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_SECTORS                      8
  59 #define MAX_SECTORS_PER_BLOCK           8
  60
  61 struct superblock {
  62         __u8 magic[8];
  63         __u8 version;
  64         __u8 log2_interleave_sectors;
  65         __u16 integrity_tag_size;
  66         __u32 journal_sections;
  67         __u64 provided_data_sectors;    /* userspace uses this value */
  68         __u32 flags;
  69         __u8 log2_sectors_per_block;
  70         __u8 log2_blocks_per_bitmap_bit;
  71         __u8 pad[2];
  72         __u64 recalc_sector;
  73 };
  74
  75 #define SB_FLAG_HAVE_JOURNAL_MAC        0x1
  76 #define SB_FLAG_RECALCULATING           0x2
  77 #define SB_FLAG_DIRTY_BITMAP            0x4
  78
  79 #define JOURNAL_ENTRY_ROUNDUP           8
  80
  81 typedef __u64 commit_id_t;
  82 #define JOURNAL_MAC_PER_SECTOR          8
  83
  84 struct journal_entry {
  85         union {
  86                 struct {
  87                         __u32 sector_lo;
  88                         __u32 sector_hi;
  89                 } s;
  90                 __u64 sector;
  91         } u;
  92         commit_id_t last_bytes[0];
  93         /* __u8 tag[0]; */
  94 };
  95
  96 #define journal_entry_tag(ic, je)               ((__u8 *)&(je)->last_bytes[(ic)->sectors_per_block])
  97
  98 #if BITS_PER_LONG == 64
  99 #define journal_entry_set_sector(je, x)         do { smp_wmb(); WRITE_ONCE((je)->u.sector, cpu_to_le64(x)); } while (0)
 100 #else
 101 #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)
 102 #endif
 103 #define journal_entry_get_sector(je)            le64_to_cpu((je)->u.sector)
 104 #define journal_entry_is_unused(je)             ((je)->u.s.sector_hi == cpu_to_le32(-1))
 105 #define journal_entry_set_unused(je)            do { ((je)->u.s.sector_hi = cpu_to_le32(-1)); } while (0)
 106 #define journal_entry_is_inprogress(je)         ((je)->u.s.sector_hi == cpu_to_le32(-2))
 107 #define journal_entry_set_inprogress(je)        do { ((je)->u.s.sector_hi = cpu_to_le32(-2)); } while (0)
 108
 109 #define JOURNAL_BLOCK_SECTORS           8
 110 #define JOURNAL_SECTOR_DATA             ((1 << SECTOR_SHIFT) - sizeof(commit_id_t))
 111 #define JOURNAL_MAC_SIZE                (JOURNAL_MAC_PER_SECTOR * JOURNAL_BLOCK_SECTORS)
 112
 113 struct journal_sector {
 114         __u8 entries[JOURNAL_SECTOR_DATA - JOURNAL_MAC_PER_SECTOR];
 115         __u8 mac[JOURNAL_MAC_PER_SECTOR];
 116         commit_id_t commit_id;
 117 };
 118
 119 #define MAX_TAG_SIZE                    (JOURNAL_SECTOR_DATA - JOURNAL_MAC_PER_SECTOR - offsetof(struct journal_entry, last_bytes[MAX_SECTORS_PER_BLOCK]))
 120
 121 #define METADATA_PADDING_SECTORS        8
 122
 123 #define N_COMMIT_IDS                    4
 124
 125 static unsigned char prev_commit_seq(unsigned char seq)
 126 {
 127         return (seq + N_COMMIT_IDS - 1) % N_COMMIT_IDS;
 128 }
 129
 130 static unsigned char next_commit_seq(unsigned char seq)
 131 {
 132         return (seq + 1) % N_COMMIT_IDS;
 133 }
 134
 135 /*
 136  * In-memory structures
 137  */
 138
 139 struct journal_node {
 140         struct rb_node node;
 141         sector_t sector;
 142 };
 143
 144 struct alg_spec {
 145         char *alg_string;
 146         char *key_string;
 147         __u8 *key;
 148         unsigned key_size;
 149 };
 150
 151 struct dm_integrity_c {
 152         struct dm_dev *dev;
 153         struct dm_dev *meta_dev;
 154         unsigned tag_size;
 155         __s8 log2_tag_size;
 156         sector_t start;
 157         mempool_t journal_io_mempool;
 158         struct dm_io_client *io;
 159         struct dm_bufio_client *bufio;
 160         struct workqueue_struct *metadata_wq;
 161         struct superblock *sb;
 162         unsigned journal_pages;
 163         unsigned n_bitmap_blocks;
 164
 165         struct page_list *journal;
 166         struct page_list *journal_io;
 167         struct page_list *journal_xor;
 168         struct page_list *recalc_bitmap;
 169         struct page_list *may_write_bitmap;
 170         struct bitmap_block_status *bbs;
 171         unsigned bitmap_flush_interval;
 172         int synchronous_mode;
 173         struct bio_list synchronous_bios;
 174         struct delayed_work bitmap_flush_work;
 175
 176         struct crypto_skcipher *journal_crypt;
 177         struct scatterlist **journal_scatterlist;
 178         struct scatterlist **journal_io_scatterlist;
 179         struct skcipher_request **sk_requests;
 180
 181         struct crypto_shash *journal_mac;
 182
 183         struct journal_node *journal_tree;
 184         struct rb_root journal_tree_root;
 185
 186         sector_t provided_data_sectors;
 187
 188         unsigned short journal_entry_size;
 189         unsigned char journal_entries_per_sector;
 190         unsigned char journal_section_entries;
 191         unsigned short journal_section_sectors;
 192         unsigned journal_sections;
 193         unsigned journal_entries;
 194         sector_t data_device_sectors;
 195         sector_t meta_device_sectors;
 196         unsigned initial_sectors;
 197         unsigned metadata_run;
 198         __s8 log2_metadata_run;
 199         __u8 log2_buffer_sectors;
 200         __u8 sectors_per_block;
 201         __u8 log2_blocks_per_bitmap_bit;
 202
 203         unsigned char mode;
 204
 205         int failed;
 206
 207         struct crypto_shash *internal_hash;
 208
 209         struct dm_target *ti;
 210
 211         /* these variables are locked with endio_wait.lock */
 212         struct rb_root in_progress;
 213         struct list_head wait_list;
 214         wait_queue_head_t endio_wait;
 215         struct workqueue_struct *wait_wq;
 216         struct workqueue_struct *offload_wq;
 217
 218         unsigned char commit_seq;
 219         commit_id_t commit_ids[N_COMMIT_IDS];
 220
 221         unsigned committed_section;
 222         unsigned n_committed_sections;
 223
 224         unsigned uncommitted_section;
 225         unsigned n_uncommitted_sections;
 226
 227         unsigned free_section;
 228         unsigned char free_section_entry;
 229         unsigned free_sectors;
 230
 231         unsigned free_sectors_threshold;
 232
 233         struct workqueue_struct *commit_wq;
 234         struct work_struct commit_work;
 235
 236         struct workqueue_struct *writer_wq;
 237         struct work_struct writer_work;
 238
 239         struct workqueue_struct *recalc_wq;
 240         struct work_struct recalc_work;
 241         u8 *recalc_buffer;
 242         u8 *recalc_tags;
 243
 244         struct bio_list flush_bio_list;
 245
 246         unsigned long autocommit_jiffies;
 247         struct timer_list autocommit_timer;
 248         unsigned autocommit_msec;
 249
 250         wait_queue_head_t copy_to_journal_wait;
 251
 252         struct completion crypto_backoff;
 253
 254         bool journal_uptodate;
 255         bool just_formatted;
 256         bool recalculate_flag;
 257         bool legacy_recalculate;
 258
 259         struct alg_spec internal_hash_alg;
 260         struct alg_spec journal_crypt_alg;
 261         struct alg_spec journal_mac_alg;
 262
 263         atomic64_t number_of_mismatches;
 264
 265         struct notifier_block reboot_notifier;
 266 };
 267
 268 struct dm_integrity_range {
 269         sector_t logical_sector;
 270         sector_t n_sectors;
 271         bool waiting;
 272         union {
 273                 struct rb_node node;
 274                 struct {
 275                         struct task_struct *task;
 276                         struct list_head wait_entry;
 277                 };
 278         };
 279 };
 280
 281 struct dm_integrity_io {
 282         struct work_struct work;
 283
 284         struct dm_integrity_c *ic;
 285         bool write;
 286         bool fua;
 287
 288         struct dm_integrity_range range;
 289
 290         sector_t metadata_block;
 291         unsigned metadata_offset;
 292
 293         atomic_t in_flight;
 294         blk_status_t bi_status;
 295
 296         struct completion *completion;
 297
 298         struct dm_bio_details bio_details;
 299 };
 300
 301 struct journal_completion {
 302         struct dm_integrity_c *ic;
 303         atomic_t in_flight;
 304         struct completion comp;
 305 };
 306
 307 struct journal_io {
 308         struct dm_integrity_range range;
 309         struct journal_completion *comp;
 310 };
 311
 312 struct bitmap_block_status {
 313         struct work_struct work;
 314         struct dm_integrity_c *ic;
 315         unsigned idx;
 316         unsigned long *bitmap;
 317         struct bio_list bio_queue;
 318         spinlock_t bio_queue_lock;
 319
 320 };
 321
 322 static struct kmem_cache *journal_io_cache;
 323
 324 #define JOURNAL_IO_MEMPOOL      32
 325
 326 #ifdef DEBUG_PRINT
 327 #define DEBUG_print(x, ...)     printk(KERN_DEBUG x, ##__VA_ARGS__)
 328 static void __DEBUG_bytes(__u8 *bytes, size_t len, const char *msg, ...)
 329 {
 330         va_list args;
 331         va_start(args, msg);
 332         vprintk(msg, args);
 333         va_end(args);
 334         if (len)
 335                 pr_cont(":");
 336         while (len) {
 337                 pr_cont(" %02x", *bytes);
 338                 bytes++;
 339                 len--;
 340         }
 341         pr_cont("\n");
 342 }
 343 #define DEBUG_bytes(bytes, len, msg, ...)       __DEBUG_bytes(bytes, len, KERN_DEBUG msg, ##__VA_ARGS__)
 344 #else
 345 #define DEBUG_print(x, ...)                     do { } while (0)
 346 #define DEBUG_bytes(bytes, len, msg, ...)       do { } while (0)
 347 #endif
 348
 349 static void dm_integrity_prepare(struct request *rq)
 350 {
 351 }
 352
 353 static void dm_integrity_complete(struct request *rq, unsigned int nr_bytes)
 354 {
 355 }
 356
 357 /*
 358  * DM Integrity profile, protection is performed layer above (dm-crypt)
 359  */
 360 static const struct blk_integrity_profile dm_integrity_profile = {
 361         .name                   = "DM-DIF-EXT-TAG",
 362         .generate_fn            = NULL,
 363         .verify_fn              = NULL,
 364         .prepare_fn             = dm_integrity_prepare,
 365         .complete_fn            = dm_integrity_complete,
 366 };
 367
 368 static void dm_integrity_map_continue(struct dm_integrity_io *dio, bool from_map);
 369 static void integrity_bio_wait(struct work_struct *w);
 370 static void dm_integrity_dtr(struct dm_target *ti);
 371
 372 static void dm_integrity_io_error(struct dm_integrity_c *ic, const char *msg, int err)
 373 {
 374         if (err == -EILSEQ)
 375                 atomic64_inc(&ic->number_of_mismatches);
 376         if (!cmpxchg(&ic->failed, 0, err))
 377                 DMERR("Error on %s: %d", msg, err);
 378 }
 379
 380 static int dm_integrity_failed(struct dm_integrity_c *ic)
 381 {
 382         return READ_ONCE(ic->failed);
 383 }
 384
 385 static bool dm_integrity_disable_recalculate(struct dm_integrity_c *ic)
 386 {
 387         if ((ic->internal_hash_alg.key || ic->journal_mac_alg.key) &&
 388             !ic->legacy_recalculate)
 389                 return true;
 390         return false;
 391 }
 392
 393 static commit_id_t dm_integrity_commit_id(struct dm_integrity_c *ic, unsigned i,
 394                                           unsigned j, unsigned char seq)
 395 {
 396         /*
 397          * Xor the number with section and sector, so that if a piece of
 398          * journal is written at wrong place, it is detected.
 399          */
 400         return ic->commit_ids[seq] ^ cpu_to_le64(((__u64)i << 32) ^ j);
 401 }
 402
 403 static void get_area_and_offset(struct dm_integrity_c *ic, sector_t data_sector,
 404                                 sector_t *area, sector_t *offset)
 405 {
 406         if (!ic->meta_dev) {
 407                 __u8 log2_interleave_sectors = ic->sb->log2_interleave_sectors;
 408                 *area = data_sector >> log2_interleave_sectors;
 409                 *offset = (unsigned)data_sector & ((1U << log2_interleave_sectors) - 1);
 410         } else {
 411                 *area = 0;
 412                 *offset = data_sector;
 413         }
 414 }
 415
 416 #define sector_to_block(ic, n)                                          \
 417 do {                                                                    \
 418         BUG_ON((n) & (unsigned)((ic)->sectors_per_block - 1));          \
 419         (n) >>= (ic)->sb->log2_sectors_per_block;                       \
 420 } while (0)
 421
 422 static __u64 get_metadata_sector_and_offset(struct dm_integrity_c *ic, sector_t area,
 423                                             sector_t offset, unsigned *metadata_offset)
 424 {
 425         __u64 ms;
 426         unsigned mo;
 427
 428         ms = area << ic->sb->log2_interleave_sectors;
 429         if (likely(ic->log2_metadata_run >= 0))
 430                 ms += area << ic->log2_metadata_run;
 431         else
 432                 ms += area * ic->metadata_run;
 433         ms >>= ic->log2_buffer_sectors;
 434
 435         sector_to_block(ic, offset);
 436
 437         if (likely(ic->log2_tag_size >= 0)) {
 438                 ms += offset >> (SECTOR_SHIFT + ic->log2_buffer_sectors - ic->log2_tag_size);
 439                 mo = (offset << ic->log2_tag_size) & ((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - 1);
 440         } else {
 441                 ms += (__u64)offset * ic->tag_size >> (SECTOR_SHIFT + ic->log2_buffer_sectors);
 442                 mo = (offset * ic->tag_size) & ((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - 1);
 443         }
 444         *metadata_offset = mo;
 445         return ms;
 446 }
 447
 448 static sector_t get_data_sector(struct dm_integrity_c *ic, sector_t area, sector_t offset)
 449 {
 450         sector_t result;
 451
 452         if (ic->meta_dev)
 453                 return offset;
 454
 455         result = area << ic->sb->log2_interleave_sectors;
 456         if (likely(ic->log2_metadata_run >= 0))
 457                 result += (area + 1) << ic->log2_metadata_run;
 458         else
 459                 result += (area + 1) * ic->metadata_run;
 460
 461         result += (sector_t)ic->initial_sectors + offset;
 462         result += ic->start;
 463
 464         return result;
 465 }
 466
 467 static void wraparound_section(struct dm_integrity_c *ic, unsigned *sec_ptr)
 468 {
 469         if (unlikely(*sec_ptr >= ic->journal_sections))
 470                 *sec_ptr -= ic->journal_sections;
 471 }
 472
 473 static void sb_set_version(struct dm_integrity_c *ic)
 474 {
 475         if (ic->mode == 'B' || ic->sb->flags & cpu_to_le32(SB_FLAG_DIRTY_BITMAP))
 476                 ic->sb->version = SB_VERSION_3;
 477         else if (ic->meta_dev || ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
 478                 ic->sb->version = SB_VERSION_2;
 479         else
 480                 ic->sb->version = SB_VERSION_1;
 481 }
 482
 483 static int sync_rw_sb(struct dm_integrity_c *ic, int op, int op_flags)
 484 {
 485         struct dm_io_request io_req;
 486         struct dm_io_region io_loc;
 487
 488         io_req.bi_op = op;
 489         io_req.bi_op_flags = op_flags;
 490         io_req.mem.type = DM_IO_KMEM;
 491         io_req.mem.ptr.addr = ic->sb;
 492         io_req.notify.fn = NULL;
 493         io_req.client = ic->io;
 494         io_loc.bdev = ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev;
 495         io_loc.sector = ic->start;
 496         io_loc.count = SB_SECTORS;
 497
 498         if (op == REQ_OP_WRITE)
 499                 sb_set_version(ic);
 500
 501         return dm_io(&io_req, 1, &io_loc, NULL);
 502 }
 503
 504 #define BITMAP_OP_TEST_ALL_SET          0
 505 #define BITMAP_OP_TEST_ALL_CLEAR        1
 506 #define BITMAP_OP_SET                   2
 507 #define BITMAP_OP_CLEAR                 3
 508
 509 static bool block_bitmap_op(struct dm_integrity_c *ic, struct page_list *bitmap,
 510                             sector_t sector, sector_t n_sectors, int mode)
 511 {
 512         unsigned long bit, end_bit, this_end_bit, page, end_page;
 513         unsigned long *data;
 514
 515         if (unlikely(((sector | n_sectors) & ((1 << ic->sb->log2_sectors_per_block) - 1)) != 0)) {
 516                 DMCRIT("invalid bitmap access (%llx,%llx,%d,%d,%d)",
 517                         (unsigned long long)sector,
 518                         (unsigned long long)n_sectors,
 519                         ic->sb->log2_sectors_per_block,
 520                         ic->log2_blocks_per_bitmap_bit,
 521                         mode);
 522                 BUG();
 523         }
 524
 525         if (unlikely(!n_sectors))
 526                 return true;
 527
 528         bit = sector >> (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
 529         end_bit = (sector + n_sectors - 1) >>
 530                 (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
 531
 532         page = bit / (PAGE_SIZE * 8);
 533         bit %= PAGE_SIZE * 8;
 534
 535         end_page = end_bit / (PAGE_SIZE * 8);
 536         end_bit %= PAGE_SIZE * 8;
 537
 538 repeat:
 539         if (page < end_page) {
 540                 this_end_bit = PAGE_SIZE * 8 - 1;
 541         } else {
 542                 this_end_bit = end_bit;
 543         }
 544
 545         data = lowmem_page_address(bitmap[page].page);
 546
 547         if (mode == BITMAP_OP_TEST_ALL_SET) {
 548                 while (bit <= this_end_bit) {
 549                         if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
 550                                 do {
 551                                         if (data[bit / BITS_PER_LONG] != -1)
 552                                                 return false;
 553                                         bit += BITS_PER_LONG;
 554                                 } while (this_end_bit >= bit + BITS_PER_LONG - 1);
 555                                 continue;
 556                         }
 557                         if (!test_bit(bit, data))
 558                                 return false;
 559                         bit++;
 560                 }
 561         } else if (mode == BITMAP_OP_TEST_ALL_CLEAR) {
 562                 while (bit <= this_end_bit) {
 563                         if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
 564                                 do {
 565                                         if (data[bit / BITS_PER_LONG] != 0)
 566                                                 return false;
 567                                         bit += BITS_PER_LONG;
 568                                 } while (this_end_bit >= bit + BITS_PER_LONG - 1);
 569                                 continue;
 570                         }
 571                         if (test_bit(bit, data))
 572                                 return false;
 573                         bit++;
 574                 }
 575         } else if (mode == BITMAP_OP_SET) {
 576                 while (bit <= this_end_bit) {
 577                         if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
 578                                 do {
 579                                         data[bit / BITS_PER_LONG] = -1;
 580                                         bit += BITS_PER_LONG;
 581                                 } while (this_end_bit >= bit + BITS_PER_LONG - 1);
 582                                 continue;
 583                         }
 584                         __set_bit(bit, data);
 585                         bit++;
 586                 }
 587         } else if (mode == BITMAP_OP_CLEAR) {
 588                 if (!bit && this_end_bit == PAGE_SIZE * 8 - 1)
 589                         clear_page(data);
 590                 else while (bit <= this_end_bit) {
 591                         if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
 592                                 do {
 593                                         data[bit / BITS_PER_LONG] = 0;
 594                                         bit += BITS_PER_LONG;
 595                                 } while (this_end_bit >= bit + BITS_PER_LONG - 1);
 596                                 continue;
 597                         }
 598                         __clear_bit(bit, data);
 599                         bit++;
 600                 }
 601         } else {
 602                 BUG();
 603         }
 604
 605         if (unlikely(page < end_page)) {
 606                 bit = 0;
 607                 page++;
 608                 goto repeat;
 609         }
 610
 611         return true;
 612 }
 613
 614 static void block_bitmap_copy(struct dm_integrity_c *ic, struct page_list *dst, struct page_list *src)
 615 {
 616         unsigned n_bitmap_pages = DIV_ROUND_UP(ic->n_bitmap_blocks, PAGE_SIZE / BITMAP_BLOCK_SIZE);
 617         unsigned i;
 618
 619         for (i = 0; i < n_bitmap_pages; i++) {
 620                 unsigned long *dst_data = lowmem_page_address(dst[i].page);
 621                 unsigned long *src_data = lowmem_page_address(src[i].page);
 622                 copy_page(dst_data, src_data);
 623         }
 624 }
 625
 626 static struct bitmap_block_status *sector_to_bitmap_block(struct dm_integrity_c *ic, sector_t sector)
 627 {
 628         unsigned bit = sector >> (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
 629         unsigned bitmap_block = bit / (BITMAP_BLOCK_SIZE * 8);
 630
 631         BUG_ON(bitmap_block >= ic->n_bitmap_blocks);
 632         return &ic->bbs[bitmap_block];
 633 }
 634
 635 static void access_journal_check(struct dm_integrity_c *ic, unsigned section, unsigned offset,
 636                                  bool e, const char *function)
 637 {
 638 #if defined(CONFIG_DM_DEBUG) || defined(INTERNAL_VERIFY)
 639         unsigned limit = e ? ic->journal_section_entries : ic->journal_section_sectors;
 640
 641         if (unlikely(section >= ic->journal_sections) ||
 642             unlikely(offset >= limit)) {
 643                 DMCRIT("%s: invalid access at (%u,%u), limit (%u,%u)",
 644                        function, section, offset, ic->journal_sections, limit);
 645                 BUG();
 646         }
 647 #endif
 648 }
 649
 650 static void page_list_location(struct dm_integrity_c *ic, unsigned section, unsigned offset,
 651                                unsigned *pl_index, unsigned *pl_offset)
 652 {
 653         unsigned sector;
 654
 655         access_journal_check(ic, section, offset, false, "page_list_location");
 656
 657         sector = section * ic->journal_section_sectors + offset;
 658
 659         *pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
 660         *pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
 661 }
 662
 663 static struct journal_sector *access_page_list(struct dm_integrity_c *ic, struct page_list *pl,
 664                                                unsigned section, unsigned offset, unsigned *n_sectors)
 665 {
 666         unsigned pl_index, pl_offset;
 667         char *va;
 668
 669         page_list_location(ic, section, offset, &pl_index, &pl_offset);
 670
 671         if (n_sectors)
 672                 *n_sectors = (PAGE_SIZE - pl_offset) >> SECTOR_SHIFT;
 673
 674         va = lowmem_page_address(pl[pl_index].page);
 675
 676         return (struct journal_sector *)(va + pl_offset);
 677 }
 678
 679 static struct journal_sector *access_journal(struct dm_integrity_c *ic, unsigned section, unsigned offset)
 680 {
 681         return access_page_list(ic, ic->journal, section, offset, NULL);
 682 }
 683
 684 static struct journal_entry *access_journal_entry(struct dm_integrity_c *ic, unsigned section, unsigned n)
 685 {
 686         unsigned rel_sector, offset;
 687         struct journal_sector *js;
 688
 689         access_journal_check(ic, section, n, true, "access_journal_entry");
 690
 691         rel_sector = n % JOURNAL_BLOCK_SECTORS;
 692         offset = n / JOURNAL_BLOCK_SECTORS;
 693
 694         js = access_journal(ic, section, rel_sector);
 695         return (struct journal_entry *)((char *)js + offset * ic->journal_entry_size);
 696 }
 697
 698 static struct journal_sector *access_journal_data(struct dm_integrity_c *ic, unsigned section, unsigned n)
 699 {
 700         n <<= ic->sb->log2_sectors_per_block;
 701
 702         n += JOURNAL_BLOCK_SECTORS;
 703
 704         access_journal_check(ic, section, n, false, "access_journal_data");
 705
 706         return access_journal(ic, section, n);
 707 }
 708
 709 static void section_mac(struct dm_integrity_c *ic, unsigned section, __u8 result[JOURNAL_MAC_SIZE])
 710 {
 711         SHASH_DESC_ON_STACK(desc, ic->journal_mac);
 712         int r;
 713         unsigned j, size;
 714
 715         desc->tfm = ic->journal_mac;
 716
 717         r = crypto_shash_init(desc);
 718         if (unlikely(r)) {
 719                 dm_integrity_io_error(ic, "crypto_shash_init", r);
 720                 goto err;
 721         }
 722
 723         for (j = 0; j < ic->journal_section_entries; j++) {
 724                 struct journal_entry *je = access_journal_entry(ic, section, j);
 725                 r = crypto_shash_update(desc, (__u8 *)&je->u.sector, sizeof je->u.sector);
 726                 if (unlikely(r)) {
 727                         dm_integrity_io_error(ic, "crypto_shash_update", r);
 728                         goto err;
 729                 }
 730         }
 731
 732         size = crypto_shash_digestsize(ic->journal_mac);
 733
 734         if (likely(size <= JOURNAL_MAC_SIZE)) {
 735                 r = crypto_shash_final(desc, result);
 736                 if (unlikely(r)) {
 737                         dm_integrity_io_error(ic, "crypto_shash_final", r);
 738                         goto err;
 739                 }
 740                 memset(result + size, 0, JOURNAL_MAC_SIZE - size);
 741         } else {
 742                 __u8 digest[HASH_MAX_DIGESTSIZE];
 743
 744                 if (WARN_ON(size > sizeof(digest))) {
 745                         dm_integrity_io_error(ic, "digest_size", -EINVAL);
 746                         goto err;
 747                 }
 748                 r = crypto_shash_final(desc, digest);
 749                 if (unlikely(r)) {
 750                         dm_integrity_io_error(ic, "crypto_shash_final", r);
 751                         goto err;
 752                 }
 753                 memcpy(result, digest, JOURNAL_MAC_SIZE);
 754         }
 755
 756         return;
 757 err:
 758         memset(result, 0, JOURNAL_MAC_SIZE);
 759 }
 760
 761 static void rw_section_mac(struct dm_integrity_c *ic, unsigned section, bool wr)
 762 {
 763         __u8 result[JOURNAL_MAC_SIZE];
 764         unsigned j;
 765
 766         if (!ic->journal_mac)
 767                 return;
 768
 769         section_mac(ic, section, result);
 770
 771         for (j = 0; j < JOURNAL_BLOCK_SECTORS; j++) {
 772                 struct journal_sector *js = access_journal(ic, section, j);
 773
 774                 if (likely(wr))
 775                         memcpy(&js->mac, result + (j * JOURNAL_MAC_PER_SECTOR), JOURNAL_MAC_PER_SECTOR);
 776                 else {
 777                         if (memcmp(&js->mac, result + (j * JOURNAL_MAC_PER_SECTOR), JOURNAL_MAC_PER_SECTOR))
 778                                 dm_integrity_io_error(ic, "journal mac", -EILSEQ);
 779                 }
 780         }
 781 }
 782
 783 static void complete_journal_op(void *context)
 784 {
 785         struct journal_completion *comp = context;
 786         BUG_ON(!atomic_read(&comp->in_flight));
 787         if (likely(atomic_dec_and_test(&comp->in_flight)))
 788                 complete(&comp->comp);
 789 }
 790
 791 static void xor_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
 792                         unsigned n_sections, struct journal_completion *comp)
 793 {
 794         struct async_submit_ctl submit;
 795         size_t n_bytes = (size_t)(n_sections * ic->journal_section_sectors) << SECTOR_SHIFT;
 796         unsigned pl_index, pl_offset, section_index;
 797         struct page_list *source_pl, *target_pl;
 798
 799         if (likely(encrypt)) {
 800                 source_pl = ic->journal;
 801                 target_pl = ic->journal_io;
 802         } else {
 803                 source_pl = ic->journal_io;
 804                 target_pl = ic->journal;
 805         }
 806
 807         page_list_location(ic, section, 0, &pl_index, &pl_offset);
 808
 809         atomic_add(roundup(pl_offset + n_bytes, PAGE_SIZE) >> PAGE_SHIFT, &comp->in_flight);
 810
 811         init_async_submit(&submit, ASYNC_TX_XOR_ZERO_DST, NULL, complete_journal_op, comp, NULL);
 812
 813         section_index = pl_index;
 814
 815         do {
 816                 size_t this_step;
 817                 struct page *src_pages[2];
 818                 struct page *dst_page;
 819
 820                 while (unlikely(pl_index == section_index)) {
 821                         unsigned dummy;
 822                         if (likely(encrypt))
 823                                 rw_section_mac(ic, section, true);
 824                         section++;
 825                         n_sections--;
 826                         if (!n_sections)
 827                                 break;
 828                         page_list_location(ic, section, 0, &section_index, &dummy);
 829                 }
 830
 831                 this_step = min(n_bytes, (size_t)PAGE_SIZE - pl_offset);
 832                 dst_page = target_pl[pl_index].page;
 833                 src_pages[0] = source_pl[pl_index].page;
 834                 src_pages[1] = ic->journal_xor[pl_index].page;
 835
 836                 async_xor(dst_page, src_pages, pl_offset, 2, this_step, &submit);
 837
 838                 pl_index++;
 839                 pl_offset = 0;
 840                 n_bytes -= this_step;
 841         } while (n_bytes);
 842
 843         BUG_ON(n_sections);
 844
 845         async_tx_issue_pending_all();
 846 }
 847
 848 static void complete_journal_encrypt(struct crypto_async_request *req, int err)
 849 {
 850         struct journal_completion *comp = req->data;
 851         if (unlikely(err)) {
 852                 if (likely(err == -EINPROGRESS)) {
 853                         complete(&comp->ic->crypto_backoff);
 854                         return;
 855                 }
 856                 dm_integrity_io_error(comp->ic, "asynchronous encrypt", err);
 857         }
 858         complete_journal_op(comp);
 859 }
 860
 861 static bool do_crypt(bool encrypt, struct skcipher_request *req, struct journal_completion *comp)
 862 {
 863         int r;
 864         skcipher_request_set_callback(req, CRYPTO_TFM_REQ_MAY_BACKLOG,
 865                                       complete_journal_encrypt, comp);
 866         if (likely(encrypt))
 867                 r = crypto_skcipher_encrypt(req);
 868         else
 869                 r = crypto_skcipher_decrypt(req);
 870         if (likely(!r))
 871                 return false;
 872         if (likely(r == -EINPROGRESS))
 873                 return true;
 874         if (likely(r == -EBUSY)) {
 875                 wait_for_completion(&comp->ic->crypto_backoff);
 876                 reinit_completion(&comp->ic->crypto_backoff);
 877                 return true;
 878         }
 879         dm_integrity_io_error(comp->ic, "encrypt", r);
 880         return false;
 881 }
 882
 883 static void crypt_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
 884                           unsigned n_sections, struct journal_completion *comp)
 885 {
 886         struct scatterlist **source_sg;
 887         struct scatterlist **target_sg;
 888
 889         atomic_add(2, &comp->in_flight);
 890
 891         if (likely(encrypt)) {
 892                 source_sg = ic->journal_scatterlist;
 893                 target_sg = ic->journal_io_scatterlist;
 894         } else {
 895                 source_sg = ic->journal_io_scatterlist;
 896                 target_sg = ic->journal_scatterlist;
 897         }
 898
 899         do {
 900                 struct skcipher_request *req;
 901                 unsigned ivsize;
 902                 char *iv;
 903
 904                 if (likely(encrypt))
 905                         rw_section_mac(ic, section, true);
 906
 907                 req = ic->sk_requests[section];
 908                 ivsize = crypto_skcipher_ivsize(ic->journal_crypt);
 909                 iv = req->iv;
 910
 911                 memcpy(iv, iv + ivsize, ivsize);
 912
 913                 req->src = source_sg[section];
 914                 req->dst = target_sg[section];
 915
 916                 if (unlikely(do_crypt(encrypt, req, comp)))
 917                         atomic_inc(&comp->in_flight);
 918
 919                 section++;
 920                 n_sections--;
 921         } while (n_sections);
 922
 923         atomic_dec(&comp->in_flight);
 924         complete_journal_op(comp);
 925 }
 926
 927 static void encrypt_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
 928                             unsigned n_sections, struct journal_completion *comp)
 929 {
 930         if (ic->journal_xor)
 931                 return xor_journal(ic, encrypt, section, n_sections, comp);
 932         else
 933                 return crypt_journal(ic, encrypt, section, n_sections, comp);
 934 }
 935
 936 static void complete_journal_io(unsigned long error, void *context)
 937 {
 938         struct journal_completion *comp = context;
 939         if (unlikely(error != 0))
 940                 dm_integrity_io_error(comp->ic, "writing journal", -EIO);
 941         complete_journal_op(comp);
 942 }
 943
 944 static void rw_journal_sectors(struct dm_integrity_c *ic, int op, int op_flags,
 945                                unsigned sector, unsigned n_sectors, struct journal_completion *comp)
 946 {
 947         struct dm_io_request io_req;
 948         struct dm_io_region io_loc;
 949         unsigned pl_index, pl_offset;
 950         int r;
 951
 952         if (unlikely(dm_integrity_failed(ic))) {
 953                 if (comp)
 954                         complete_journal_io(-1UL, comp);
 955                 return;
 956         }
 957
 958         pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
 959         pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
 960
 961         io_req.bi_op = op;
 962         io_req.bi_op_flags = op_flags;
 963         io_req.mem.type = DM_IO_PAGE_LIST;
 964         if (ic->journal_io)
 965                 io_req.mem.ptr.pl = &ic->journal_io[pl_index];
 966         else
 967                 io_req.mem.ptr.pl = &ic->journal[pl_index];
 968         io_req.mem.offset = pl_offset;
 969         if (likely(comp != NULL)) {
 970                 io_req.notify.fn = complete_journal_io;
 971                 io_req.notify.context = comp;
 972         } else {
 973                 io_req.notify.fn = NULL;
 974         }
 975         io_req.client = ic->io;
 976         io_loc.bdev = ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev;
 977         io_loc.sector = ic->start + SB_SECTORS + sector;
 978         io_loc.count = n_sectors;
 979
 980         r = dm_io(&io_req, 1, &io_loc, NULL);
 981         if (unlikely(r)) {
 982                 dm_integrity_io_error(ic, op == REQ_OP_READ ? "reading journal" : "writing journal", r);
 983                 if (comp) {
 984                         WARN_ONCE(1, "asynchronous dm_io failed: %d", r);
 985                         complete_journal_io(-1UL, comp);
 986                 }
 987         }
 988 }
 989
 990 static void rw_journal(struct dm_integrity_c *ic, int op, int op_flags, unsigned section,
 991                        unsigned n_sections, struct journal_completion *comp)
 992 {
 993         unsigned sector, n_sectors;
 994
 995         sector = section * ic->journal_section_sectors;
 996         n_sectors = n_sections * ic->journal_section_sectors;
 997
 998         rw_journal_sectors(ic, op, op_flags, sector, n_sectors, comp);
 999 }
1000
1001 static void write_journal(struct dm_integrity_c *ic, unsigned commit_start, unsigned commit_sections)
1002 {
1003         struct journal_completion io_comp;
1004         struct journal_completion crypt_comp_1;
1005         struct journal_completion crypt_comp_2;
1006         unsigned i;
1007
1008         io_comp.ic = ic;
1009         init_completion(&io_comp.comp);
1010
1011         if (commit_start + commit_sections <= ic->journal_sections) {
1012                 io_comp.in_flight = (atomic_t)ATOMIC_INIT(1);
1013                 if (ic->journal_io) {
1014                         crypt_comp_1.ic = ic;
1015                         init_completion(&crypt_comp_1.comp);
1016                         crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
1017                         encrypt_journal(ic, true, commit_start, commit_sections, &crypt_comp_1);
1018                         wait_for_completion_io(&crypt_comp_1.comp);
1019                 } else {
1020                         for (i = 0; i < commit_sections; i++)
1021                                 rw_section_mac(ic, commit_start + i, true);
1022                 }
1023                 rw_journal(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, commit_start,
1024                            commit_sections, &io_comp);
1025         } else {
1026                 unsigned to_end;
1027                 io_comp.in_flight = (atomic_t)ATOMIC_INIT(2);
1028                 to_end = ic->journal_sections - commit_start;
1029                 if (ic->journal_io) {
1030                         crypt_comp_1.ic = ic;
1031                         init_completion(&crypt_comp_1.comp);
1032                         crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
1033                         encrypt_journal(ic, true, commit_start, to_end, &crypt_comp_1);
1034                         if (try_wait_for_completion(&crypt_comp_1.comp)) {
1035                                 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
1036                                 reinit_completion(&crypt_comp_1.comp);
1037                                 crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
1038                                 encrypt_journal(ic, true, 0, commit_sections - to_end, &crypt_comp_1);
1039                                 wait_for_completion_io(&crypt_comp_1.comp);
1040                         } else {
1041                                 crypt_comp_2.ic = ic;
1042                                 init_completion(&crypt_comp_2.comp);
1043                                 crypt_comp_2.in_flight = (atomic_t)ATOMIC_INIT(0);
1044                                 encrypt_journal(ic, true, 0, commit_sections - to_end, &crypt_comp_2);
1045                                 wait_for_completion_io(&crypt_comp_1.comp);
1046                                 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
1047                                 wait_for_completion_io(&crypt_comp_2.comp);
1048                         }
1049                 } else {
1050                         for (i = 0; i < to_end; i++)
1051                                 rw_section_mac(ic, commit_start + i, true);
1052                         rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
1053                         for (i = 0; i < commit_sections - to_end; i++)
1054                                 rw_section_mac(ic, i, true);
1055                 }
1056                 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, 0, commit_sections - to_end, &io_comp);
1057         }
1058
1059         wait_for_completion_io(&io_comp.comp);
1060 }
1061
1062 static void copy_from_journal(struct dm_integrity_c *ic, unsigned section, unsigned offset,
1063                               unsigned n_sectors, sector_t target, io_notify_fn fn, void *data)
1064 {
1065         struct dm_io_request io_req;
1066         struct dm_io_region io_loc;
1067         int r;
1068         unsigned sector, pl_index, pl_offset;
1069
1070         BUG_ON((target | n_sectors | offset) & (unsigned)(ic->sectors_per_block - 1));
1071
1072         if (unlikely(dm_integrity_failed(ic))) {
1073                 fn(-1UL, data);
1074                 return;
1075         }
1076
1077         sector = section * ic->journal_section_sectors + JOURNAL_BLOCK_SECTORS + offset;
1078
1079         pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
1080         pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
1081
1082         io_req.bi_op = REQ_OP_WRITE;
1083         io_req.bi_op_flags = 0;
1084         io_req.mem.type = DM_IO_PAGE_LIST;
1085         io_req.mem.ptr.pl = &ic->journal[pl_index];
1086         io_req.mem.offset = pl_offset;
1087         io_req.notify.fn = fn;
1088         io_req.notify.context = data;
1089         io_req.client = ic->io;
1090         io_loc.bdev = ic->dev->bdev;
1091         io_loc.sector = target;
1092         io_loc.count = n_sectors;
1093
1094         r = dm_io(&io_req, 1, &io_loc, NULL);
1095         if (unlikely(r)) {
1096                 WARN_ONCE(1, "asynchronous dm_io failed: %d", r);
1097                 fn(-1UL, data);
1098         }
1099 }
1100
1101 static bool ranges_overlap(struct dm_integrity_range *range1, struct dm_integrity_range *range2)
1102 {
1103         return range1->logical_sector < range2->logical_sector + range2->n_sectors &&
1104                range1->logical_sector + range1->n_sectors > range2->logical_sector;
1105 }
1106
1107 static bool add_new_range(struct dm_integrity_c *ic, struct dm_integrity_range *new_range, bool check_waiting)
1108 {
1109         struct rb_node **n = &ic->in_progress.rb_node;
1110         struct rb_node *parent;
1111
1112         BUG_ON((new_range->logical_sector | new_range->n_sectors) & (unsigned)(ic->sectors_per_block - 1));
1113
1114         if (likely(check_waiting)) {
1115                 struct dm_integrity_range *range;
1116                 list_for_each_entry(range, &ic->wait_list, wait_entry) {
1117                         if (unlikely(ranges_overlap(range, new_range)))
1118                                 return false;
1119                 }
1120         }
1121
1122         parent = NULL;
1123
1124         while (*n) {
1125                 struct dm_integrity_range *range = container_of(*n, struct dm_integrity_range, node);
1126
1127                 parent = *n;
1128                 if (new_range->logical_sector + new_range->n_sectors <= range->logical_sector) {
1129                         n = &range->node.rb_left;
1130                 } else if (new_range->logical_sector >= range->logical_sector + range->n_sectors) {
1131                         n = &range->node.rb_right;
1132                 } else {
1133                         return false;
1134                 }
1135         }
1136
1137         rb_link_node(&new_range->node, parent, n);
1138         rb_insert_color(&new_range->node, &ic->in_progress);
1139
1140         return true;
1141 }
1142
1143 static void remove_range_unlocked(struct dm_integrity_c *ic, struct dm_integrity_range *range)
1144 {
1145         rb_erase(&range->node, &ic->in_progress);
1146         while (unlikely(!list_empty(&ic->wait_list))) {
1147                 struct dm_integrity_range *last_range =
1148                         list_first_entry(&ic->wait_list, struct dm_integrity_range, wait_entry);
1149                 struct task_struct *last_range_task;
1150                 last_range_task = last_range->task;
1151                 list_del(&last_range->wait_entry);
1152                 if (!add_new_range(ic, last_range, false)) {
1153                         last_range->task = last_range_task;
1154                         list_add(&last_range->wait_entry, &ic->wait_list);
1155                         break;
1156                 }
1157                 last_range->waiting = false;
1158                 wake_up_process(last_range_task);
1159         }
1160 }
1161
1162 static void remove_range(struct dm_integrity_c *ic, struct dm_integrity_range *range)
1163 {
1164         unsigned long flags;
1165
1166         spin_lock_irqsave(&ic->endio_wait.lock, flags);
1167         remove_range_unlocked(ic, range);
1168         spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
1169 }
1170
1171 static void wait_and_add_new_range(struct dm_integrity_c *ic, struct dm_integrity_range *new_range)
1172 {
1173         new_range->waiting = true;
1174         list_add_tail(&new_range->wait_entry, &ic->wait_list);
1175         new_range->task = current;
1176         do {
1177                 __set_current_state(TASK_UNINTERRUPTIBLE);
1178                 spin_unlock_irq(&ic->endio_wait.lock);
1179                 io_schedule();
1180                 spin_lock_irq(&ic->endio_wait.lock);
1181         } while (unlikely(new_range->waiting));
1182 }
1183
1184 static void add_new_range_and_wait(struct dm_integrity_c *ic, struct dm_integrity_range *new_range)
1185 {
1186         if (unlikely(!add_new_range(ic, new_range, true)))
1187                 wait_and_add_new_range(ic, new_range);
1188 }
1189
1190 static void init_journal_node(struct journal_node *node)
1191 {
1192         RB_CLEAR_NODE(&node->node);
1193         node->sector = (sector_t)-1;
1194 }
1195
1196 static void add_journal_node(struct dm_integrity_c *ic, struct journal_node *node, sector_t sector)
1197 {
1198         struct rb_node **link;
1199         struct rb_node *parent;
1200
1201         node->sector = sector;
1202         BUG_ON(!RB_EMPTY_NODE(&node->node));
1203
1204         link = &ic->journal_tree_root.rb_node;
1205         parent = NULL;
1206
1207         while (*link) {
1208                 struct journal_node *j;
1209                 parent = *link;
1210                 j = container_of(parent, struct journal_node, node);
1211                 if (sector < j->sector)
1212                         link = &j->node.rb_left;
1213                 else
1214                         link = &j->node.rb_right;
1215         }
1216
1217         rb_link_node(&node->node, parent, link);
1218         rb_insert_color(&node->node, &ic->journal_tree_root);
1219 }
1220
1221 static void remove_journal_node(struct dm_integrity_c *ic, struct journal_node *node)
1222 {
1223         BUG_ON(RB_EMPTY_NODE(&node->node));
1224         rb_erase(&node->node, &ic->journal_tree_root);
1225         init_journal_node(node);
1226 }
1227
1228 #define NOT_FOUND       (-1U)
1229
1230 static unsigned find_journal_node(struct dm_integrity_c *ic, sector_t sector, sector_t *next_sector)
1231 {
1232         struct rb_node *n = ic->journal_tree_root.rb_node;
1233         unsigned found = NOT_FOUND;
1234         *next_sector = (sector_t)-1;
1235         while (n) {
1236                 struct journal_node *j = container_of(n, struct journal_node, node);
1237                 if (sector == j->sector) {
1238                         found = j - ic->journal_tree;
1239                 }
1240                 if (sector < j->sector) {
1241                         *next_sector = j->sector;
1242                         n = j->node.rb_left;
1243                 } else {
1244                         n = j->node.rb_right;
1245                 }
1246         }
1247
1248         return found;
1249 }
1250
1251 static bool test_journal_node(struct dm_integrity_c *ic, unsigned pos, sector_t sector)
1252 {
1253         struct journal_node *node, *next_node;
1254         struct rb_node *next;
1255
1256         if (unlikely(pos >= ic->journal_entries))
1257                 return false;
1258         node = &ic->journal_tree[pos];
1259         if (unlikely(RB_EMPTY_NODE(&node->node)))
1260                 return false;
1261         if (unlikely(node->sector != sector))
1262                 return false;
1263
1264         next = rb_next(&node->node);
1265         if (unlikely(!next))
1266                 return true;
1267
1268         next_node = container_of(next, struct journal_node, node);
1269         return next_node->sector != sector;
1270 }
1271
1272 static bool find_newer_committed_node(struct dm_integrity_c *ic, struct journal_node *node)
1273 {
1274         struct rb_node *next;
1275         struct journal_node *next_node;
1276         unsigned next_section;
1277
1278         BUG_ON(RB_EMPTY_NODE(&node->node));
1279
1280         next = rb_next(&node->node);
1281         if (unlikely(!next))
1282                 return false;
1283
1284         next_node = container_of(next, struct journal_node, node);
1285
1286         if (next_node->sector != node->sector)
1287                 return false;
1288
1289         next_section = (unsigned)(next_node - ic->journal_tree) / ic->journal_section_entries;
1290         if (next_section >= ic->committed_section &&
1291             next_section < ic->committed_section + ic->n_committed_sections)
1292                 return true;
1293         if (next_section + ic->journal_sections < ic->committed_section + ic->n_committed_sections)
1294                 return true;
1295
1296         return false;
1297 }
1298
1299 #define TAG_READ        0
1300 #define TAG_WRITE       1
1301 #define TAG_CMP         2
1302
1303 static int dm_integrity_rw_tag(struct dm_integrity_c *ic, unsigned char *tag, sector_t *metadata_block,
1304                                unsigned *metadata_offset, unsigned total_size, int op)
1305 {
1306         do {
1307                 unsigned char *data, *dp;
1308                 struct dm_buffer *b;
1309                 unsigned to_copy;
1310                 int r;
1311
1312                 r = dm_integrity_failed(ic);
1313                 if (unlikely(r))
1314                         return r;
1315
1316                 data = dm_bufio_read(ic->bufio, *metadata_block, &b);
1317                 if (IS_ERR(data))
1318                         return PTR_ERR(data);
1319
1320                 to_copy = min((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - *metadata_offset, total_size);
1321                 dp = data + *metadata_offset;
1322                 if (op == TAG_READ) {
1323                         memcpy(tag, dp, to_copy);
1324                 } else if (op == TAG_WRITE) {
1325                         memcpy(dp, tag, to_copy);
1326                         dm_bufio_mark_partial_buffer_dirty(b, *metadata_offset, *metadata_offset + to_copy);
1327                 } else  {
1328                         /* e.g.: op == TAG_CMP */
1329                         if (unlikely(memcmp(dp, tag, to_copy))) {
1330                                 unsigned i;
1331
1332                                 for (i = 0; i < to_copy; i++) {
1333                                         if (dp[i] != tag[i])
1334                                                 break;
1335                                         total_size--;
1336                                 }
1337                                 dm_bufio_release(b);
1338                                 return total_size;
1339                         }
1340                 }
1341                 dm_bufio_release(b);
1342
1343                 tag += to_copy;
1344                 *metadata_offset += to_copy;
1345                 if (unlikely(*metadata_offset == 1U << SECTOR_SHIFT << ic->log2_buffer_sectors)) {
1346                         (*metadata_block)++;
1347                         *metadata_offset = 0;
1348                 }
1349                 total_size -= to_copy;
1350         } while (unlikely(total_size));
1351
1352         return 0;
1353 }
1354
1355 struct flush_request {
1356         struct dm_io_request io_req;
1357         struct dm_io_region io_reg;
1358         struct dm_integrity_c *ic;
1359         struct completion comp;
1360 };
1361
1362 static void flush_notify(unsigned long error, void *fr_)
1363 {
1364         struct flush_request *fr = fr_;
1365         if (unlikely(error != 0))
1366                 dm_integrity_io_error(fr->ic, "flusing disk cache", -EIO);
1367         complete(&fr->comp);
1368 }
1369
1370 static void dm_integrity_flush_buffers(struct dm_integrity_c *ic, bool flush_data)
1371 {
1372         int r;
1373
1374         struct flush_request fr;
1375
1376         if (!ic->meta_dev)
1377                 flush_data = false;
1378         if (flush_data) {
1379                 fr.io_req.bi_op = REQ_OP_WRITE,
1380                 fr.io_req.bi_op_flags = REQ_PREFLUSH | REQ_SYNC,
1381                 fr.io_req.mem.type = DM_IO_KMEM,
1382                 fr.io_req.mem.ptr.addr = NULL,
1383                 fr.io_req.notify.fn = flush_notify,
1384                 fr.io_req.notify.context = &fr;
1385                 fr.io_req.client = dm_bufio_get_dm_io_client(ic->bufio),
1386                 fr.io_reg.bdev = ic->dev->bdev,
1387                 fr.io_reg.sector = 0,
1388                 fr.io_reg.count = 0,
1389                 fr.ic = ic;
1390                 init_completion(&fr.comp);
1391                 r = dm_io(&fr.io_req, 1, &fr.io_reg, NULL);
1392                 BUG_ON(r);
1393         }
1394
1395         r = dm_bufio_write_dirty_buffers(ic->bufio);
1396         if (unlikely(r))
1397                 dm_integrity_io_error(ic, "writing tags", r);
1398
1399         if (flush_data)
1400                 wait_for_completion(&fr.comp);
1401 }
1402
1403 static void sleep_on_endio_wait(struct dm_integrity_c *ic)
1404 {
1405         DECLARE_WAITQUEUE(wait, current);
1406         __add_wait_queue(&ic->endio_wait, &wait);
1407         __set_current_state(TASK_UNINTERRUPTIBLE);
1408         spin_unlock_irq(&ic->endio_wait.lock);
1409         io_schedule();
1410         spin_lock_irq(&ic->endio_wait.lock);
1411         __remove_wait_queue(&ic->endio_wait, &wait);
1412 }
1413
1414 static void autocommit_fn(struct timer_list *t)
1415 {
1416         struct dm_integrity_c *ic = from_timer(ic, t, autocommit_timer);
1417
1418         if (likely(!dm_integrity_failed(ic)))
1419                 queue_work(ic->commit_wq, &ic->commit_work);
1420 }
1421
1422 static void schedule_autocommit(struct dm_integrity_c *ic)
1423 {
1424         if (!timer_pending(&ic->autocommit_timer))
1425                 mod_timer(&ic->autocommit_timer, jiffies + ic->autocommit_jiffies);
1426 }
1427
1428 static void submit_flush_bio(struct dm_integrity_c *ic, struct dm_integrity_io *dio)
1429 {
1430         struct bio *bio;
1431         unsigned long flags;
1432
1433         spin_lock_irqsave(&ic->endio_wait.lock, flags);
1434         bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1435         bio_list_add(&ic->flush_bio_list, bio);
1436         spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
1437
1438         queue_work(ic->commit_wq, &ic->commit_work);
1439 }
1440
1441 static void do_endio(struct dm_integrity_c *ic, struct bio *bio)
1442 {
1443         int r = dm_integrity_failed(ic);
1444         if (unlikely(r) && !bio->bi_status)
1445                 bio->bi_status = errno_to_blk_status(r);
1446         if (unlikely(ic->synchronous_mode) && bio_op(bio) == REQ_OP_WRITE) {
1447                 unsigned long flags;
1448                 spin_lock_irqsave(&ic->endio_wait.lock, flags);
1449                 bio_list_add(&ic->synchronous_bios, bio);
1450                 queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, 0);
1451                 spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
1452                 return;
1453         }
1454         bio_endio(bio);
1455 }
1456
1457 static void do_endio_flush(struct dm_integrity_c *ic, struct dm_integrity_io *dio)
1458 {
1459         struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1460
1461         if (unlikely(dio->fua) && likely(!bio->bi_status) && likely(!dm_integrity_failed(ic)))
1462                 submit_flush_bio(ic, dio);
1463         else
1464                 do_endio(ic, bio);
1465 }
1466
1467 static void dec_in_flight(struct dm_integrity_io *dio)
1468 {
1469         if (atomic_dec_and_test(&dio->in_flight)) {
1470                 struct dm_integrity_c *ic = dio->ic;
1471                 struct bio *bio;
1472
1473                 remove_range(ic, &dio->range);
1474
1475                 if (unlikely(dio->write))
1476                         schedule_autocommit(ic);
1477
1478                 bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1479
1480                 if (unlikely(dio->bi_status) && !bio->bi_status)
1481                         bio->bi_status = dio->bi_status;
1482                 if (likely(!bio->bi_status) && unlikely(bio_sectors(bio) != dio->range.n_sectors)) {
1483                         dio->range.logical_sector += dio->range.n_sectors;
1484                         bio_advance(bio, dio->range.n_sectors << SECTOR_SHIFT);
1485                         INIT_WORK(&dio->work, integrity_bio_wait);
1486                         queue_work(ic->offload_wq, &dio->work);
1487                         return;
1488                 }
1489                 do_endio_flush(ic, dio);
1490         }
1491 }
1492
1493 static void integrity_end_io(struct bio *bio)
1494 {
1495         struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
1496
1497         dm_bio_restore(&dio->bio_details, bio);
1498         if (bio->bi_integrity)
1499                 bio->bi_opf |= REQ_INTEGRITY;
1500
1501         if (dio->completion)
1502                 complete(dio->completion);
1503
1504         dec_in_flight(dio);
1505 }
1506
1507 static void integrity_sector_checksum(struct dm_integrity_c *ic, sector_t sector,
1508                                       const char *data, char *result)
1509 {
1510         __u64 sector_le = cpu_to_le64(sector);
1511         SHASH_DESC_ON_STACK(req, ic->internal_hash);
1512         int r;
1513         unsigned digest_size;
1514
1515         req->tfm = ic->internal_hash;
1516
1517         r = crypto_shash_init(req);
1518         if (unlikely(r < 0)) {
1519                 dm_integrity_io_error(ic, "crypto_shash_init", r);
1520                 goto failed;
1521         }
1522
1523         r = crypto_shash_update(req, (const __u8 *)&sector_le, sizeof sector_le);
1524         if (unlikely(r < 0)) {
1525                 dm_integrity_io_error(ic, "crypto_shash_update", r);
1526                 goto failed;
1527         }
1528
1529         r = crypto_shash_update(req, data, ic->sectors_per_block << SECTOR_SHIFT);
1530         if (unlikely(r < 0)) {
1531                 dm_integrity_io_error(ic, "crypto_shash_update", r);
1532                 goto failed;
1533         }
1534
1535         r = crypto_shash_final(req, result);
1536         if (unlikely(r < 0)) {
1537                 dm_integrity_io_error(ic, "crypto_shash_final", r);
1538                 goto failed;
1539         }
1540
1541         digest_size = crypto_shash_digestsize(ic->internal_hash);
1542         if (unlikely(digest_size < ic->tag_size))
1543                 memset(result + digest_size, 0, ic->tag_size - digest_size);
1544
1545         return;
1546
1547 failed:
1548         /* this shouldn't happen anyway, the hash functions have no reason to fail */
1549         get_random_bytes(result, ic->tag_size);
1550 }
1551
1552 static void integrity_metadata(struct work_struct *w)
1553 {
1554         struct dm_integrity_io *dio = container_of(w, struct dm_integrity_io, work);
1555         struct dm_integrity_c *ic = dio->ic;
1556
1557         int r;
1558
1559         if (ic->internal_hash) {
1560                 struct bvec_iter iter;
1561                 struct bio_vec bv;
1562                 unsigned digest_size = crypto_shash_digestsize(ic->internal_hash);
1563                 struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1564                 char *checksums;
1565                 unsigned extra_space = unlikely(digest_size > ic->tag_size) ? digest_size - ic->tag_size : 0;
1566                 char checksums_onstack[max((size_t)HASH_MAX_DIGESTSIZE, MAX_TAG_SIZE)];
1567                 unsigned sectors_to_process = dio->range.n_sectors;
1568                 sector_t sector = dio->range.logical_sector;
1569
1570                 if (unlikely(ic->mode == 'R'))
1571                         goto skip_io;
1572
1573                 checksums = kmalloc((PAGE_SIZE >> SECTOR_SHIFT >> ic->sb->log2_sectors_per_block) * ic->tag_size + extra_space,
1574                                     GFP_NOIO | __GFP_NORETRY | __GFP_NOWARN);
1575                 if (!checksums) {
1576                         checksums = checksums_onstack;
1577                         if (WARN_ON(extra_space &&
1578                                     digest_size > sizeof(checksums_onstack))) {
1579                                 r = -EINVAL;
1580                                 goto error;
1581                         }
1582                 }
1583
1584                 __bio_for_each_segment(bv, bio, iter, dio->bio_details.bi_iter) {
1585                         unsigned pos;
1586                         char *mem, *checksums_ptr;
1587
1588 again:
1589                         mem = (char *)kmap_atomic(bv.bv_page) + bv.bv_offset;
1590                         pos = 0;
1591                         checksums_ptr = checksums;
1592                         do {
1593                                 integrity_sector_checksum(ic, sector, mem + pos, checksums_ptr);
1594                                 checksums_ptr += ic->tag_size;
1595                                 sectors_to_process -= ic->sectors_per_block;
1596                                 pos += ic->sectors_per_block << SECTOR_SHIFT;
1597                                 sector += ic->sectors_per_block;
1598                         } while (pos < bv.bv_len && sectors_to_process && checksums != checksums_onstack);
1599                         kunmap_atomic(mem);
1600
1601                         r = dm_integrity_rw_tag(ic, checksums, &dio->metadata_block, &dio->metadata_offset,
1602                                                 checksums_ptr - checksums, !dio->write ? TAG_CMP : TAG_WRITE);
1603                         if (unlikely(r)) {
1604                                 if (r > 0) {
1605                                         DMERR_LIMIT("Checksum failed at sector 0x%llx",
1606                                                     (unsigned long long)(sector - ((r + ic->tag_size - 1) / ic->tag_size)));
1607                                         r = -EILSEQ;
1608                                         atomic64_inc(&ic->number_of_mismatches);
1609                                 }
1610                                 if (likely(checksums != checksums_onstack))
1611                                         kfree(checksums);
1612                                 goto error;
1613                         }
1614
1615                         if (!sectors_to_process)
1616                                 break;
1617
1618                         if (unlikely(pos < bv.bv_len)) {
1619                                 bv.bv_offset += pos;
1620                                 bv.bv_len -= pos;
1621                                 goto again;
1622                         }
1623                 }
1624
1625                 if (likely(checksums != checksums_onstack))
1626                         kfree(checksums);
1627         } else {
1628                 struct bio_integrity_payload *bip = dio->bio_details.bi_integrity;
1629
1630                 if (bip) {
1631                         struct bio_vec biv;
1632                         struct bvec_iter iter;
1633                         unsigned data_to_process = dio->range.n_sectors;
1634                         sector_to_block(ic, data_to_process);
1635                         data_to_process *= ic->tag_size;
1636
1637                         bip_for_each_vec(biv, bip, iter) {
1638                                 unsigned char *tag;
1639                                 unsigned this_len;
1640
1641                                 BUG_ON(PageHighMem(biv.bv_page));
1642                                 tag = lowmem_page_address(biv.bv_page) + biv.bv_offset;
1643                                 this_len = min(biv.bv_len, data_to_process);
1644                                 r = dm_integrity_rw_tag(ic, tag, &dio->metadata_block, &dio->metadata_offset,
1645                                                         this_len, !dio->write ? TAG_READ : TAG_WRITE);
1646                                 if (unlikely(r))
1647                                         goto error;
1648                                 data_to_process -= this_len;
1649                                 if (!data_to_process)
1650                                         break;
1651                         }
1652                 }
1653         }
1654 skip_io:
1655         dec_in_flight(dio);
1656         return;
1657 error:
1658         dio->bi_status = errno_to_blk_status(r);
1659         dec_in_flight(dio);
1660 }
1661
1662 static int dm_integrity_map(struct dm_target *ti, struct bio *bio)
1663 {
1664         struct dm_integrity_c *ic = ti->private;
1665         struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
1666         struct bio_integrity_payload *bip;
1667
1668         sector_t area, offset;
1669
1670         dio->ic = ic;
1671         dio->bi_status = 0;
1672
1673         if (unlikely(bio->bi_opf & REQ_PREFLUSH)) {
1674                 submit_flush_bio(ic, dio);
1675                 return DM_MAPIO_SUBMITTED;
1676         }
1677
1678         dio->range.logical_sector = dm_target_offset(ti, bio->bi_iter.bi_sector);
1679         dio->write = bio_op(bio) == REQ_OP_WRITE;
1680         dio->fua = dio->write && bio->bi_opf & REQ_FUA;
1681         if (unlikely(dio->fua)) {
1682                 /*
1683                  * Don't pass down the FUA flag because we have to flush
1684                  * disk cache anyway.
1685                  */
1686                 bio->bi_opf &= ~REQ_FUA;
1687         }
1688         if (unlikely(dio->range.logical_sector + bio_sectors(bio) > ic->provided_data_sectors)) {
1689                 DMERR("Too big sector number: 0x%llx + 0x%x > 0x%llx",
1690                       (unsigned long long)dio->range.logical_sector, bio_sectors(bio),
1691                       (unsigned long long)ic->provided_data_sectors);
1692                 return DM_MAPIO_KILL;
1693         }
1694         if (unlikely((dio->range.logical_sector | bio_sectors(bio)) & (unsigned)(ic->sectors_per_block - 1))) {
1695                 DMERR("Bio not aligned on %u sectors: 0x%llx, 0x%x",
1696                       ic->sectors_per_block,
1697                       (unsigned long long)dio->range.logical_sector, bio_sectors(bio));
1698                 return DM_MAPIO_KILL;
1699         }
1700
1701         if (ic->sectors_per_block > 1) {
1702                 struct bvec_iter iter;
1703                 struct bio_vec bv;
1704                 bio_for_each_segment(bv, bio, iter) {
1705                         if (unlikely(bv.bv_len & ((ic->sectors_per_block << SECTOR_SHIFT) - 1))) {
1706                                 DMERR("Bio vector (%u,%u) is not aligned on %u-sector boundary",
1707                                         bv.bv_offset, bv.bv_len, ic->sectors_per_block);
1708                                 return DM_MAPIO_KILL;
1709                         }
1710                 }
1711         }
1712
1713         bip = bio_integrity(bio);
1714         if (!ic->internal_hash) {
1715                 if (bip) {
1716                         unsigned wanted_tag_size = bio_sectors(bio) >> ic->sb->log2_sectors_per_block;
1717                         if (ic->log2_tag_size >= 0)
1718                                 wanted_tag_size <<= ic->log2_tag_size;
1719                         else
1720                                 wanted_tag_size *= ic->tag_size;
1721                         if (unlikely(wanted_tag_size != bip->bip_iter.bi_size)) {
1722                                 DMERR("Invalid integrity data size %u, expected %u",
1723                                       bip->bip_iter.bi_size, wanted_tag_size);
1724                                 return DM_MAPIO_KILL;
1725                         }
1726                 }
1727         } else {
1728                 if (unlikely(bip != NULL)) {
1729                         DMERR("Unexpected integrity data when using internal hash");
1730                         return DM_MAPIO_KILL;
1731                 }
1732         }
1733
1734         if (unlikely(ic->mode == 'R') && unlikely(dio->write))
1735                 return DM_MAPIO_KILL;
1736
1737         get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
1738         dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset, &dio->metadata_offset);
1739         bio->bi_iter.bi_sector = get_data_sector(ic, area, offset);
1740
1741         dm_integrity_map_continue(dio, true);
1742         return DM_MAPIO_SUBMITTED;
1743 }
1744
1745 static bool __journal_read_write(struct dm_integrity_io *dio, struct bio *bio,
1746                                  unsigned journal_section, unsigned journal_entry)
1747 {
1748         struct dm_integrity_c *ic = dio->ic;
1749         sector_t logical_sector;
1750         unsigned n_sectors;
1751
1752         logical_sector = dio->range.logical_sector;
1753         n_sectors = dio->range.n_sectors;
1754         do {
1755                 struct bio_vec bv = bio_iovec(bio);
1756                 char *mem;
1757
1758                 if (unlikely(bv.bv_len >> SECTOR_SHIFT > n_sectors))
1759                         bv.bv_len = n_sectors << SECTOR_SHIFT;
1760                 n_sectors -= bv.bv_len >> SECTOR_SHIFT;
1761                 bio_advance_iter(bio, &bio->bi_iter, bv.bv_len);
1762 retry_kmap:
1763                 mem = kmap_atomic(bv.bv_page);
1764                 if (likely(dio->write))
1765                         flush_dcache_page(bv.bv_page);
1766
1767                 do {
1768                         struct journal_entry *je = access_journal_entry(ic, journal_section, journal_entry);
1769
1770                         if (unlikely(!dio->write)) {
1771                                 struct journal_sector *js;
1772                                 char *mem_ptr;
1773                                 unsigned s;
1774
1775                                 if (unlikely(journal_entry_is_inprogress(je))) {
1776                                         flush_dcache_page(bv.bv_page);
1777                                         kunmap_atomic(mem);
1778
1779                                         __io_wait_event(ic->copy_to_journal_wait, !journal_entry_is_inprogress(je));
1780                                         goto retry_kmap;
1781                                 }
1782                                 smp_rmb();
1783                                 BUG_ON(journal_entry_get_sector(je) != logical_sector);
1784                                 js = access_journal_data(ic, journal_section, journal_entry);
1785                                 mem_ptr = mem + bv.bv_offset;
1786                                 s = 0;
1787                                 do {
1788                                         memcpy(mem_ptr, js, JOURNAL_SECTOR_DATA);
1789                                         *(commit_id_t *)(mem_ptr + JOURNAL_SECTOR_DATA) = je->last_bytes[s];
1790                                         js++;
1791                                         mem_ptr += 1 << SECTOR_SHIFT;
1792                                 } while (++s < ic->sectors_per_block);
1793 #ifdef INTERNAL_VERIFY
1794                                 if (ic->internal_hash) {
1795                                         char checksums_onstack[max((size_t)HASH_MAX_DIGESTSIZE, MAX_TAG_SIZE)];
1796
1797                                         integrity_sector_checksum(ic, logical_sector, mem + bv.bv_offset, checksums_onstack);
1798                                         if (unlikely(memcmp(checksums_onstack, journal_entry_tag(ic, je), ic->tag_size))) {
1799                                                 DMERR_LIMIT("Checksum failed when reading from journal, at sector 0x%llx",
1800                                                             (unsigned long long)logical_sector);
1801                                         }
1802                                 }
1803 #endif
1804                         }
1805
1806                         if (!ic->internal_hash) {
1807                                 struct bio_integrity_payload *bip = bio_integrity(bio);
1808                                 unsigned tag_todo = ic->tag_size;
1809                                 char *tag_ptr = journal_entry_tag(ic, je);
1810
1811                                 if (bip) do {
1812                                         struct bio_vec biv = bvec_iter_bvec(bip->bip_vec, bip->bip_iter);
1813                                         unsigned tag_now = min(biv.bv_len, tag_todo);
1814                                         char *tag_addr;
1815                                         BUG_ON(PageHighMem(biv.bv_page));
1816                                         tag_addr = lowmem_page_address(biv.bv_page) + biv.bv_offset;
1817                                         if (likely(dio->write))
1818                                                 memcpy(tag_ptr, tag_addr, tag_now);
1819                                         else
1820                                                 memcpy(tag_addr, tag_ptr, tag_now);
1821                                         bvec_iter_advance(bip->bip_vec, &bip->bip_iter, tag_now);
1822                                         tag_ptr += tag_now;
1823                                         tag_todo -= tag_now;
1824                                 } while (unlikely(tag_todo)); else {
1825                                         if (likely(dio->write))
1826                                                 memset(tag_ptr, 0, tag_todo);
1827                                 }
1828                         }
1829
1830                         if (likely(dio->write)) {
1831                                 struct journal_sector *js;
1832                                 unsigned s;
1833
1834                                 js = access_journal_data(ic, journal_section, journal_entry);
1835                                 memcpy(js, mem + bv.bv_offset, ic->sectors_per_block << SECTOR_SHIFT);
1836
1837                                 s = 0;
1838                                 do {
1839                                         je->last_bytes[s] = js[s].commit_id;
1840                                 } while (++s < ic->sectors_per_block);
1841
1842                                 if (ic->internal_hash) {
1843                                         unsigned digest_size = crypto_shash_digestsize(ic->internal_hash);
1844                                         if (unlikely(digest_size > ic->tag_size)) {
1845                                                 char checksums_onstack[HASH_MAX_DIGESTSIZE];
1846                                                 integrity_sector_checksum(ic, logical_sector, (char *)js, checksums_onstack);
1847                                                 memcpy(journal_entry_tag(ic, je), checksums_onstack, ic->tag_size);
1848                                         } else
1849                                                 integrity_sector_checksum(ic, logical_sector, (char *)js, journal_entry_tag(ic, je));
1850                                 }
1851
1852                                 journal_entry_set_sector(je, logical_sector);
1853                         }
1854                         logical_sector += ic->sectors_per_block;
1855
1856                         journal_entry++;
1857                         if (unlikely(journal_entry == ic->journal_section_entries)) {
1858                                 journal_entry = 0;
1859                                 journal_section++;
1860                                 wraparound_section(ic, &journal_section);
1861                         }
1862
1863                         bv.bv_offset += ic->sectors_per_block << SECTOR_SHIFT;
1864                 } while (bv.bv_len -= ic->sectors_per_block << SECTOR_SHIFT);
1865
1866                 if (unlikely(!dio->write))
1867                         flush_dcache_page(bv.bv_page);
1868                 kunmap_atomic(mem);
1869         } while (n_sectors);
1870
1871         if (likely(dio->write)) {
1872                 smp_mb();
1873                 if (unlikely(waitqueue_active(&ic->copy_to_journal_wait)))
1874                         wake_up(&ic->copy_to_journal_wait);
1875                 if (READ_ONCE(ic->free_sectors) <= ic->free_sectors_threshold) {
1876                         queue_work(ic->commit_wq, &ic->commit_work);
1877                 } else {
1878                         schedule_autocommit(ic);
1879                 }
1880         } else {
1881                 remove_range(ic, &dio->range);
1882         }
1883
1884         if (unlikely(bio->bi_iter.bi_size)) {
1885                 sector_t area, offset;
1886
1887                 dio->range.logical_sector = logical_sector;
1888                 get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
1889                 dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset, &dio->metadata_offset);
1890                 return true;
1891         }
1892
1893         return false;
1894 }
1895
1896 static void dm_integrity_map_continue(struct dm_integrity_io *dio, bool from_map)
1897 {
1898         struct dm_integrity_c *ic = dio->ic;
1899         struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1900         unsigned journal_section, journal_entry;
1901         unsigned journal_read_pos;
1902         struct completion read_comp;
1903         bool need_sync_io = ic->internal_hash && !dio->write;
1904
1905         if (need_sync_io && from_map) {
1906                 INIT_WORK(&dio->work, integrity_bio_wait);
1907                 queue_work(ic->offload_wq, &dio->work);
1908                 return;
1909         }
1910
1911 lock_retry:
1912         spin_lock_irq(&ic->endio_wait.lock);
1913 retry:
1914         if (unlikely(dm_integrity_failed(ic))) {
1915                 spin_unlock_irq(&ic->endio_wait.lock);
1916                 do_endio(ic, bio);
1917                 return;
1918         }
1919         dio->range.n_sectors = bio_sectors(bio);
1920         journal_read_pos = NOT_FOUND;
1921         if (likely(ic->mode == 'J')) {
1922                 if (dio->write) {
1923                         unsigned next_entry, i, pos;
1924                         unsigned ws, we, range_sectors;
1925
1926                         dio->range.n_sectors = min(dio->range.n_sectors,
1927                                                    (sector_t)ic->free_sectors << ic->sb->log2_sectors_per_block);
1928                         if (unlikely(!dio->range.n_sectors)) {
1929                                 if (from_map)
1930                                         goto offload_to_thread;
1931                                 sleep_on_endio_wait(ic);
1932                                 goto retry;
1933                         }
1934                         range_sectors = dio->range.n_sectors >> ic->sb->log2_sectors_per_block;
1935                         ic->free_sectors -= range_sectors;
1936                         journal_section = ic->free_section;
1937                         journal_entry = ic->free_section_entry;
1938
1939                         next_entry = ic->free_section_entry + range_sectors;
1940                         ic->free_section_entry = next_entry % ic->journal_section_entries;
1941                         ic->free_section += next_entry / ic->journal_section_entries;
1942                         ic->n_uncommitted_sections += next_entry / ic->journal_section_entries;
1943                         wraparound_section(ic, &ic->free_section);
1944
1945                         pos = journal_section * ic->journal_section_entries + journal_entry;
1946                         ws = journal_section;
1947                         we = journal_entry;
1948                         i = 0;
1949                         do {
1950                                 struct journal_entry *je;
1951
1952                                 add_journal_node(ic, &ic->journal_tree[pos], dio->range.logical_sector + i);
1953                                 pos++;
1954                                 if (unlikely(pos >= ic->journal_entries))
1955                                         pos = 0;
1956
1957                                 je = access_journal_entry(ic, ws, we);
1958                                 BUG_ON(!journal_entry_is_unused(je));
1959                                 journal_entry_set_inprogress(je);
1960                                 we++;
1961                                 if (unlikely(we == ic->journal_section_entries)) {
1962                                         we = 0;
1963                                         ws++;
1964                                         wraparound_section(ic, &ws);
1965                                 }
1966                         } while ((i += ic->sectors_per_block) < dio->range.n_sectors);
1967
1968                         spin_unlock_irq(&ic->endio_wait.lock);
1969                         goto journal_read_write;
1970                 } else {
1971                         sector_t next_sector;
1972                         journal_read_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
1973                         if (likely(journal_read_pos == NOT_FOUND)) {
1974                                 if (unlikely(dio->range.n_sectors > next_sector - dio->range.logical_sector))
1975                                         dio->range.n_sectors = next_sector - dio->range.logical_sector;
1976                         } else {
1977                                 unsigned i;
1978                                 unsigned jp = journal_read_pos + 1;
1979                                 for (i = ic->sectors_per_block; i < dio->range.n_sectors; i += ic->sectors_per_block, jp++) {
1980                                         if (!test_journal_node(ic, jp, dio->range.logical_sector + i))
1981                                                 break;
1982                                 }
1983                                 dio->range.n_sectors = i;
1984                         }
1985                 }
1986         }
1987         if (unlikely(!add_new_range(ic, &dio->range, true))) {
1988                 /*
1989                  * We must not sleep in the request routine because it could
1990                  * stall bios on current->bio_list.
1991                  * So, we offload the bio to a workqueue if we have to sleep.
1992                  */
1993                 if (from_map) {
1994 offload_to_thread:
1995                         spin_unlock_irq(&ic->endio_wait.lock);
1996                         INIT_WORK(&dio->work, integrity_bio_wait);
1997                         queue_work(ic->wait_wq, &dio->work);
1998                         return;
1999                 }
2000                 if (journal_read_pos != NOT_FOUND)
2001                         dio->range.n_sectors = ic->sectors_per_block;
2002                 wait_and_add_new_range(ic, &dio->range);
2003                 /*
2004                  * wait_and_add_new_range drops the spinlock, so the journal
2005                  * may have been changed arbitrarily. We need to recheck.
2006                  * To simplify the code, we restrict I/O size to just one block.
2007                  */
2008                 if (journal_read_pos != NOT_FOUND) {
2009                         sector_t next_sector;
2010                         unsigned new_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
2011                         if (unlikely(new_pos != journal_read_pos)) {
2012                                 remove_range_unlocked(ic, &dio->range);
2013                                 goto retry;
2014                         }
2015                 }
2016         }
2017         spin_unlock_irq(&ic->endio_wait.lock);
2018
2019         if (unlikely(journal_read_pos != NOT_FOUND)) {
2020                 journal_section = journal_read_pos / ic->journal_section_entries;
2021                 journal_entry = journal_read_pos % ic->journal_section_entries;
2022                 goto journal_read_write;
2023         }
2024
2025         if (ic->mode == 'B' && dio->write) {
2026                 if (!block_bitmap_op(ic, ic->may_write_bitmap, dio->range.logical_sector,
2027                                      dio->range.n_sectors, BITMAP_OP_TEST_ALL_SET)) {
2028                         struct bitmap_block_status *bbs;
2029
2030                         bbs = sector_to_bitmap_block(ic, dio->range.logical_sector);
2031                         spin_lock(&bbs->bio_queue_lock);
2032                         bio_list_add(&bbs->bio_queue, bio);
2033                         spin_unlock(&bbs->bio_queue_lock);
2034                         queue_work(ic->writer_wq, &bbs->work);
2035                         return;
2036                 }
2037         }
2038
2039         dio->in_flight = (atomic_t)ATOMIC_INIT(2);
2040
2041         if (need_sync_io) {
2042                 init_completion(&read_comp);
2043                 dio->completion = &read_comp;
2044         } else
2045                 dio->completion = NULL;
2046
2047         dm_bio_record(&dio->bio_details, bio);
2048         bio_set_dev(bio, ic->dev->bdev);
2049         bio->bi_integrity = NULL;
2050         bio->bi_opf &= ~REQ_INTEGRITY;
2051         bio->bi_end_io = integrity_end_io;
2052         bio->bi_iter.bi_size = dio->range.n_sectors << SECTOR_SHIFT;
2053
2054         generic_make_request(bio);
2055
2056         if (need_sync_io) {
2057                 wait_for_completion_io(&read_comp);
2058                 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING) &&
2059                     dio->range.logical_sector + dio->range.n_sectors > le64_to_cpu(ic->sb->recalc_sector))
2060                         goto skip_check;
2061                 if (ic->mode == 'B') {
2062                         if (!block_bitmap_op(ic, ic->recalc_bitmap, dio->range.logical_sector,
2063                                              dio->range.n_sectors, BITMAP_OP_TEST_ALL_CLEAR))
2064                                 goto skip_check;
2065                 }
2066
2067                 if (likely(!bio->bi_status))
2068                         integrity_metadata(&dio->work);
2069                 else
2070 skip_check:
2071                         dec_in_flight(dio);
2072
2073         } else {
2074                 INIT_WORK(&dio->work, integrity_metadata);
2075                 queue_work(ic->metadata_wq, &dio->work);
2076         }
2077
2078         return;
2079
2080 journal_read_write:
2081         if (unlikely(__journal_read_write(dio, bio, journal_section, journal_entry)))
2082                 goto lock_retry;
2083
2084         do_endio_flush(ic, dio);
2085 }
2086
2087
2088 static void integrity_bio_wait(struct work_struct *w)
2089 {
2090         struct dm_integrity_io *dio = container_of(w, struct dm_integrity_io, work);
2091
2092         dm_integrity_map_continue(dio, false);
2093 }
2094
2095 static void pad_uncommitted(struct dm_integrity_c *ic)
2096 {
2097         if (ic->free_section_entry) {
2098                 ic->free_sectors -= ic->journal_section_entries - ic->free_section_entry;
2099                 ic->free_section_entry = 0;
2100                 ic->free_section++;
2101                 wraparound_section(ic, &ic->free_section);
2102                 ic->n_uncommitted_sections++;
2103         }
2104         if (WARN_ON(ic->journal_sections * ic->journal_section_entries !=
2105                     (ic->n_uncommitted_sections + ic->n_committed_sections) *
2106                     ic->journal_section_entries + ic->free_sectors)) {
2107                 DMCRIT("journal_sections %u, journal_section_entries %u, "
2108                        "n_uncommitted_sections %u, n_committed_sections %u, "
2109                        "journal_section_entries %u, free_sectors %u",
2110                        ic->journal_sections, ic->journal_section_entries,
2111                        ic->n_uncommitted_sections, ic->n_committed_sections,
2112                        ic->journal_section_entries, ic->free_sectors);
2113         }
2114 }
2115
2116 static void integrity_commit(struct work_struct *w)
2117 {
2118         struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, commit_work);
2119         unsigned commit_start, commit_sections;
2120         unsigned i, j, n;
2121         struct bio *flushes;
2122
2123         del_timer(&ic->autocommit_timer);
2124
2125         spin_lock_irq(&ic->endio_wait.lock);
2126         flushes = bio_list_get(&ic->flush_bio_list);
2127         if (unlikely(ic->mode != 'J')) {
2128                 spin_unlock_irq(&ic->endio_wait.lock);
2129                 dm_integrity_flush_buffers(ic, true);
2130                 goto release_flush_bios;
2131         }
2132
2133         pad_uncommitted(ic);
2134         commit_start = ic->uncommitted_section;
2135         commit_sections = ic->n_uncommitted_sections;
2136         spin_unlock_irq(&ic->endio_wait.lock);
2137
2138         if (!commit_sections)
2139                 goto release_flush_bios;
2140
2141         i = commit_start;
2142         for (n = 0; n < commit_sections; n++) {
2143                 for (j = 0; j < ic->journal_section_entries; j++) {
2144                         struct journal_entry *je;
2145                         je = access_journal_entry(ic, i, j);
2146                         io_wait_event(ic->copy_to_journal_wait, !journal_entry_is_inprogress(je));
2147                 }
2148                 for (j = 0; j < ic->journal_section_sectors; j++) {
2149                         struct journal_sector *js;
2150                         js = access_journal(ic, i, j);
2151                         js->commit_id = dm_integrity_commit_id(ic, i, j, ic->commit_seq);
2152                 }
2153                 i++;
2154                 if (unlikely(i >= ic->journal_sections))
2155                         ic->commit_seq = next_commit_seq(ic->commit_seq);
2156                 wraparound_section(ic, &i);
2157         }
2158         smp_rmb();
2159
2160         write_journal(ic, commit_start, commit_sections);
2161
2162         spin_lock_irq(&ic->endio_wait.lock);
2163         ic->uncommitted_section += commit_sections;
2164         wraparound_section(ic, &ic->uncommitted_section);
2165         ic->n_uncommitted_sections -= commit_sections;
2166         ic->n_committed_sections += commit_sections;
2167         spin_unlock_irq(&ic->endio_wait.lock);
2168
2169         if (READ_ONCE(ic->free_sectors) <= ic->free_sectors_threshold)
2170                 queue_work(ic->writer_wq, &ic->writer_work);
2171
2172 release_flush_bios:
2173         while (flushes) {
2174                 struct bio *next = flushes->bi_next;
2175                 flushes->bi_next = NULL;
2176                 do_endio(ic, flushes);
2177                 flushes = next;
2178         }
2179 }
2180
2181 static void complete_copy_from_journal(unsigned long error, void *context)
2182 {
2183         struct journal_io *io = context;
2184         struct journal_completion *comp = io->comp;
2185         struct dm_integrity_c *ic = comp->ic;
2186         remove_range(ic, &io->range);
2187         mempool_free(io, &ic->journal_io_mempool);
2188         if (unlikely(error != 0))
2189                 dm_integrity_io_error(ic, "copying from journal", -EIO);
2190         complete_journal_op(comp);
2191 }
2192
2193 static void restore_last_bytes(struct dm_integrity_c *ic, struct journal_sector *js,
2194                                struct journal_entry *je)
2195 {
2196         unsigned s = 0;
2197         do {
2198                 js->commit_id = je->last_bytes[s];
2199                 js++;
2200         } while (++s < ic->sectors_per_block);
2201 }
2202
2203 static void do_journal_write(struct dm_integrity_c *ic, unsigned write_start,
2204                              unsigned write_sections, bool from_replay)
2205 {
2206         unsigned i, j, n;
2207         struct journal_completion comp;
2208         struct blk_plug plug;
2209
2210         blk_start_plug(&plug);
2211
2212         comp.ic = ic;
2213         comp.in_flight = (atomic_t)ATOMIC_INIT(1);
2214         init_completion(&comp.comp);
2215
2216         i = write_start;
2217         for (n = 0; n < write_sections; n++, i++, wraparound_section(ic, &i)) {
2218 #ifndef INTERNAL_VERIFY
2219                 if (unlikely(from_replay))
2220 #endif
2221                         rw_section_mac(ic, i, false);
2222                 for (j = 0; j < ic->journal_section_entries; j++) {
2223                         struct journal_entry *je = access_journal_entry(ic, i, j);
2224                         sector_t sec, area, offset;
2225                         unsigned k, l, next_loop;
2226                         sector_t metadata_block;
2227                         unsigned metadata_offset;
2228                         struct journal_io *io;
2229
2230                         if (journal_entry_is_unused(je))
2231                                 continue;
2232                         BUG_ON(unlikely(journal_entry_is_inprogress(je)) && !from_replay);
2233                         sec = journal_entry_get_sector(je);
2234                         if (unlikely(from_replay)) {
2235                                 if (unlikely(sec & (unsigned)(ic->sectors_per_block - 1))) {
2236                                         dm_integrity_io_error(ic, "invalid sector in journal", -EIO);
2237                                         sec &= ~(sector_t)(ic->sectors_per_block - 1);
2238                                 }
2239                         }
2240                         get_area_and_offset(ic, sec, &area, &offset);
2241                         restore_last_bytes(ic, access_journal_data(ic, i, j), je);
2242                         for (k = j + 1; k < ic->journal_section_entries; k++) {
2243                                 struct journal_entry *je2 = access_journal_entry(ic, i, k);
2244                                 sector_t sec2, area2, offset2;
2245                                 if (journal_entry_is_unused(je2))
2246                                         break;
2247                                 BUG_ON(unlikely(journal_entry_is_inprogress(je2)) && !from_replay);
2248                                 sec2 = journal_entry_get_sector(je2);
2249                                 get_area_and_offset(ic, sec2, &area2, &offset2);
2250                                 if (area2 != area || offset2 != offset + ((k - j) << ic->sb->log2_sectors_per_block))
2251                                         break;
2252                                 restore_last_bytes(ic, access_journal_data(ic, i, k), je2);
2253                         }
2254                         next_loop = k - 1;
2255
2256                         io = mempool_alloc(&ic->journal_io_mempool, GFP_NOIO);
2257                         io->comp = &comp;
2258                         io->range.logical_sector = sec;
2259                         io->range.n_sectors = (k - j) << ic->sb->log2_sectors_per_block;
2260
2261                         spin_lock_irq(&ic->endio_wait.lock);
2262                         add_new_range_and_wait(ic, &io->range);
2263
2264                         if (likely(!from_replay)) {
2265                                 struct journal_node *section_node = &ic->journal_tree[i * ic->journal_section_entries];
2266
2267                                 /* don't write if there is newer committed sector */
2268                                 while (j < k && find_newer_committed_node(ic, &section_node[j])) {
2269                                         struct journal_entry *je2 = access_journal_entry(ic, i, j);
2270
2271                                         journal_entry_set_unused(je2);
2272                                         remove_journal_node(ic, &section_node[j]);
2273                                         j++;
2274                                         sec += ic->sectors_per_block;
2275                                         offset += ic->sectors_per_block;
2276                                 }
2277                                 while (j < k && find_newer_committed_node(ic, &section_node[k - 1])) {
2278                                         struct journal_entry *je2 = access_journal_entry(ic, i, k - 1);
2279
2280                                         journal_entry_set_unused(je2);
2281                                         remove_journal_node(ic, &section_node[k - 1]);
2282                                         k--;
2283                                 }
2284                                 if (j == k) {
2285                                         remove_range_unlocked(ic, &io->range);
2286                                         spin_unlock_irq(&ic->endio_wait.lock);
2287                                         mempool_free(io, &ic->journal_io_mempool);
2288                                         goto skip_io;
2289                                 }
2290                                 for (l = j; l < k; l++) {
2291                                         remove_journal_node(ic, &section_node[l]);
2292                                 }
2293                         }
2294                         spin_unlock_irq(&ic->endio_wait.lock);
2295
2296                         metadata_block = get_metadata_sector_and_offset(ic, area, offset, &metadata_offset);
2297                         for (l = j; l < k; l++) {
2298                                 int r;
2299                                 struct journal_entry *je2 = access_journal_entry(ic, i, l);
2300
2301                                 if (
2302 #ifndef INTERNAL_VERIFY
2303                                     unlikely(from_replay) &&
2304 #endif
2305                                     ic->internal_hash) {
2306                                         char test_tag[max_t(size_t, HASH_MAX_DIGESTSIZE, MAX_TAG_SIZE)];
2307
2308                                         integrity_sector_checksum(ic, sec + ((l - j) << ic->sb->log2_sectors_per_block),
2309                                                                   (char *)access_journal_data(ic, i, l), test_tag);
2310                                         if (unlikely(memcmp(test_tag, journal_entry_tag(ic, je2), ic->tag_size)))
2311                                                 dm_integrity_io_error(ic, "tag mismatch when replaying journal", -EILSEQ);
2312                                 }
2313
2314                                 journal_entry_set_unused(je2);
2315                                 r = dm_integrity_rw_tag(ic, journal_entry_tag(ic, je2), &metadata_block, &metadata_offset,
2316                                                         ic->tag_size, TAG_WRITE);
2317                                 if (unlikely(r)) {
2318                                         dm_integrity_io_error(ic, "reading tags", r);
2319                                 }
2320                         }
2321
2322                         atomic_inc(&comp.in_flight);
2323                         copy_from_journal(ic, i, j << ic->sb->log2_sectors_per_block,
2324                                           (k - j) << ic->sb->log2_sectors_per_block,
2325                                           get_data_sector(ic, area, offset),
2326                                           complete_copy_from_journal, io);
2327 skip_io:
2328                         j = next_loop;
2329                 }
2330         }
2331
2332         dm_bufio_write_dirty_buffers_async(ic->bufio);
2333
2334         blk_finish_plug(&plug);
2335
2336         complete_journal_op(&comp);
2337         wait_for_completion_io(&comp.comp);
2338
2339         dm_integrity_flush_buffers(ic, true);
2340 }
2341
2342 static void integrity_writer(struct work_struct *w)
2343 {
2344         struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, writer_work);
2345         unsigned write_start, write_sections;
2346
2347         unsigned prev_free_sectors;
2348
2349         spin_lock_irq(&ic->endio_wait.lock);
2350         write_start = ic->committed_section;
2351         write_sections = ic->n_committed_sections;
2352         spin_unlock_irq(&ic->endio_wait.lock);
2353
2354         if (!write_sections)
2355                 return;
2356
2357         do_journal_write(ic, write_start, write_sections, false);
2358
2359         spin_lock_irq(&ic->endio_wait.lock);
2360
2361         ic->committed_section += write_sections;
2362         wraparound_section(ic, &ic->committed_section);
2363         ic->n_committed_sections -= write_sections;
2364
2365         prev_free_sectors = ic->free_sectors;
2366         ic->free_sectors += write_sections * ic->journal_section_entries;
2367         if (unlikely(!prev_free_sectors))
2368                 wake_up_locked(&ic->endio_wait);
2369
2370         spin_unlock_irq(&ic->endio_wait.lock);
2371 }
2372
2373 static void recalc_write_super(struct dm_integrity_c *ic)
2374 {
2375         int r;
2376
2377         dm_integrity_flush_buffers(ic, false);
2378         if (dm_integrity_failed(ic))
2379                 return;
2380
2381         r = sync_rw_sb(ic, REQ_OP_WRITE, 0);
2382         if (unlikely(r))
2383                 dm_integrity_io_error(ic, "writing superblock", r);
2384 }
2385
2386 static void integrity_recalc(struct work_struct *w)
2387 {
2388         struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, recalc_work);
2389         struct dm_integrity_range range;
2390         struct dm_io_request io_req;
2391         struct dm_io_region io_loc;
2392         sector_t area, offset;
2393         sector_t metadata_block;
2394         unsigned metadata_offset;
2395         sector_t logical_sector, n_sectors;
2396         __u8 *t;
2397         unsigned i;
2398         int r;
2399         unsigned super_counter = 0;
2400
2401         DEBUG_print("start recalculation... (position %llx)\n", le64_to_cpu(ic->sb->recalc_sector));
2402
2403         spin_lock_irq(&ic->endio_wait.lock);
2404
2405 next_chunk:
2406
2407         if (unlikely(dm_post_suspending(ic->ti)))
2408                 goto unlock_ret;
2409
2410         range.logical_sector = le64_to_cpu(ic->sb->recalc_sector);
2411         if (unlikely(range.logical_sector >= ic->provided_data_sectors)) {
2412                 if (ic->mode == 'B') {
2413                         block_bitmap_op(ic, ic->recalc_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_CLEAR);
2414                         DEBUG_print("queue_delayed_work: bitmap_flush_work\n");
2415                         queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, 0);
2416                 }
2417                 goto unlock_ret;
2418         }
2419
2420         get_area_and_offset(ic, range.logical_sector, &area, &offset);
2421         range.n_sectors = min((sector_t)RECALC_SECTORS, ic->provided_data_sectors - range.logical_sector);
2422         if (!ic->meta_dev)
2423                 range.n_sectors = min(range.n_sectors, ((sector_t)1U << ic->sb->log2_interleave_sectors) - (unsigned)offset);
2424
2425         add_new_range_and_wait(ic, &range);
2426         spin_unlock_irq(&ic->endio_wait.lock);
2427         logical_sector = range.logical_sector;
2428         n_sectors = range.n_sectors;
2429
2430         if (ic->mode == 'B') {
2431                 if (block_bitmap_op(ic, ic->recalc_bitmap, logical_sector, n_sectors, BITMAP_OP_TEST_ALL_CLEAR)) {
2432                         goto advance_and_next;
2433                 }
2434                 while (block_bitmap_op(ic, ic->recalc_bitmap, logical_sector,
2435                                        ic->sectors_per_block, BITMAP_OP_TEST_ALL_CLEAR)) {
2436                         logical_sector += ic->sectors_per_block;
2437                         n_sectors -= ic->sectors_per_block;
2438                         cond_resched();
2439                 }
2440                 while (block_bitmap_op(ic, ic->recalc_bitmap, logical_sector + n_sectors - ic->sectors_per_block,
2441                                        ic->sectors_per_block, BITMAP_OP_TEST_ALL_CLEAR)) {
2442                         n_sectors -= ic->sectors_per_block;
2443                         cond_resched();
2444                 }
2445                 get_area_and_offset(ic, logical_sector, &area, &offset);
2446         }
2447
2448         DEBUG_print("recalculating: %lx, %lx\n", logical_sector, n_sectors);
2449
2450         if (unlikely(++super_counter == RECALC_WRITE_SUPER)) {
2451                 recalc_write_super(ic);
2452                 if (ic->mode == 'B') {
2453                         queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, ic->bitmap_flush_interval);
2454                 }
2455                 super_counter = 0;
2456         }
2457
2458         if (unlikely(dm_integrity_failed(ic)))
2459                 goto err;
2460
2461         io_req.bi_op = REQ_OP_READ;
2462         io_req.bi_op_flags = 0;
2463         io_req.mem.type = DM_IO_VMA;
2464         io_req.mem.ptr.addr = ic->recalc_buffer;
2465         io_req.notify.fn = NULL;
2466         io_req.client = ic->io;
2467         io_loc.bdev = ic->dev->bdev;
2468         io_loc.sector = get_data_sector(ic, area, offset);
2469         io_loc.count = n_sectors;
2470
2471         r = dm_io(&io_req, 1, &io_loc, NULL);
2472         if (unlikely(r)) {
2473                 dm_integrity_io_error(ic, "reading data", r);
2474                 goto err;
2475         }
2476
2477         t = ic->recalc_tags;
2478         for (i = 0; i < n_sectors; i += ic->sectors_per_block) {
2479                 integrity_sector_checksum(ic, logical_sector + i, ic->recalc_buffer + (i << SECTOR_SHIFT), t);
2480                 t += ic->tag_size;
2481         }
2482
2483         metadata_block = get_metadata_sector_and_offset(ic, area, offset, &metadata_offset);
2484
2485         r = dm_integrity_rw_tag(ic, ic->recalc_tags, &metadata_block, &metadata_offset, t - ic->recalc_tags, TAG_WRITE);
2486         if (unlikely(r)) {
2487                 dm_integrity_io_error(ic, "writing tags", r);
2488                 goto err;
2489         }
2490
2491         if (ic->mode == 'B') {
2492                 sector_t start, end;
2493                 start = (range.logical_sector >>
2494                          (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit)) <<
2495                         (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
2496                 end = ((range.logical_sector + range.n_sectors) >>
2497                        (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit)) <<
2498                         (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
2499                 block_bitmap_op(ic, ic->recalc_bitmap, start, end - start, BITMAP_OP_CLEAR);
2500         }
2501
2502 advance_and_next:
2503         cond_resched();
2504
2505         spin_lock_irq(&ic->endio_wait.lock);
2506         remove_range_unlocked(ic, &range);
2507         ic->sb->recalc_sector = cpu_to_le64(range.logical_sector + range.n_sectors);
2508         goto next_chunk;
2509
2510 err:
2511         remove_range(ic, &range);
2512         return;
2513
2514 unlock_ret:
2515         spin_unlock_irq(&ic->endio_wait.lock);
2516
2517         recalc_write_super(ic);
2518 }
2519
2520 static void bitmap_block_work(struct work_struct *w)
2521 {
2522         struct bitmap_block_status *bbs = container_of(w, struct bitmap_block_status, work);
2523         struct dm_integrity_c *ic = bbs->ic;
2524         struct bio *bio;
2525         struct bio_list bio_queue;
2526         struct bio_list waiting;
2527
2528         bio_list_init(&waiting);
2529
2530         spin_lock(&bbs->bio_queue_lock);
2531         bio_queue = bbs->bio_queue;
2532         bio_list_init(&bbs->bio_queue);
2533         spin_unlock(&bbs->bio_queue_lock);
2534
2535         while ((bio = bio_list_pop(&bio_queue))) {
2536                 struct dm_integrity_io *dio;
2537
2538                 dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
2539
2540                 if (block_bitmap_op(ic, ic->may_write_bitmap, dio->range.logical_sector,
2541                                     dio->range.n_sectors, BITMAP_OP_TEST_ALL_SET)) {
2542                         remove_range(ic, &dio->range);
2543                         INIT_WORK(&dio->work, integrity_bio_wait);
2544                         queue_work(ic->offload_wq, &dio->work);
2545                 } else {
2546                         block_bitmap_op(ic, ic->journal, dio->range.logical_sector,
2547                                         dio->range.n_sectors, BITMAP_OP_SET);
2548                         bio_list_add(&waiting, bio);
2549                 }
2550         }
2551
2552         if (bio_list_empty(&waiting))
2553                 return;
2554
2555         rw_journal_sectors(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC,
2556                            bbs->idx * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT),
2557                            BITMAP_BLOCK_SIZE >> SECTOR_SHIFT, NULL);
2558
2559         while ((bio = bio_list_pop(&waiting))) {
2560                 struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
2561
2562                 block_bitmap_op(ic, ic->may_write_bitmap, dio->range.logical_sector,
2563                                 dio->range.n_sectors, BITMAP_OP_SET);
2564
2565                 remove_range(ic, &dio->range);
2566                 INIT_WORK(&dio->work, integrity_bio_wait);
2567                 queue_work(ic->offload_wq, &dio->work);
2568         }
2569
2570         queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, ic->bitmap_flush_interval);
2571 }
2572
2573 static void bitmap_flush_work(struct work_struct *work)
2574 {
2575         struct dm_integrity_c *ic = container_of(work, struct dm_integrity_c, bitmap_flush_work.work);
2576         struct dm_integrity_range range;
2577         unsigned long limit;
2578         struct bio *bio;
2579
2580         dm_integrity_flush_buffers(ic, false);
2581
2582         range.logical_sector = 0;
2583         range.n_sectors = ic->provided_data_sectors;
2584
2585         spin_lock_irq(&ic->endio_wait.lock);
2586         add_new_range_and_wait(ic, &range);
2587         spin_unlock_irq(&ic->endio_wait.lock);
2588
2589         dm_integrity_flush_buffers(ic, true);
2590         if (ic->meta_dev)
2591                 blkdev_issue_flush(ic->dev->bdev, GFP_NOIO, NULL);
2592
2593         limit = ic->provided_data_sectors;
2594         if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
2595                 limit = le64_to_cpu(ic->sb->recalc_sector)
2596                         >> (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit)
2597                         << (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
2598         }
2599         /*DEBUG_print("zeroing journal\n");*/
2600         block_bitmap_op(ic, ic->journal, 0, limit, BITMAP_OP_CLEAR);
2601         block_bitmap_op(ic, ic->may_write_bitmap, 0, limit, BITMAP_OP_CLEAR);
2602
2603         rw_journal_sectors(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, 0,
2604                            ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
2605
2606         spin_lock_irq(&ic->endio_wait.lock);
2607         remove_range_unlocked(ic, &range);
2608         while (unlikely((bio = bio_list_pop(&ic->synchronous_bios)) != NULL)) {
2609                 bio_endio(bio);
2610                 spin_unlock_irq(&ic->endio_wait.lock);
2611                 spin_lock_irq(&ic->endio_wait.lock);
2612         }
2613         spin_unlock_irq(&ic->endio_wait.lock);
2614 }
2615
2616
2617 static void init_journal(struct dm_integrity_c *ic, unsigned start_section,
2618                          unsigned n_sections, unsigned char commit_seq)
2619 {
2620         unsigned i, j, n;
2621
2622         if (!n_sections)
2623                 return;
2624
2625         for (n = 0; n < n_sections; n++) {
2626                 i = start_section + n;
2627                 wraparound_section(ic, &i);
2628                 for (j = 0; j < ic->journal_section_sectors; j++) {
2629                         struct journal_sector *js = access_journal(ic, i, j);
2630                         memset(&js->entries, 0, JOURNAL_SECTOR_DATA);
2631                         js->commit_id = dm_integrity_commit_id(ic, i, j, commit_seq);
2632                 }
2633                 for (j = 0; j < ic->journal_section_entries; j++) {
2634                         struct journal_entry *je = access_journal_entry(ic, i, j);
2635                         journal_entry_set_unused(je);
2636                 }
2637         }
2638
2639         write_journal(ic, start_section, n_sections);
2640 }
2641
2642 static int find_commit_seq(struct dm_integrity_c *ic, unsigned i, unsigned j, commit_id_t id)
2643 {
2644         unsigned char k;
2645         for (k = 0; k < N_COMMIT_IDS; k++) {
2646                 if (dm_integrity_commit_id(ic, i, j, k) == id)
2647                         return k;
2648         }
2649         dm_integrity_io_error(ic, "journal commit id", -EIO);
2650         return -EIO;
2651 }
2652
2653 static void replay_journal(struct dm_integrity_c *ic)
2654 {
2655         unsigned i, j;
2656         bool used_commit_ids[N_COMMIT_IDS];
2657         unsigned max_commit_id_sections[N_COMMIT_IDS];
2658         unsigned write_start, write_sections;
2659         unsigned continue_section;
2660         bool journal_empty;
2661         unsigned char unused, last_used, want_commit_seq;
2662
2663         if (ic->mode == 'R')
2664                 return;
2665
2666         if (ic->journal_uptodate)
2667                 return;
2668
2669         last_used = 0;
2670         write_start = 0;
2671
2672         if (!ic->just_formatted) {
2673                 DEBUG_print("reading journal\n");
2674                 rw_journal(ic, REQ_OP_READ, 0, 0, ic->journal_sections, NULL);
2675                 if (ic->journal_io)
2676                         DEBUG_bytes(lowmem_page_address(ic->journal_io[0].page), 64, "read journal");
2677                 if (ic->journal_io) {
2678                         struct journal_completion crypt_comp;
2679                         crypt_comp.ic = ic;
2680                         init_completion(&crypt_comp.comp);
2681                         crypt_comp.in_flight = (atomic_t)ATOMIC_INIT(0);
2682                         encrypt_journal(ic, false, 0, ic->journal_sections, &crypt_comp);
2683                         wait_for_completion(&crypt_comp.comp);
2684                 }
2685                 DEBUG_bytes(lowmem_page_address(ic->journal[0].page), 64, "decrypted journal");
2686         }
2687
2688         if (dm_integrity_failed(ic))
2689                 goto clear_journal;
2690
2691         journal_empty = true;
2692         memset(used_commit_ids, 0, sizeof used_commit_ids);
2693         memset(max_commit_id_sections, 0, sizeof max_commit_id_sections);
2694         for (i = 0; i < ic->journal_sections; i++) {
2695                 for (j = 0; j < ic->journal_section_sectors; j++) {
2696                         int k;
2697                         struct journal_sector *js = access_journal(ic, i, j);
2698                         k = find_commit_seq(ic, i, j, js->commit_id);
2699                         if (k < 0)
2700                                 goto clear_journal;
2701                         used_commit_ids[k] = true;
2702                         max_commit_id_sections[k] = i;
2703                 }
2704                 if (journal_empty) {
2705                         for (j = 0; j < ic->journal_section_entries; j++) {
2706                                 struct journal_entry *je = access_journal_entry(ic, i, j);
2707                                 if (!journal_entry_is_unused(je)) {
2708                                         journal_empty = false;
2709                                         break;
2710                                 }
2711                         }
2712                 }
2713         }
2714
2715         if (!used_commit_ids[N_COMMIT_IDS - 1]) {
2716                 unused = N_COMMIT_IDS - 1;
2717                 while (unused && !used_commit_ids[unused - 1])
2718                         unused--;
2719         } else {
2720                 for (unused = 0; unused < N_COMMIT_IDS; unused++)
2721                         if (!used_commit_ids[unused])
2722                                 break;
2723                 if (unused == N_COMMIT_IDS) {
2724                         dm_integrity_io_error(ic, "journal commit ids", -EIO);
2725                         goto clear_journal;
2726                 }
2727         }
2728         DEBUG_print("first unused commit seq %d [%d,%d,%d,%d]\n",
2729                     unused, used_commit_ids[0], used_commit_ids[1],
2730                     used_commit_ids[2], used_commit_ids[3]);
2731
2732         last_used = prev_commit_seq(unused);
2733         want_commit_seq = prev_commit_seq(last_used);
2734
2735         if (!used_commit_ids[want_commit_seq] && used_commit_ids[prev_commit_seq(want_commit_seq)])
2736                 journal_empty = true;
2737
2738         write_start = max_commit_id_sections[last_used] + 1;
2739         if (unlikely(write_start >= ic->journal_sections))
2740                 want_commit_seq = next_commit_seq(want_commit_seq);
2741         wraparound_section(ic, &write_start);
2742
2743         i = write_start;
2744         for (write_sections = 0; write_sections < ic->journal_sections; write_sections++) {
2745                 for (j = 0; j < ic->journal_section_sectors; j++) {
2746                         struct journal_sector *js = access_journal(ic, i, j);
2747
2748                         if (js->commit_id != dm_integrity_commit_id(ic, i, j, want_commit_seq)) {
2749                                 /*
2750                                  * This could be caused by crash during writing.
2751                                  * We won't replay the inconsistent part of the
2752                                  * journal.
2753                                  */
2754                                 DEBUG_print("commit id mismatch at position (%u, %u): %d != %d\n",
2755                                             i, j, find_commit_seq(ic, i, j, js->commit_id), want_commit_seq);
2756                                 goto brk;
2757                         }
2758                 }
2759                 i++;
2760                 if (unlikely(i >= ic->journal_sections))
2761                         want_commit_seq = next_commit_seq(want_commit_seq);
2762                 wraparound_section(ic, &i);
2763         }
2764 brk:
2765
2766         if (!journal_empty) {
2767                 DEBUG_print("replaying %u sections, starting at %u, commit seq %d\n",
2768                             write_sections, write_start, want_commit_seq);
2769                 do_journal_write(ic, write_start, write_sections, true);
2770         }
2771
2772         if (write_sections == ic->journal_sections && (ic->mode == 'J' || journal_empty)) {
2773                 continue_section = write_start;
2774                 ic->commit_seq = want_commit_seq;
2775                 DEBUG_print("continuing from section %u, commit seq %d\n", write_start, ic->commit_seq);
2776         } else {
2777                 unsigned s;
2778                 unsigned char erase_seq;
2779 clear_journal:
2780                 DEBUG_print("clearing journal\n");
2781
2782                 erase_seq = prev_commit_seq(prev_commit_seq(last_used));
2783                 s = write_start;
2784                 init_journal(ic, s, 1, erase_seq);
2785                 s++;
2786                 wraparound_section(ic, &s);
2787                 if (ic->journal_sections >= 2) {
2788                         init_journal(ic, s, ic->journal_sections - 2, erase_seq);
2789                         s += ic->journal_sections - 2;
2790                         wraparound_section(ic, &s);
2791                         init_journal(ic, s, 1, erase_seq);
2792                 }
2793
2794                 continue_section = 0;
2795                 ic->commit_seq = next_commit_seq(erase_seq);
2796         }
2797
2798         ic->committed_section = continue_section;
2799         ic->n_committed_sections = 0;
2800
2801         ic->uncommitted_section = continue_section;
2802         ic->n_uncommitted_sections = 0;
2803
2804         ic->free_section = continue_section;
2805         ic->free_section_entry = 0;
2806         ic->free_sectors = ic->journal_entries;
2807
2808         ic->journal_tree_root = RB_ROOT;
2809         for (i = 0; i < ic->journal_entries; i++)
2810                 init_journal_node(&ic->journal_tree[i]);
2811 }
2812
2813 static void dm_integrity_enter_synchronous_mode(struct dm_integrity_c *ic)
2814 {
2815         DEBUG_print("dm_integrity_enter_synchronous_mode\n");
2816
2817         if (ic->mode == 'B') {
2818                 ic->bitmap_flush_interval = msecs_to_jiffies(10) + 1;
2819                 ic->synchronous_mode = 1;
2820
2821                 cancel_delayed_work_sync(&ic->bitmap_flush_work);
2822                 queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, 0);
2823                 flush_workqueue(ic->commit_wq);
2824         }
2825 }
2826
2827 static int dm_integrity_reboot(struct notifier_block *n, unsigned long code, void *x)
2828 {
2829         struct dm_integrity_c *ic = container_of(n, struct dm_integrity_c, reboot_notifier);
2830
2831         DEBUG_print("dm_integrity_reboot\n");
2832
2833         dm_integrity_enter_synchronous_mode(ic);
2834
2835         return NOTIFY_DONE;
2836 }
2837
2838 static void dm_integrity_postsuspend(struct dm_target *ti)
2839 {
2840         struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
2841         int r;
2842
2843         WARN_ON(unregister_reboot_notifier(&ic->reboot_notifier));
2844
2845         del_timer_sync(&ic->autocommit_timer);
2846
2847         if (ic->recalc_wq)
2848                 drain_workqueue(ic->recalc_wq);
2849
2850         if (ic->mode == 'B')
2851                 cancel_delayed_work_sync(&ic->bitmap_flush_work);
2852
2853         queue_work(ic->commit_wq, &ic->commit_work);
2854         drain_workqueue(ic->commit_wq);
2855
2856         if (ic->mode == 'J') {
2857                 queue_work(ic->writer_wq, &ic->writer_work);
2858                 drain_workqueue(ic->writer_wq);
2859                 dm_integrity_flush_buffers(ic, true);
2860         }
2861
2862         if (ic->mode == 'B') {
2863                 dm_integrity_flush_buffers(ic, true);
2864 #if 1
2865                 /* set to 0 to test bitmap replay code */
2866                 init_journal(ic, 0, ic->journal_sections, 0);
2867                 ic->sb->flags &= ~cpu_to_le32(SB_FLAG_DIRTY_BITMAP);
2868                 r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
2869                 if (unlikely(r))
2870                         dm_integrity_io_error(ic, "writing superblock", r);
2871 #endif
2872         }
2873
2874         BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
2875
2876         ic->journal_uptodate = true;
2877 }
2878
2879 static void dm_integrity_resume(struct dm_target *ti)
2880 {
2881         struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
2882         int r;
2883         DEBUG_print("resume\n");
2884
2885         if (ic->sb->flags & cpu_to_le32(SB_FLAG_DIRTY_BITMAP)) {
2886                 DEBUG_print("resume dirty_bitmap\n");
2887                 rw_journal_sectors(ic, REQ_OP_READ, 0, 0,
2888                                    ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
2889                 if (ic->mode == 'B') {
2890                         if (ic->sb->log2_blocks_per_bitmap_bit == ic->log2_blocks_per_bitmap_bit) {
2891                                 block_bitmap_copy(ic, ic->recalc_bitmap, ic->journal);
2892                                 block_bitmap_copy(ic, ic->may_write_bitmap, ic->journal);
2893                                 if (!block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors,
2894                                                      BITMAP_OP_TEST_ALL_CLEAR)) {
2895                                         ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
2896                                         ic->sb->recalc_sector = cpu_to_le64(0);
2897                                 }
2898                         } else {
2899                                 DEBUG_print("non-matching blocks_per_bitmap_bit: %u, %u\n",
2900                                             ic->sb->log2_blocks_per_bitmap_bit, ic->log2_blocks_per_bitmap_bit);
2901                                 ic->sb->log2_blocks_per_bitmap_bit = ic->log2_blocks_per_bitmap_bit;
2902                                 block_bitmap_op(ic, ic->recalc_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_SET);
2903                                 block_bitmap_op(ic, ic->may_write_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_SET);
2904                                 block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors, BITMAP_OP_SET);
2905                                 rw_journal_sectors(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, 0,
2906                                                    ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
2907                                 ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
2908                                 ic->sb->recalc_sector = cpu_to_le64(0);
2909                         }
2910                 } else {
2911                         if (!(ic->sb->log2_blocks_per_bitmap_bit == ic->log2_blocks_per_bitmap_bit &&
2912                               block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors, BITMAP_OP_TEST_ALL_CLEAR))) {
2913                                 ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
2914                                 ic->sb->recalc_sector = cpu_to_le64(0);
2915                         }
2916                         init_journal(ic, 0, ic->journal_sections, 0);
2917                         replay_journal(ic);
2918                         ic->sb->flags &= ~cpu_to_le32(SB_FLAG_DIRTY_BITMAP);
2919                 }
2920                 r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
2921                 if (unlikely(r))
2922                         dm_integrity_io_error(ic, "writing superblock", r);
2923         } else {
2924                 replay_journal(ic);
2925                 if (ic->mode == 'B') {
2926                         ic->sb->flags |= cpu_to_le32(SB_FLAG_DIRTY_BITMAP);
2927                         ic->sb->log2_blocks_per_bitmap_bit = ic->log2_blocks_per_bitmap_bit;
2928                         r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
2929                         if (unlikely(r))
2930                                 dm_integrity_io_error(ic, "writing superblock", r);
2931
2932                         block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors, BITMAP_OP_CLEAR);
2933                         block_bitmap_op(ic, ic->recalc_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_CLEAR);
2934                         block_bitmap_op(ic, ic->may_write_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_CLEAR);
2935                         if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING) &&
2936                             le64_to_cpu(ic->sb->recalc_sector) < ic->provided_data_sectors) {
2937                                 block_bitmap_op(ic, ic->journal, le64_to_cpu(ic->sb->recalc_sector),
2938                                                 ic->provided_data_sectors - le64_to_cpu(ic->sb->recalc_sector), BITMAP_OP_SET);
2939                                 block_bitmap_op(ic, ic->recalc_bitmap, le64_to_cpu(ic->sb->recalc_sector),
2940                                                 ic->provided_data_sectors - le64_to_cpu(ic->sb->recalc_sector), BITMAP_OP_SET);
2941                                 block_bitmap_op(ic, ic->may_write_bitmap, le64_to_cpu(ic->sb->recalc_sector),
2942                                                 ic->provided_data_sectors - le64_to_cpu(ic->sb->recalc_sector), BITMAP_OP_SET);
2943                         }
2944                         rw_journal_sectors(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, 0,
2945                                            ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
2946                 }
2947         }
2948
2949         DEBUG_print("testing recalc: %x\n", ic->sb->flags);
2950         if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
2951                 __u64 recalc_pos = le64_to_cpu(ic->sb->recalc_sector);
2952                 DEBUG_print("recalc pos: %lx / %lx\n", (long)recalc_pos, ic->provided_data_sectors);
2953                 if (recalc_pos < ic->provided_data_sectors) {
2954                         queue_work(ic->recalc_wq, &ic->recalc_work);
2955                 } else if (recalc_pos > ic->provided_data_sectors) {
2956                         ic->sb->recalc_sector = cpu_to_le64(ic->provided_data_sectors);
2957                         recalc_write_super(ic);
2958                 }
2959         }
2960
2961         ic->reboot_notifier.notifier_call = dm_integrity_reboot;
2962         ic->reboot_notifier.next = NULL;
2963         ic->reboot_notifier.priority = INT_MAX - 1;     /* be notified after md and before hardware drivers */
2964         WARN_ON(register_reboot_notifier(&ic->reboot_notifier));
2965
2966 #if 0
2967         /* set to 1 to stress test synchronous mode */
2968         dm_integrity_enter_synchronous_mode(ic);
2969 #endif
2970 }
2971
2972 static void dm_integrity_status(struct dm_target *ti, status_type_t type,
2973                                 unsigned status_flags, char *result, unsigned maxlen)
2974 {
2975         struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
2976         unsigned arg_count;
2977         size_t sz = 0;
2978
2979         switch (type) {
2980         case STATUSTYPE_INFO:
2981                 DMEMIT("%llu %llu",
2982                         (unsigned long long)atomic64_read(&ic->number_of_mismatches),
2983                         (unsigned long long)ic->provided_data_sectors);
2984                 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
2985                         DMEMIT(" %llu", (unsigned long long)le64_to_cpu(ic->sb->recalc_sector));
2986                 else
2987                         DMEMIT(" -");
2988                 break;
2989
2990         case STATUSTYPE_TABLE: {
2991                 __u64 watermark_percentage = (__u64)(ic->journal_entries - ic->free_sectors_threshold) * 100;
2992                 watermark_percentage += ic->journal_entries / 2;
2993                 do_div(watermark_percentage, ic->journal_entries);
2994                 arg_count = 3;
2995                 arg_count += !!ic->meta_dev;
2996                 arg_count += ic->sectors_per_block != 1;
2997                 arg_count += !!(ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING));
2998                 arg_count += ic->mode == 'J';
2999                 arg_count += ic->mode == 'J';
3000                 arg_count += ic->mode == 'B';
3001                 arg_count += ic->mode == 'B';
3002                 arg_count += !!ic->internal_hash_alg.alg_string;
3003                 arg_count += !!ic->journal_crypt_alg.alg_string;
3004                 arg_count += !!ic->journal_mac_alg.alg_string;
3005                 arg_count += ic->legacy_recalculate;
3006                 DMEMIT("%s %llu %u %c %u", ic->dev->name, (unsigned long long)ic->start,
3007                        ic->tag_size, ic->mode, arg_count);
3008                 if (ic->meta_dev)
3009                         DMEMIT(" meta_device:%s", ic->meta_dev->name);
3010                 if (ic->sectors_per_block != 1)
3011                         DMEMIT(" block_size:%u", ic->sectors_per_block << SECTOR_SHIFT);
3012                 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
3013                         DMEMIT(" recalculate");
3014                 DMEMIT(" journal_sectors:%u", ic->initial_sectors - SB_SECTORS);
3015                 DMEMIT(" interleave_sectors:%u", 1U << ic->sb->log2_interleave_sectors);
3016                 DMEMIT(" buffer_sectors:%u", 1U << ic->log2_buffer_sectors);
3017                 if (ic->mode == 'J') {
3018                         DMEMIT(" journal_watermark:%u", (unsigned)watermark_percentage);
3019                         DMEMIT(" commit_time:%u", ic->autocommit_msec);
3020                 }
3021                 if (ic->mode == 'B') {
3022                         DMEMIT(" sectors_per_bit:%llu", (unsigned long long)ic->sectors_per_block << ic->log2_blocks_per_bitmap_bit);
3023                         DMEMIT(" bitmap_flush_interval:%u", jiffies_to_msecs(ic->bitmap_flush_interval));
3024                 }
3025                 if (ic->legacy_recalculate)
3026                         DMEMIT(" legacy_recalculate");
3027
3028 #define EMIT_ALG(a, n)                                                  \
3029                 do {                                                    \
3030                         if (ic->a.alg_string) {                         \
3031                                 DMEMIT(" %s:%s", n, ic->a.alg_string);  \
3032                                 if (ic->a.key_string)                   \
3033                                         DMEMIT(":%s", ic->a.key_string);\
3034                         }                                               \
3035                 } while (0)
3036                 EMIT_ALG(internal_hash_alg, "internal_hash");
3037                 EMIT_ALG(journal_crypt_alg, "journal_crypt");
3038                 EMIT_ALG(journal_mac_alg, "journal_mac");
3039                 break;
3040         }
3041         }
3042 }
3043
3044 static int dm_integrity_iterate_devices(struct dm_target *ti,
3045                                         iterate_devices_callout_fn fn, void *data)
3046 {
3047         struct dm_integrity_c *ic = ti->private;
3048
3049         if (!ic->meta_dev)
3050                 return fn(ti, ic->dev, ic->start + ic->initial_sectors + ic->metadata_run, ti->len, data);
3051         else
3052                 return fn(ti, ic->dev, 0, ti->len, data);
3053 }
3054
3055 static void dm_integrity_io_hints(struct dm_target *ti, struct queue_limits *limits)
3056 {
3057         struct dm_integrity_c *ic = ti->private;
3058
3059         if (ic->sectors_per_block > 1) {
3060                 limits->logical_block_size = ic->sectors_per_block << SECTOR_SHIFT;
3061                 limits->physical_block_size = ic->sectors_per_block << SECTOR_SHIFT;
3062                 blk_limits_io_min(limits, ic->sectors_per_block << SECTOR_SHIFT);
3063         }
3064 }
3065
3066 static void calculate_journal_section_size(struct dm_integrity_c *ic)
3067 {
3068         unsigned sector_space = JOURNAL_SECTOR_DATA;
3069
3070         ic->journal_sections = le32_to_cpu(ic->sb->journal_sections);
3071         ic->journal_entry_size = roundup(offsetof(struct journal_entry, last_bytes[ic->sectors_per_block]) + ic->tag_size,
3072                                          JOURNAL_ENTRY_ROUNDUP);
3073
3074         if (ic->sb->flags & cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC))
3075                 sector_space -= JOURNAL_MAC_PER_SECTOR;
3076         ic->journal_entries_per_sector = sector_space / ic->journal_entry_size;
3077         ic->journal_section_entries = ic->journal_entries_per_sector * JOURNAL_BLOCK_SECTORS;
3078         ic->journal_section_sectors = (ic->journal_section_entries << ic->sb->log2_sectors_per_block) + JOURNAL_BLOCK_SECTORS;
3079         ic->journal_entries = ic->journal_section_entries * ic->journal_sections;
3080 }
3081
3082 static int calculate_device_limits(struct dm_integrity_c *ic)
3083 {
3084         __u64 initial_sectors;
3085
3086         calculate_journal_section_size(ic);
3087         initial_sectors = SB_SECTORS + (__u64)ic->journal_section_sectors * ic->journal_sections;
3088         if (initial_sectors + METADATA_PADDING_SECTORS >= ic->meta_device_sectors || initial_sectors > UINT_MAX)
3089                 return -EINVAL;
3090         ic->initial_sectors = initial_sectors;
3091
3092         if (!ic->meta_dev) {
3093                 sector_t last_sector, last_area, last_offset;
3094
3095                 ic->metadata_run = roundup((__u64)ic->tag_size << (ic->sb->log2_interleave_sectors - ic->sb->log2_sectors_per_block),
3096                                            (__u64)(1 << SECTOR_SHIFT << METADATA_PADDING_SECTORS)) >> SECTOR_SHIFT;
3097                 if (!(ic->metadata_run & (ic->metadata_run - 1)))
3098                         ic->log2_metadata_run = __ffs(ic->metadata_run);
3099                 else
3100                         ic->log2_metadata_run = -1;
3101
3102                 get_area_and_offset(ic, ic->provided_data_sectors - 1, &last_area, &last_offset);
3103                 last_sector = get_data_sector(ic, last_area, last_offset);
3104                 if (last_sector < ic->start || last_sector >= ic->meta_device_sectors)
3105                         return -EINVAL;
3106         } else {
3107                 __u64 meta_size = (ic->provided_data_sectors >> ic->sb->log2_sectors_per_block) * ic->tag_size;
3108                 meta_size = (meta_size + ((1U << (ic->log2_buffer_sectors + SECTOR_SHIFT)) - 1))
3109                                 >> (ic->log2_buffer_sectors + SECTOR_SHIFT);
3110                 meta_size <<= ic->log2_buffer_sectors;
3111                 if (ic->initial_sectors + meta_size < ic->initial_sectors ||
3112                     ic->initial_sectors + meta_size > ic->meta_device_sectors)
3113                         return -EINVAL;
3114                 ic->metadata_run = 1;
3115                 ic->log2_metadata_run = 0;
3116         }
3117
3118         return 0;
3119 }
3120
3121 static int initialize_superblock(struct dm_integrity_c *ic, unsigned journal_sectors, unsigned interleave_sectors)
3122 {
3123         unsigned journal_sections;
3124         int test_bit;
3125
3126         memset(ic->sb, 0, SB_SECTORS << SECTOR_SHIFT);
3127         memcpy(ic->sb->magic, SB_MAGIC, 8);
3128         ic->sb->integrity_tag_size = cpu_to_le16(ic->tag_size);
3129         ic->sb->log2_sectors_per_block = __ffs(ic->sectors_per_block);
3130         if (ic->journal_mac_alg.alg_string)
3131                 ic->sb->flags |= cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC);
3132
3133         calculate_journal_section_size(ic);
3134         journal_sections = journal_sectors / ic->journal_section_sectors;
3135         if (!journal_sections)
3136                 journal_sections = 1;
3137
3138         if (!ic->meta_dev) {
3139                 ic->sb->journal_sections = cpu_to_le32(journal_sections);
3140                 if (!interleave_sectors)
3141                         interleave_sectors = DEFAULT_INTERLEAVE_SECTORS;
3142                 ic->sb->log2_interleave_sectors = __fls(interleave_sectors);
3143                 ic->sb->log2_interleave_sectors = max((__u8)MIN_LOG2_INTERLEAVE_SECTORS, ic->sb->log2_interleave_sectors);
3144                 ic->sb->log2_interleave_sectors = min((__u8)MAX_LOG2_INTERLEAVE_SECTORS, ic->sb->log2_interleave_sectors);
3145
3146                 ic->provided_data_sectors = 0;
3147                 for (test_bit = fls64(ic->meta_device_sectors) - 1; test_bit >= 3; test_bit--) {
3148                         __u64 prev_data_sectors = ic->provided_data_sectors;
3149
3150                         ic->provided_data_sectors |= (sector_t)1 << test_bit;
3151                         if (calculate_device_limits(ic))
3152                                 ic->provided_data_sectors = prev_data_sectors;
3153                 }
3154                 if (!ic->provided_data_sectors)
3155                         return -EINVAL;
3156         } else {
3157                 ic->sb->log2_interleave_sectors = 0;
3158                 ic->provided_data_sectors = ic->data_device_sectors;
3159                 ic->provided_data_sectors &= ~(sector_t)(ic->sectors_per_block - 1);
3160
3161 try_smaller_buffer:
3162                 ic->sb->journal_sections = cpu_to_le32(0);
3163                 for (test_bit = fls(journal_sections) - 1; test_bit >= 0; test_bit--) {
3164                         __u32 prev_journal_sections = le32_to_cpu(ic->sb->journal_sections);
3165                         __u32 test_journal_sections = prev_journal_sections | (1U << test_bit);
3166                         if (test_journal_sections > journal_sections)
3167                                 continue;
3168                         ic->sb->journal_sections = cpu_to_le32(test_journal_sections);
3169                         if (calculate_device_limits(ic))
3170                                 ic->sb->journal_sections = cpu_to_le32(prev_journal_sections);
3171
3172                 }
3173                 if (!le32_to_cpu(ic->sb->journal_sections)) {
3174                         if (ic->log2_buffer_sectors > 3) {
3175                                 ic->log2_buffer_sectors--;
3176                                 goto try_smaller_buffer;
3177                         }
3178                         return -EINVAL;
3179                 }
3180         }
3181
3182         ic->sb->provided_data_sectors = cpu_to_le64(ic->provided_data_sectors);
3183
3184         sb_set_version(ic);
3185
3186         return 0;
3187 }
3188
3189 static void dm_integrity_set(struct dm_target *ti, struct dm_integrity_c *ic)
3190 {
3191         struct gendisk *disk = dm_disk(dm_table_get_md(ti->table));
3192         struct blk_integrity bi;
3193
3194         memset(&bi, 0, sizeof(bi));
3195         bi.profile = &dm_integrity_profile;
3196         bi.tuple_size = ic->tag_size;
3197         bi.tag_size = bi.tuple_size;
3198         bi.interval_exp = ic->sb->log2_sectors_per_block + SECTOR_SHIFT;
3199
3200         blk_integrity_register(disk, &bi);
3201         blk_queue_max_integrity_segments(disk->queue, UINT_MAX);
3202 }
3203
3204 static void dm_integrity_free_page_list(struct page_list *pl)
3205 {
3206         unsigned i;
3207
3208         if (!pl)
3209                 return;
3210         for (i = 0; pl[i].page; i++)
3211                 __free_page(pl[i].page);
3212         kvfree(pl);
3213 }
3214
3215 static struct page_list *dm_integrity_alloc_page_list(unsigned n_pages)
3216 {
3217         struct page_list *pl;
3218         unsigned i;
3219
3220         pl = kvmalloc_array(n_pages + 1, sizeof(struct page_list), GFP_KERNEL | __GFP_ZERO);
3221         if (!pl)
3222                 return NULL;
3223
3224         for (i = 0; i < n_pages; i++) {
3225                 pl[i].page = alloc_page(GFP_KERNEL);
3226                 if (!pl[i].page) {
3227                         dm_integrity_free_page_list(pl);
3228                         return NULL;
3229                 }
3230                 if (i)
3231                         pl[i - 1].next = &pl[i];
3232         }
3233         pl[i].page = NULL;
3234         pl[i].next = NULL;
3235
3236         return pl;
3237 }
3238
3239 static void dm_integrity_free_journal_scatterlist(struct dm_integrity_c *ic, struct scatterlist **sl)
3240 {
3241         unsigned i;
3242         for (i = 0; i < ic->journal_sections; i++)
3243                 kvfree(sl[i]);
3244         kvfree(sl);
3245 }
3246
3247 static struct scatterlist **dm_integrity_alloc_journal_scatterlist(struct dm_integrity_c *ic,
3248                                                                    struct page_list *pl)
3249 {
3250         struct scatterlist **sl;
3251         unsigned i;
3252
3253         sl = kvmalloc_array(ic->journal_sections,
3254                             sizeof(struct scatterlist *),
3255                             GFP_KERNEL | __GFP_ZERO);
3256         if (!sl)
3257                 return NULL;
3258
3259         for (i = 0; i < ic->journal_sections; i++) {
3260                 struct scatterlist *s;
3261                 unsigned start_index, start_offset;
3262                 unsigned end_index, end_offset;
3263                 unsigned n_pages;
3264                 unsigned idx;
3265
3266                 page_list_location(ic, i, 0, &start_index, &start_offset);
3267                 page_list_location(ic, i, ic->journal_section_sectors - 1,
3268                                    &end_index, &end_offset);
3269
3270                 n_pages = (end_index - start_index + 1);
3271
3272                 s = kvmalloc_array(n_pages, sizeof(struct scatterlist),
3273                                    GFP_KERNEL);
3274                 if (!s) {
3275                         dm_integrity_free_journal_scatterlist(ic, sl);
3276                         return NULL;
3277                 }
3278
3279                 sg_init_table(s, n_pages);
3280                 for (idx = start_index; idx <= end_index; idx++) {
3281                         char *va = lowmem_page_address(pl[idx].page);
3282                         unsigned start = 0, end = PAGE_SIZE;
3283                         if (idx == start_index)
3284                                 start = start_offset;
3285                         if (idx == end_index)
3286                                 end = end_offset + (1 << SECTOR_SHIFT);
3287                         sg_set_buf(&s[idx - start_index], va + start, end - start);
3288                 }
3289
3290                 sl[i] = s;
3291         }
3292
3293         return sl;
3294 }
3295
3296 static void free_alg(struct alg_spec *a)
3297 {
3298         kzfree(a->alg_string);
3299         kzfree(a->key);
3300         memset(a, 0, sizeof *a);
3301 }
3302
3303 static int get_alg_and_key(const char *arg, struct alg_spec *a, char **error, char *error_inval)
3304 {
3305         char *k;
3306
3307         free_alg(a);
3308
3309         a->alg_string = kstrdup(strchr(arg, ':') + 1, GFP_KERNEL);
3310         if (!a->alg_string)
3311                 goto nomem;
3312
3313         k = strchr(a->alg_string, ':');
3314         if (k) {
3315                 *k = 0;
3316                 a->key_string = k + 1;
3317                 if (strlen(a->key_string) & 1)
3318                         goto inval;
3319
3320                 a->key_size = strlen(a->key_string) / 2;
3321                 a->key = kmalloc(a->key_size, GFP_KERNEL);
3322                 if (!a->key)
3323                         goto nomem;
3324                 if (hex2bin(a->key, a->key_string, a->key_size))
3325                         goto inval;
3326         }
3327
3328         return 0;
3329 inval:
3330         *error = error_inval;
3331         return -EINVAL;
3332 nomem:
3333         *error = "Out of memory for an argument";
3334         return -ENOMEM;
3335 }
3336
3337 static int get_mac(struct crypto_shash **hash, struct alg_spec *a, char **error,
3338                    char *error_alg, char *error_key)
3339 {
3340         int r;
3341
3342         if (a->alg_string) {
3343                 *hash = crypto_alloc_shash(a->alg_string, 0, 0);
3344                 if (IS_ERR(*hash)) {
3345                         *error = error_alg;
3346                         r = PTR_ERR(*hash);
3347                         *hash = NULL;
3348                         return r;
3349                 }
3350
3351                 if (a->key) {
3352                         r = crypto_shash_setkey(*hash, a->key, a->key_size);
3353                         if (r) {
3354                                 *error = error_key;
3355                                 return r;
3356                         }
3357                 } else if (crypto_shash_get_flags(*hash) & CRYPTO_TFM_NEED_KEY) {
3358                         *error = error_key;
3359                         return -ENOKEY;
3360                 }
3361         }
3362
3363         return 0;
3364 }
3365
3366 static int create_journal(struct dm_integrity_c *ic, char **error)
3367 {
3368         int r = 0;
3369         unsigned i;
3370         __u64 journal_pages, journal_desc_size, journal_tree_size;
3371         unsigned char *crypt_data = NULL, *crypt_iv = NULL;
3372         struct skcipher_request *req = NULL;
3373
3374         ic->commit_ids[0] = cpu_to_le64(0x1111111111111111ULL);
3375         ic->commit_ids[1] = cpu_to_le64(0x2222222222222222ULL);
3376         ic->commit_ids[2] = cpu_to_le64(0x3333333333333333ULL);
3377         ic->commit_ids[3] = cpu_to_le64(0x4444444444444444ULL);
3378
3379         journal_pages = roundup((__u64)ic->journal_sections * ic->journal_section_sectors,
3380                                 PAGE_SIZE >> SECTOR_SHIFT) >> (PAGE_SHIFT - SECTOR_SHIFT);
3381         journal_desc_size = journal_pages * sizeof(struct page_list);
3382         if (journal_pages >= totalram_pages() - totalhigh_pages() || journal_desc_size > ULONG_MAX) {
3383                 *error = "Journal doesn't fit into memory";
3384                 r = -ENOMEM;
3385                 goto bad;
3386         }
3387         ic->journal_pages = journal_pages;
3388
3389         ic->journal = dm_integrity_alloc_page_list(ic->journal_pages);
3390         if (!ic->journal) {
3391                 *error = "Could not allocate memory for journal";
3392                 r = -ENOMEM;
3393                 goto bad;
3394         }
3395         if (ic->journal_crypt_alg.alg_string) {
3396                 unsigned ivsize, blocksize;
3397                 struct journal_completion comp;
3398
3399                 comp.ic = ic;
3400                 ic->journal_crypt = crypto_alloc_skcipher(ic->journal_crypt_alg.alg_string, 0, 0);
3401                 if (IS_ERR(ic->journal_crypt)) {
3402                         *error = "Invalid journal cipher";
3403                         r = PTR_ERR(ic->journal_crypt);
3404                         ic->journal_crypt = NULL;
3405                         goto bad;
3406                 }
3407                 ivsize = crypto_skcipher_ivsize(ic->journal_crypt);
3408                 blocksize = crypto_skcipher_blocksize(ic->journal_crypt);
3409
3410                 if (ic->journal_crypt_alg.key) {
3411                         r = crypto_skcipher_setkey(ic->journal_crypt, ic->journal_crypt_alg.key,
3412                                                    ic->journal_crypt_alg.key_size);
3413                         if (r) {
3414                                 *error = "Error setting encryption key";
3415                                 goto bad;
3416                         }
3417                 }
3418                 DEBUG_print("cipher %s, block size %u iv size %u\n",
3419                             ic->journal_crypt_alg.alg_string, blocksize, ivsize);
3420
3421                 ic->journal_io = dm_integrity_alloc_page_list(ic->journal_pages);
3422                 if (!ic->journal_io) {
3423                         *error = "Could not allocate memory for journal io";
3424                         r = -ENOMEM;
3425                         goto bad;
3426                 }
3427
3428                 if (blocksize == 1) {
3429                         struct scatterlist *sg;
3430
3431                         req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
3432                         if (!req) {
3433                                 *error = "Could not allocate crypt request";
3434                                 r = -ENOMEM;
3435                                 goto bad;
3436                         }
3437
3438                         crypt_iv = kzalloc(ivsize, GFP_KERNEL);
3439                         if (!crypt_iv) {
3440                                 *error = "Could not allocate iv";
3441                                 r = -ENOMEM;
3442                                 goto bad;
3443                         }
3444
3445                         ic->journal_xor = dm_integrity_alloc_page_list(ic->journal_pages);
3446                         if (!ic->journal_xor) {
3447                                 *error = "Could not allocate memory for journal xor";
3448                                 r = -ENOMEM;
3449                                 goto bad;
3450                         }
3451
3452                         sg = kvmalloc_array(ic->journal_pages + 1,
3453                                             sizeof(struct scatterlist),
3454                                             GFP_KERNEL);
3455                         if (!sg) {
3456                                 *error = "Unable to allocate sg list";
3457                                 r = -ENOMEM;
3458                                 goto bad;
3459                         }
3460                         sg_init_table(sg, ic->journal_pages + 1);
3461                         for (i = 0; i < ic->journal_pages; i++) {
3462                                 char *va = lowmem_page_address(ic->journal_xor[i].page);
3463                                 clear_page(va);
3464                                 sg_set_buf(&sg[i], va, PAGE_SIZE);
3465                         }
3466                         sg_set_buf(&sg[i], &ic->commit_ids, sizeof ic->commit_ids);
3467
3468                         skcipher_request_set_crypt(req, sg, sg,
3469                                                    PAGE_SIZE * ic->journal_pages + sizeof ic->commit_ids, crypt_iv);
3470                         init_completion(&comp.comp);
3471                         comp.in_flight = (atomic_t)ATOMIC_INIT(1);
3472                         if (do_crypt(true, req, &comp))
3473                                 wait_for_completion(&comp.comp);
3474                         kvfree(sg);
3475                         r = dm_integrity_failed(ic);
3476                         if (r) {
3477                                 *error = "Unable to encrypt journal";
3478                                 goto bad;
3479                         }
3480                         DEBUG_bytes(lowmem_page_address(ic->journal_xor[0].page), 64, "xor data");
3481
3482                         crypto_free_skcipher(ic->journal_crypt);
3483                         ic->journal_crypt = NULL;
3484                 } else {
3485                         unsigned crypt_len = roundup(ivsize, blocksize);
3486
3487                         req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
3488                         if (!req) {
3489                                 *error = "Could not allocate crypt request";
3490                                 r = -ENOMEM;
3491                                 goto bad;
3492                         }
3493
3494                         crypt_iv = kmalloc(ivsize, GFP_KERNEL);
3495                         if (!crypt_iv) {
3496                                 *error = "Could not allocate iv";
3497                                 r = -ENOMEM;
3498                                 goto bad;
3499                         }
3500
3501                         crypt_data = kmalloc(crypt_len, GFP_KERNEL);
3502                         if (!crypt_data) {
3503                                 *error = "Unable to allocate crypt data";
3504                                 r = -ENOMEM;
3505                                 goto bad;
3506                         }
3507
3508                         ic->journal_scatterlist = dm_integrity_alloc_journal_scatterlist(ic, ic->journal);
3509                         if (!ic->journal_scatterlist) {
3510                                 *error = "Unable to allocate sg list";
3511                                 r = -ENOMEM;
3512                                 goto bad;
3513                         }
3514                         ic->journal_io_scatterlist = dm_integrity_alloc_journal_scatterlist(ic, ic->journal_io);
3515                         if (!ic->journal_io_scatterlist) {
3516                                 *error = "Unable to allocate sg list";
3517                                 r = -ENOMEM;
3518                                 goto bad;
3519                         }
3520                         ic->sk_requests = kvmalloc_array(ic->journal_sections,
3521                                                          sizeof(struct skcipher_request *),
3522                                                          GFP_KERNEL | __GFP_ZERO);
3523                         if (!ic->sk_requests) {
3524                                 *error = "Unable to allocate sk requests";
3525                                 r = -ENOMEM;
3526                                 goto bad;
3527                         }
3528                         for (i = 0; i < ic->journal_sections; i++) {
3529                                 struct scatterlist sg;
3530                                 struct skcipher_request *section_req;
3531                                 __u32 section_le = cpu_to_le32(i);
3532
3533                                 memset(crypt_iv, 0x00, ivsize);
3534                                 memset(crypt_data, 0x00, crypt_len);
3535                                 memcpy(crypt_data, &section_le, min((size_t)crypt_len, sizeof(section_le)));
3536
3537                                 sg_init_one(&sg, crypt_data, crypt_len);
3538                                 skcipher_request_set_crypt(req, &sg, &sg, crypt_len, crypt_iv);
3539                                 init_completion(&comp.comp);
3540                                 comp.in_flight = (atomic_t)ATOMIC_INIT(1);
3541                                 if (do_crypt(true, req, &comp))
3542                                         wait_for_completion(&comp.comp);
3543
3544                                 r = dm_integrity_failed(ic);
3545                                 if (r) {
3546                                         *error = "Unable to generate iv";
3547                                         goto bad;
3548                                 }
3549
3550                                 section_req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
3551                                 if (!section_req) {
3552                                         *error = "Unable to allocate crypt request";
3553                                         r = -ENOMEM;
3554                                         goto bad;
3555                                 }
3556                                 section_req->iv = kmalloc_array(ivsize, 2,
3557                                                                 GFP_KERNEL);
3558                                 if (!section_req->iv) {
3559                                         skcipher_request_free(section_req);
3560                                         *error = "Unable to allocate iv";
3561                                         r = -ENOMEM;
3562                                         goto bad;
3563                                 }
3564                                 memcpy(section_req->iv + ivsize, crypt_data, ivsize);
3565                                 section_req->cryptlen = (size_t)ic->journal_section_sectors << SECTOR_SHIFT;
3566                                 ic->sk_requests[i] = section_req;
3567                                 DEBUG_bytes(crypt_data, ivsize, "iv(%u)", i);
3568                         }
3569                 }
3570         }
3571
3572         for (i = 0; i < N_COMMIT_IDS; i++) {
3573                 unsigned j;
3574 retest_commit_id:
3575                 for (j = 0; j < i; j++) {
3576                         if (ic->commit_ids[j] == ic->commit_ids[i]) {
3577                                 ic->commit_ids[i] = cpu_to_le64(le64_to_cpu(ic->commit_ids[i]) + 1);
3578                                 goto retest_commit_id;
3579                         }
3580                 }
3581                 DEBUG_print("commit id %u: %016llx\n", i, ic->commit_ids[i]);
3582         }
3583
3584         journal_tree_size = (__u64)ic->journal_entries * sizeof(struct journal_node);
3585         if (journal_tree_size > ULONG_MAX) {
3586                 *error = "Journal doesn't fit into memory";
3587                 r = -ENOMEM;
3588                 goto bad;
3589         }
3590         ic->journal_tree = kvmalloc(journal_tree_size, GFP_KERNEL);
3591         if (!ic->journal_tree) {
3592                 *error = "Could not allocate memory for journal tree";
3593                 r = -ENOMEM;
3594         }
3595 bad:
3596         kfree(crypt_data);
3597         kfree(crypt_iv);
3598         skcipher_request_free(req);
3599
3600         return r;
3601 }
3602
3603 /*
3604  * Construct a integrity mapping
3605  *
3606  * Arguments:
3607  *      device
3608  *      offset from the start of the device
3609  *      tag size
3610  *      D - direct writes, J - journal writes, B - bitmap mode, R - recovery mode
3611  *      number of optional arguments
3612  *      optional arguments:
3613  *              journal_sectors
3614  *              interleave_sectors
3615  *              buffer_sectors
3616  *              journal_watermark
3617  *              commit_time
3618  *              meta_device
3619  *              block_size
3620  *              sectors_per_bit
3621  *              bitmap_flush_interval
3622  *              internal_hash
3623  *              journal_crypt
3624  *              journal_mac
3625  *              recalculate
3626  */
3627 static int dm_integrity_ctr(struct dm_target *ti, unsigned argc, char **argv)
3628 {
3629         struct dm_integrity_c *ic;
3630         char dummy;
3631         int r;
3632         unsigned extra_args;
3633         struct dm_arg_set as;
3634         static const struct dm_arg _args[] = {
3635                 {0, 14, "Invalid number of feature args"},
3636         };
3637         unsigned journal_sectors, interleave_sectors, buffer_sectors, journal_watermark, sync_msec;
3638         bool should_write_sb;
3639         __u64 threshold;
3640         unsigned long long start;
3641         __s8 log2_sectors_per_bitmap_bit = -1;
3642         __s8 log2_blocks_per_bitmap_bit;
3643         __u64 bits_in_journal;
3644         __u64 n_bitmap_bits;
3645
3646 #define DIRECT_ARGUMENTS        4
3647
3648         if (argc <= DIRECT_ARGUMENTS) {
3649                 ti->error = "Invalid argument count";
3650                 return -EINVAL;
3651         }
3652
3653         ic = kzalloc(sizeof(struct dm_integrity_c), GFP_KERNEL);
3654         if (!ic) {
3655                 ti->error = "Cannot allocate integrity context";
3656                 return -ENOMEM;
3657         }
3658         ti->private = ic;
3659         ti->per_io_data_size = sizeof(struct dm_integrity_io);
3660         ic->ti = ti;
3661
3662         ic->in_progress = RB_ROOT;
3663         INIT_LIST_HEAD(&ic->wait_list);
3664         init_waitqueue_head(&ic->endio_wait);
3665         bio_list_init(&ic->flush_bio_list);
3666         init_waitqueue_head(&ic->copy_to_journal_wait);
3667         init_completion(&ic->crypto_backoff);
3668         atomic64_set(&ic->number_of_mismatches, 0);
3669         ic->bitmap_flush_interval = BITMAP_FLUSH_INTERVAL;
3670
3671         r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &ic->dev);
3672         if (r) {
3673                 ti->error = "Device lookup failed";
3674                 goto bad;
3675         }
3676
3677         if (sscanf(argv[1], "%llu%c", &start, &dummy) != 1 || start != (sector_t)start) {
3678                 ti->error = "Invalid starting offset";
3679                 r = -EINVAL;
3680                 goto bad;
3681         }
3682         ic->start = start;
3683
3684         if (strcmp(argv[2], "-")) {
3685                 if (sscanf(argv[2], "%u%c", &ic->tag_size, &dummy) != 1 || !ic->tag_size) {
3686                         ti->error = "Invalid tag size";
3687                         r = -EINVAL;
3688                         goto bad;
3689                 }
3690         }
3691
3692         if (!strcmp(argv[3], "J") || !strcmp(argv[3], "B") ||
3693             !strcmp(argv[3], "D") || !strcmp(argv[3], "R")) {
3694                 ic->mode = argv[3][0];
3695         } else {
3696                 ti->error = "Invalid mode (expecting J, B, D, R)";
3697                 r = -EINVAL;
3698                 goto bad;
3699         }
3700
3701         journal_sectors = 0;
3702         interleave_sectors = DEFAULT_INTERLEAVE_SECTORS;
3703         buffer_sectors = DEFAULT_BUFFER_SECTORS;
3704         journal_watermark = DEFAULT_JOURNAL_WATERMARK;
3705         sync_msec = DEFAULT_SYNC_MSEC;
3706         ic->sectors_per_block = 1;
3707
3708         as.argc = argc - DIRECT_ARGUMENTS;
3709         as.argv = argv + DIRECT_ARGUMENTS;
3710         r = dm_read_arg_group(_args, &as, &extra_args, &ti->error);
3711         if (r)
3712                 goto bad;
3713
3714         while (extra_args--) {
3715                 const char *opt_string;
3716                 unsigned val;
3717                 unsigned long long llval;
3718                 opt_string = dm_shift_arg(&as);
3719                 if (!opt_string) {
3720                         r = -EINVAL;
3721                         ti->error = "Not enough feature arguments";
3722                         goto bad;
3723                 }
3724                 if (sscanf(opt_string, "journal_sectors:%u%c", &val, &dummy) == 1)
3725                         journal_sectors = val ? val : 1;
3726                 else if (sscanf(opt_string, "interleave_sectors:%u%c", &val, &dummy) == 1)
3727                         interleave_sectors = val;
3728                 else if (sscanf(opt_string, "buffer_sectors:%u%c", &val, &dummy) == 1)
3729                         buffer_sectors = val;
3730                 else if (sscanf(opt_string, "journal_watermark:%u%c", &val, &dummy) == 1 && val <= 100)
3731                         journal_watermark = val;
3732                 else if (sscanf(opt_string, "commit_time:%u%c", &val, &dummy) == 1)
3733                         sync_msec = val;
3734                 else if (!strncmp(opt_string, "meta_device:", strlen("meta_device:"))) {
3735                         if (ic->meta_dev) {
3736                                 dm_put_device(ti, ic->meta_dev);
3737                                 ic->meta_dev = NULL;
3738                         }
3739                         r = dm_get_device(ti, strchr(opt_string, ':') + 1,
3740                                           dm_table_get_mode(ti->table), &ic->meta_dev);
3741                         if (r) {
3742                                 ti->error = "Device lookup failed";
3743                                 goto bad;
3744                         }
3745                 } else if (sscanf(opt_string, "block_size:%u%c", &val, &dummy) == 1) {
3746                         if (val < 1 << SECTOR_SHIFT ||
3747                             val > MAX_SECTORS_PER_BLOCK << SECTOR_SHIFT ||
3748                             (val & (val -1))) {
3749                                 r = -EINVAL;
3750                                 ti->error = "Invalid block_size argument";
3751                                 goto bad;
3752                         }
3753                         ic->sectors_per_block = val >> SECTOR_SHIFT;
3754                 } else if (sscanf(opt_string, "sectors_per_bit:%llu%c", &llval, &dummy) == 1) {
3755                         log2_sectors_per_bitmap_bit = !llval ? 0 : __ilog2_u64(llval);
3756                 } else if (sscanf(opt_string, "bitmap_flush_interval:%u%c", &val, &dummy) == 1) {
3757                         if (val >= (uint64_t)UINT_MAX * 1000 / HZ) {
3758                                 r = -EINVAL;
3759                                 ti->error = "Invalid bitmap_flush_interval argument";
3760                                 goto bad;
3761                         }
3762                         ic->bitmap_flush_interval = msecs_to_jiffies(val);
3763                 } else if (!strncmp(opt_string, "internal_hash:", strlen("internal_hash:"))) {
3764                         r = get_alg_and_key(opt_string, &ic->internal_hash_alg, &ti->error,
3765                                             "Invalid internal_hash argument");
3766                         if (r)
3767                                 goto bad;
3768                 } else if (!strncmp(opt_string, "journal_crypt:", strlen("journal_crypt:"))) {
3769                         r = get_alg_and_key(opt_string, &ic->journal_crypt_alg, &ti->error,
3770                                             "Invalid journal_crypt argument");
3771                         if (r)
3772                                 goto bad;
3773                 } else if (!strncmp(opt_string, "journal_mac:", strlen("journal_mac:"))) {
3774                         r = get_alg_and_key(opt_string, &ic->journal_mac_alg,  &ti->error,
3775                                             "Invalid journal_mac argument");
3776                         if (r)
3777                                 goto bad;
3778                 } else if (!strcmp(opt_string, "recalculate")) {
3779                         ic->recalculate_flag = true;
3780                 } else if (!strcmp(opt_string, "legacy_recalculate")) {
3781                         ic->legacy_recalculate = true;
3782                 } else {
3783                         r = -EINVAL;
3784                         ti->error = "Invalid argument";
3785                         goto bad;
3786                 }
3787         }
3788
3789         ic->data_device_sectors = i_size_read(ic->dev->bdev->bd_inode) >> SECTOR_SHIFT;
3790         if (!ic->meta_dev)
3791                 ic->meta_device_sectors = ic->data_device_sectors;
3792         else
3793                 ic->meta_device_sectors = i_size_read(ic->meta_dev->bdev->bd_inode) >> SECTOR_SHIFT;
3794
3795         if (!journal_sectors) {
3796                 journal_sectors = min((sector_t)DEFAULT_MAX_JOURNAL_SECTORS,
3797                                       ic->data_device_sectors >> DEFAULT_JOURNAL_SIZE_FACTOR);
3798         }
3799
3800         if (!buffer_sectors)
3801                 buffer_sectors = 1;
3802         ic->log2_buffer_sectors = min((int)__fls(buffer_sectors), 31 - SECTOR_SHIFT);
3803
3804         r = get_mac(&ic->internal_hash, &ic->internal_hash_alg, &ti->error,
3805                     "Invalid internal hash", "Error setting internal hash key");
3806         if (r)
3807                 goto bad;
3808
3809         r = get_mac(&ic->journal_mac, &ic->journal_mac_alg, &ti->error,
3810                     "Invalid journal mac", "Error setting journal mac key");
3811         if (r)
3812                 goto bad;
3813
3814         if (!ic->tag_size) {
3815                 if (!ic->internal_hash) {
3816                         ti->error = "Unknown tag size";
3817                         r = -EINVAL;
3818                         goto bad;
3819                 }
3820                 ic->tag_size = crypto_shash_digestsize(ic->internal_hash);
3821         }
3822         if (ic->tag_size > MAX_TAG_SIZE) {
3823                 ti->error = "Too big tag size";
3824                 r = -EINVAL;
3825                 goto bad;
3826         }
3827         if (!(ic->tag_size & (ic->tag_size - 1)))
3828                 ic->log2_tag_size = __ffs(ic->tag_size);
3829         else
3830                 ic->log2_tag_size = -1;
3831
3832         if (ic->mode == 'B' && !ic->internal_hash) {
3833                 r = -EINVAL;
3834                 ti->error = "Bitmap mode can be only used with internal hash";
3835                 goto bad;
3836         }
3837
3838         ic->autocommit_jiffies = msecs_to_jiffies(sync_msec);
3839         ic->autocommit_msec = sync_msec;
3840         timer_setup(&ic->autocommit_timer, autocommit_fn, 0);
3841
3842         ic->io = dm_io_client_create();
3843         if (IS_ERR(ic->io)) {
3844                 r = PTR_ERR(ic->io);
3845                 ic->io = NULL;
3846                 ti->error = "Cannot allocate dm io";
3847                 goto bad;
3848         }
3849
3850         r = mempool_init_slab_pool(&ic->journal_io_mempool, JOURNAL_IO_MEMPOOL, journal_io_cache);
3851         if (r) {
3852                 ti->error = "Cannot allocate mempool";
3853                 goto bad;
3854         }
3855
3856         ic->metadata_wq = alloc_workqueue("dm-integrity-metadata",
3857                                           WQ_MEM_RECLAIM, METADATA_WORKQUEUE_MAX_ACTIVE);
3858         if (!ic->metadata_wq) {
3859                 ti->error = "Cannot allocate workqueue";
3860                 r = -ENOMEM;
3861                 goto bad;
3862         }
3863
3864         /*
3865          * If this workqueue were percpu, it would cause bio reordering
3866          * and reduced performance.
3867          */
3868         ic->wait_wq = alloc_workqueue("dm-integrity-wait", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3869         if (!ic->wait_wq) {
3870                 ti->error = "Cannot allocate workqueue";
3871                 r = -ENOMEM;
3872                 goto bad;
3873         }
3874
3875         ic->offload_wq = alloc_workqueue("dm-integrity-offload", WQ_MEM_RECLAIM,
3876                                           METADATA_WORKQUEUE_MAX_ACTIVE);
3877         if (!ic->offload_wq) {
3878                 ti->error = "Cannot allocate workqueue";
3879                 r = -ENOMEM;
3880                 goto bad;
3881         }
3882
3883         ic->commit_wq = alloc_workqueue("dm-integrity-commit", WQ_MEM_RECLAIM, 1);
3884         if (!ic->commit_wq) {
3885                 ti->error = "Cannot allocate workqueue";
3886                 r = -ENOMEM;
3887                 goto bad;
3888         }
3889         INIT_WORK(&ic->commit_work, integrity_commit);
3890
3891         if (ic->mode == 'J' || ic->mode == 'B') {
3892                 ic->writer_wq = alloc_workqueue("dm-integrity-writer", WQ_MEM_RECLAIM, 1);
3893                 if (!ic->writer_wq) {
3894                         ti->error = "Cannot allocate workqueue";
3895                         r = -ENOMEM;
3896                         goto bad;
3897                 }
3898                 INIT_WORK(&ic->writer_work, integrity_writer);
3899         }
3900
3901         ic->sb = alloc_pages_exact(SB_SECTORS << SECTOR_SHIFT, GFP_KERNEL);
3902         if (!ic->sb) {
3903                 r = -ENOMEM;
3904                 ti->error = "Cannot allocate superblock area";
3905                 goto bad;
3906         }
3907
3908         r = sync_rw_sb(ic, REQ_OP_READ, 0);
3909         if (r) {
3910                 ti->error = "Error reading superblock";
3911                 goto bad;
3912         }
3913         should_write_sb = false;
3914         if (memcmp(ic->sb->magic, SB_MAGIC, 8)) {
3915                 if (ic->mode != 'R') {
3916                         if (memchr_inv(ic->sb, 0, SB_SECTORS << SECTOR_SHIFT)) {
3917                                 r = -EINVAL;
3918                                 ti->error = "The device is not initialized";
3919                                 goto bad;
3920                         }
3921                 }
3922
3923                 r = initialize_superblock(ic, journal_sectors, interleave_sectors);
3924                 if (r) {
3925                         ti->error = "Could not initialize superblock";
3926                         goto bad;
3927                 }
3928                 if (ic->mode != 'R')
3929                         should_write_sb = true;
3930         }
3931
3932         if (!ic->sb->version || ic->sb->version > SB_VERSION_3) {
3933                 r = -EINVAL;
3934                 ti->error = "Unknown version";
3935                 goto bad;
3936         }
3937         if (le16_to_cpu(ic->sb->integrity_tag_size) != ic->tag_size) {
3938                 r = -EINVAL;
3939                 ti->error = "Tag size doesn't match the information in superblock";
3940                 goto bad;
3941         }
3942         if (ic->sb->log2_sectors_per_block != __ffs(ic->sectors_per_block)) {
3943                 r = -EINVAL;
3944                 ti->error = "Block size doesn't match the information in superblock";
3945                 goto bad;
3946         }
3947         if (!le32_to_cpu(ic->sb->journal_sections)) {
3948                 r = -EINVAL;
3949                 ti->error = "Corrupted superblock, journal_sections is 0";
3950                 goto bad;
3951         }
3952         /* make sure that ti->max_io_len doesn't overflow */
3953         if (!ic->meta_dev) {
3954                 if (ic->sb->log2_interleave_sectors < MIN_LOG2_INTERLEAVE_SECTORS ||
3955                     ic->sb->log2_interleave_sectors > MAX_LOG2_INTERLEAVE_SECTORS) {
3956                         r = -EINVAL;
3957                         ti->error = "Invalid interleave_sectors in the superblock";
3958                         goto bad;
3959                 }
3960         } else {
3961                 if (ic->sb->log2_interleave_sectors) {
3962                         r = -EINVAL;
3963                         ti->error = "Invalid interleave_sectors in the superblock";
3964                         goto bad;
3965                 }
3966         }
3967         ic->provided_data_sectors = le64_to_cpu(ic->sb->provided_data_sectors);
3968         if (ic->provided_data_sectors != le64_to_cpu(ic->sb->provided_data_sectors)) {
3969                 /* test for overflow */
3970                 r = -EINVAL;
3971                 ti->error = "The superblock has 64-bit device size, but the kernel was compiled with 32-bit sectors";
3972                 goto bad;
3973         }
3974         if (!!(ic->sb->flags & cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC)) != !!ic->journal_mac_alg.alg_string) {
3975                 r = -EINVAL;
3976                 ti->error = "Journal mac mismatch";
3977                 goto bad;
3978         }
3979
3980 try_smaller_buffer:
3981         r = calculate_device_limits(ic);
3982         if (r) {
3983                 if (ic->meta_dev) {
3984                         if (ic->log2_buffer_sectors > 3) {
3985                                 ic->log2_buffer_sectors--;
3986                                 goto try_smaller_buffer;
3987                         }
3988                 }
3989                 ti->error = "The device is too small";
3990                 goto bad;
3991         }
3992
3993         if (log2_sectors_per_bitmap_bit < 0)
3994                 log2_sectors_per_bitmap_bit = __fls(DEFAULT_SECTORS_PER_BITMAP_BIT);
3995         if (log2_sectors_per_bitmap_bit < ic->sb->log2_sectors_per_block)
3996                 log2_sectors_per_bitmap_bit = ic->sb->log2_sectors_per_block;
3997
3998         bits_in_journal = ((__u64)ic->journal_section_sectors * ic->journal_sections) << (SECTOR_SHIFT + 3);
3999         if (bits_in_journal > UINT_MAX)
4000                 bits_in_journal = UINT_MAX;
4001         while (bits_in_journal < (ic->provided_data_sectors + ((sector_t)1 << log2_sectors_per_bitmap_bit) - 1) >> log2_sectors_per_bitmap_bit)
4002                 log2_sectors_per_bitmap_bit++;
4003
4004         log2_blocks_per_bitmap_bit = log2_sectors_per_bitmap_bit - ic->sb->log2_sectors_per_block;
4005         ic->log2_blocks_per_bitmap_bit = log2_blocks_per_bitmap_bit;
4006         if (should_write_sb) {
4007                 ic->sb->log2_blocks_per_bitmap_bit = log2_blocks_per_bitmap_bit;
4008         }
4009         n_bitmap_bits = ((ic->provided_data_sectors >> ic->sb->log2_sectors_per_block)
4010                                 + (((sector_t)1 << log2_blocks_per_bitmap_bit) - 1)) >> log2_blocks_per_bitmap_bit;
4011         ic->n_bitmap_blocks = DIV_ROUND_UP(n_bitmap_bits, BITMAP_BLOCK_SIZE * 8);
4012
4013         if (!ic->meta_dev)
4014                 ic->log2_buffer_sectors = min(ic->log2_buffer_sectors, (__u8)__ffs(ic->metadata_run));
4015
4016         if (ti->len > ic->provided_data_sectors) {
4017                 r = -EINVAL;
4018                 ti->error = "Not enough provided sectors for requested mapping size";
4019                 goto bad;
4020         }
4021
4022
4023         threshold = (__u64)ic->journal_entries * (100 - journal_watermark);
4024         threshold += 50;
4025         do_div(threshold, 100);
4026         ic->free_sectors_threshold = threshold;
4027
4028         DEBUG_print("initialized:\n");
4029         DEBUG_print("   integrity_tag_size %u\n", le16_to_cpu(ic->sb->integrity_tag_size));
4030         DEBUG_print("   journal_entry_size %u\n", ic->journal_entry_size);
4031         DEBUG_print("   journal_entries_per_sector %u\n", ic->journal_entries_per_sector);
4032         DEBUG_print("   journal_section_entries %u\n", ic->journal_section_entries);
4033         DEBUG_print("   journal_section_sectors %u\n", ic->journal_section_sectors);
4034         DEBUG_print("   journal_sections %u\n", (unsigned)le32_to_cpu(ic->sb->journal_sections));
4035         DEBUG_print("   journal_entries %u\n", ic->journal_entries);
4036         DEBUG_print("   log2_interleave_sectors %d\n", ic->sb->log2_interleave_sectors);
4037         DEBUG_print("   data_device_sectors 0x%llx\n", i_size_read(ic->dev->bdev->bd_inode) >> SECTOR_SHIFT);
4038         DEBUG_print("   initial_sectors 0x%x\n", ic->initial_sectors);
4039         DEBUG_print("   metadata_run 0x%x\n", ic->metadata_run);
4040         DEBUG_print("   log2_metadata_run %d\n", ic->log2_metadata_run);
4041         DEBUG_print("   provided_data_sectors 0x%llx (%llu)\n", (unsigned long long)ic->provided_data_sectors,
4042                     (unsigned long long)ic->provided_data_sectors);
4043         DEBUG_print("   log2_buffer_sectors %u\n", ic->log2_buffer_sectors);
4044         DEBUG_print("   bits_in_journal %llu\n", (unsigned long long)bits_in_journal);
4045
4046         if (ic->recalculate_flag && !(ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))) {
4047                 ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
4048                 ic->sb->recalc_sector = cpu_to_le64(0);
4049         }
4050
4051         if (ic->internal_hash) {
4052                 size_t recalc_tags_size;
4053                 ic->recalc_wq = alloc_workqueue("dm-integrity-recalc", WQ_MEM_RECLAIM, 1);
4054                 if (!ic->recalc_wq ) {
4055                         ti->error = "Cannot allocate workqueue";
4056                         r = -ENOMEM;
4057                         goto bad;
4058                 }
4059                 INIT_WORK(&ic->recalc_work, integrity_recalc);
4060                 ic->recalc_buffer = vmalloc(RECALC_SECTORS << SECTOR_SHIFT);
4061                 if (!ic->recalc_buffer) {
4062                         ti->error = "Cannot allocate buffer for recalculating";
4063                         r = -ENOMEM;
4064                         goto bad;
4065                 }
4066                 recalc_tags_size = (RECALC_SECTORS >> ic->sb->log2_sectors_per_block) * ic->tag_size;
4067                 if (crypto_shash_digestsize(ic->internal_hash) > ic->tag_size)
4068                         recalc_tags_size += crypto_shash_digestsize(ic->internal_hash) - ic->tag_size;
4069                 ic->recalc_tags = kvmalloc(recalc_tags_size, GFP_KERNEL);
4070                 if (!ic->recalc_tags) {
4071                         ti->error = "Cannot allocate tags for recalculating";
4072                         r = -ENOMEM;
4073                         goto bad;
4074                 }
4075         } else {
4076                 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
4077                         ti->error = "Recalculate can only be specified with internal_hash";
4078                         r = -EINVAL;
4079                         goto bad;
4080                 }
4081         }
4082
4083         if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING) &&
4084             le64_to_cpu(ic->sb->recalc_sector) < ic->provided_data_sectors &&
4085             dm_integrity_disable_recalculate(ic)) {
4086                 ti->error = "Recalculating with HMAC is disabled for security reasons - if you really need it, use the argument \"legacy_recalculate\"";
4087                 r = -EOPNOTSUPP;
4088                 goto bad;
4089         }
4090
4091         ic->bufio = dm_bufio_client_create(ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev,
4092                         1U << (SECTOR_SHIFT + ic->log2_buffer_sectors), 1, 0, NULL, NULL);
4093         if (IS_ERR(ic->bufio)) {
4094                 r = PTR_ERR(ic->bufio);
4095                 ti->error = "Cannot initialize dm-bufio";
4096                 ic->bufio = NULL;
4097                 goto bad;
4098         }
4099         dm_bufio_set_sector_offset(ic->bufio, ic->start + ic->initial_sectors);
4100
4101         if (ic->mode != 'R') {
4102                 r = create_journal(ic, &ti->error);
4103                 if (r)
4104                         goto bad;
4105
4106         }
4107
4108         if (ic->mode == 'B') {
4109                 unsigned i;
4110                 unsigned n_bitmap_pages = DIV_ROUND_UP(ic->n_bitmap_blocks, PAGE_SIZE / BITMAP_BLOCK_SIZE);
4111
4112                 ic->recalc_bitmap = dm_integrity_alloc_page_list(n_bitmap_pages);
4113                 if (!ic->recalc_bitmap) {
4114                         r = -ENOMEM;
4115                         goto bad;
4116                 }
4117                 ic->may_write_bitmap = dm_integrity_alloc_page_list(n_bitmap_pages);
4118                 if (!ic->may_write_bitmap) {
4119                         r = -ENOMEM;
4120                         goto bad;
4121                 }
4122                 ic->bbs = kvmalloc_array(ic->n_bitmap_blocks, sizeof(struct bitmap_block_status), GFP_KERNEL);
4123                 if (!ic->bbs) {
4124                         r = -ENOMEM;
4125                         goto bad;
4126                 }
4127                 INIT_DELAYED_WORK(&ic->bitmap_flush_work, bitmap_flush_work);
4128                 for (i = 0; i < ic->n_bitmap_blocks; i++) {
4129                         struct bitmap_block_status *bbs = &ic->bbs[i];
4130                         unsigned sector, pl_index, pl_offset;
4131
4132                         INIT_WORK(&bbs->work, bitmap_block_work);
4133                         bbs->ic = ic;
4134                         bbs->idx = i;
4135                         bio_list_init(&bbs->bio_queue);
4136                         spin_lock_init(&bbs->bio_queue_lock);
4137
4138                         sector = i * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT);
4139                         pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
4140                         pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
4141
4142                         bbs->bitmap = lowmem_page_address(ic->journal[pl_index].page) + pl_offset;
4143                 }
4144         }
4145
4146         if (should_write_sb) {
4147                 init_journal(ic, 0, ic->journal_sections, 0);
4148                 r = dm_integrity_failed(ic);
4149                 if (unlikely(r)) {
4150                         ti->error = "Error initializing journal";
4151                         goto bad;
4152                 }
4153                 r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
4154                 if (r) {
4155                         ti->error = "Error initializing superblock";
4156                         goto bad;
4157                 }
4158                 ic->just_formatted = true;
4159         }
4160
4161         if (!ic->meta_dev) {
4162                 r = dm_set_target_max_io_len(ti, 1U << ic->sb->log2_interleave_sectors);
4163                 if (r)
4164                         goto bad;
4165         }
4166         if (ic->mode == 'B') {
4167                 unsigned max_io_len = ((sector_t)ic->sectors_per_block << ic->log2_blocks_per_bitmap_bit) * (BITMAP_BLOCK_SIZE * 8);
4168                 if (!max_io_len)
4169                         max_io_len = 1U << 31;
4170                 DEBUG_print("max_io_len: old %u, new %u\n", ti->max_io_len, max_io_len);
4171                 if (!ti->max_io_len || ti->max_io_len > max_io_len) {
4172                         r = dm_set_target_max_io_len(ti, max_io_len);
4173                         if (r)
4174                                 goto bad;
4175                 }
4176         }
4177
4178         if (!ic->internal_hash)
4179                 dm_integrity_set(ti, ic);
4180
4181         ti->num_flush_bios = 1;
4182         ti->flush_supported = true;
4183
4184         return 0;
4185
4186 bad:
4187         dm_integrity_dtr(ti);
4188         return r;
4189 }
4190
4191 static void dm_integrity_dtr(struct dm_target *ti)
4192 {
4193         struct dm_integrity_c *ic = ti->private;
4194
4195         BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
4196         BUG_ON(!list_empty(&ic->wait_list));
4197
4198         if (ic->mode == 'B')
4199                 cancel_delayed_work_sync(&ic->bitmap_flush_work);
4200         if (ic->metadata_wq)
4201                 destroy_workqueue(ic->metadata_wq);
4202         if (ic->wait_wq)
4203                 destroy_workqueue(ic->wait_wq);
4204         if (ic->offload_wq)
4205                 destroy_workqueue(ic->offload_wq);
4206         if (ic->commit_wq)
4207                 destroy_workqueue(ic->commit_wq);
4208         if (ic->writer_wq)
4209                 destroy_workqueue(ic->writer_wq);
4210         if (ic->recalc_wq)
4211                 destroy_workqueue(ic->recalc_wq);
4212         vfree(ic->recalc_buffer);
4213         kvfree(ic->recalc_tags);
4214         kvfree(ic->bbs);
4215         if (ic->bufio)
4216                 dm_bufio_client_destroy(ic->bufio);
4217         mempool_exit(&ic->journal_io_mempool);
4218         if (ic->io)
4219                 dm_io_client_destroy(ic->io);
4220         if (ic->dev)
4221                 dm_put_device(ti, ic->dev);
4222         if (ic->meta_dev)
4223                 dm_put_device(ti, ic->meta_dev);
4224         dm_integrity_free_page_list(ic->journal);
4225         dm_integrity_free_page_list(ic->journal_io);
4226         dm_integrity_free_page_list(ic->journal_xor);
4227         dm_integrity_free_page_list(ic->recalc_bitmap);
4228         dm_integrity_free_page_list(ic->may_write_bitmap);
4229         if (ic->journal_scatterlist)
4230                 dm_integrity_free_journal_scatterlist(ic, ic->journal_scatterlist);
4231         if (ic->journal_io_scatterlist)
4232                 dm_integrity_free_journal_scatterlist(ic, ic->journal_io_scatterlist);
4233         if (ic->sk_requests) {
4234                 unsigned i;
4235
4236                 for (i = 0; i < ic->journal_sections; i++) {
4237                         struct skcipher_request *req = ic->sk_requests[i];
4238                         if (req) {
4239                                 kzfree(req->iv);
4240                                 skcipher_request_free(req);
4241                         }
4242                 }
4243                 kvfree(ic->sk_requests);
4244         }
4245         kvfree(ic->journal_tree);
4246         if (ic->sb)
4247                 free_pages_exact(ic->sb, SB_SECTORS << SECTOR_SHIFT);
4248
4249         if (ic->internal_hash)
4250                 crypto_free_shash(ic->internal_hash);
4251         free_alg(&ic->internal_hash_alg);
4252
4253         if (ic->journal_crypt)
4254                 crypto_free_skcipher(ic->journal_crypt);
4255         free_alg(&ic->journal_crypt_alg);
4256
4257         if (ic->journal_mac)
4258                 crypto_free_shash(ic->journal_mac);
4259         free_alg(&ic->journal_mac_alg);
4260
4261         kfree(ic);
4262 }
4263
4264 static struct target_type integrity_target = {
4265         .name                   = "integrity",
4266         .version                = {1, 3, 0},
4267         .module                 = THIS_MODULE,
4268         .features               = DM_TARGET_SINGLETON | DM_TARGET_INTEGRITY,
4269         .ctr                    = dm_integrity_ctr,
4270         .dtr                    = dm_integrity_dtr,
4271         .map                    = dm_integrity_map,
4272         .postsuspend            = dm_integrity_postsuspend,
4273         .resume                 = dm_integrity_resume,
4274         .status                 = dm_integrity_status,
4275         .iterate_devices        = dm_integrity_iterate_devices,
4276         .io_hints               = dm_integrity_io_hints,
4277 };
4278
4279 static int __init dm_integrity_init(void)
4280 {
4281         int r;
4282
4283         journal_io_cache = kmem_cache_create("integrity_journal_io",
4284                                              sizeof(struct journal_io), 0, 0, NULL);
4285         if (!journal_io_cache) {
4286                 DMERR("can't allocate journal io cache");
4287                 return -ENOMEM;
4288         }
4289
4290         r = dm_register_target(&integrity_target);
4291
4292         if (r < 0)
4293                 DMERR("register failed %d", r);
4294
4295         return r;
4296 }
4297
4298 static void __exit dm_integrity_exit(void)
4299 {
4300         dm_unregister_target(&integrity_target);
4301         kmem_cache_destroy(journal_io_cache);
4302 }
4303
4304 module_init(dm_integrity_init);
4305 module_exit(dm_integrity_exit);
4306
4307 MODULE_AUTHOR("Milan Broz");
4308 MODULE_AUTHOR("Mikulas Patocka");
4309 MODULE_DESCRIPTION(DM_NAME " target for integrity tags extension");
4310 MODULE_LICENSE("GPL");