (!max_io_len)
4110                         max_io_len = 1U << 31;
4111                 DEBUG_print("max_io_len: old %u, new %u\n", ti->max_io_len, max_io_len);
4112                 if (!ti->max_io_len || ti->max_io_len > max_io_len) {
4113                         r = dm_set_target_max_io_len(ti, max_io_len);
4114                         if (r)
4115                                 goto bad;
4116                 }
4117         }
4118
4119         if (!ic->internal_hash)
4120                 dm_integrity_set(ti, ic);
4121
4122         ti->num_flush_bios = 1;
4123         ti->flush_supported = true;
4124
4125         return 0;
4126
4127 bad:
4128         dm_integrity_dtr(ti);
4129         return r;
4130 }
4131
4132 static void dm_integrity_dtr(struct dm_target *ti)
4133 {
4134         struct dm_integrity_c *ic = ti->private;
4135
4136         BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
4137         BUG_ON(!list_empty(&ic->wait_list));
4138
4139         if (ic->metadata_wq)
4140                 destroy_workqueue(ic->metadata_wq);
4141         if (ic->wait_wq)
4142                 destroy_workqueue(ic->wait_wq);
4143         if (ic->offload_wq)
4144                 destroy_workqueue(ic->offload_wq);
4145         if (ic->commit_wq)
4146                 destroy_workqueue(ic->commit_wq);
4147         if (ic->writer_wq)
4148                 destroy_workqueue(ic->writer_wq);
4149         if (ic->recalc_wq)
4150                 destroy_workqueue(ic->recalc_wq);
4151         vfree(ic->recalc_buffer);
4152         kvfree(ic->recalc_tags);
4153         kvfree(ic->bbs);
4154         if (ic->bufio)
4155                 dm_bufio_client_destroy(ic->bufio);
4156         mempool_exit(&ic->journal_io_mempool);
4157         if (ic->io)
4158                 dm_io_client_destroy(ic->io);
4159         if (ic->dev)
4160                 dm_put_device(ti, ic->dev);
4161         if (ic->meta_dev)
4162                 dm_put_device(ti, ic->meta_dev);
4163         dm_integrity_free_page_list(ic->journal);
4164         dm_integrity_free_page_list(ic->journal_io);
4165         dm_integrity_free_page_list(ic->journal_xor);
4166         dm_integrity_free_page_list(ic->recalc_bitmap);
4167         dm_integrity_free_page_list(ic->may_write_bitmap);
4168         if (ic->journal_scatterlist)
4169                 dm_integrity_free_journal_scatterlist(ic, ic->journal_scatterlist);
4170         if (ic->journal_io_scatterlist)
4171                 dm_integrity_free_journal_scatterlist(ic, ic->journal_io_scatterlist);
4172         if (ic->sk_requests) {
4173                 unsigned i;
4174
4175                 for (i = 0; i < ic->journal_sections; i++) {
4176                         struct skcipher_request *req = ic->sk_requests[i];
4177                         if (req) {
4178                                 kzfree(req->iv);
4179                                 skcipher_request_free(req);
4180                         }
4181                 }
4182                 kvfree(ic->sk_requests);
4183         }
4184         kvfree(ic->journal_tree);
4185         if (ic->sb)
4186                 free_pages_exact(ic->sb, SB_SECTORS << SECTOR_SHIFT);
4187
4188         if (ic->internal_hash)
4189                 crypto_free_shash(ic->internal_hash);
4190         free_alg(&ic->internal_hash_alg);
4191
4192         if (ic->journal_crypt)
4193                 crypto_free_skcipher(ic->journal_crypt);
4194         free_alg(&ic->journal_crypt_alg);
4195
4196         if (ic->journal_mac)
4197                 crypto_free_shash(ic->journal_mac);
4198         free_alg(&ic->journal_mac_alg);
4199
4200         kfree(ic);
4201 }
4202
4203 static struct target_type integrity_target = {
4204         .name                   = "integrity",
4205         .version                = {1, 5, 0},
4206         .module                 = THIS_MODULE,
4207         .features               = DM_TARGET_SINGLETON | DM_TARGET_INTEGRITY,
4208         .ctr                    = dm_integrity_ctr,
4209         .dtr                    = dm_integrity_dtr,
4210         .map                    = dm_integrity_map,
4211         .postsuspend            = dm_integrity_postsuspend,
4212         .resume                 = dm_integrity_resume,
4213         .status                 = dm_integrity_status,
4214         .iterate_devices        = dm_integrity_iterate_devices,
4215         .io_hints               = dm_integrity_io_hints,
4216 };
4217
4218 static int __init dm_integrity_init(void)
4219 {
4220         int r;
4221
4222         journal_io_cache = kmem_cache_create("integrity_journal_io",
4223                                              sizeof(struct journal_io), 0, 0, NULL);
4224         if (!journal_io_cache) {
4225                 DMERR("can't allocate journal io cache");
4226                 return -ENOMEM;
4227         }
4228
4229         r = dm_register_target(&integrity_target);
4230
4231         if (r < 0)
4232                 DMERR("register failed %d", r);
4233
4234         return r;
4235 }
4236
4237 static void __exit dm_integrity_exit(void)
4238 {
4239         dm_unregister_target(&integrity_target);
4240         kmem_cache_destroy(journal_io_cache);
4241 }
4242
4243 module_init(dm_integrity_init);
4244 module_exit(dm_integrity_exit);
4245
4246 MODULE_AUTHOR("Milan Broz");
4247 MODULE_AUTHOR("Mikulas Patocka");
4248 MODULE_DESCRIPTION(DM_NAME " target for integrity tags extension");
4249 MODULE_LICENSE("GPL");