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
6 * This file is released under the GPL.
9 #include "dm-bio-record.h"
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>
26 #define DM_MSG_PREFIX "integrity"
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 #define DISCARD_FILLER 0xf6
46 * Warning - DEBUG_PRINT prints security-sensitive data to the log,
47 * so it should not be enabled in the official kernel
50 //#define INTERNAL_VERIFY
56 #define SB_MAGIC "integrt"
57 #define SB_VERSION_1 1
58 #define SB_VERSION_2 2
59 #define SB_VERSION_3 3
60 #define SB_VERSION_4 4
61 #define SB_VERSION_5 5
63 #define MAX_SECTORS_PER_BLOCK 8
68 __u8 log2_interleave_sectors;
69 __u16 integrity_tag_size;
70 __u32 journal_sections;
71 __u64 provided_data_sectors; /* userspace uses this value */
73 __u8 log2_sectors_per_block;
74 __u8 log2_blocks_per_bitmap_bit;
81 #define SB_FLAG_HAVE_JOURNAL_MAC 0x1
82 #define SB_FLAG_RECALCULATING 0x2
83 #define SB_FLAG_DIRTY_BITMAP 0x4
84 #define SB_FLAG_FIXED_PADDING 0x8
85 #define SB_FLAG_FIXED_HMAC 0x10
87 #define JOURNAL_ENTRY_ROUNDUP 8
89 typedef __u64 commit_id_t;
90 #define JOURNAL_MAC_PER_SECTOR 8
92 struct journal_entry {
100 commit_id_t last_bytes[];
104 #define journal_entry_tag(ic, je) ((__u8 *)&(je)->last_bytes[(ic)->sectors_per_block])
106 #if BITS_PER_LONG == 64
107 #define journal_entry_set_sector(je, x) do { smp_wmb(); WRITE_ONCE((je)->u.sector, cpu_to_le64(x)); } while (0)
109 #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)
111 #define journal_entry_get_sector(je) le64_to_cpu((je)->u.sector)
112 #define journal_entry_is_unused(je) ((je)->u.s.sector_hi == cpu_to_le32(-1))
113 #define journal_entry_set_unused(je) do { ((je)->u.s.sector_hi = cpu_to_le32(-1)); } while (0)
114 #define journal_entry_is_inprogress(je) ((je)->u.s.sector_hi == cpu_to_le32(-2))
115 #define journal_entry_set_inprogress(je) do { ((je)->u.s.sector_hi = cpu_to_le32(-2)); } while (0)
117 #define JOURNAL_BLOCK_SECTORS 8
118 #define JOURNAL_SECTOR_DATA ((1 << SECTOR_SHIFT) - sizeof(commit_id_t))
119 #define JOURNAL_MAC_SIZE (JOURNAL_MAC_PER_SECTOR * JOURNAL_BLOCK_SECTORS)
121 struct journal_sector {
122 __u8 entries[JOURNAL_SECTOR_DATA - JOURNAL_MAC_PER_SECTOR];
123 __u8 mac[JOURNAL_MAC_PER_SECTOR];
124 commit_id_t commit_id;
127 #define MAX_TAG_SIZE (JOURNAL_SECTOR_DATA - JOURNAL_MAC_PER_SECTOR - offsetof(struct journal_entry, last_bytes[MAX_SECTORS_PER_BLOCK]))
129 #define METADATA_PADDING_SECTORS 8
131 #define N_COMMIT_IDS 4
133 static unsigned char prev_commit_seq(unsigned char seq)
135 return (seq + N_COMMIT_IDS - 1) % N_COMMIT_IDS;
138 static unsigned char next_commit_seq(unsigned char seq)
140 return (seq + 1) % N_COMMIT_IDS;
144 * In-memory structures
147 struct journal_node {
159 struct dm_integrity_c {
161 struct dm_dev *meta_dev;
165 mempool_t journal_io_mempool;
166 struct dm_io_client *io;
167 struct dm_bufio_client *bufio;
168 struct workqueue_struct *metadata_wq;
169 struct superblock *sb;
170 unsigned journal_pages;
171 unsigned n_bitmap_blocks;
173 struct page_list *journal;
174 struct page_list *journal_io;
175 struct page_list *journal_xor;
176 struct page_list *recalc_bitmap;
177 struct page_list *may_write_bitmap;
178 struct bitmap_block_status *bbs;
179 unsigned bitmap_flush_interval;
180 int synchronous_mode;
181 struct bio_list synchronous_bios;
182 struct delayed_work bitmap_flush_work;
184 struct crypto_skcipher *journal_crypt;
185 struct scatterlist **journal_scatterlist;
186 struct scatterlist **journal_io_scatterlist;
187 struct skcipher_request **sk_requests;
189 struct crypto_shash *journal_mac;
191 struct journal_node *journal_tree;
192 struct rb_root journal_tree_root;
194 sector_t provided_data_sectors;
196 unsigned short journal_entry_size;
197 unsigned char journal_entries_per_sector;
198 unsigned char journal_section_entries;
199 unsigned short journal_section_sectors;
200 unsigned journal_sections;
201 unsigned journal_entries;
202 sector_t data_device_sectors;
203 sector_t meta_device_sectors;
204 unsigned initial_sectors;
205 unsigned metadata_run;
206 __s8 log2_metadata_run;
207 __u8 log2_buffer_sectors;
208 __u8 sectors_per_block;
209 __u8 log2_blocks_per_bitmap_bit;
215 struct crypto_shash *internal_hash;
217 struct dm_target *ti;
219 /* these variables are locked with endio_wait.lock */
220 struct rb_root in_progress;
221 struct list_head wait_list;
222 wait_queue_head_t endio_wait;
223 struct workqueue_struct *wait_wq;
224 struct workqueue_struct *offload_wq;
226 unsigned char commit_seq;
227 commit_id_t commit_ids[N_COMMIT_IDS];
229 unsigned committed_section;
230 unsigned n_committed_sections;
232 unsigned uncommitted_section;
233 unsigned n_uncommitted_sections;
235 unsigned free_section;
236 unsigned char free_section_entry;
237 unsigned free_sectors;
239 unsigned free_sectors_threshold;
241 struct workqueue_struct *commit_wq;
242 struct work_struct commit_work;
244 struct workqueue_struct *writer_wq;
245 struct work_struct writer_work;
247 struct workqueue_struct *recalc_wq;
248 struct work_struct recalc_work;
252 struct bio_list flush_bio_list;
254 unsigned long autocommit_jiffies;
255 struct timer_list autocommit_timer;
256 unsigned autocommit_msec;
258 wait_queue_head_t copy_to_journal_wait;
260 struct completion crypto_backoff;
262 bool journal_uptodate;
264 bool recalculate_flag;
268 bool legacy_recalculate;
270 struct alg_spec internal_hash_alg;
271 struct alg_spec journal_crypt_alg;
272 struct alg_spec journal_mac_alg;
274 atomic64_t number_of_mismatches;
276 struct notifier_block reboot_notifier;
279 struct dm_integrity_range {
280 sector_t logical_sector;
286 struct task_struct *task;
287 struct list_head wait_entry;
292 struct dm_integrity_io {
293 struct work_struct work;
295 struct dm_integrity_c *ic;
299 struct dm_integrity_range range;
301 sector_t metadata_block;
302 unsigned metadata_offset;
305 blk_status_t bi_status;
307 struct completion *completion;
309 struct dm_bio_details bio_details;
312 struct journal_completion {
313 struct dm_integrity_c *ic;
315 struct completion comp;
319 struct dm_integrity_range range;
320 struct journal_completion *comp;
323 struct bitmap_block_status {
324 struct work_struct work;
325 struct dm_integrity_c *ic;
327 unsigned long *bitmap;
328 struct bio_list bio_queue;
329 spinlock_t bio_queue_lock;
333 static struct kmem_cache *journal_io_cache;
335 #define JOURNAL_IO_MEMPOOL 32
338 #define DEBUG_print(x, ...) printk(KERN_DEBUG x, ##__VA_ARGS__)
339 static void __DEBUG_bytes(__u8 *bytes, size_t len, const char *msg, ...)
348 pr_cont(" %02x", *bytes);
354 #define DEBUG_bytes(bytes, len, msg, ...) __DEBUG_bytes(bytes, len, KERN_DEBUG msg, ##__VA_ARGS__)
356 #define DEBUG_print(x, ...) do { } while (0)
357 #define DEBUG_bytes(bytes, len, msg, ...) do { } while (0)
360 static void dm_integrity_prepare(struct request *rq)
364 static void dm_integrity_complete(struct request *rq, unsigned int nr_bytes)
369 * DM Integrity profile, protection is performed layer above (dm-crypt)
371 static const struct blk_integrity_profile dm_integrity_profile = {
372 .name = "DM-DIF-EXT-TAG",
375 .prepare_fn = dm_integrity_prepare,
376 .complete_fn = dm_integrity_complete,
379 static void dm_integrity_map_continue(struct dm_integrity_io *dio, bool from_map);
380 static void integrity_bio_wait(struct work_struct *w);
381 static void dm_integrity_dtr(struct dm_target *ti);
383 static void dm_integrity_io_error(struct dm_integrity_c *ic, const char *msg, int err)
386 atomic64_inc(&ic->number_of_mismatches);
387 if (!cmpxchg(&ic->failed, 0, err))
388 DMERR("Error on %s: %d", msg, err);
391 static int dm_integrity_failed(struct dm_integrity_c *ic)
393 return READ_ONCE(ic->failed);
396 static bool dm_integrity_disable_recalculate(struct dm_integrity_c *ic)
398 if (ic->legacy_recalculate)
400 if (!(ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) ?
401 ic->internal_hash_alg.key || ic->journal_mac_alg.key :
402 ic->internal_hash_alg.key && !ic->journal_mac_alg.key)
407 static commit_id_t dm_integrity_commit_id(struct dm_integrity_c *ic, unsigned i,
408 unsigned j, unsigned char seq)
411 * Xor the number with section and sector, so that if a piece of
412 * journal is written at wrong place, it is detected.
414 return ic->commit_ids[seq] ^ cpu_to_le64(((__u64)i << 32) ^ j);
417 static void get_area_and_offset(struct dm_integrity_c *ic, sector_t data_sector,
418 sector_t *area, sector_t *offset)
421 __u8 log2_interleave_sectors = ic->sb->log2_interleave_sectors;
422 *area = data_sector >> log2_interleave_sectors;
423 *offset = (unsigned)data_sector & ((1U << log2_interleave_sectors) - 1);
426 *offset = data_sector;
430 #define sector_to_block(ic, n) \
432 BUG_ON((n) & (unsigned)((ic)->sectors_per_block - 1)); \
433 (n) >>= (ic)->sb->log2_sectors_per_block; \
436 static __u64 get_metadata_sector_and_offset(struct dm_integrity_c *ic, sector_t area,
437 sector_t offset, unsigned *metadata_offset)
442 ms = area << ic->sb->log2_interleave_sectors;
443 if (likely(ic->log2_metadata_run >= 0))
444 ms += area << ic->log2_metadata_run;
446 ms += area * ic->metadata_run;
447 ms >>= ic->log2_buffer_sectors;
449 sector_to_block(ic, offset);
451 if (likely(ic->log2_tag_size >= 0)) {
452 ms += offset >> (SECTOR_SHIFT + ic->log2_buffer_sectors - ic->log2_tag_size);
453 mo = (offset << ic->log2_tag_size) & ((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - 1);
455 ms += (__u64)offset * ic->tag_size >> (SECTOR_SHIFT + ic->log2_buffer_sectors);
456 mo = (offset * ic->tag_size) & ((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - 1);
458 *metadata_offset = mo;
462 static sector_t get_data_sector(struct dm_integrity_c *ic, sector_t area, sector_t offset)
469 result = area << ic->sb->log2_interleave_sectors;
470 if (likely(ic->log2_metadata_run >= 0))
471 result += (area + 1) << ic->log2_metadata_run;
473 result += (area + 1) * ic->metadata_run;
475 result += (sector_t)ic->initial_sectors + offset;
481 static void wraparound_section(struct dm_integrity_c *ic, unsigned *sec_ptr)
483 if (unlikely(*sec_ptr >= ic->journal_sections))
484 *sec_ptr -= ic->journal_sections;
487 static void sb_set_version(struct dm_integrity_c *ic)
489 if (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC))
490 ic->sb->version = SB_VERSION_5;
491 else if (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING))
492 ic->sb->version = SB_VERSION_4;
493 else if (ic->mode == 'B' || ic->sb->flags & cpu_to_le32(SB_FLAG_DIRTY_BITMAP))
494 ic->sb->version = SB_VERSION_3;
495 else if (ic->meta_dev || ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
496 ic->sb->version = SB_VERSION_2;
498 ic->sb->version = SB_VERSION_1;
501 static int sb_mac(struct dm_integrity_c *ic, bool wr)
503 SHASH_DESC_ON_STACK(desc, ic->journal_mac);
505 unsigned size = crypto_shash_digestsize(ic->journal_mac);
507 if (sizeof(struct superblock) + size > 1 << SECTOR_SHIFT) {
508 dm_integrity_io_error(ic, "digest is too long", -EINVAL);
512 desc->tfm = ic->journal_mac;
514 r = crypto_shash_init(desc);
515 if (unlikely(r < 0)) {
516 dm_integrity_io_error(ic, "crypto_shash_init", r);
520 r = crypto_shash_update(desc, (__u8 *)ic->sb, (1 << SECTOR_SHIFT) - size);
521 if (unlikely(r < 0)) {
522 dm_integrity_io_error(ic, "crypto_shash_update", r);
527 r = crypto_shash_final(desc, (__u8 *)ic->sb + (1 << SECTOR_SHIFT) - size);
528 if (unlikely(r < 0)) {
529 dm_integrity_io_error(ic, "crypto_shash_final", r);
533 __u8 result[HASH_MAX_DIGESTSIZE];
534 r = crypto_shash_final(desc, result);
535 if (unlikely(r < 0)) {
536 dm_integrity_io_error(ic, "crypto_shash_final", r);
539 if (memcmp((__u8 *)ic->sb + (1 << SECTOR_SHIFT) - size, result, size)) {
540 dm_integrity_io_error(ic, "superblock mac", -EILSEQ);
548 static int sync_rw_sb(struct dm_integrity_c *ic, int op, int op_flags)
550 struct dm_io_request io_req;
551 struct dm_io_region io_loc;
555 io_req.bi_op_flags = op_flags;
556 io_req.mem.type = DM_IO_KMEM;
557 io_req.mem.ptr.addr = ic->sb;
558 io_req.notify.fn = NULL;
559 io_req.client = ic->io;
560 io_loc.bdev = ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev;
561 io_loc.sector = ic->start;
562 io_loc.count = SB_SECTORS;
564 if (op == REQ_OP_WRITE) {
566 if (ic->journal_mac && ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) {
567 r = sb_mac(ic, true);
573 r = dm_io(&io_req, 1, &io_loc, NULL);
577 if (op == REQ_OP_READ) {
578 if (ic->mode != 'R' && ic->journal_mac && ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) {
579 r = sb_mac(ic, false);
588 #define BITMAP_OP_TEST_ALL_SET 0
589 #define BITMAP_OP_TEST_ALL_CLEAR 1
590 #define BITMAP_OP_SET 2
591 #define BITMAP_OP_CLEAR 3
593 static bool block_bitmap_op(struct dm_integrity_c *ic, struct page_list *bitmap,
594 sector_t sector, sector_t n_sectors, int mode)
596 unsigned long bit, end_bit, this_end_bit, page, end_page;
599 if (unlikely(((sector | n_sectors) & ((1 << ic->sb->log2_sectors_per_block) - 1)) != 0)) {
600 DMCRIT("invalid bitmap access (%llx,%llx,%d,%d,%d)",
603 ic->sb->log2_sectors_per_block,
604 ic->log2_blocks_per_bitmap_bit,
609 if (unlikely(!n_sectors))
612 bit = sector >> (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
613 end_bit = (sector + n_sectors - 1) >>
614 (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
616 page = bit / (PAGE_SIZE * 8);
617 bit %= PAGE_SIZE * 8;
619 end_page = end_bit / (PAGE_SIZE * 8);
620 end_bit %= PAGE_SIZE * 8;
623 if (page < end_page) {
624 this_end_bit = PAGE_SIZE * 8 - 1;
626 this_end_bit = end_bit;
629 data = lowmem_page_address(bitmap[page].page);
631 if (mode == BITMAP_OP_TEST_ALL_SET) {
632 while (bit <= this_end_bit) {
633 if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
635 if (data[bit / BITS_PER_LONG] != -1)
637 bit += BITS_PER_LONG;
638 } while (this_end_bit >= bit + BITS_PER_LONG - 1);
641 if (!test_bit(bit, data))
645 } else if (mode == BITMAP_OP_TEST_ALL_CLEAR) {
646 while (bit <= this_end_bit) {
647 if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
649 if (data[bit / BITS_PER_LONG] != 0)
651 bit += BITS_PER_LONG;
652 } while (this_end_bit >= bit + BITS_PER_LONG - 1);
655 if (test_bit(bit, data))
659 } else if (mode == BITMAP_OP_SET) {
660 while (bit <= this_end_bit) {
661 if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
663 data[bit / BITS_PER_LONG] = -1;
664 bit += BITS_PER_LONG;
665 } while (this_end_bit >= bit + BITS_PER_LONG - 1);
668 __set_bit(bit, data);
671 } else if (mode == BITMAP_OP_CLEAR) {
672 if (!bit && this_end_bit == PAGE_SIZE * 8 - 1)
674 else while (bit <= this_end_bit) {
675 if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
677 data[bit / BITS_PER_LONG] = 0;
678 bit += BITS_PER_LONG;
679 } while (this_end_bit >= bit + BITS_PER_LONG - 1);
682 __clear_bit(bit, data);
689 if (unlikely(page < end_page)) {
698 static void block_bitmap_copy(struct dm_integrity_c *ic, struct page_list *dst, struct page_list *src)
700 unsigned n_bitmap_pages = DIV_ROUND_UP(ic->n_bitmap_blocks, PAGE_SIZE / BITMAP_BLOCK_SIZE);
703 for (i = 0; i < n_bitmap_pages; i++) {
704 unsigned long *dst_data = lowmem_page_address(dst[i].page);
705 unsigned long *src_data = lowmem_page_address(src[i].page);
706 copy_page(dst_data, src_data);
710 static struct bitmap_block_status *sector_to_bitmap_block(struct dm_integrity_c *ic, sector_t sector)
712 unsigned bit = sector >> (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
713 unsigned bitmap_block = bit / (BITMAP_BLOCK_SIZE * 8);
715 BUG_ON(bitmap_block >= ic->n_bitmap_blocks);
716 return &ic->bbs[bitmap_block];
719 static void access_journal_check(struct dm_integrity_c *ic, unsigned section, unsigned offset,
720 bool e, const char *function)
722 #if defined(CONFIG_DM_DEBUG) || defined(INTERNAL_VERIFY)
723 unsigned limit = e ? ic->journal_section_entries : ic->journal_section_sectors;
725 if (unlikely(section >= ic->journal_sections) ||
726 unlikely(offset >= limit)) {
727 DMCRIT("%s: invalid access at (%u,%u), limit (%u,%u)",
728 function, section, offset, ic->journal_sections, limit);
734 static void page_list_location(struct dm_integrity_c *ic, unsigned section, unsigned offset,
735 unsigned *pl_index, unsigned *pl_offset)
739 access_journal_check(ic, section, offset, false, "page_list_location");
741 sector = section * ic->journal_section_sectors + offset;
743 *pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
744 *pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
747 static struct journal_sector *access_page_list(struct dm_integrity_c *ic, struct page_list *pl,
748 unsigned section, unsigned offset, unsigned *n_sectors)
750 unsigned pl_index, pl_offset;
753 page_list_location(ic, section, offset, &pl_index, &pl_offset);
756 *n_sectors = (PAGE_SIZE - pl_offset) >> SECTOR_SHIFT;
758 va = lowmem_page_address(pl[pl_index].page);
760 return (struct journal_sector *)(va + pl_offset);
763 static struct journal_sector *access_journal(struct dm_integrity_c *ic, unsigned section, unsigned offset)
765 return access_page_list(ic, ic->journal, section, offset, NULL);
768 static struct journal_entry *access_journal_entry(struct dm_integrity_c *ic, unsigned section, unsigned n)
770 unsigned rel_sector, offset;
771 struct journal_sector *js;
773 access_journal_check(ic, section, n, true, "access_journal_entry");
775 rel_sector = n % JOURNAL_BLOCK_SECTORS;
776 offset = n / JOURNAL_BLOCK_SECTORS;
778 js = access_journal(ic, section, rel_sector);
779 return (struct journal_entry *)((char *)js + offset * ic->journal_entry_size);
782 static struct journal_sector *access_journal_data(struct dm_integrity_c *ic, unsigned section, unsigned n)
784 n <<= ic->sb->log2_sectors_per_block;
786 n += JOURNAL_BLOCK_SECTORS;
788 access_journal_check(ic, section, n, false, "access_journal_data");
790 return access_journal(ic, section, n);
793 static void section_mac(struct dm_integrity_c *ic, unsigned section, __u8 result[JOURNAL_MAC_SIZE])
795 SHASH_DESC_ON_STACK(desc, ic->journal_mac);
799 desc->tfm = ic->journal_mac;
801 r = crypto_shash_init(desc);
802 if (unlikely(r < 0)) {
803 dm_integrity_io_error(ic, "crypto_shash_init", r);
807 if (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) {
810 r = crypto_shash_update(desc, (__u8 *)&ic->sb->salt, SALT_SIZE);
811 if (unlikely(r < 0)) {
812 dm_integrity_io_error(ic, "crypto_shash_update", r);
816 section_le = cpu_to_le64(section);
817 r = crypto_shash_update(desc, (__u8 *)§ion_le, sizeof section_le);
818 if (unlikely(r < 0)) {
819 dm_integrity_io_error(ic, "crypto_shash_update", r);
824 for (j = 0; j < ic->journal_section_entries; j++) {
825 struct journal_entry *je = access_journal_entry(ic, section, j);
826 r = crypto_shash_update(desc, (__u8 *)&je->u.sector, sizeof je->u.sector);
827 if (unlikely(r < 0)) {
828 dm_integrity_io_error(ic, "crypto_shash_update", r);
833 size = crypto_shash_digestsize(ic->journal_mac);
835 if (likely(size <= JOURNAL_MAC_SIZE)) {
836 r = crypto_shash_final(desc, result);
837 if (unlikely(r < 0)) {
838 dm_integrity_io_error(ic, "crypto_shash_final", r);
841 memset(result + size, 0, JOURNAL_MAC_SIZE - size);
843 __u8 digest[HASH_MAX_DIGESTSIZE];
845 if (WARN_ON(size > sizeof(digest))) {
846 dm_integrity_io_error(ic, "digest_size", -EINVAL);
849 r = crypto_shash_final(desc, digest);
850 if (unlikely(r < 0)) {
851 dm_integrity_io_error(ic, "crypto_shash_final", r);
854 memcpy(result, digest, JOURNAL_MAC_SIZE);
859 memset(result, 0, JOURNAL_MAC_SIZE);
862 static void rw_section_mac(struct dm_integrity_c *ic, unsigned section, bool wr)
864 __u8 result[JOURNAL_MAC_SIZE];
867 if (!ic->journal_mac)
870 section_mac(ic, section, result);
872 for (j = 0; j < JOURNAL_BLOCK_SECTORS; j++) {
873 struct journal_sector *js = access_journal(ic, section, j);
876 memcpy(&js->mac, result + (j * JOURNAL_MAC_PER_SECTOR), JOURNAL_MAC_PER_SECTOR);
878 if (memcmp(&js->mac, result + (j * JOURNAL_MAC_PER_SECTOR), JOURNAL_MAC_PER_SECTOR))
879 dm_integrity_io_error(ic, "journal mac", -EILSEQ);
884 static void complete_journal_op(void *context)
886 struct journal_completion *comp = context;
887 BUG_ON(!atomic_read(&comp->in_flight));
888 if (likely(atomic_dec_and_test(&comp->in_flight)))
889 complete(&comp->comp);
892 static void xor_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
893 unsigned n_sections, struct journal_completion *comp)
895 struct async_submit_ctl submit;
896 size_t n_bytes = (size_t)(n_sections * ic->journal_section_sectors) << SECTOR_SHIFT;
897 unsigned pl_index, pl_offset, section_index;
898 struct page_list *source_pl, *target_pl;
900 if (likely(encrypt)) {
901 source_pl = ic->journal;
902 target_pl = ic->journal_io;
904 source_pl = ic->journal_io;
905 target_pl = ic->journal;
908 page_list_location(ic, section, 0, &pl_index, &pl_offset);
910 atomic_add(roundup(pl_offset + n_bytes, PAGE_SIZE) >> PAGE_SHIFT, &comp->in_flight);
912 init_async_submit(&submit, ASYNC_TX_XOR_ZERO_DST, NULL, complete_journal_op, comp, NULL);
914 section_index = pl_index;
918 struct page *src_pages[2];
919 struct page *dst_page;
921 while (unlikely(pl_index == section_index)) {
924 rw_section_mac(ic, section, true);
929 page_list_location(ic, section, 0, §ion_index, &dummy);
932 this_step = min(n_bytes, (size_t)PAGE_SIZE - pl_offset);
933 dst_page = target_pl[pl_index].page;
934 src_pages[0] = source_pl[pl_index].page;
935 src_pages[1] = ic->journal_xor[pl_index].page;
937 async_xor(dst_page, src_pages, pl_offset, 2, this_step, &submit);
941 n_bytes -= this_step;
946 async_tx_issue_pending_all();
949 static void complete_journal_encrypt(struct crypto_async_request *req, int err)
951 struct journal_completion *comp = req->data;
953 if (likely(err == -EINPROGRESS)) {
954 complete(&comp->ic->crypto_backoff);
957 dm_integrity_io_error(comp->ic, "asynchronous encrypt", err);
959 complete_journal_op(comp);
962 static bool do_crypt(bool encrypt, struct skcipher_request *req, struct journal_completion *comp)
965 skcipher_request_set_callback(req, CRYPTO_TFM_REQ_MAY_BACKLOG,
966 complete_journal_encrypt, comp);
968 r = crypto_skcipher_encrypt(req);
970 r = crypto_skcipher_decrypt(req);
973 if (likely(r == -EINPROGRESS))
975 if (likely(r == -EBUSY)) {
976 wait_for_completion(&comp->ic->crypto_backoff);
977 reinit_completion(&comp->ic->crypto_backoff);
980 dm_integrity_io_error(comp->ic, "encrypt", r);
984 static void crypt_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
985 unsigned n_sections, struct journal_completion *comp)
987 struct scatterlist **source_sg;
988 struct scatterlist **target_sg;
990 atomic_add(2, &comp->in_flight);
992 if (likely(encrypt)) {
993 source_sg = ic->journal_scatterlist;
994 target_sg = ic->journal_io_scatterlist;
996 source_sg = ic->journal_io_scatterlist;
997 target_sg = ic->journal_scatterlist;
1001 struct skcipher_request *req;
1005 if (likely(encrypt))
1006 rw_section_mac(ic, section, true);
1008 req = ic->sk_requests[section];
1009 ivsize = crypto_skcipher_ivsize(ic->journal_crypt);
1012 memcpy(iv, iv + ivsize, ivsize);
1014 req->src = source_sg[section];
1015 req->dst = target_sg[section];
1017 if (unlikely(do_crypt(encrypt, req, comp)))
1018 atomic_inc(&comp->in_flight);
1022 } while (n_sections);
1024 atomic_dec(&comp->in_flight);
1025 complete_journal_op(comp);
1028 static void encrypt_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
1029 unsigned n_sections, struct journal_completion *comp)
1031 if (ic->journal_xor)
1032 return xor_journal(ic, encrypt, section, n_sections, comp);
1034 return crypt_journal(ic, encrypt, section, n_sections, comp);
1037 static void complete_journal_io(unsigned long error, void *context)
1039 struct journal_completion *comp = context;
1040 if (unlikely(error != 0))
1041 dm_integrity_io_error(comp->ic, "writing journal", -EIO);
1042 complete_journal_op(comp);
1045 static void rw_journal_sectors(struct dm_integrity_c *ic, int op, int op_flags,
1046 unsigned sector, unsigned n_sectors, struct journal_completion *comp)
1048 struct dm_io_request io_req;
1049 struct dm_io_region io_loc;
1050 unsigned pl_index, pl_offset;
1053 if (unlikely(dm_integrity_failed(ic))) {
1055 complete_journal_io(-1UL, comp);
1059 pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
1060 pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
1063 io_req.bi_op_flags = op_flags;
1064 io_req.mem.type = DM_IO_PAGE_LIST;
1066 io_req.mem.ptr.pl = &ic->journal_io[pl_index];
1068 io_req.mem.ptr.pl = &ic->journal[pl_index];
1069 io_req.mem.offset = pl_offset;
1070 if (likely(comp != NULL)) {
1071 io_req.notify.fn = complete_journal_io;
1072 io_req.notify.context = comp;
1074 io_req.notify.fn = NULL;
1076 io_req.client = ic->io;
1077 io_loc.bdev = ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev;
1078 io_loc.sector = ic->start + SB_SECTORS + sector;
1079 io_loc.count = n_sectors;
1081 r = dm_io(&io_req, 1, &io_loc, NULL);
1083 dm_integrity_io_error(ic, op == REQ_OP_READ ? "reading journal" : "writing journal", r);
1085 WARN_ONCE(1, "asynchronous dm_io failed: %d", r);
1086 complete_journal_io(-1UL, comp);
1091 static void rw_journal(struct dm_integrity_c *ic, int op, int op_flags, unsigned section,
1092 unsigned n_sections, struct journal_completion *comp)
1094 unsigned sector, n_sectors;
1096 sector = section * ic->journal_section_sectors;
1097 n_sectors = n_sections * ic->journal_section_sectors;
1099 rw_journal_sectors(ic, op, op_flags, sector, n_sectors, comp);
1102 static void write_journal(struct dm_integrity_c *ic, unsigned commit_start, unsigned commit_sections)
1104 struct journal_completion io_comp;
1105 struct journal_completion crypt_comp_1;
1106 struct journal_completion crypt_comp_2;
1110 init_completion(&io_comp.comp);
1112 if (commit_start + commit_sections <= ic->journal_sections) {
1113 io_comp.in_flight = (atomic_t)ATOMIC_INIT(1);
1114 if (ic->journal_io) {
1115 crypt_comp_1.ic = ic;
1116 init_completion(&crypt_comp_1.comp);
1117 crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
1118 encrypt_journal(ic, true, commit_start, commit_sections, &crypt_comp_1);
1119 wait_for_completion_io(&crypt_comp_1.comp);
1121 for (i = 0; i < commit_sections; i++)
1122 rw_section_mac(ic, commit_start + i, true);
1124 rw_journal(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, commit_start,
1125 commit_sections, &io_comp);
1128 io_comp.in_flight = (atomic_t)ATOMIC_INIT(2);
1129 to_end = ic->journal_sections - commit_start;
1130 if (ic->journal_io) {
1131 crypt_comp_1.ic = ic;
1132 init_completion(&crypt_comp_1.comp);
1133 crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
1134 encrypt_journal(ic, true, commit_start, to_end, &crypt_comp_1);
1135 if (try_wait_for_completion(&crypt_comp_1.comp)) {
1136 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
1137 reinit_completion(&crypt_comp_1.comp);
1138 crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
1139 encrypt_journal(ic, true, 0, commit_sections - to_end, &crypt_comp_1);
1140 wait_for_completion_io(&crypt_comp_1.comp);
1142 crypt_comp_2.ic = ic;
1143 init_completion(&crypt_comp_2.comp);
1144 crypt_comp_2.in_flight = (atomic_t)ATOMIC_INIT(0);
1145 encrypt_journal(ic, true, 0, commit_sections - to_end, &crypt_comp_2);
1146 wait_for_completion_io(&crypt_comp_1.comp);
1147 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
1148 wait_for_completion_io(&crypt_comp_2.comp);
1151 for (i = 0; i < to_end; i++)
1152 rw_section_mac(ic, commit_start + i, true);
1153 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
1154 for (i = 0; i < commit_sections - to_end; i++)
1155 rw_section_mac(ic, i, true);
1157 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, 0, commit_sections - to_end, &io_comp);
1160 wait_for_completion_io(&io_comp.comp);
1163 static void copy_from_journal(struct dm_integrity_c *ic, unsigned section, unsigned offset,
1164 unsigned n_sectors, sector_t target, io_notify_fn fn, void *data)
1166 struct dm_io_request io_req;
1167 struct dm_io_region io_loc;
1169 unsigned sector, pl_index, pl_offset;
1171 BUG_ON((target | n_sectors | offset) & (unsigned)(ic->sectors_per_block - 1));
1173 if (unlikely(dm_integrity_failed(ic))) {
1178 sector = section * ic->journal_section_sectors + JOURNAL_BLOCK_SECTORS + offset;
1180 pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
1181 pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
1183 io_req.bi_op = REQ_OP_WRITE;
1184 io_req.bi_op_flags = 0;
1185 io_req.mem.type = DM_IO_PAGE_LIST;
1186 io_req.mem.ptr.pl = &ic->journal[pl_index];
1187 io_req.mem.offset = pl_offset;
1188 io_req.notify.fn = fn;
1189 io_req.notify.context = data;
1190 io_req.client = ic->io;
1191 io_loc.bdev = ic->dev->bdev;
1192 io_loc.sector = target;
1193 io_loc.count = n_sectors;
1195 r = dm_io(&io_req, 1, &io_loc, NULL);
1197 WARN_ONCE(1, "asynchronous dm_io failed: %d", r);
1202 static bool ranges_overlap(struct dm_integrity_range *range1, struct dm_integrity_range *range2)
1204 return range1->logical_sector < range2->logical_sector + range2->n_sectors &&
1205 range1->logical_sector + range1->n_sectors > range2->logical_sector;
1208 static bool add_new_range(struct dm_integrity_c *ic, struct dm_integrity_range *new_range, bool check_waiting)
1210 struct rb_node **n = &ic->in_progress.rb_node;
1211 struct rb_node *parent;
1213 BUG_ON((new_range->logical_sector | new_range->n_sectors) & (unsigned)(ic->sectors_per_block - 1));
1215 if (likely(check_waiting)) {
1216 struct dm_integrity_range *range;
1217 list_for_each_entry(range, &ic->wait_list, wait_entry) {
1218 if (unlikely(ranges_overlap(range, new_range)))
1226 struct dm_integrity_range *range = container_of(*n, struct dm_integrity_range, node);
1229 if (new_range->logical_sector + new_range->n_sectors <= range->logical_sector) {
1230 n = &range->node.rb_left;
1231 } else if (new_range->logical_sector >= range->logical_sector + range->n_sectors) {
1232 n = &range->node.rb_right;
1238 rb_link_node(&new_range->node, parent, n);
1239 rb_insert_color(&new_range->node, &ic->in_progress);
1244 static void remove_range_unlocked(struct dm_integrity_c *ic, struct dm_integrity_range *range)
1246 rb_erase(&range->node, &ic->in_progress);
1247 while (unlikely(!list_empty(&ic->wait_list))) {
1248 struct dm_integrity_range *last_range =
1249 list_first_entry(&ic->wait_list, struct dm_integrity_range, wait_entry);
1250 struct task_struct *last_range_task;
1251 last_range_task = last_range->task;
1252 list_del(&last_range->wait_entry);
1253 if (!add_new_range(ic, last_range, false)) {
1254 last_range->task = last_range_task;
1255 list_add(&last_range->wait_entry, &ic->wait_list);
1258 last_range->waiting = false;
1259 wake_up_process(last_range_task);
1263 static void remove_range(struct dm_integrity_c *ic, struct dm_integrity_range *range)
1265 unsigned long flags;
1267 spin_lock_irqsave(&ic->endio_wait.lock, flags);
1268 remove_range_unlocked(ic, range);
1269 spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
1272 static void wait_and_add_new_range(struct dm_integrity_c *ic, struct dm_integrity_range *new_range)
1274 new_range->waiting = true;
1275 list_add_tail(&new_range->wait_entry, &ic->wait_list);
1276 new_range->task = current;
1278 __set_current_state(TASK_UNINTERRUPTIBLE);
1279 spin_unlock_irq(&ic->endio_wait.lock);
1281 spin_lock_irq(&ic->endio_wait.lock);
1282 } while (unlikely(new_range->waiting));
1285 static void add_new_range_and_wait(struct dm_integrity_c *ic, struct dm_integrity_range *new_range)
1287 if (unlikely(!add_new_range(ic, new_range, true)))
1288 wait_and_add_new_range(ic, new_range);
1291 static void init_journal_node(struct journal_node *node)
1293 RB_CLEAR_NODE(&node->node);
1294 node->sector = (sector_t)-1;
1297 static void add_journal_node(struct dm_integrity_c *ic, struct journal_node *node, sector_t sector)
1299 struct rb_node **link;
1300 struct rb_node *parent;
1302 node->sector = sector;
1303 BUG_ON(!RB_EMPTY_NODE(&node->node));
1305 link = &ic->journal_tree_root.rb_node;
1309 struct journal_node *j;
1311 j = container_of(parent, struct journal_node, node);
1312 if (sector < j->sector)
1313 link = &j->node.rb_left;
1315 link = &j->node.rb_right;
1318 rb_link_node(&node->node, parent, link);
1319 rb_insert_color(&node->node, &ic->journal_tree_root);
1322 static void remove_journal_node(struct dm_integrity_c *ic, struct journal_node *node)
1324 BUG_ON(RB_EMPTY_NODE(&node->node));
1325 rb_erase(&node->node, &ic->journal_tree_root);
1326 init_journal_node(node);
1329 #define NOT_FOUND (-1U)
1331 static unsigned find_journal_node(struct dm_integrity_c *ic, sector_t sector, sector_t *next_sector)
1333 struct rb_node *n = ic->journal_tree_root.rb_node;
1334 unsigned found = NOT_FOUND;
1335 *next_sector = (sector_t)-1;
1337 struct journal_node *j = container_of(n, struct journal_node, node);
1338 if (sector == j->sector) {
1339 found = j - ic->journal_tree;
1341 if (sector < j->sector) {
1342 *next_sector = j->sector;
1343 n = j->node.rb_left;
1345 n = j->node.rb_right;
1352 static bool test_journal_node(struct dm_integrity_c *ic, unsigned pos, sector_t sector)
1354 struct journal_node *node, *next_node;
1355 struct rb_node *next;
1357 if (unlikely(pos >= ic->journal_entries))
1359 node = &ic->journal_tree[pos];
1360 if (unlikely(RB_EMPTY_NODE(&node->node)))
1362 if (unlikely(node->sector != sector))
1365 next = rb_next(&node->node);
1366 if (unlikely(!next))
1369 next_node = container_of(next, struct journal_node, node);
1370 return next_node->sector != sector;
1373 static bool find_newer_committed_node(struct dm_integrity_c *ic, struct journal_node *node)
1375 struct rb_node *next;
1376 struct journal_node *next_node;
1377 unsigned next_section;
1379 BUG_ON(RB_EMPTY_NODE(&node->node));
1381 next = rb_next(&node->node);
1382 if (unlikely(!next))
1385 next_node = container_of(next, struct journal_node, node);
1387 if (next_node->sector != node->sector)
1390 next_section = (unsigned)(next_node - ic->journal_tree) / ic->journal_section_entries;
1391 if (next_section >= ic->committed_section &&
1392 next_section < ic->committed_section + ic->n_committed_sections)
1394 if (next_section + ic->journal_sections < ic->committed_section + ic->n_committed_sections)
1404 static int dm_integrity_rw_tag(struct dm_integrity_c *ic, unsigned char *tag, sector_t *metadata_block,
1405 unsigned *metadata_offset, unsigned total_size, int op)
1407 #define MAY_BE_FILLER 1
1408 #define MAY_BE_HASH 2
1409 unsigned hash_offset = 0;
1410 unsigned may_be = MAY_BE_HASH | (ic->discard ? MAY_BE_FILLER : 0);
1413 unsigned char *data, *dp;
1414 struct dm_buffer *b;
1418 r = dm_integrity_failed(ic);
1422 data = dm_bufio_read(ic->bufio, *metadata_block, &b);
1424 return PTR_ERR(data);
1426 to_copy = min((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - *metadata_offset, total_size);
1427 dp = data + *metadata_offset;
1428 if (op == TAG_READ) {
1429 memcpy(tag, dp, to_copy);
1430 } else if (op == TAG_WRITE) {
1431 memcpy(dp, tag, to_copy);
1432 dm_bufio_mark_partial_buffer_dirty(b, *metadata_offset, *metadata_offset + to_copy);
1434 /* e.g.: op == TAG_CMP */
1436 if (likely(is_power_of_2(ic->tag_size))) {
1437 if (unlikely(memcmp(dp, tag, to_copy)))
1438 if (unlikely(!ic->discard) ||
1439 unlikely(memchr_inv(dp, DISCARD_FILLER, to_copy) != NULL)) {
1447 for (i = 0; i < to_copy; i++, ts--) {
1448 if (unlikely(dp[i] != tag[i]))
1449 may_be &= ~MAY_BE_HASH;
1450 if (likely(dp[i] != DISCARD_FILLER))
1451 may_be &= ~MAY_BE_FILLER;
1453 if (unlikely(hash_offset == ic->tag_size)) {
1454 if (unlikely(!may_be)) {
1455 dm_bufio_release(b);
1459 may_be = MAY_BE_HASH | (ic->discard ? MAY_BE_FILLER : 0);
1464 dm_bufio_release(b);
1467 *metadata_offset += to_copy;
1468 if (unlikely(*metadata_offset == 1U << SECTOR_SHIFT << ic->log2_buffer_sectors)) {
1469 (*metadata_block)++;
1470 *metadata_offset = 0;
1473 if (unlikely(!is_power_of_2(ic->tag_size))) {
1474 hash_offset = (hash_offset + to_copy) % ic->tag_size;
1477 total_size -= to_copy;
1478 } while (unlikely(total_size));
1481 #undef MAY_BE_FILLER
1485 struct flush_request {
1486 struct dm_io_request io_req;
1487 struct dm_io_region io_reg;
1488 struct dm_integrity_c *ic;
1489 struct completion comp;
1492 static void flush_notify(unsigned long error, void *fr_)
1494 struct flush_request *fr = fr_;
1495 if (unlikely(error != 0))
1496 dm_integrity_io_error(fr->ic, "flushing disk cache", -EIO);
1497 complete(&fr->comp);
1500 static void dm_integrity_flush_buffers(struct dm_integrity_c *ic, bool flush_data)
1504 struct flush_request fr;
1509 fr.io_req.bi_op = REQ_OP_WRITE,
1510 fr.io_req.bi_op_flags = REQ_PREFLUSH | REQ_SYNC,
1511 fr.io_req.mem.type = DM_IO_KMEM,
1512 fr.io_req.mem.ptr.addr = NULL,
1513 fr.io_req.notify.fn = flush_notify,
1514 fr.io_req.notify.context = &fr;
1515 fr.io_req.client = dm_bufio_get_dm_io_client(ic->bufio),
1516 fr.io_reg.bdev = ic->dev->bdev,
1517 fr.io_reg.sector = 0,
1518 fr.io_reg.count = 0,
1520 init_completion(&fr.comp);
1521 r = dm_io(&fr.io_req, 1, &fr.io_reg, NULL);
1525 r = dm_bufio_write_dirty_buffers(ic->bufio);
1527 dm_integrity_io_error(ic, "writing tags", r);
1530 wait_for_completion(&fr.comp);
1533 static void sleep_on_endio_wait(struct dm_integrity_c *ic)
1535 DECLARE_WAITQUEUE(wait, current);
1536 __add_wait_queue(&ic->endio_wait, &wait);
1537 __set_current_state(TASK_UNINTERRUPTIBLE);
1538 spin_unlock_irq(&ic->endio_wait.lock);
1540 spin_lock_irq(&ic->endio_wait.lock);
1541 __remove_wait_queue(&ic->endio_wait, &wait);
1544 static void autocommit_fn(struct timer_list *t)
1546 struct dm_integrity_c *ic = from_timer(ic, t, autocommit_timer);
1548 if (likely(!dm_integrity_failed(ic)))
1549 queue_work(ic->commit_wq, &ic->commit_work);
1552 static void schedule_autocommit(struct dm_integrity_c *ic)
1554 if (!timer_pending(&ic->autocommit_timer))
1555 mod_timer(&ic->autocommit_timer, jiffies + ic->autocommit_jiffies);
1558 static void submit_flush_bio(struct dm_integrity_c *ic, struct dm_integrity_io *dio)
1561 unsigned long flags;
1563 spin_lock_irqsave(&ic->endio_wait.lock, flags);
1564 bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1565 bio_list_add(&ic->flush_bio_list, bio);
1566 spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
1568 queue_work(ic->commit_wq, &ic->commit_work);
1571 static void do_endio(struct dm_integrity_c *ic, struct bio *bio)
1573 int r = dm_integrity_failed(ic);
1574 if (unlikely(r) && !bio->bi_status)
1575 bio->bi_status = errno_to_blk_status(r);
1576 if (unlikely(ic->synchronous_mode) && bio_op(bio) == REQ_OP_WRITE) {
1577 unsigned long flags;
1578 spin_lock_irqsave(&ic->endio_wait.lock, flags);
1579 bio_list_add(&ic->synchronous_bios, bio);
1580 queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, 0);
1581 spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
1587 static void do_endio_flush(struct dm_integrity_c *ic, struct dm_integrity_io *dio)
1589 struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1591 if (unlikely(dio->fua) && likely(!bio->bi_status) && likely(!dm_integrity_failed(ic)))
1592 submit_flush_bio(ic, dio);
1597 static void dec_in_flight(struct dm_integrity_io *dio)
1599 if (atomic_dec_and_test(&dio->in_flight)) {
1600 struct dm_integrity_c *ic = dio->ic;
1603 remove_range(ic, &dio->range);
1605 if (dio->op == REQ_OP_WRITE || unlikely(dio->op == REQ_OP_DISCARD))
1606 schedule_autocommit(ic);
1608 bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1610 if (unlikely(dio->bi_status) && !bio->bi_status)
1611 bio->bi_status = dio->bi_status;
1612 if (likely(!bio->bi_status) && unlikely(bio_sectors(bio) != dio->range.n_sectors)) {
1613 dio->range.logical_sector += dio->range.n_sectors;
1614 bio_advance(bio, dio->range.n_sectors << SECTOR_SHIFT);
1615 INIT_WORK(&dio->work, integrity_bio_wait);
1616 queue_work(ic->offload_wq, &dio->work);
1619 do_endio_flush(ic, dio);
1623 static void integrity_end_io(struct bio *bio)
1625 struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
1627 dm_bio_restore(&dio->bio_details, bio);
1628 if (bio->bi_integrity)
1629 bio->bi_opf |= REQ_INTEGRITY;
1631 if (dio->completion)
1632 complete(dio->completion);
1637 static void integrity_sector_checksum(struct dm_integrity_c *ic, sector_t sector,
1638 const char *data, char *result)
1640 __u64 sector_le = cpu_to_le64(sector);
1641 SHASH_DESC_ON_STACK(req, ic->internal_hash);
1643 unsigned digest_size;
1645 req->tfm = ic->internal_hash;
1647 r = crypto_shash_init(req);
1648 if (unlikely(r < 0)) {
1649 dm_integrity_io_error(ic, "crypto_shash_init", r);
1653 if (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) {
1654 r = crypto_shash_update(req, (__u8 *)&ic->sb->salt, SALT_SIZE);
1655 if (unlikely(r < 0)) {
1656 dm_integrity_io_error(ic, "crypto_shash_update", r);
1661 r = crypto_shash_update(req, (const __u8 *)§or_le, sizeof sector_le);
1662 if (unlikely(r < 0)) {
1663 dm_integrity_io_error(ic, "crypto_shash_update", r);
1667 r = crypto_shash_update(req, data, ic->sectors_per_block << SECTOR_SHIFT);
1668 if (unlikely(r < 0)) {
1669 dm_integrity_io_error(ic, "crypto_shash_update", r);
1673 r = crypto_shash_final(req, result);
1674 if (unlikely(r < 0)) {
1675 dm_integrity_io_error(ic, "crypto_shash_final", r);
1679 digest_size = crypto_shash_digestsize(ic->internal_hash);
1680 if (unlikely(digest_size < ic->tag_size))
1681 memset(result + digest_size, 0, ic->tag_size - digest_size);
1686 /* this shouldn't happen anyway, the hash functions have no reason to fail */
1687 get_random_bytes(result, ic->tag_size);
1690 static void integrity_metadata(struct work_struct *w)
1692 struct dm_integrity_io *dio = container_of(w, struct dm_integrity_io, work);
1693 struct dm_integrity_c *ic = dio->ic;
1697 if (ic->internal_hash) {
1698 struct bvec_iter iter;
1700 unsigned digest_size = crypto_shash_digestsize(ic->internal_hash);
1701 struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1703 unsigned extra_space = unlikely(digest_size > ic->tag_size) ? digest_size - ic->tag_size : 0;
1704 char checksums_onstack[max((size_t)HASH_MAX_DIGESTSIZE, MAX_TAG_SIZE)];
1706 unsigned sectors_to_process;
1708 if (unlikely(ic->mode == 'R'))
1711 if (likely(dio->op != REQ_OP_DISCARD))
1712 checksums = kmalloc((PAGE_SIZE >> SECTOR_SHIFT >> ic->sb->log2_sectors_per_block) * ic->tag_size + extra_space,
1713 GFP_NOIO | __GFP_NORETRY | __GFP_NOWARN);
1715 checksums = kmalloc(PAGE_SIZE, GFP_NOIO | __GFP_NORETRY | __GFP_NOWARN);
1717 checksums = checksums_onstack;
1718 if (WARN_ON(extra_space &&
1719 digest_size > sizeof(checksums_onstack))) {
1725 if (unlikely(dio->op == REQ_OP_DISCARD)) {
1726 sector_t bi_sector = dio->bio_details.bi_iter.bi_sector;
1727 unsigned bi_size = dio->bio_details.bi_iter.bi_size;
1728 unsigned max_size = likely(checksums != checksums_onstack) ? PAGE_SIZE : HASH_MAX_DIGESTSIZE;
1729 unsigned max_blocks = max_size / ic->tag_size;
1730 memset(checksums, DISCARD_FILLER, max_size);
1733 unsigned this_step_blocks = bi_size >> (SECTOR_SHIFT + ic->sb->log2_sectors_per_block);
1734 this_step_blocks = min(this_step_blocks, max_blocks);
1735 r = dm_integrity_rw_tag(ic, checksums, &dio->metadata_block, &dio->metadata_offset,
1736 this_step_blocks * ic->tag_size, TAG_WRITE);
1738 if (likely(checksums != checksums_onstack))
1743 /*if (bi_size < this_step_blocks << (SECTOR_SHIFT + ic->sb->log2_sectors_per_block)) {
1744 printk("BUGG: bi_sector: %llx, bi_size: %u\n", bi_sector, bi_size);
1745 printk("BUGG: this_step_blocks: %u\n", this_step_blocks);
1748 bi_size -= this_step_blocks << (SECTOR_SHIFT + ic->sb->log2_sectors_per_block);
1749 bi_sector += this_step_blocks << ic->sb->log2_sectors_per_block;
1752 if (likely(checksums != checksums_onstack))
1757 sector = dio->range.logical_sector;
1758 sectors_to_process = dio->range.n_sectors;
1760 __bio_for_each_segment(bv, bio, iter, dio->bio_details.bi_iter) {
1762 char *mem, *checksums_ptr;
1765 mem = (char *)kmap_atomic(bv.bv_page) + bv.bv_offset;
1767 checksums_ptr = checksums;
1769 integrity_sector_checksum(ic, sector, mem + pos, checksums_ptr);
1770 checksums_ptr += ic->tag_size;
1771 sectors_to_process -= ic->sectors_per_block;
1772 pos += ic->sectors_per_block << SECTOR_SHIFT;
1773 sector += ic->sectors_per_block;
1774 } while (pos < bv.bv_len && sectors_to_process && checksums != checksums_onstack);
1777 r = dm_integrity_rw_tag(ic, checksums, &dio->metadata_block, &dio->metadata_offset,
1778 checksums_ptr - checksums, dio->op == REQ_OP_READ ? TAG_CMP : TAG_WRITE);
1781 char b[BDEVNAME_SIZE];
1782 DMERR_LIMIT("%s: Checksum failed at sector 0x%llx", bio_devname(bio, b),
1783 (sector - ((r + ic->tag_size - 1) / ic->tag_size)));
1785 atomic64_inc(&ic->number_of_mismatches);
1787 if (likely(checksums != checksums_onstack))
1792 if (!sectors_to_process)
1795 if (unlikely(pos < bv.bv_len)) {
1796 bv.bv_offset += pos;
1802 if (likely(checksums != checksums_onstack))
1805 struct bio_integrity_payload *bip = dio->bio_details.bi_integrity;
1809 struct bvec_iter iter;
1810 unsigned data_to_process = dio->range.n_sectors;
1811 sector_to_block(ic, data_to_process);
1812 data_to_process *= ic->tag_size;
1814 bip_for_each_vec(biv, bip, iter) {
1818 BUG_ON(PageHighMem(biv.bv_page));
1819 tag = lowmem_page_address(biv.bv_page) + biv.bv_offset;
1820 this_len = min(biv.bv_len, data_to_process);
1821 r = dm_integrity_rw_tag(ic, tag, &dio->metadata_block, &dio->metadata_offset,
1822 this_len, dio->op == REQ_OP_READ ? TAG_READ : TAG_WRITE);
1825 data_to_process -= this_len;
1826 if (!data_to_process)
1835 dio->bi_status = errno_to_blk_status(r);
1839 static int dm_integrity_map(struct dm_target *ti, struct bio *bio)
1841 struct dm_integrity_c *ic = ti->private;
1842 struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
1843 struct bio_integrity_payload *bip;
1845 sector_t area, offset;
1849 dio->op = bio_op(bio);
1851 if (unlikely(dio->op == REQ_OP_DISCARD)) {
1852 if (ti->max_io_len) {
1853 sector_t sec = dm_target_offset(ti, bio->bi_iter.bi_sector);
1854 unsigned log2_max_io_len = __fls(ti->max_io_len);
1855 sector_t start_boundary = sec >> log2_max_io_len;
1856 sector_t end_boundary = (sec + bio_sectors(bio) - 1) >> log2_max_io_len;
1857 if (start_boundary < end_boundary) {
1858 sector_t len = ti->max_io_len - (sec & (ti->max_io_len - 1));
1859 dm_accept_partial_bio(bio, len);
1864 if (unlikely(bio->bi_opf & REQ_PREFLUSH)) {
1865 submit_flush_bio(ic, dio);
1866 return DM_MAPIO_SUBMITTED;
1869 dio->range.logical_sector = dm_target_offset(ti, bio->bi_iter.bi_sector);
1870 dio->fua = dio->op == REQ_OP_WRITE && bio->bi_opf & REQ_FUA;
1871 if (unlikely(dio->fua)) {
1873 * Don't pass down the FUA flag because we have to flush
1874 * disk cache anyway.
1876 bio->bi_opf &= ~REQ_FUA;
1878 if (unlikely(dio->range.logical_sector + bio_sectors(bio) > ic->provided_data_sectors)) {
1879 DMERR("Too big sector number: 0x%llx + 0x%x > 0x%llx",
1880 dio->range.logical_sector, bio_sectors(bio),
1881 ic->provided_data_sectors);
1882 return DM_MAPIO_KILL;
1884 if (unlikely((dio->range.logical_sector | bio_sectors(bio)) & (unsigned)(ic->sectors_per_block - 1))) {
1885 DMERR("Bio not aligned on %u sectors: 0x%llx, 0x%x",
1886 ic->sectors_per_block,
1887 dio->range.logical_sector, bio_sectors(bio));
1888 return DM_MAPIO_KILL;
1891 if (ic->sectors_per_block > 1 && likely(dio->op != REQ_OP_DISCARD)) {
1892 struct bvec_iter iter;
1894 bio_for_each_segment(bv, bio, iter) {
1895 if (unlikely(bv.bv_len & ((ic->sectors_per_block << SECTOR_SHIFT) - 1))) {
1896 DMERR("Bio vector (%u,%u) is not aligned on %u-sector boundary",
1897 bv.bv_offset, bv.bv_len, ic->sectors_per_block);
1898 return DM_MAPIO_KILL;
1903 bip = bio_integrity(bio);
1904 if (!ic->internal_hash) {
1906 unsigned wanted_tag_size = bio_sectors(bio) >> ic->sb->log2_sectors_per_block;
1907 if (ic->log2_tag_size >= 0)
1908 wanted_tag_size <<= ic->log2_tag_size;
1910 wanted_tag_size *= ic->tag_size;
1911 if (unlikely(wanted_tag_size != bip->bip_iter.bi_size)) {
1912 DMERR("Invalid integrity data size %u, expected %u",
1913 bip->bip_iter.bi_size, wanted_tag_size);
1914 return DM_MAPIO_KILL;
1918 if (unlikely(bip != NULL)) {
1919 DMERR("Unexpected integrity data when using internal hash");
1920 return DM_MAPIO_KILL;
1924 if (unlikely(ic->mode == 'R') && unlikely(dio->op != REQ_OP_READ))
1925 return DM_MAPIO_KILL;
1927 get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
1928 dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset, &dio->metadata_offset);
1929 bio->bi_iter.bi_sector = get_data_sector(ic, area, offset);
1931 dm_integrity_map_continue(dio, true);
1932 return DM_MAPIO_SUBMITTED;
1935 static bool __journal_read_write(struct dm_integrity_io *dio, struct bio *bio,
1936 unsigned journal_section, unsigned journal_entry)
1938 struct dm_integrity_c *ic = dio->ic;
1939 sector_t logical_sector;
1942 logical_sector = dio->range.logical_sector;
1943 n_sectors = dio->range.n_sectors;
1945 struct bio_vec bv = bio_iovec(bio);
1948 if (unlikely(bv.bv_len >> SECTOR_SHIFT > n_sectors))
1949 bv.bv_len = n_sectors << SECTOR_SHIFT;
1950 n_sectors -= bv.bv_len >> SECTOR_SHIFT;
1951 bio_advance_iter(bio, &bio->bi_iter, bv.bv_len);
1953 mem = kmap_atomic(bv.bv_page);
1954 if (likely(dio->op == REQ_OP_WRITE))
1955 flush_dcache_page(bv.bv_page);
1958 struct journal_entry *je = access_journal_entry(ic, journal_section, journal_entry);
1960 if (unlikely(dio->op == REQ_OP_READ)) {
1961 struct journal_sector *js;
1965 if (unlikely(journal_entry_is_inprogress(je))) {
1966 flush_dcache_page(bv.bv_page);
1969 __io_wait_event(ic->copy_to_journal_wait, !journal_entry_is_inprogress(je));
1973 BUG_ON(journal_entry_get_sector(je) != logical_sector);
1974 js = access_journal_data(ic, journal_section, journal_entry);
1975 mem_ptr = mem + bv.bv_offset;
1978 memcpy(mem_ptr, js, JOURNAL_SECTOR_DATA);
1979 *(commit_id_t *)(mem_ptr + JOURNAL_SECTOR_DATA) = je->last_bytes[s];
1981 mem_ptr += 1 << SECTOR_SHIFT;
1982 } while (++s < ic->sectors_per_block);
1983 #ifdef INTERNAL_VERIFY
1984 if (ic->internal_hash) {
1985 char checksums_onstack[max((size_t)HASH_MAX_DIGESTSIZE, MAX_TAG_SIZE)];
1987 integrity_sector_checksum(ic, logical_sector, mem + bv.bv_offset, checksums_onstack);
1988 if (unlikely(memcmp(checksums_onstack, journal_entry_tag(ic, je), ic->tag_size))) {
1989 DMERR_LIMIT("Checksum failed when reading from journal, at sector 0x%llx",
1996 if (!ic->internal_hash) {
1997 struct bio_integrity_payload *bip = bio_integrity(bio);
1998 unsigned tag_todo = ic->tag_size;
1999 char *tag_ptr = journal_entry_tag(ic, je);
2002 struct bio_vec biv = bvec_iter_bvec(bip->bip_vec, bip->bip_iter);
2003 unsigned tag_now = min(biv.bv_len, tag_todo);
2005 BUG_ON(PageHighMem(biv.bv_page));
2006 tag_addr = lowmem_page_address(biv.bv_page) + biv.bv_offset;
2007 if (likely(dio->op == REQ_OP_WRITE))
2008 memcpy(tag_ptr, tag_addr, tag_now);
2010 memcpy(tag_addr, tag_ptr, tag_now);
2011 bvec_iter_advance(bip->bip_vec, &bip->bip_iter, tag_now);
2013 tag_todo -= tag_now;
2014 } while (unlikely(tag_todo)); else {
2015 if (likely(dio->op == REQ_OP_WRITE))
2016 memset(tag_ptr, 0, tag_todo);
2020 if (likely(dio->op == REQ_OP_WRITE)) {
2021 struct journal_sector *js;
2024 js = access_journal_data(ic, journal_section, journal_entry);
2025 memcpy(js, mem + bv.bv_offset, ic->sectors_per_block << SECTOR_SHIFT);
2029 je->last_bytes[s] = js[s].commit_id;
2030 } while (++s < ic->sectors_per_block);
2032 if (ic->internal_hash) {
2033 unsigned digest_size = crypto_shash_digestsize(ic->internal_hash);
2034 if (unlikely(digest_size > ic->tag_size)) {
2035 char checksums_onstack[HASH_MAX_DIGESTSIZE];
2036 integrity_sector_checksum(ic, logical_sector, (char *)js, checksums_onstack);
2037 memcpy(journal_entry_tag(ic, je), checksums_onstack, ic->tag_size);
2039 integrity_sector_checksum(ic, logical_sector, (char *)js, journal_entry_tag(ic, je));
2042 journal_entry_set_sector(je, logical_sector);
2044 logical_sector += ic->sectors_per_block;
2047 if (unlikely(journal_entry == ic->journal_section_entries)) {
2050 wraparound_section(ic, &journal_section);
2053 bv.bv_offset += ic->sectors_per_block << SECTOR_SHIFT;
2054 } while (bv.bv_len -= ic->sectors_per_block << SECTOR_SHIFT);
2056 if (unlikely(dio->op == REQ_OP_READ))
2057 flush_dcache_page(bv.bv_page);
2059 } while (n_sectors);
2061 if (likely(dio->op == REQ_OP_WRITE)) {
2063 if (unlikely(waitqueue_active(&ic->copy_to_journal_wait)))
2064 wake_up(&ic->copy_to_journal_wait);
2065 if (READ_ONCE(ic->free_sectors) <= ic->free_sectors_threshold) {
2066 queue_work(ic->commit_wq, &ic->commit_work);
2068 schedule_autocommit(ic);
2071 remove_range(ic, &dio->range);
2074 if (unlikely(bio->bi_iter.bi_size)) {
2075 sector_t area, offset;
2077 dio->range.logical_sector = logical_sector;
2078 get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
2079 dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset, &dio->metadata_offset);
2086 static void dm_integrity_map_continue(struct dm_integrity_io *dio, bool from_map)
2088 struct dm_integrity_c *ic = dio->ic;
2089 struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
2090 unsigned journal_section, journal_entry;
2091 unsigned journal_read_pos;
2092 struct completion read_comp;
2093 bool discard_retried = false;
2094 bool need_sync_io = ic->internal_hash && dio->op == REQ_OP_READ;
2095 if (unlikely(dio->op == REQ_OP_DISCARD) && ic->mode != 'D')
2096 need_sync_io = true;
2098 if (need_sync_io && from_map) {
2099 INIT_WORK(&dio->work, integrity_bio_wait);
2100 queue_work(ic->offload_wq, &dio->work);
2105 spin_lock_irq(&ic->endio_wait.lock);
2107 if (unlikely(dm_integrity_failed(ic))) {
2108 spin_unlock_irq(&ic->endio_wait.lock);
2112 dio->range.n_sectors = bio_sectors(bio);
2113 journal_read_pos = NOT_FOUND;
2114 if (ic->mode == 'J' && likely(dio->op != REQ_OP_DISCARD)) {
2115 if (dio->op == REQ_OP_WRITE) {
2116 unsigned next_entry, i, pos;
2117 unsigned ws, we, range_sectors;
2119 dio->range.n_sectors = min(dio->range.n_sectors,
2120 (sector_t)ic->free_sectors << ic->sb->log2_sectors_per_block);
2121 if (unlikely(!dio->range.n_sectors)) {
2123 goto offload_to_thread;
2124 sleep_on_endio_wait(ic);
2127 range_sectors = dio->range.n_sectors >> ic->sb->log2_sectors_per_block;
2128 ic->free_sectors -= range_sectors;
2129 journal_section = ic->free_section;
2130 journal_entry = ic->free_section_entry;
2132 next_entry = ic->free_section_entry + range_sectors;
2133 ic->free_section_entry = next_entry % ic->journal_section_entries;
2134 ic->free_section += next_entry / ic->journal_section_entries;
2135 ic->n_uncommitted_sections += next_entry / ic->journal_section_entries;
2136 wraparound_section(ic, &ic->free_section);
2138 pos = journal_section * ic->journal_section_entries + journal_entry;
2139 ws = journal_section;
2143 struct journal_entry *je;
2145 add_journal_node(ic, &ic->journal_tree[pos], dio->range.logical_sector + i);
2147 if (unlikely(pos >= ic->journal_entries))
2150 je = access_journal_entry(ic, ws, we);
2151 BUG_ON(!journal_entry_is_unused(je));
2152 journal_entry_set_inprogress(je);
2154 if (unlikely(we == ic->journal_section_entries)) {
2157 wraparound_section(ic, &ws);
2159 } while ((i += ic->sectors_per_block) < dio->range.n_sectors);
2161 spin_unlock_irq(&ic->endio_wait.lock);
2162 goto journal_read_write;
2164 sector_t next_sector;
2165 journal_read_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
2166 if (likely(journal_read_pos == NOT_FOUND)) {
2167 if (unlikely(dio->range.n_sectors > next_sector - dio->range.logical_sector))
2168 dio->range.n_sectors = next_sector - dio->range.logical_sector;
2171 unsigned jp = journal_read_pos + 1;
2172 for (i = ic->sectors_per_block; i < dio->range.n_sectors; i += ic->sectors_per_block, jp++) {
2173 if (!test_journal_node(ic, jp, dio->range.logical_sector + i))
2176 dio->range.n_sectors = i;
2180 if (unlikely(!add_new_range(ic, &dio->range, true))) {
2182 * We must not sleep in the request routine because it could
2183 * stall bios on current->bio_list.
2184 * So, we offload the bio to a workqueue if we have to sleep.
2188 spin_unlock_irq(&ic->endio_wait.lock);
2189 INIT_WORK(&dio->work, integrity_bio_wait);
2190 queue_work(ic->wait_wq, &dio->work);
2193 if (journal_read_pos != NOT_FOUND)
2194 dio->range.n_sectors = ic->sectors_per_block;
2195 wait_and_add_new_range(ic, &dio->range);
2197 * wait_and_add_new_range drops the spinlock, so the journal
2198 * may have been changed arbitrarily. We need to recheck.
2199 * To simplify the code, we restrict I/O size to just one block.
2201 if (journal_read_pos != NOT_FOUND) {
2202 sector_t next_sector;
2203 unsigned new_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
2204 if (unlikely(new_pos != journal_read_pos)) {
2205 remove_range_unlocked(ic, &dio->range);
2210 if (ic->mode == 'J' && likely(dio->op == REQ_OP_DISCARD) && !discard_retried) {
2211 sector_t next_sector;
2212 unsigned new_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
2213 if (unlikely(new_pos != NOT_FOUND) ||
2214 unlikely(next_sector < dio->range.logical_sector - dio->range.n_sectors)) {
2215 remove_range_unlocked(ic, &dio->range);
2216 spin_unlock_irq(&ic->endio_wait.lock);
2217 queue_work(ic->commit_wq, &ic->commit_work);
2218 flush_workqueue(ic->commit_wq);
2219 queue_work(ic->writer_wq, &ic->writer_work);
2220 flush_workqueue(ic->writer_wq);
2221 discard_retried = true;
2225 spin_unlock_irq(&ic->endio_wait.lock);
2227 if (unlikely(journal_read_pos != NOT_FOUND)) {
2228 journal_section = journal_read_pos / ic->journal_section_entries;
2229 journal_entry = journal_read_pos % ic->journal_section_entries;
2230 goto journal_read_write;
2233 if (ic->mode == 'B' && (dio->op == REQ_OP_WRITE || unlikely(dio->op == REQ_OP_DISCARD))) {
2234 if (!block_bitmap_op(ic, ic->may_write_bitmap, dio->range.logical_sector,
2235 dio->range.n_sectors, BITMAP_OP_TEST_ALL_SET)) {
2236 struct bitmap_block_status *bbs;
2238 bbs = sector_to_bitmap_block(ic, dio->range.logical_sector);
2239 spin_lock(&bbs->bio_queue_lock);
2240 bio_list_add(&bbs->bio_queue, bio);
2241 spin_unlock(&bbs->bio_queue_lock);
2242 queue_work(ic->writer_wq, &bbs->work);
2247 dio->in_flight = (atomic_t)ATOMIC_INIT(2);
2250 init_completion(&read_comp);
2251 dio->completion = &read_comp;
2253 dio->completion = NULL;
2255 dm_bio_record(&dio->bio_details, bio);
2256 bio_set_dev(bio, ic->dev->bdev);
2257 bio->bi_integrity = NULL;
2258 bio->bi_opf &= ~REQ_INTEGRITY;
2259 bio->bi_end_io = integrity_end_io;
2260 bio->bi_iter.bi_size = dio->range.n_sectors << SECTOR_SHIFT;
2262 if (unlikely(dio->op == REQ_OP_DISCARD) && likely(ic->mode != 'D')) {
2263 integrity_metadata(&dio->work);
2264 dm_integrity_flush_buffers(ic, false);
2266 dio->in_flight = (atomic_t)ATOMIC_INIT(1);
2267 dio->completion = NULL;
2269 submit_bio_noacct(bio);
2274 submit_bio_noacct(bio);
2277 wait_for_completion_io(&read_comp);
2278 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING) &&
2279 dio->range.logical_sector + dio->range.n_sectors > le64_to_cpu(ic->sb->recalc_sector))
2281 if (ic->mode == 'B') {
2282 if (!block_bitmap_op(ic, ic->recalc_bitmap, dio->range.logical_sector,
2283 dio->range.n_sectors, BITMAP_OP_TEST_ALL_CLEAR))
2287 if (likely(!bio->bi_status))
2288 integrity_metadata(&dio->work);
2294 INIT_WORK(&dio->work, integrity_metadata);
2295 queue_work(ic->metadata_wq, &dio->work);
2301 if (unlikely(__journal_read_write(dio, bio, journal_section, journal_entry)))
2304 do_endio_flush(ic, dio);
2308 static void integrity_bio_wait(struct work_struct *w)
2310 struct dm_integrity_io *dio = container_of(w, struct dm_integrity_io, work);
2312 dm_integrity_map_continue(dio, false);
2315 static void pad_uncommitted(struct dm_integrity_c *ic)
2317 if (ic->free_section_entry) {
2318 ic->free_sectors -= ic->journal_section_entries - ic->free_section_entry;
2319 ic->free_section_entry = 0;
2321 wraparound_section(ic, &ic->free_section);
2322 ic->n_uncommitted_sections++;
2324 if (WARN_ON(ic->journal_sections * ic->journal_section_entries !=
2325 (ic->n_uncommitted_sections + ic->n_committed_sections) *
2326 ic->journal_section_entries + ic->free_sectors)) {
2327 DMCRIT("journal_sections %u, journal_section_entries %u, "
2328 "n_uncommitted_sections %u, n_committed_sections %u, "
2329 "journal_section_entries %u, free_sectors %u",
2330 ic->journal_sections, ic->journal_section_entries,
2331 ic->n_uncommitted_sections, ic->n_committed_sections,
2332 ic->journal_section_entries, ic->free_sectors);
2336 static void integrity_commit(struct work_struct *w)
2338 struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, commit_work);
2339 unsigned commit_start, commit_sections;
2341 struct bio *flushes;
2343 del_timer(&ic->autocommit_timer);
2345 spin_lock_irq(&ic->endio_wait.lock);
2346 flushes = bio_list_get(&ic->flush_bio_list);
2347 if (unlikely(ic->mode != 'J')) {
2348 spin_unlock_irq(&ic->endio_wait.lock);
2349 dm_integrity_flush_buffers(ic, true);
2350 goto release_flush_bios;
2353 pad_uncommitted(ic);
2354 commit_start = ic->uncommitted_section;
2355 commit_sections = ic->n_uncommitted_sections;
2356 spin_unlock_irq(&ic->endio_wait.lock);
2358 if (!commit_sections)
2359 goto release_flush_bios;
2362 for (n = 0; n < commit_sections; n++) {
2363 for (j = 0; j < ic->journal_section_entries; j++) {
2364 struct journal_entry *je;
2365 je = access_journal_entry(ic, i, j);
2366 io_wait_event(ic->copy_to_journal_wait, !journal_entry_is_inprogress(je));
2368 for (j = 0; j < ic->journal_section_sectors; j++) {
2369 struct journal_sector *js;
2370 js = access_journal(ic, i, j);
2371 js->commit_id = dm_integrity_commit_id(ic, i, j, ic->commit_seq);
2374 if (unlikely(i >= ic->journal_sections))
2375 ic->commit_seq = next_commit_seq(ic->commit_seq);
2376 wraparound_section(ic, &i);
2380 write_journal(ic, commit_start, commit_sections);
2382 spin_lock_irq(&ic->endio_wait.lock);
2383 ic->uncommitted_section += commit_sections;
2384 wraparound_section(ic, &ic->uncommitted_section);
2385 ic->n_uncommitted_sections -= commit_sections;
2386 ic->n_committed_sections += commit_sections;
2387 spin_unlock_irq(&ic->endio_wait.lock);
2389 if (READ_ONCE(ic->free_sectors) <= ic->free_sectors_threshold)
2390 queue_work(ic->writer_wq, &ic->writer_work);
2394 struct bio *next = flushes->bi_next;
2395 flushes->bi_next = NULL;
2396 do_endio(ic, flushes);
2401 static void complete_copy_from_journal(unsigned long error, void *context)
2403 struct journal_io *io = context;
2404 struct journal_completion *comp = io->comp;
2405 struct dm_integrity_c *ic = comp->ic;
2406 remove_range(ic, &io->range);
2407 mempool_free(io, &ic->journal_io_mempool);
2408 if (unlikely(error != 0))
2409 dm_integrity_io_error(ic, "copying from journal", -EIO);
2410 complete_journal_op(comp);
2413 static void restore_last_bytes(struct dm_integrity_c *ic, struct journal_sector *js,
2414 struct journal_entry *je)
2418 js->commit_id = je->last_bytes[s];
2420 } while (++s < ic->sectors_per_block);
2423 static void do_journal_write(struct dm_integrity_c *ic, unsigned write_start,
2424 unsigned write_sections, bool from_replay)
2427 struct journal_completion comp;
2428 struct blk_plug plug;
2430 blk_start_plug(&plug);
2433 comp.in_flight = (atomic_t)ATOMIC_INIT(1);
2434 init_completion(&comp.comp);
2437 for (n = 0; n < write_sections; n++, i++, wraparound_section(ic, &i)) {
2438 #ifndef INTERNAL_VERIFY
2439 if (unlikely(from_replay))
2441 rw_section_mac(ic, i, false);
2442 for (j = 0; j < ic->journal_section_entries; j++) {
2443 struct journal_entry *je = access_journal_entry(ic, i, j);
2444 sector_t sec, area, offset;
2445 unsigned k, l, next_loop;
2446 sector_t metadata_block;
2447 unsigned metadata_offset;
2448 struct journal_io *io;
2450 if (journal_entry_is_unused(je))
2452 BUG_ON(unlikely(journal_entry_is_inprogress(je)) && !from_replay);
2453 sec = journal_entry_get_sector(je);
2454 if (unlikely(from_replay)) {
2455 if (unlikely(sec & (unsigned)(ic->sectors_per_block - 1))) {
2456 dm_integrity_io_error(ic, "invalid sector in journal", -EIO);
2457 sec &= ~(sector_t)(ic->sectors_per_block - 1);
2460 if (unlikely(sec >= ic->provided_data_sectors))
2462 get_area_and_offset(ic, sec, &area, &offset);
2463 restore_last_bytes(ic, access_journal_data(ic, i, j), je);
2464 for (k = j + 1; k < ic->journal_section_entries; k++) {
2465 struct journal_entry *je2 = access_journal_entry(ic, i, k);
2466 sector_t sec2, area2, offset2;
2467 if (journal_entry_is_unused(je2))
2469 BUG_ON(unlikely(journal_entry_is_inprogress(je2)) && !from_replay);
2470 sec2 = journal_entry_get_sector(je2);
2471 if (unlikely(sec2 >= ic->provided_data_sectors))
2473 get_area_and_offset(ic, sec2, &area2, &offset2);
2474 if (area2 != area || offset2 != offset + ((k - j) << ic->sb->log2_sectors_per_block))
2476 restore_last_bytes(ic, access_journal_data(ic, i, k), je2);
2480 io = mempool_alloc(&ic->journal_io_mempool, GFP_NOIO);
2482 io->range.logical_sector = sec;
2483 io->range.n_sectors = (k - j) << ic->sb->log2_sectors_per_block;
2485 spin_lock_irq(&ic->endio_wait.lock);
2486 add_new_range_and_wait(ic, &io->range);
2488 if (likely(!from_replay)) {
2489 struct journal_node *section_node = &ic->journal_tree[i * ic->journal_section_entries];
2491 /* don't write if there is newer committed sector */
2492 while (j < k && find_newer_committed_node(ic, §ion_node[j])) {
2493 struct journal_entry *je2 = access_journal_entry(ic, i, j);
2495 journal_entry_set_unused(je2);
2496 remove_journal_node(ic, §ion_node[j]);
2498 sec += ic->sectors_per_block;
2499 offset += ic->sectors_per_block;
2501 while (j < k && find_newer_committed_node(ic, §ion_node[k - 1])) {
2502 struct journal_entry *je2 = access_journal_entry(ic, i, k - 1);
2504 journal_entry_set_unused(je2);
2505 remove_journal_node(ic, §ion_node[k - 1]);
2509 remove_range_unlocked(ic, &io->range);
2510 spin_unlock_irq(&ic->endio_wait.lock);
2511 mempool_free(io, &ic->journal_io_mempool);
2514 for (l = j; l < k; l++) {
2515 remove_journal_node(ic, §ion_node[l]);
2518 spin_unlock_irq(&ic->endio_wait.lock);
2520 metadata_block = get_metadata_sector_and_offset(ic, area, offset, &metadata_offset);
2521 for (l = j; l < k; l++) {
2523 struct journal_entry *je2 = access_journal_entry(ic, i, l);
2526 #ifndef INTERNAL_VERIFY
2527 unlikely(from_replay) &&
2529 ic->internal_hash) {
2530 char test_tag[max_t(size_t, HASH_MAX_DIGESTSIZE, MAX_TAG_SIZE)];
2532 integrity_sector_checksum(ic, sec + ((l - j) << ic->sb->log2_sectors_per_block),
2533 (char *)access_journal_data(ic, i, l), test_tag);
2534 if (unlikely(memcmp(test_tag, journal_entry_tag(ic, je2), ic->tag_size)))
2535 dm_integrity_io_error(ic, "tag mismatch when replaying journal", -EILSEQ);
2538 journal_entry_set_unused(je2);
2539 r = dm_integrity_rw_tag(ic, journal_entry_tag(ic, je2), &metadata_block, &metadata_offset,
2540 ic->tag_size, TAG_WRITE);
2542 dm_integrity_io_error(ic, "reading tags", r);
2546 atomic_inc(&comp.in_flight);
2547 copy_from_journal(ic, i, j << ic->sb->log2_sectors_per_block,
2548 (k - j) << ic->sb->log2_sectors_per_block,
2549 get_data_sector(ic, area, offset),
2550 complete_copy_from_journal, io);
2556 dm_bufio_write_dirty_buffers_async(ic->bufio);
2558 blk_finish_plug(&plug);
2560 complete_journal_op(&comp);
2561 wait_for_completion_io(&comp.comp);
2563 dm_integrity_flush_buffers(ic, true);
2566 static void integrity_writer(struct work_struct *w)
2568 struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, writer_work);
2569 unsigned write_start, write_sections;
2571 unsigned prev_free_sectors;
2573 /* the following test is not needed, but it tests the replay code */
2574 if (unlikely(dm_post_suspending(ic->ti)) && !ic->meta_dev)
2577 spin_lock_irq(&ic->endio_wait.lock);
2578 write_start = ic->committed_section;
2579 write_sections = ic->n_committed_sections;
2580 spin_unlock_irq(&ic->endio_wait.lock);
2582 if (!write_sections)
2585 do_journal_write(ic, write_start, write_sections, false);
2587 spin_lock_irq(&ic->endio_wait.lock);
2589 ic->committed_section += write_sections;
2590 wraparound_section(ic, &ic->committed_section);
2591 ic->n_committed_sections -= write_sections;
2593 prev_free_sectors = ic->free_sectors;
2594 ic->free_sectors += write_sections * ic->journal_section_entries;
2595 if (unlikely(!prev_free_sectors))
2596 wake_up_locked(&ic->endio_wait);
2598 spin_unlock_irq(&ic->endio_wait.lock);
2601 static void recalc_write_super(struct dm_integrity_c *ic)
2605 dm_integrity_flush_buffers(ic, false);
2606 if (dm_integrity_failed(ic))
2609 r = sync_rw_sb(ic, REQ_OP_WRITE, 0);
2611 dm_integrity_io_error(ic, "writing superblock", r);
2614 static void integrity_recalc(struct work_struct *w)
2616 struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, recalc_work);
2617 struct dm_integrity_range range;
2618 struct dm_io_request io_req;
2619 struct dm_io_region io_loc;
2620 sector_t area, offset;
2621 sector_t metadata_block;
2622 unsigned metadata_offset;
2623 sector_t logical_sector, n_sectors;
2627 unsigned super_counter = 0;
2629 DEBUG_print("start recalculation... (position %llx)\n", le64_to_cpu(ic->sb->recalc_sector));
2631 spin_lock_irq(&ic->endio_wait.lock);
2635 if (unlikely(dm_post_suspending(ic->ti)))
2638 range.logical_sector = le64_to_cpu(ic->sb->recalc_sector);
2639 if (unlikely(range.logical_sector >= ic->provided_data_sectors)) {
2640 if (ic->mode == 'B') {
2641 block_bitmap_op(ic, ic->recalc_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_CLEAR);
2642 DEBUG_print("queue_delayed_work: bitmap_flush_work\n");
2643 queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, 0);
2648 get_area_and_offset(ic, range.logical_sector, &area, &offset);
2649 range.n_sectors = min((sector_t)RECALC_SECTORS, ic->provided_data_sectors - range.logical_sector);
2651 range.n_sectors = min(range.n_sectors, ((sector_t)1U << ic->sb->log2_interleave_sectors) - (unsigned)offset);
2653 add_new_range_and_wait(ic, &range);
2654 spin_unlock_irq(&ic->endio_wait.lock);
2655 logical_sector = range.logical_sector;
2656 n_sectors = range.n_sectors;
2658 if (ic->mode == 'B') {
2659 if (block_bitmap_op(ic, ic->recalc_bitmap, logical_sector, n_sectors, BITMAP_OP_TEST_ALL_CLEAR)) {
2660 goto advance_and_next;
2662 while (block_bitmap_op(ic, ic->recalc_bitmap, logical_sector,
2663 ic->sectors_per_block, BITMAP_OP_TEST_ALL_CLEAR)) {
2664 logical_sector += ic->sectors_per_block;
2665 n_sectors -= ic->sectors_per_block;
2668 while (block_bitmap_op(ic, ic->recalc_bitmap, logical_sector + n_sectors - ic->sectors_per_block,
2669 ic->sectors_per_block, BITMAP_OP_TEST_ALL_CLEAR)) {
2670 n_sectors -= ic->sectors_per_block;
2673 get_area_and_offset(ic, logical_sector, &area, &offset);
2676 DEBUG_print("recalculating: %llx, %llx\n", logical_sector, n_sectors);
2678 if (unlikely(++super_counter == RECALC_WRITE_SUPER)) {
2679 recalc_write_super(ic);
2680 if (ic->mode == 'B') {
2681 queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, ic->bitmap_flush_interval);
2686 if (unlikely(dm_integrity_failed(ic)))
2689 io_req.bi_op = REQ_OP_READ;
2690 io_req.bi_op_flags = 0;
2691 io_req.mem.type = DM_IO_VMA;
2692 io_req.mem.ptr.addr = ic->recalc_buffer;
2693 io_req.notify.fn = NULL;
2694 io_req.client = ic->io;
2695 io_loc.bdev = ic->dev->bdev;
2696 io_loc.sector = get_data_sector(ic, area, offset);
2697 io_loc.count = n_sectors;
2699 r = dm_io(&io_req, 1, &io_loc, NULL);
2701 dm_integrity_io_error(ic, "reading data", r);
2705 t = ic->recalc_tags;
2706 for (i = 0; i < n_sectors; i += ic->sectors_per_block) {
2707 integrity_sector_checksum(ic, logical_sector + i, ic->recalc_buffer + (i << SECTOR_SHIFT), t);
2711 metadata_block = get_metadata_sector_and_offset(ic, area, offset, &metadata_offset);
2713 r = dm_integrity_rw_tag(ic, ic->recalc_tags, &metadata_block, &metadata_offset, t - ic->recalc_tags, TAG_WRITE);
2715 dm_integrity_io_error(ic, "writing tags", r);
2719 if (ic->mode == 'B') {
2720 sector_t start, end;
2721 start = (range.logical_sector >>
2722 (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit)) <<
2723 (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
2724 end = ((range.logical_sector + range.n_sectors) >>
2725 (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit)) <<
2726 (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
2727 block_bitmap_op(ic, ic->recalc_bitmap, start, end - start, BITMAP_OP_CLEAR);
2733 spin_lock_irq(&ic->endio_wait.lock);
2734 remove_range_unlocked(ic, &range);
2735 ic->sb->recalc_sector = cpu_to_le64(range.logical_sector + range.n_sectors);
2739 remove_range(ic, &range);
2743 spin_unlock_irq(&ic->endio_wait.lock);
2745 recalc_write_super(ic);
2748 static void bitmap_block_work(struct work_struct *w)
2750 struct bitmap_block_status *bbs = container_of(w, struct bitmap_block_status, work);
2751 struct dm_integrity_c *ic = bbs->ic;
2753 struct bio_list bio_queue;
2754 struct bio_list waiting;
2756 bio_list_init(&waiting);
2758 spin_lock(&bbs->bio_queue_lock);
2759 bio_queue = bbs->bio_queue;
2760 bio_list_init(&bbs->bio_queue);
2761 spin_unlock(&bbs->bio_queue_lock);
2763 while ((bio = bio_list_pop(&bio_queue))) {
2764 struct dm_integrity_io *dio;
2766 dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
2768 if (block_bitmap_op(ic, ic->may_write_bitmap, dio->range.logical_sector,
2769 dio->range.n_sectors, BITMAP_OP_TEST_ALL_SET)) {
2770 remove_range(ic, &dio->range);
2771 INIT_WORK(&dio->work, integrity_bio_wait);
2772 queue_work(ic->offload_wq, &dio->work);
2774 block_bitmap_op(ic, ic->journal, dio->range.logical_sector,
2775 dio->range.n_sectors, BITMAP_OP_SET);
2776 bio_list_add(&waiting, bio);
2780 if (bio_list_empty(&waiting))
2783 rw_journal_sectors(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC,
2784 bbs->idx * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT),
2785 BITMAP_BLOCK_SIZE >> SECTOR_SHIFT, NULL);
2787 while ((bio = bio_list_pop(&waiting))) {
2788 struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
2790 block_bitmap_op(ic, ic->may_write_bitmap, dio->range.logical_sector,
2791 dio->range.n_sectors, BITMAP_OP_SET);
2793 remove_range(ic, &dio->range);
2794 INIT_WORK(&dio->work, integrity_bio_wait);
2795 queue_work(ic->offload_wq, &dio->work);
2798 queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, ic->bitmap_flush_interval);
2801 static void bitmap_flush_work(struct work_struct *work)
2803 struct dm_integrity_c *ic = container_of(work, struct dm_integrity_c, bitmap_flush_work.work);
2804 struct dm_integrity_range range;
2805 unsigned long limit;
2808 dm_integrity_flush_buffers(ic, false);
2810 range.logical_sector = 0;
2811 range.n_sectors = ic->provided_data_sectors;
2813 spin_lock_irq(&ic->endio_wait.lock);
2814 add_new_range_and_wait(ic, &range);
2815 spin_unlock_irq(&ic->endio_wait.lock);
2817 dm_integrity_flush_buffers(ic, true);
2819 limit = ic->provided_data_sectors;
2820 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
2821 limit = le64_to_cpu(ic->sb->recalc_sector)
2822 >> (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit)
2823 << (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
2825 /*DEBUG_print("zeroing journal\n");*/
2826 block_bitmap_op(ic, ic->journal, 0, limit, BITMAP_OP_CLEAR);
2827 block_bitmap_op(ic, ic->may_write_bitmap, 0, limit, BITMAP_OP_CLEAR);
2829 rw_journal_sectors(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, 0,
2830 ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
2832 spin_lock_irq(&ic->endio_wait.lock);
2833 remove_range_unlocked(ic, &range);
2834 while (unlikely((bio = bio_list_pop(&ic->synchronous_bios)) != NULL)) {
2836 spin_unlock_irq(&ic->endio_wait.lock);
2837 spin_lock_irq(&ic->endio_wait.lock);
2839 spin_unlock_irq(&ic->endio_wait.lock);
2843 static void init_journal(struct dm_integrity_c *ic, unsigned start_section,
2844 unsigned n_sections, unsigned char commit_seq)
2851 for (n = 0; n < n_sections; n++) {
2852 i = start_section + n;
2853 wraparound_section(ic, &i);
2854 for (j = 0; j < ic->journal_section_sectors; j++) {
2855 struct journal_sector *js = access_journal(ic, i, j);
2856 memset(&js->entries, 0, JOURNAL_SECTOR_DATA);
2857 js->commit_id = dm_integrity_commit_id(ic, i, j, commit_seq);
2859 for (j = 0; j < ic->journal_section_entries; j++) {
2860 struct journal_entry *je = access_journal_entry(ic, i, j);
2861 journal_entry_set_unused(je);
2865 write_journal(ic, start_section, n_sections);
2868 static int find_commit_seq(struct dm_integrity_c *ic, unsigned i, unsigned j, commit_id_t id)
2871 for (k = 0; k < N_COMMIT_IDS; k++) {
2872 if (dm_integrity_commit_id(ic, i, j, k) == id)
2875 dm_integrity_io_error(ic, "journal commit id", -EIO);
2879 static void replay_journal(struct dm_integrity_c *ic)
2882 bool used_commit_ids[N_COMMIT_IDS];
2883 unsigned max_commit_id_sections[N_COMMIT_IDS];
2884 unsigned write_start, write_sections;
2885 unsigned continue_section;
2887 unsigned char unused, last_used, want_commit_seq;
2889 if (ic->mode == 'R')
2892 if (ic->journal_uptodate)
2898 if (!ic->just_formatted) {
2899 DEBUG_print("reading journal\n");
2900 rw_journal(ic, REQ_OP_READ, 0, 0, ic->journal_sections, NULL);
2902 DEBUG_bytes(lowmem_page_address(ic->journal_io[0].page), 64, "read journal");
2903 if (ic->journal_io) {
2904 struct journal_completion crypt_comp;
2906 init_completion(&crypt_comp.comp);
2907 crypt_comp.in_flight = (atomic_t)ATOMIC_INIT(0);
2908 encrypt_journal(ic, false, 0, ic->journal_sections, &crypt_comp);
2909 wait_for_completion(&crypt_comp.comp);
2911 DEBUG_bytes(lowmem_page_address(ic->journal[0].page), 64, "decrypted journal");
2914 if (dm_integrity_failed(ic))
2917 journal_empty = true;
2918 memset(used_commit_ids, 0, sizeof used_commit_ids);
2919 memset(max_commit_id_sections, 0, sizeof max_commit_id_sections);
2920 for (i = 0; i < ic->journal_sections; i++) {
2921 for (j = 0; j < ic->journal_section_sectors; j++) {
2923 struct journal_sector *js = access_journal(ic, i, j);
2924 k = find_commit_seq(ic, i, j, js->commit_id);
2927 used_commit_ids[k] = true;
2928 max_commit_id_sections[k] = i;
2930 if (journal_empty) {
2931 for (j = 0; j < ic->journal_section_entries; j++) {
2932 struct journal_entry *je = access_journal_entry(ic, i, j);
2933 if (!journal_entry_is_unused(je)) {
2934 journal_empty = false;
2941 if (!used_commit_ids[N_COMMIT_IDS - 1]) {
2942 unused = N_COMMIT_IDS - 1;
2943 while (unused && !used_commit_ids[unused - 1])
2946 for (unused = 0; unused < N_COMMIT_IDS; unused++)
2947 if (!used_commit_ids[unused])
2949 if (unused == N_COMMIT_IDS) {
2950 dm_integrity_io_error(ic, "journal commit ids", -EIO);
2954 DEBUG_print("first unused commit seq %d [%d,%d,%d,%d]\n",
2955 unused, used_commit_ids[0], used_commit_ids[1],
2956 used_commit_ids[2], used_commit_ids[3]);
2958 last_used = prev_commit_seq(unused);
2959 want_commit_seq = prev_commit_seq(last_used);
2961 if (!used_commit_ids[want_commit_seq] && used_commit_ids[prev_commit_seq(want_commit_seq)])
2962 journal_empty = true;
2964 write_start = max_commit_id_sections[last_used] + 1;
2965 if (unlikely(write_start >= ic->journal_sections))
2966 want_commit_seq = next_commit_seq(want_commit_seq);
2967 wraparound_section(ic, &write_start);
2970 for (write_sections = 0; write_sections < ic->journal_sections; write_sections++) {
2971 for (j = 0; j < ic->journal_section_sectors; j++) {
2972 struct journal_sector *js = access_journal(ic, i, j);
2974 if (js->commit_id != dm_integrity_commit_id(ic, i, j, want_commit_seq)) {
2976 * This could be caused by crash during writing.
2977 * We won't replay the inconsistent part of the
2980 DEBUG_print("commit id mismatch at position (%u, %u): %d != %d\n",
2981 i, j, find_commit_seq(ic, i, j, js->commit_id), want_commit_seq);
2986 if (unlikely(i >= ic->journal_sections))
2987 want_commit_seq = next_commit_seq(want_commit_seq);
2988 wraparound_section(ic, &i);
2992 if (!journal_empty) {
2993 DEBUG_print("replaying %u sections, starting at %u, commit seq %d\n",
2994 write_sections, write_start, want_commit_seq);
2995 do_journal_write(ic, write_start, write_sections, true);
2998 if (write_sections == ic->journal_sections && (ic->mode == 'J' || journal_empty)) {
2999 continue_section = write_start;
3000 ic->commit_seq = want_commit_seq;
3001 DEBUG_print("continuing from section %u, commit seq %d\n", write_start, ic->commit_seq);
3004 unsigned char erase_seq;
3006 DEBUG_print("clearing journal\n");
3008 erase_seq = prev_commit_seq(prev_commit_seq(last_used));
3010 init_journal(ic, s, 1, erase_seq);
3012 wraparound_section(ic, &s);
3013 if (ic->journal_sections >= 2) {
3014 init_journal(ic, s, ic->journal_sections - 2, erase_seq);
3015 s += ic->journal_sections - 2;
3016 wraparound_section(ic, &s);
3017 init_journal(ic, s, 1, erase_seq);
3020 continue_section = 0;
3021 ic->commit_seq = next_commit_seq(erase_seq);
3024 ic->committed_section = continue_section;
3025 ic->n_committed_sections = 0;
3027 ic->uncommitted_section = continue_section;
3028 ic->n_uncommitted_sections = 0;
3030 ic->free_section = continue_section;
3031 ic->free_section_entry = 0;
3032 ic->free_sectors = ic->journal_entries;
3034 ic->journal_tree_root = RB_ROOT;
3035 for (i = 0; i < ic->journal_entries; i++)
3036 init_journal_node(&ic->journal_tree[i]);
3039 static void dm_integrity_enter_synchronous_mode(struct dm_integrity_c *ic)
3041 DEBUG_print("dm_integrity_enter_synchronous_mode\n");
3043 if (ic->mode == 'B') {
3044 ic->bitmap_flush_interval = msecs_to_jiffies(10) + 1;
3045 ic->synchronous_mode = 1;
3047 cancel_delayed_work_sync(&ic->bitmap_flush_work);
3048 queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, 0);
3049 flush_workqueue(ic->commit_wq);
3053 static int dm_integrity_reboot(struct notifier_block *n, unsigned long code, void *x)
3055 struct dm_integrity_c *ic = container_of(n, struct dm_integrity_c, reboot_notifier);
3057 DEBUG_print("dm_integrity_reboot\n");
3059 dm_integrity_enter_synchronous_mode(ic);
3064 static void dm_integrity_postsuspend(struct dm_target *ti)
3066 struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
3069 WARN_ON(unregister_reboot_notifier(&ic->reboot_notifier));
3071 del_timer_sync(&ic->autocommit_timer);
3074 drain_workqueue(ic->recalc_wq);
3076 if (ic->mode == 'B')
3077 cancel_delayed_work_sync(&ic->bitmap_flush_work);
3079 queue_work(ic->commit_wq, &ic->commit_work);
3080 drain_workqueue(ic->commit_wq);
3082 if (ic->mode == 'J') {
3084 queue_work(ic->writer_wq, &ic->writer_work);
3085 drain_workqueue(ic->writer_wq);
3086 dm_integrity_flush_buffers(ic, true);
3089 if (ic->mode == 'B') {
3090 dm_integrity_flush_buffers(ic, true);
3092 /* set to 0 to test bitmap replay code */
3093 init_journal(ic, 0, ic->journal_sections, 0);
3094 ic->sb->flags &= ~cpu_to_le32(SB_FLAG_DIRTY_BITMAP);
3095 r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
3097 dm_integrity_io_error(ic, "writing superblock", r);
3101 BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
3103 ic->journal_uptodate = true;
3106 static void dm_integrity_resume(struct dm_target *ti)
3108 struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
3109 __u64 old_provided_data_sectors = le64_to_cpu(ic->sb->provided_data_sectors);
3112 DEBUG_print("resume\n");
3114 if (ic->provided_data_sectors != old_provided_data_sectors) {
3115 if (ic->provided_data_sectors > old_provided_data_sectors &&
3117 ic->sb->log2_blocks_per_bitmap_bit == ic->log2_blocks_per_bitmap_bit) {
3118 rw_journal_sectors(ic, REQ_OP_READ, 0, 0,
3119 ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3120 block_bitmap_op(ic, ic->journal, old_provided_data_sectors,
3121 ic->provided_data_sectors - old_provided_data_sectors, BITMAP_OP_SET);
3122 rw_journal_sectors(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, 0,
3123 ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3126 ic->sb->provided_data_sectors = cpu_to_le64(ic->provided_data_sectors);
3127 r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
3129 dm_integrity_io_error(ic, "writing superblock", r);
3132 if (ic->sb->flags & cpu_to_le32(SB_FLAG_DIRTY_BITMAP)) {
3133 DEBUG_print("resume dirty_bitmap\n");
3134 rw_journal_sectors(ic, REQ_OP_READ, 0, 0,
3135 ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3136 if (ic->mode == 'B') {
3137 if (ic->sb->log2_blocks_per_bitmap_bit == ic->log2_blocks_per_bitmap_bit) {
3138 block_bitmap_copy(ic, ic->recalc_bitmap, ic->journal);
3139 block_bitmap_copy(ic, ic->may_write_bitmap, ic->journal);
3140 if (!block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors,
3141 BITMAP_OP_TEST_ALL_CLEAR)) {
3142 ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
3143 ic->sb->recalc_sector = cpu_to_le64(0);
3146 DEBUG_print("non-matching blocks_per_bitmap_bit: %u, %u\n",
3147 ic->sb->log2_blocks_per_bitmap_bit, ic->log2_blocks_per_bitmap_bit);
3148 ic->sb->log2_blocks_per_bitmap_bit = ic->log2_blocks_per_bitmap_bit;
3149 block_bitmap_op(ic, ic->recalc_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_SET);
3150 block_bitmap_op(ic, ic->may_write_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_SET);
3151 block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors, BITMAP_OP_SET);
3152 rw_journal_sectors(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, 0,
3153 ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3154 ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
3155 ic->sb->recalc_sector = cpu_to_le64(0);
3158 if (!(ic->sb->log2_blocks_per_bitmap_bit == ic->log2_blocks_per_bitmap_bit &&
3159 block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors, BITMAP_OP_TEST_ALL_CLEAR))) {
3160 ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
3161 ic->sb->recalc_sector = cpu_to_le64(0);
3163 init_journal(ic, 0, ic->journal_sections, 0);
3165 ic->sb->flags &= ~cpu_to_le32(SB_FLAG_DIRTY_BITMAP);
3167 r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
3169 dm_integrity_io_error(ic, "writing superblock", r);
3172 if (ic->mode == 'B') {
3173 ic->sb->flags |= cpu_to_le32(SB_FLAG_DIRTY_BITMAP);
3174 ic->sb->log2_blocks_per_bitmap_bit = ic->log2_blocks_per_bitmap_bit;
3175 r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
3177 dm_integrity_io_error(ic, "writing superblock", r);
3179 block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors, BITMAP_OP_CLEAR);
3180 block_bitmap_op(ic, ic->recalc_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_CLEAR);
3181 block_bitmap_op(ic, ic->may_write_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_CLEAR);
3182 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING) &&
3183 le64_to_cpu(ic->sb->recalc_sector) < ic->provided_data_sectors) {
3184 block_bitmap_op(ic, ic->journal, le64_to_cpu(ic->sb->recalc_sector),
3185 ic->provided_data_sectors - le64_to_cpu(ic->sb->recalc_sector), BITMAP_OP_SET);
3186 block_bitmap_op(ic, ic->recalc_bitmap, le64_to_cpu(ic->sb->recalc_sector),
3187 ic->provided_data_sectors - le64_to_cpu(ic->sb->recalc_sector), BITMAP_OP_SET);
3188 block_bitmap_op(ic, ic->may_write_bitmap, le64_to_cpu(ic->sb->recalc_sector),
3189 ic->provided_data_sectors - le64_to_cpu(ic->sb->recalc_sector), BITMAP_OP_SET);
3191 rw_journal_sectors(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, 0,
3192 ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3196 DEBUG_print("testing recalc: %x\n", ic->sb->flags);
3197 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
3198 __u64 recalc_pos = le64_to_cpu(ic->sb->recalc_sector);
3199 DEBUG_print("recalc pos: %llx / %llx\n", recalc_pos, ic->provided_data_sectors);
3200 if (recalc_pos < ic->provided_data_sectors) {
3201 queue_work(ic->recalc_wq, &ic->recalc_work);
3202 } else if (recalc_pos > ic->provided_data_sectors) {
3203 ic->sb->recalc_sector = cpu_to_le64(ic->provided_data_sectors);
3204 recalc_write_super(ic);
3208 ic->reboot_notifier.notifier_call = dm_integrity_reboot;
3209 ic->reboot_notifier.next = NULL;
3210 ic->reboot_notifier.priority = INT_MAX - 1; /* be notified after md and before hardware drivers */
3211 WARN_ON(register_reboot_notifier(&ic->reboot_notifier));
3214 /* set to 1 to stress test synchronous mode */
3215 dm_integrity_enter_synchronous_mode(ic);
3219 static void dm_integrity_status(struct dm_target *ti, status_type_t type,
3220 unsigned status_flags, char *result, unsigned maxlen)
3222 struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
3227 case STATUSTYPE_INFO:
3229 (unsigned long long)atomic64_read(&ic->number_of_mismatches),
3230 ic->provided_data_sectors);
3231 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
3232 DMEMIT(" %llu", le64_to_cpu(ic->sb->recalc_sector));
3237 case STATUSTYPE_TABLE: {
3238 __u64 watermark_percentage = (__u64)(ic->journal_entries - ic->free_sectors_threshold) * 100;
3239 watermark_percentage += ic->journal_entries / 2;
3240 do_div(watermark_percentage, ic->journal_entries);
3242 arg_count += !!ic->meta_dev;
3243 arg_count += ic->sectors_per_block != 1;
3244 arg_count += !!(ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING));
3245 arg_count += ic->discard;
3246 arg_count += ic->mode == 'J';
3247 arg_count += ic->mode == 'J';
3248 arg_count += ic->mode == 'B';
3249 arg_count += ic->mode == 'B';
3250 arg_count += !!ic->internal_hash_alg.alg_string;
3251 arg_count += !!ic->journal_crypt_alg.alg_string;
3252 arg_count += !!ic->journal_mac_alg.alg_string;
3253 arg_count += (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING)) != 0;
3254 arg_count += (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) != 0;
3255 arg_count += ic->legacy_recalculate;
3256 DMEMIT("%s %llu %u %c %u", ic->dev->name, ic->start,
3257 ic->tag_size, ic->mode, arg_count);
3259 DMEMIT(" meta_device:%s", ic->meta_dev->name);
3260 if (ic->sectors_per_block != 1)
3261 DMEMIT(" block_size:%u", ic->sectors_per_block << SECTOR_SHIFT);
3262 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
3263 DMEMIT(" recalculate");
3265 DMEMIT(" allow_discards");
3266 DMEMIT(" journal_sectors:%u", ic->initial_sectors - SB_SECTORS);
3267 DMEMIT(" interleave_sectors:%u", 1U << ic->sb->log2_interleave_sectors);
3268 DMEMIT(" buffer_sectors:%u", 1U << ic->log2_buffer_sectors);
3269 if (ic->mode == 'J') {
3270 DMEMIT(" journal_watermark:%u", (unsigned)watermark_percentage);
3271 DMEMIT(" commit_time:%u", ic->autocommit_msec);
3273 if (ic->mode == 'B') {
3274 DMEMIT(" sectors_per_bit:%llu", (sector_t)ic->sectors_per_block << ic->log2_blocks_per_bitmap_bit);
3275 DMEMIT(" bitmap_flush_interval:%u", jiffies_to_msecs(ic->bitmap_flush_interval));
3277 if ((ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING)) != 0)
3278 DMEMIT(" fix_padding");
3279 if ((ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) != 0)
3280 DMEMIT(" fix_hmac");
3281 if (ic->legacy_recalculate)
3282 DMEMIT(" legacy_recalculate");
3284 #define EMIT_ALG(a, n) \
3286 if (ic->a.alg_string) { \
3287 DMEMIT(" %s:%s", n, ic->a.alg_string); \
3288 if (ic->a.key_string) \
3289 DMEMIT(":%s", ic->a.key_string);\
3292 EMIT_ALG(internal_hash_alg, "internal_hash");
3293 EMIT_ALG(journal_crypt_alg, "journal_crypt");
3294 EMIT_ALG(journal_mac_alg, "journal_mac");
3300 static int dm_integrity_iterate_devices(struct dm_target *ti,
3301 iterate_devices_callout_fn fn, void *data)
3303 struct dm_integrity_c *ic = ti->private;
3306 return fn(ti, ic->dev, ic->start + ic->initial_sectors + ic->metadata_run, ti->len, data);
3308 return fn(ti, ic->dev, 0, ti->len, data);
3311 static void dm_integrity_io_hints(struct dm_target *ti, struct queue_limits *limits)
3313 struct dm_integrity_c *ic = ti->private;
3315 if (ic->sectors_per_block > 1) {
3316 limits->logical_block_size = ic->sectors_per_block << SECTOR_SHIFT;
3317 limits->physical_block_size = ic->sectors_per_block << SECTOR_SHIFT;
3318 blk_limits_io_min(limits, ic->sectors_per_block << SECTOR_SHIFT);
3322 static void calculate_journal_section_size(struct dm_integrity_c *ic)
3324 unsigned sector_space = JOURNAL_SECTOR_DATA;
3326 ic->journal_sections = le32_to_cpu(ic->sb->journal_sections);
3327 ic->journal_entry_size = roundup(offsetof(struct journal_entry, last_bytes[ic->sectors_per_block]) + ic->tag_size,
3328 JOURNAL_ENTRY_ROUNDUP);
3330 if (ic->sb->flags & cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC))
3331 sector_space -= JOURNAL_MAC_PER_SECTOR;
3332 ic->journal_entries_per_sector = sector_space / ic->journal_entry_size;
3333 ic->journal_section_entries = ic->journal_entries_per_sector * JOURNAL_BLOCK_SECTORS;
3334 ic->journal_section_sectors = (ic->journal_section_entries << ic->sb->log2_sectors_per_block) + JOURNAL_BLOCK_SECTORS;
3335 ic->journal_entries = ic->journal_section_entries * ic->journal_sections;
3338 static int calculate_device_limits(struct dm_integrity_c *ic)
3340 __u64 initial_sectors;
3342 calculate_journal_section_size(ic);
3343 initial_sectors = SB_SECTORS + (__u64)ic->journal_section_sectors * ic->journal_sections;
3344 if (initial_sectors + METADATA_PADDING_SECTORS >= ic->meta_device_sectors || initial_sectors > UINT_MAX)
3346 ic->initial_sectors = initial_sectors;
3348 if (!ic->meta_dev) {
3349 sector_t last_sector, last_area, last_offset;
3351 /* we have to maintain excessive padding for compatibility with existing volumes */
3352 __u64 metadata_run_padding =
3353 ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING) ?
3354 (__u64)(METADATA_PADDING_SECTORS << SECTOR_SHIFT) :
3355 (__u64)(1 << SECTOR_SHIFT << METADATA_PADDING_SECTORS);
3357 ic->metadata_run = round_up((__u64)ic->tag_size << (ic->sb->log2_interleave_sectors - ic->sb->log2_sectors_per_block),
3358 metadata_run_padding) >> SECTOR_SHIFT;
3359 if (!(ic->metadata_run & (ic->metadata_run - 1)))
3360 ic->log2_metadata_run = __ffs(ic->metadata_run);
3362 ic->log2_metadata_run = -1;
3364 get_area_and_offset(ic, ic->provided_data_sectors - 1, &last_area, &last_offset);
3365 last_sector = get_data_sector(ic, last_area, last_offset);
3366 if (last_sector < ic->start || last_sector >= ic->meta_device_sectors)
3369 __u64 meta_size = (ic->provided_data_sectors >> ic->sb->log2_sectors_per_block) * ic->tag_size;
3370 meta_size = (meta_size + ((1U << (ic->log2_buffer_sectors + SECTOR_SHIFT)) - 1))
3371 >> (ic->log2_buffer_sectors + SECTOR_SHIFT);
3372 meta_size <<= ic->log2_buffer_sectors;
3373 if (ic->initial_sectors + meta_size < ic->initial_sectors ||
3374 ic->initial_sectors + meta_size > ic->meta_device_sectors)
3376 ic->metadata_run = 1;
3377 ic->log2_metadata_run = 0;
3383 static void get_provided_data_sectors(struct dm_integrity_c *ic)
3385 if (!ic->meta_dev) {
3387 ic->provided_data_sectors = 0;
3388 for (test_bit = fls64(ic->meta_device_sectors) - 1; test_bit >= 3; test_bit--) {
3389 __u64 prev_data_sectors = ic->provided_data_sectors;
3391 ic->provided_data_sectors |= (sector_t)1 << test_bit;
3392 if (calculate_device_limits(ic))
3393 ic->provided_data_sectors = prev_data_sectors;
3396 ic->provided_data_sectors = ic->data_device_sectors;
3397 ic->provided_data_sectors &= ~(sector_t)(ic->sectors_per_block - 1);
3401 static int initialize_superblock(struct dm_integrity_c *ic, unsigned journal_sectors, unsigned interleave_sectors)
3403 unsigned journal_sections;
3406 memset(ic->sb, 0, SB_SECTORS << SECTOR_SHIFT);
3407 memcpy(ic->sb->magic, SB_MAGIC, 8);
3408 ic->sb->integrity_tag_size = cpu_to_le16(ic->tag_size);
3409 ic->sb->log2_sectors_per_block = __ffs(ic->sectors_per_block);
3410 if (ic->journal_mac_alg.alg_string)
3411 ic->sb->flags |= cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC);
3413 calculate_journal_section_size(ic);
3414 journal_sections = journal_sectors / ic->journal_section_sectors;
3415 if (!journal_sections)
3416 journal_sections = 1;
3418 if (ic->fix_hmac && (ic->internal_hash_alg.alg_string || ic->journal_mac_alg.alg_string)) {
3419 ic->sb->flags |= cpu_to_le32(SB_FLAG_FIXED_HMAC);
3420 get_random_bytes(ic->sb->salt, SALT_SIZE);
3423 if (!ic->meta_dev) {
3424 if (ic->fix_padding)
3425 ic->sb->flags |= cpu_to_le32(SB_FLAG_FIXED_PADDING);
3426 ic->sb->journal_sections = cpu_to_le32(journal_sections);
3427 if (!interleave_sectors)
3428 interleave_sectors = DEFAULT_INTERLEAVE_SECTORS;
3429 ic->sb->log2_interleave_sectors = __fls(interleave_sectors);
3430 ic->sb->log2_interleave_sectors = max((__u8)MIN_LOG2_INTERLEAVE_SECTORS, ic->sb->log2_interleave_sectors);
3431 ic->sb->log2_interleave_sectors = min((__u8)MAX_LOG2_INTERLEAVE_SECTORS, ic->sb->log2_interleave_sectors);
3433 get_provided_data_sectors(ic);
3434 if (!ic->provided_data_sectors)
3437 ic->sb->log2_interleave_sectors = 0;
3439 get_provided_data_sectors(ic);
3440 if (!ic->provided_data_sectors)
3444 ic->sb->journal_sections = cpu_to_le32(0);
3445 for (test_bit = fls(journal_sections) - 1; test_bit >= 0; test_bit--) {
3446 __u32 prev_journal_sections = le32_to_cpu(ic->sb->journal_sections);
3447 __u32 test_journal_sections = prev_journal_sections | (1U << test_bit);
3448 if (test_journal_sections > journal_sections)
3450 ic->sb->journal_sections = cpu_to_le32(test_journal_sections);
3451 if (calculate_device_limits(ic))
3452 ic->sb->journal_sections = cpu_to_le32(prev_journal_sections);
3455 if (!le32_to_cpu(ic->sb->journal_sections)) {
3456 if (ic->log2_buffer_sectors > 3) {
3457 ic->log2_buffer_sectors--;
3458 goto try_smaller_buffer;
3464 ic->sb->provided_data_sectors = cpu_to_le64(ic->provided_data_sectors);
3471 static void dm_integrity_set(struct dm_target *ti, struct dm_integrity_c *ic)
3473 struct gendisk *disk = dm_disk(dm_table_get_md(ti->table));
3474 struct blk_integrity bi;
3476 memset(&bi, 0, sizeof(bi));
3477 bi.profile = &dm_integrity_profile;
3478 bi.tuple_size = ic->tag_size;
3479 bi.tag_size = bi.tuple_size;
3480 bi.interval_exp = ic->sb->log2_sectors_per_block + SECTOR_SHIFT;
3482 blk_integrity_register(disk, &bi);
3483 blk_queue_max_integrity_segments(disk->queue, UINT_MAX);
3486 static void dm_integrity_free_page_list(struct page_list *pl)
3492 for (i = 0; pl[i].page; i++)
3493 __free_page(pl[i].page);
3497 static struct page_list *dm_integrity_alloc_page_list(unsigned n_pages)
3499 struct page_list *pl;
3502 pl = kvmalloc_array(n_pages + 1, sizeof(struct page_list), GFP_KERNEL | __GFP_ZERO);
3506 for (i = 0; i < n_pages; i++) {
3507 pl[i].page = alloc_page(GFP_KERNEL);
3509 dm_integrity_free_page_list(pl);
3513 pl[i - 1].next = &pl[i];
3521 static void dm_integrity_free_journal_scatterlist(struct dm_integrity_c *ic, struct scatterlist **sl)
3524 for (i = 0; i < ic->journal_sections; i++)
3529 static struct scatterlist **dm_integrity_alloc_journal_scatterlist(struct dm_integrity_c *ic,
3530 struct page_list *pl)
3532 struct scatterlist **sl;
3535 sl = kvmalloc_array(ic->journal_sections,
3536 sizeof(struct scatterlist *),
3537 GFP_KERNEL | __GFP_ZERO);
3541 for (i = 0; i < ic->journal_sections; i++) {
3542 struct scatterlist *s;
3543 unsigned start_index, start_offset;
3544 unsigned end_index, end_offset;
3548 page_list_location(ic, i, 0, &start_index, &start_offset);
3549 page_list_location(ic, i, ic->journal_section_sectors - 1,
3550 &end_index, &end_offset);
3552 n_pages = (end_index - start_index + 1);
3554 s = kvmalloc_array(n_pages, sizeof(struct scatterlist),
3557 dm_integrity_free_journal_scatterlist(ic, sl);
3561 sg_init_table(s, n_pages);
3562 for (idx = start_index; idx <= end_index; idx++) {
3563 char *va = lowmem_page_address(pl[idx].page);
3564 unsigned start = 0, end = PAGE_SIZE;
3565 if (idx == start_index)
3566 start = start_offset;
3567 if (idx == end_index)
3568 end = end_offset + (1 << SECTOR_SHIFT);
3569 sg_set_buf(&s[idx - start_index], va + start, end - start);
3578 static void free_alg(struct alg_spec *a)
3580 kfree_sensitive(a->alg_string);
3581 kfree_sensitive(a->key);
3582 memset(a, 0, sizeof *a);
3585 static int get_alg_and_key(const char *arg, struct alg_spec *a, char **error, char *error_inval)
3591 a->alg_string = kstrdup(strchr(arg, ':') + 1, GFP_KERNEL);
3595 k = strchr(a->alg_string, ':');
3598 a->key_string = k + 1;
3599 if (strlen(a->key_string) & 1)
3602 a->key_size = strlen(a->key_string) / 2;
3603 a->key = kmalloc(a->key_size, GFP_KERNEL);
3606 if (hex2bin(a->key, a->key_string, a->key_size))
3612 *error = error_inval;
3615 *error = "Out of memory for an argument";
3619 static int get_mac(struct crypto_shash **hash, struct alg_spec *a, char **error,
3620 char *error_alg, char *error_key)
3624 if (a->alg_string) {
3625 *hash = crypto_alloc_shash(a->alg_string, 0, CRYPTO_ALG_ALLOCATES_MEMORY);
3626 if (IS_ERR(*hash)) {
3634 r = crypto_shash_setkey(*hash, a->key, a->key_size);
3639 } else if (crypto_shash_get_flags(*hash) & CRYPTO_TFM_NEED_KEY) {
3648 static int create_journal(struct dm_integrity_c *ic, char **error)
3652 __u64 journal_pages, journal_desc_size, journal_tree_size;
3653 unsigned char *crypt_data = NULL, *crypt_iv = NULL;
3654 struct skcipher_request *req = NULL;
3656 ic->commit_ids[0] = cpu_to_le64(0x1111111111111111ULL);
3657 ic->commit_ids[1] = cpu_to_le64(0x2222222222222222ULL);
3658 ic->commit_ids[2] = cpu_to_le64(0x3333333333333333ULL);
3659 ic->commit_ids[3] = cpu_to_le64(0x4444444444444444ULL);
3661 journal_pages = roundup((__u64)ic->journal_sections * ic->journal_section_sectors,
3662 PAGE_SIZE >> SECTOR_SHIFT) >> (PAGE_SHIFT - SECTOR_SHIFT);
3663 journal_desc_size = journal_pages * sizeof(struct page_list);
3664 if (journal_pages >= totalram_pages() - totalhigh_pages() || journal_desc_size > ULONG_MAX) {
3665 *error = "Journal doesn't fit into memory";
3669 ic->journal_pages = journal_pages;
3671 ic->journal = dm_integrity_alloc_page_list(ic->journal_pages);
3673 *error = "Could not allocate memory for journal";
3677 if (ic->journal_crypt_alg.alg_string) {
3678 unsigned ivsize, blocksize;
3679 struct journal_completion comp;
3682 ic->journal_crypt = crypto_alloc_skcipher(ic->journal_crypt_alg.alg_string, 0, CRYPTO_ALG_ALLOCATES_MEMORY);
3683 if (IS_ERR(ic->journal_crypt)) {
3684 *error = "Invalid journal cipher";
3685 r = PTR_ERR(ic->journal_crypt);
3686 ic->journal_crypt = NULL;
3689 ivsize = crypto_skcipher_ivsize(ic->journal_crypt);
3690 blocksize = crypto_skcipher_blocksize(ic->journal_crypt);
3692 if (ic->journal_crypt_alg.key) {
3693 r = crypto_skcipher_setkey(ic->journal_crypt, ic->journal_crypt_alg.key,
3694 ic->journal_crypt_alg.key_size);
3696 *error = "Error setting encryption key";
3700 DEBUG_print("cipher %s, block size %u iv size %u\n",
3701 ic->journal_crypt_alg.alg_string, blocksize, ivsize);
3703 ic->journal_io = dm_integrity_alloc_page_list(ic->journal_pages);
3704 if (!ic->journal_io) {
3705 *error = "Could not allocate memory for journal io";
3710 if (blocksize == 1) {
3711 struct scatterlist *sg;
3713 req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
3715 *error = "Could not allocate crypt request";
3720 crypt_iv = kzalloc(ivsize, GFP_KERNEL);
3722 *error = "Could not allocate iv";
3727 ic->journal_xor = dm_integrity_alloc_page_list(ic->journal_pages);
3728 if (!ic->journal_xor) {
3729 *error = "Could not allocate memory for journal xor";
3734 sg = kvmalloc_array(ic->journal_pages + 1,
3735 sizeof(struct scatterlist),
3738 *error = "Unable to allocate sg list";
3742 sg_init_table(sg, ic->journal_pages + 1);
3743 for (i = 0; i < ic->journal_pages; i++) {
3744 char *va = lowmem_page_address(ic->journal_xor[i].page);
3746 sg_set_buf(&sg[i], va, PAGE_SIZE);
3748 sg_set_buf(&sg[i], &ic->commit_ids, sizeof ic->commit_ids);
3750 skcipher_request_set_crypt(req, sg, sg,
3751 PAGE_SIZE * ic->journal_pages + sizeof ic->commit_ids, crypt_iv);
3752 init_completion(&comp.comp);
3753 comp.in_flight = (atomic_t)ATOMIC_INIT(1);
3754 if (do_crypt(true, req, &comp))
3755 wait_for_completion(&comp.comp);
3757 r = dm_integrity_failed(ic);
3759 *error = "Unable to encrypt journal";
3762 DEBUG_bytes(lowmem_page_address(ic->journal_xor[0].page), 64, "xor data");
3764 crypto_free_skcipher(ic->journal_crypt);
3765 ic->journal_crypt = NULL;
3767 unsigned crypt_len = roundup(ivsize, blocksize);
3769 req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
3771 *error = "Could not allocate crypt request";
3776 crypt_iv = kmalloc(ivsize, GFP_KERNEL);
3778 *error = "Could not allocate iv";
3783 crypt_data = kmalloc(crypt_len, GFP_KERNEL);
3785 *error = "Unable to allocate crypt data";
3790 ic->journal_scatterlist = dm_integrity_alloc_journal_scatterlist(ic, ic->journal);
3791 if (!ic->journal_scatterlist) {
3792 *error = "Unable to allocate sg list";
3796 ic->journal_io_scatterlist = dm_integrity_alloc_journal_scatterlist(ic, ic->journal_io);
3797 if (!ic->journal_io_scatterlist) {
3798 *error = "Unable to allocate sg list";
3802 ic->sk_requests = kvmalloc_array(ic->journal_sections,
3803 sizeof(struct skcipher_request *),
3804 GFP_KERNEL | __GFP_ZERO);
3805 if (!ic->sk_requests) {
3806 *error = "Unable to allocate sk requests";
3810 for (i = 0; i < ic->journal_sections; i++) {
3811 struct scatterlist sg;
3812 struct skcipher_request *section_req;
3813 __u32 section_le = cpu_to_le32(i);
3815 memset(crypt_iv, 0x00, ivsize);
3816 memset(crypt_data, 0x00, crypt_len);
3817 memcpy(crypt_data, §ion_le, min((size_t)crypt_len, sizeof(section_le)));
3819 sg_init_one(&sg, crypt_data, crypt_len);
3820 skcipher_request_set_crypt(req, &sg, &sg, crypt_len, crypt_iv);
3821 init_completion(&comp.comp);
3822 comp.in_flight = (atomic_t)ATOMIC_INIT(1);
3823 if (do_crypt(true, req, &comp))
3824 wait_for_completion(&comp.comp);
3826 r = dm_integrity_failed(ic);
3828 *error = "Unable to generate iv";
3832 section_req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
3834 *error = "Unable to allocate crypt request";
3838 section_req->iv = kmalloc_array(ivsize, 2,
3840 if (!section_req->iv) {
3841 skcipher_request_free(section_req);
3842 *error = "Unable to allocate iv";
3846 memcpy(section_req->iv + ivsize, crypt_data, ivsize);
3847 section_req->cryptlen = (size_t)ic->journal_section_sectors << SECTOR_SHIFT;
3848 ic->sk_requests[i] = section_req;
3849 DEBUG_bytes(crypt_data, ivsize, "iv(%u)", i);
3854 for (i = 0; i < N_COMMIT_IDS; i++) {
3857 for (j = 0; j < i; j++) {
3858 if (ic->commit_ids[j] == ic->commit_ids[i]) {
3859 ic->commit_ids[i] = cpu_to_le64(le64_to_cpu(ic->commit_ids[i]) + 1);
3860 goto retest_commit_id;
3863 DEBUG_print("commit id %u: %016llx\n", i, ic->commit_ids[i]);
3866 journal_tree_size = (__u64)ic->journal_entries * sizeof(struct journal_node);
3867 if (journal_tree_size > ULONG_MAX) {
3868 *error = "Journal doesn't fit into memory";
3872 ic->journal_tree = kvmalloc(journal_tree_size, GFP_KERNEL);
3873 if (!ic->journal_tree) {
3874 *error = "Could not allocate memory for journal tree";
3880 skcipher_request_free(req);
3886 * Construct a integrity mapping
3890 * offset from the start of the device
3892 * D - direct writes, J - journal writes, B - bitmap mode, R - recovery mode
3893 * number of optional arguments
3894 * optional arguments:
3896 * interleave_sectors
3903 * bitmap_flush_interval
3909 static int dm_integrity_ctr(struct dm_target *ti, unsigned argc, char **argv)
3911 struct dm_integrity_c *ic;
3914 unsigned extra_args;
3915 struct dm_arg_set as;
3916 static const struct dm_arg _args[] = {
3917 {0, 17, "Invalid number of feature args"},
3919 unsigned journal_sectors, interleave_sectors, buffer_sectors, journal_watermark, sync_msec;
3920 bool should_write_sb;
3922 unsigned long long start;
3923 __s8 log2_sectors_per_bitmap_bit = -1;
3924 __s8 log2_blocks_per_bitmap_bit;
3925 __u64 bits_in_journal;
3926 __u64 n_bitmap_bits;
3928 #define DIRECT_ARGUMENTS 4
3930 if (argc <= DIRECT_ARGUMENTS) {
3931 ti->error = "Invalid argument count";
3935 ic = kzalloc(sizeof(struct dm_integrity_c), GFP_KERNEL);
3937 ti->error = "Cannot allocate integrity context";
3941 ti->per_io_data_size = sizeof(struct dm_integrity_io);
3944 ic->in_progress = RB_ROOT;
3945 INIT_LIST_HEAD(&ic->wait_list);
3946 init_waitqueue_head(&ic->endio_wait);
3947 bio_list_init(&ic->flush_bio_list);
3948 init_waitqueue_head(&ic->copy_to_journal_wait);
3949 init_completion(&ic->crypto_backoff);
3950 atomic64_set(&ic->number_of_mismatches, 0);
3951 ic->bitmap_flush_interval = BITMAP_FLUSH_INTERVAL;
3953 r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &ic->dev);
3955 ti->error = "Device lookup failed";
3959 if (sscanf(argv[1], "%llu%c", &start, &dummy) != 1 || start != (sector_t)start) {
3960 ti->error = "Invalid starting offset";
3966 if (strcmp(argv[2], "-")) {
3967 if (sscanf(argv[2], "%u%c", &ic->tag_size, &dummy) != 1 || !ic->tag_size) {
3968 ti->error = "Invalid tag size";
3974 if (!strcmp(argv[3], "J") || !strcmp(argv[3], "B") ||
3975 !strcmp(argv[3], "D") || !strcmp(argv[3], "R")) {
3976 ic->mode = argv[3][0];
3978 ti->error = "Invalid mode (expecting J, B, D, R)";
3983 journal_sectors = 0;
3984 interleave_sectors = DEFAULT_INTERLEAVE_SECTORS;
3985 buffer_sectors = DEFAULT_BUFFER_SECTORS;
3986 journal_watermark = DEFAULT_JOURNAL_WATERMARK;
3987 sync_msec = DEFAULT_SYNC_MSEC;
3988 ic->sectors_per_block = 1;
3990 as.argc = argc - DIRECT_ARGUMENTS;
3991 as.argv = argv + DIRECT_ARGUMENTS;
3992 r = dm_read_arg_group(_args, &as, &extra_args, &ti->error);
3996 while (extra_args--) {
3997 const char *opt_string;
3999 unsigned long long llval;
4000 opt_string = dm_shift_arg(&as);
4003 ti->error = "Not enough feature arguments";
4006 if (sscanf(opt_string, "journal_sectors:%u%c", &val, &dummy) == 1)
4007 journal_sectors = val ? val : 1;
4008 else if (sscanf(opt_string, "interleave_sectors:%u%c", &val, &dummy) == 1)
4009 interleave_sectors = val;
4010 else if (sscanf(opt_string, "buffer_sectors:%u%c", &val, &dummy) == 1)
4011 buffer_sectors = val;
4012 else if (sscanf(opt_string, "journal_watermark:%u%c", &val, &dummy) == 1 && val <= 100)
4013 journal_watermark = val;
4014 else if (sscanf(opt_string, "commit_time:%u%c", &val, &dummy) == 1)
4016 else if (!strncmp(opt_string, "meta_device:", strlen("meta_device:"))) {
4018 dm_put_device(ti, ic->meta_dev);
4019 ic->meta_dev = NULL;
4021 r = dm_get_device(ti, strchr(opt_string, ':') + 1,
4022 dm_table_get_mode(ti->table), &ic->meta_dev);
4024 ti->error = "Device lookup failed";
4027 } else if (sscanf(opt_string, "block_size:%u%c", &val, &dummy) == 1) {
4028 if (val < 1 << SECTOR_SHIFT ||
4029 val > MAX_SECTORS_PER_BLOCK << SECTOR_SHIFT ||
4032 ti->error = "Invalid block_size argument";
4035 ic->sectors_per_block = val >> SECTOR_SHIFT;
4036 } else if (sscanf(opt_string, "sectors_per_bit:%llu%c", &llval, &dummy) == 1) {
4037 log2_sectors_per_bitmap_bit = !llval ? 0 : __ilog2_u64(llval);
4038 } else if (sscanf(opt_string, "bitmap_flush_interval:%u%c", &val, &dummy) == 1) {
4039 if (val >= (uint64_t)UINT_MAX * 1000 / HZ) {
4041 ti->error = "Invalid bitmap_flush_interval argument";
4043 ic->bitmap_flush_interval = msecs_to_jiffies(val);
4044 } else if (!strncmp(opt_string, "internal_hash:", strlen("internal_hash:"))) {
4045 r = get_alg_and_key(opt_string, &ic->internal_hash_alg, &ti->error,
4046 "Invalid internal_hash argument");
4049 } else if (!strncmp(opt_string, "journal_crypt:", strlen("journal_crypt:"))) {
4050 r = get_alg_and_key(opt_string, &ic->journal_crypt_alg, &ti->error,
4051 "Invalid journal_crypt argument");
4054 } else if (!strncmp(opt_string, "journal_mac:", strlen("journal_mac:"))) {
4055 r = get_alg_and_key(opt_string, &ic->journal_mac_alg, &ti->error,
4056 "Invalid journal_mac argument");
4059 } else if (!strcmp(opt_string, "recalculate")) {
4060 ic->recalculate_flag = true;
4061 } else if (!strcmp(opt_string, "allow_discards")) {
4063 } else if (!strcmp(opt_string, "fix_padding")) {
4064 ic->fix_padding = true;
4065 } else if (!strcmp(opt_string, "fix_hmac")) {
4066 ic->fix_hmac = true;
4067 } else if (!strcmp(opt_string, "legacy_recalculate")) {
4068 ic->legacy_recalculate = true;
4071 ti->error = "Invalid argument";
4076 ic->data_device_sectors = i_size_read(ic->dev->bdev->bd_inode) >> SECTOR_SHIFT;
4078 ic->meta_device_sectors = ic->data_device_sectors;
4080 ic->meta_device_sectors = i_size_read(ic->meta_dev->bdev->bd_inode) >> SECTOR_SHIFT;
4082 if (!journal_sectors) {
4083 journal_sectors = min((sector_t)DEFAULT_MAX_JOURNAL_SECTORS,
4084 ic->data_device_sectors >> DEFAULT_JOURNAL_SIZE_FACTOR);
4087 if (!buffer_sectors)
4089 ic->log2_buffer_sectors = min((int)__fls(buffer_sectors), 31 - SECTOR_SHIFT);
4091 r = get_mac(&ic->internal_hash, &ic->internal_hash_alg, &ti->error,
4092 "Invalid internal hash", "Error setting internal hash key");
4096 r = get_mac(&ic->journal_mac, &ic->journal_mac_alg, &ti->error,
4097 "Invalid journal mac", "Error setting journal mac key");
4101 if (!ic->tag_size) {
4102 if (!ic->internal_hash) {
4103 ti->error = "Unknown tag size";
4107 ic->tag_size = crypto_shash_digestsize(ic->internal_hash);
4109 if (ic->tag_size > MAX_TAG_SIZE) {
4110 ti->error = "Too big tag size";
4114 if (!(ic->tag_size & (ic->tag_size - 1)))
4115 ic->log2_tag_size = __ffs(ic->tag_size);
4117 ic->log2_tag_size = -1;
4119 if (ic->mode == 'B' && !ic->internal_hash) {
4121 ti->error = "Bitmap mode can be only used with internal hash";
4125 if (ic->discard && !ic->internal_hash) {
4127 ti->error = "Discard can be only used with internal hash";
4131 ic->autocommit_jiffies = msecs_to_jiffies(sync_msec);
4132 ic->autocommit_msec = sync_msec;
4133 timer_setup(&ic->autocommit_timer, autocommit_fn, 0);
4135 ic->io = dm_io_client_create();
4136 if (IS_ERR(ic->io)) {
4137 r = PTR_ERR(ic->io);
4139 ti->error = "Cannot allocate dm io";
4143 r = mempool_init_slab_pool(&ic->journal_io_mempool, JOURNAL_IO_MEMPOOL, journal_io_cache);
4145 ti->error = "Cannot allocate mempool";
4149 ic->metadata_wq = alloc_workqueue("dm-integrity-metadata",
4150 WQ_MEM_RECLAIM, METADATA_WORKQUEUE_MAX_ACTIVE);
4151 if (!ic->metadata_wq) {
4152 ti->error = "Cannot allocate workqueue";
4158 * If this workqueue were percpu, it would cause bio reordering
4159 * and reduced performance.
4161 ic->wait_wq = alloc_workqueue("dm-integrity-wait", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
4163 ti->error = "Cannot allocate workqueue";
4168 ic->offload_wq = alloc_workqueue("dm-integrity-offload", WQ_MEM_RECLAIM,
4169 METADATA_WORKQUEUE_MAX_ACTIVE);
4170 if (!ic->offload_wq) {
4171 ti->error = "Cannot allocate workqueue";
4176 ic->commit_wq = alloc_workqueue("dm-integrity-commit", WQ_MEM_RECLAIM, 1);
4177 if (!ic->commit_wq) {
4178 ti->error = "Cannot allocate workqueue";
4182 INIT_WORK(&ic->commit_work, integrity_commit);
4184 if (ic->mode == 'J' || ic->mode == 'B') {
4185 ic->writer_wq = alloc_workqueue("dm-integrity-writer", WQ_MEM_RECLAIM, 1);
4186 if (!ic->writer_wq) {
4187 ti->error = "Cannot allocate workqueue";
4191 INIT_WORK(&ic->writer_work, integrity_writer);
4194 ic->sb = alloc_pages_exact(SB_SECTORS << SECTOR_SHIFT, GFP_KERNEL);
4197 ti->error = "Cannot allocate superblock area";
4201 r = sync_rw_sb(ic, REQ_OP_READ, 0);
4203 ti->error = "Error reading superblock";
4206 should_write_sb = false;
4207 if (memcmp(ic->sb->magic, SB_MAGIC, 8)) {
4208 if (ic->mode != 'R') {
4209 if (memchr_inv(ic->sb, 0, SB_SECTORS << SECTOR_SHIFT)) {
4211 ti->error = "The device is not initialized";
4216 r = initialize_superblock(ic, journal_sectors, interleave_sectors);
4218 ti->error = "Could not initialize superblock";
4221 if (ic->mode != 'R')
4222 should_write_sb = true;
4225 if (!ic->sb->version || ic->sb->version > SB_VERSION_5) {
4227 ti->error = "Unknown version";
4230 if (le16_to_cpu(ic->sb->integrity_tag_size) != ic->tag_size) {
4232 ti->error = "Tag size doesn't match the information in superblock";
4235 if (ic->sb->log2_sectors_per_block != __ffs(ic->sectors_per_block)) {
4237 ti->error = "Block size doesn't match the information in superblock";
4240 if (!le32_to_cpu(ic->sb->journal_sections)) {
4242 ti->error = "Corrupted superblock, journal_sections is 0";
4245 /* make sure that ti->max_io_len doesn't overflow */
4246 if (!ic->meta_dev) {
4247 if (ic->sb->log2_interleave_sectors < MIN_LOG2_INTERLEAVE_SECTORS ||
4248 ic->sb->log2_interleave_sectors > MAX_LOG2_INTERLEAVE_SECTORS) {
4250 ti->error = "Invalid interleave_sectors in the superblock";
4254 if (ic->sb->log2_interleave_sectors) {
4256 ti->error = "Invalid interleave_sectors in the superblock";
4260 if (!!(ic->sb->flags & cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC)) != !!ic->journal_mac_alg.alg_string) {
4262 ti->error = "Journal mac mismatch";
4266 get_provided_data_sectors(ic);
4267 if (!ic->provided_data_sectors) {
4269 ti->error = "The device is too small";
4274 r = calculate_device_limits(ic);
4277 if (ic->log2_buffer_sectors > 3) {
4278 ic->log2_buffer_sectors--;
4279 goto try_smaller_buffer;
4282 ti->error = "The device is too small";
4286 if (log2_sectors_per_bitmap_bit < 0)
4287 log2_sectors_per_bitmap_bit = __fls(DEFAULT_SECTORS_PER_BITMAP_BIT);
4288 if (log2_sectors_per_bitmap_bit < ic->sb->log2_sectors_per_block)
4289 log2_sectors_per_bitmap_bit = ic->sb->log2_sectors_per_block;
4291 bits_in_journal = ((__u64)ic->journal_section_sectors * ic->journal_sections) << (SECTOR_SHIFT + 3);
4292 if (bits_in_journal > UINT_MAX)
4293 bits_in_journal = UINT_MAX;
4294 while (bits_in_journal < (ic->provided_data_sectors + ((sector_t)1 << log2_sectors_per_bitmap_bit) - 1) >> log2_sectors_per_bitmap_bit)
4295 log2_sectors_per_bitmap_bit++;
4297 log2_blocks_per_bitmap_bit = log2_sectors_per_bitmap_bit - ic->sb->log2_sectors_per_block;
4298 ic->log2_blocks_per_bitmap_bit = log2_blocks_per_bitmap_bit;
4299 if (should_write_sb) {
4300 ic->sb->log2_blocks_per_bitmap_bit = log2_blocks_per_bitmap_bit;
4302 n_bitmap_bits = ((ic->provided_data_sectors >> ic->sb->log2_sectors_per_block)
4303 + (((sector_t)1 << log2_blocks_per_bitmap_bit) - 1)) >> log2_blocks_per_bitmap_bit;
4304 ic->n_bitmap_blocks = DIV_ROUND_UP(n_bitmap_bits, BITMAP_BLOCK_SIZE * 8);
4307 ic->log2_buffer_sectors = min(ic->log2_buffer_sectors, (__u8)__ffs(ic->metadata_run));
4309 if (ti->len > ic->provided_data_sectors) {
4311 ti->error = "Not enough provided sectors for requested mapping size";
4316 threshold = (__u64)ic->journal_entries * (100 - journal_watermark);
4318 do_div(threshold, 100);
4319 ic->free_sectors_threshold = threshold;
4321 DEBUG_print("initialized:\n");
4322 DEBUG_print(" integrity_tag_size %u\n", le16_to_cpu(ic->sb->integrity_tag_size));
4323 DEBUG_print(" journal_entry_size %u\n", ic->journal_entry_size);
4324 DEBUG_print(" journal_entries_per_sector %u\n", ic->journal_entries_per_sector);
4325 DEBUG_print(" journal_section_entries %u\n", ic->journal_section_entries);
4326 DEBUG_print(" journal_section_sectors %u\n", ic->journal_section_sectors);
4327 DEBUG_print(" journal_sections %u\n", (unsigned)le32_to_cpu(ic->sb->journal_sections));
4328 DEBUG_print(" journal_entries %u\n", ic->journal_entries);
4329 DEBUG_print(" log2_interleave_sectors %d\n", ic->sb->log2_interleave_sectors);
4330 DEBUG_print(" data_device_sectors 0x%llx\n", i_size_read(ic->dev->bdev->bd_inode) >> SECTOR_SHIFT);
4331 DEBUG_print(" initial_sectors 0x%x\n", ic->initial_sectors);
4332 DEBUG_print(" metadata_run 0x%x\n", ic->metadata_run);
4333 DEBUG_print(" log2_metadata_run %d\n", ic->log2_metadata_run);
4334 DEBUG_print(" provided_data_sectors 0x%llx (%llu)\n", ic->provided_data_sectors, ic->provided_data_sectors);
4335 DEBUG_print(" log2_buffer_sectors %u\n", ic->log2_buffer_sectors);
4336 DEBUG_print(" bits_in_journal %llu\n", bits_in_journal);
4338 if (ic->recalculate_flag && !(ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))) {
4339 ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
4340 ic->sb->recalc_sector = cpu_to_le64(0);
4343 if (ic->internal_hash) {
4344 ic->recalc_wq = alloc_workqueue("dm-integrity-recalc", WQ_MEM_RECLAIM, 1);
4345 if (!ic->recalc_wq ) {
4346 ti->error = "Cannot allocate workqueue";
4350 INIT_WORK(&ic->recalc_work, integrity_recalc);
4351 ic->recalc_buffer = vmalloc(RECALC_SECTORS << SECTOR_SHIFT);
4352 if (!ic->recalc_buffer) {
4353 ti->error = "Cannot allocate buffer for recalculating";
4357 ic->recalc_tags = kvmalloc_array(RECALC_SECTORS >> ic->sb->log2_sectors_per_block,
4358 ic->tag_size, GFP_KERNEL);
4359 if (!ic->recalc_tags) {
4360 ti->error = "Cannot allocate tags for recalculating";
4365 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
4366 ti->error = "Recalculate can only be specified with internal_hash";
4372 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING) &&
4373 le64_to_cpu(ic->sb->recalc_sector) < ic->provided_data_sectors &&
4374 dm_integrity_disable_recalculate(ic)) {
4375 ti->error = "Recalculating with HMAC is disabled for security reasons - if you really need it, use the argument \"legacy_recalculate\"";
4380 ic->bufio = dm_bufio_client_create(ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev,
4381 1U << (SECTOR_SHIFT + ic->log2_buffer_sectors), 1, 0, NULL, NULL);
4382 if (IS_ERR(ic->bufio)) {
4383 r = PTR_ERR(ic->bufio);
4384 ti->error = "Cannot initialize dm-bufio";
4388 dm_bufio_set_sector_offset(ic->bufio, ic->start + ic->initial_sectors);
4390 if (ic->mode != 'R') {
4391 r = create_journal(ic, &ti->error);
4397 if (ic->mode == 'B') {
4399 unsigned n_bitmap_pages = DIV_ROUND_UP(ic->n_bitmap_blocks, PAGE_SIZE / BITMAP_BLOCK_SIZE);
4401 ic->recalc_bitmap = dm_integrity_alloc_page_list(n_bitmap_pages);
4402 if (!ic->recalc_bitmap) {
4406 ic->may_write_bitmap = dm_integrity_alloc_page_list(n_bitmap_pages);
4407 if (!ic->may_write_bitmap) {
4411 ic->bbs = kvmalloc_array(ic->n_bitmap_blocks, sizeof(struct bitmap_block_status), GFP_KERNEL);
4416 INIT_DELAYED_WORK(&ic->bitmap_flush_work, bitmap_flush_work);
4417 for (i = 0; i < ic->n_bitmap_blocks; i++) {
4418 struct bitmap_block_status *bbs = &ic->bbs[i];
4419 unsigned sector, pl_index, pl_offset;
4421 INIT_WORK(&bbs->work, bitmap_block_work);
4424 bio_list_init(&bbs->bio_queue);
4425 spin_lock_init(&bbs->bio_queue_lock);
4427 sector = i * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT);
4428 pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
4429 pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
4431 bbs->bitmap = lowmem_page_address(ic->journal[pl_index].page) + pl_offset;
4435 if (should_write_sb) {
4438 init_journal(ic, 0, ic->journal_sections, 0);
4439 r = dm_integrity_failed(ic);
4441 ti->error = "Error initializing journal";
4444 r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
4446 ti->error = "Error initializing superblock";
4449 ic->just_formatted = true;
4452 if (!ic->meta_dev) {
4453 r = dm_set_target_max_io_len(ti, 1U << ic->sb->log2_interleave_sectors);
4457 if (ic->mode == 'B') {
4458 unsigned max_io_len = ((sector_t)ic->sectors_per_block << ic->log2_blocks_per_bitmap_bit) * (BITMAP_BLOCK_SIZE * 8);
4460 max_io_len = 1U << 31;
4461 DEBUG_print("max_io_len: old %u, new %u\n", ti->max_io_len, max_io_len);
4462 if (!ti->max_io_len || ti->max_io_len > max_io_len) {
4463 r = dm_set_target_max_io_len(ti, max_io_len);
4469 if (!ic->internal_hash)
4470 dm_integrity_set(ti, ic);
4472 ti->num_flush_bios = 1;
4473 ti->flush_supported = true;
4475 ti->num_discard_bios = 1;
4480 dm_integrity_dtr(ti);
4484 static void dm_integrity_dtr(struct dm_target *ti)
4486 struct dm_integrity_c *ic = ti->private;
4488 BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
4489 BUG_ON(!list_empty(&ic->wait_list));
4491 if (ic->metadata_wq)
4492 destroy_workqueue(ic->metadata_wq);
4494 destroy_workqueue(ic->wait_wq);
4496 destroy_workqueue(ic->offload_wq);
4498 destroy_workqueue(ic->commit_wq);
4500 destroy_workqueue(ic->writer_wq);
4502 destroy_workqueue(ic->recalc_wq);
4503 vfree(ic->recalc_buffer);
4504 kvfree(ic->recalc_tags);
4507 dm_bufio_client_destroy(ic->bufio);
4508 mempool_exit(&ic->journal_io_mempool);
4510 dm_io_client_destroy(ic->io);
4512 dm_put_device(ti, ic->dev);
4514 dm_put_device(ti, ic->meta_dev);
4515 dm_integrity_free_page_list(ic->journal);
4516 dm_integrity_free_page_list(ic->journal_io);
4517 dm_integrity_free_page_list(ic->journal_xor);
4518 dm_integrity_free_page_list(ic->recalc_bitmap);
4519 dm_integrity_free_page_list(ic->may_write_bitmap);
4520 if (ic->journal_scatterlist)
4521 dm_integrity_free_journal_scatterlist(ic, ic->journal_scatterlist);
4522 if (ic->journal_io_scatterlist)
4523 dm_integrity_free_journal_scatterlist(ic, ic->journal_io_scatterlist);
4524 if (ic->sk_requests) {
4527 for (i = 0; i < ic->journal_sections; i++) {
4528 struct skcipher_request *req = ic->sk_requests[i];
4530 kfree_sensitive(req->iv);
4531 skcipher_request_free(req);
4534 kvfree(ic->sk_requests);
4536 kvfree(ic->journal_tree);
4538 free_pages_exact(ic->sb, SB_SECTORS << SECTOR_SHIFT);
4540 if (ic->internal_hash)
4541 crypto_free_shash(ic->internal_hash);
4542 free_alg(&ic->internal_hash_alg);
4544 if (ic->journal_crypt)
4545 crypto_free_skcipher(ic->journal_crypt);
4546 free_alg(&ic->journal_crypt_alg);
4548 if (ic->journal_mac)
4549 crypto_free_shash(ic->journal_mac);
4550 free_alg(&ic->journal_mac_alg);
4555 static struct target_type integrity_target = {
4556 .name = "integrity",
4557 .version = {1, 7, 0},
4558 .module = THIS_MODULE,
4559 .features = DM_TARGET_SINGLETON | DM_TARGET_INTEGRITY,
4560 .ctr = dm_integrity_ctr,
4561 .dtr = dm_integrity_dtr,
4562 .map = dm_integrity_map,
4563 .postsuspend = dm_integrity_postsuspend,
4564 .resume = dm_integrity_resume,
4565 .status = dm_integrity_status,
4566 .iterate_devices = dm_integrity_iterate_devices,
4567 .io_hints = dm_integrity_io_hints,
4570 static int __init dm_integrity_init(void)
4574 journal_io_cache = kmem_cache_create("integrity_journal_io",
4575 sizeof(struct journal_io), 0, 0, NULL);
4576 if (!journal_io_cache) {
4577 DMERR("can't allocate journal io cache");
4581 r = dm_register_target(&integrity_target);
4584 DMERR("register failed %d", r);
4589 static void __exit dm_integrity_exit(void)
4591 dm_unregister_target(&integrity_target);
4592 kmem_cache_destroy(journal_io_cache);
4595 module_init(dm_integrity_init);
4596 module_exit(dm_integrity_exit);
4598 MODULE_AUTHOR("Milan Broz");
4599 MODULE_AUTHOR("Mikulas Patocka");
4600 MODULE_DESCRIPTION(DM_NAME " target for integrity tags extension");
4601 MODULE_LICENSE("GPL");