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>
28 #define DM_MSG_PREFIX "integrity"
30 #define DEFAULT_INTERLEAVE_SECTORS 32768
31 #define DEFAULT_JOURNAL_SIZE_FACTOR 7
32 #define DEFAULT_SECTORS_PER_BITMAP_BIT 32768
33 #define DEFAULT_BUFFER_SECTORS 128
34 #define DEFAULT_JOURNAL_WATERMARK 50
35 #define DEFAULT_SYNC_MSEC 10000
36 #define DEFAULT_MAX_JOURNAL_SECTORS 131072
37 #define MIN_LOG2_INTERLEAVE_SECTORS 3
38 #define MAX_LOG2_INTERLEAVE_SECTORS 31
39 #define METADATA_WORKQUEUE_MAX_ACTIVE 16
40 #define RECALC_SECTORS 32768
41 #define RECALC_WRITE_SUPER 16
42 #define BITMAP_BLOCK_SIZE 4096 /* don't change it */
43 #define BITMAP_FLUSH_INTERVAL (10 * HZ)
44 #define DISCARD_FILLER 0xf6
48 * Warning - DEBUG_PRINT prints security-sensitive data to the log,
49 * so it should not be enabled in the official kernel
52 //#define INTERNAL_VERIFY
58 #define SB_MAGIC "integrt"
59 #define SB_VERSION_1 1
60 #define SB_VERSION_2 2
61 #define SB_VERSION_3 3
62 #define SB_VERSION_4 4
63 #define SB_VERSION_5 5
65 #define MAX_SECTORS_PER_BLOCK 8
70 __u8 log2_interleave_sectors;
71 __le16 integrity_tag_size;
72 __le32 journal_sections;
73 __le64 provided_data_sectors; /* userspace uses this value */
75 __u8 log2_sectors_per_block;
76 __u8 log2_blocks_per_bitmap_bit;
83 #define SB_FLAG_HAVE_JOURNAL_MAC 0x1
84 #define SB_FLAG_RECALCULATING 0x2
85 #define SB_FLAG_DIRTY_BITMAP 0x4
86 #define SB_FLAG_FIXED_PADDING 0x8
87 #define SB_FLAG_FIXED_HMAC 0x10
89 #define JOURNAL_ENTRY_ROUNDUP 8
91 typedef __le64 commit_id_t;
92 #define JOURNAL_MAC_PER_SECTOR 8
94 struct journal_entry {
102 commit_id_t last_bytes[];
106 #define journal_entry_tag(ic, je) ((__u8 *)&(je)->last_bytes[(ic)->sectors_per_block])
108 #if BITS_PER_LONG == 64
109 #define journal_entry_set_sector(je, x) do { smp_wmb(); WRITE_ONCE((je)->u.sector, cpu_to_le64(x)); } while (0)
111 #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)
113 #define journal_entry_get_sector(je) le64_to_cpu((je)->u.sector)
114 #define journal_entry_is_unused(je) ((je)->u.s.sector_hi == cpu_to_le32(-1))
115 #define journal_entry_set_unused(je) do { ((je)->u.s.sector_hi = cpu_to_le32(-1)); } while (0)
116 #define journal_entry_is_inprogress(je) ((je)->u.s.sector_hi == cpu_to_le32(-2))
117 #define journal_entry_set_inprogress(je) do { ((je)->u.s.sector_hi = cpu_to_le32(-2)); } while (0)
119 #define JOURNAL_BLOCK_SECTORS 8
120 #define JOURNAL_SECTOR_DATA ((1 << SECTOR_SHIFT) - sizeof(commit_id_t))
121 #define JOURNAL_MAC_SIZE (JOURNAL_MAC_PER_SECTOR * JOURNAL_BLOCK_SECTORS)
123 struct journal_sector {
124 struct_group(sectors,
125 __u8 entries[JOURNAL_SECTOR_DATA - JOURNAL_MAC_PER_SECTOR];
126 __u8 mac[JOURNAL_MAC_PER_SECTOR];
128 commit_id_t commit_id;
131 #define MAX_TAG_SIZE (JOURNAL_SECTOR_DATA - JOURNAL_MAC_PER_SECTOR - offsetof(struct journal_entry, last_bytes[MAX_SECTORS_PER_BLOCK]))
133 #define METADATA_PADDING_SECTORS 8
135 #define N_COMMIT_IDS 4
137 static unsigned char prev_commit_seq(unsigned char seq)
139 return (seq + N_COMMIT_IDS - 1) % N_COMMIT_IDS;
142 static unsigned char next_commit_seq(unsigned char seq)
144 return (seq + 1) % N_COMMIT_IDS;
148 * In-memory structures
151 struct journal_node {
163 struct dm_integrity_c {
165 struct dm_dev *meta_dev;
169 mempool_t journal_io_mempool;
170 struct dm_io_client *io;
171 struct dm_bufio_client *bufio;
172 struct workqueue_struct *metadata_wq;
173 struct superblock *sb;
174 unsigned journal_pages;
175 unsigned n_bitmap_blocks;
177 struct page_list *journal;
178 struct page_list *journal_io;
179 struct page_list *journal_xor;
180 struct page_list *recalc_bitmap;
181 struct page_list *may_write_bitmap;
182 struct bitmap_block_status *bbs;
183 unsigned bitmap_flush_interval;
184 int synchronous_mode;
185 struct bio_list synchronous_bios;
186 struct delayed_work bitmap_flush_work;
188 struct crypto_skcipher *journal_crypt;
189 struct scatterlist **journal_scatterlist;
190 struct scatterlist **journal_io_scatterlist;
191 struct skcipher_request **sk_requests;
193 struct crypto_shash *journal_mac;
195 struct journal_node *journal_tree;
196 struct rb_root journal_tree_root;
198 sector_t provided_data_sectors;
200 unsigned short journal_entry_size;
201 unsigned char journal_entries_per_sector;
202 unsigned char journal_section_entries;
203 unsigned short journal_section_sectors;
204 unsigned journal_sections;
205 unsigned journal_entries;
206 sector_t data_device_sectors;
207 sector_t meta_device_sectors;
208 unsigned initial_sectors;
209 unsigned metadata_run;
210 __s8 log2_metadata_run;
211 __u8 log2_buffer_sectors;
212 __u8 sectors_per_block;
213 __u8 log2_blocks_per_bitmap_bit;
219 struct crypto_shash *internal_hash;
221 struct dm_target *ti;
223 /* these variables are locked with endio_wait.lock */
224 struct rb_root in_progress;
225 struct list_head wait_list;
226 wait_queue_head_t endio_wait;
227 struct workqueue_struct *wait_wq;
228 struct workqueue_struct *offload_wq;
230 unsigned char commit_seq;
231 commit_id_t commit_ids[N_COMMIT_IDS];
233 unsigned committed_section;
234 unsigned n_committed_sections;
236 unsigned uncommitted_section;
237 unsigned n_uncommitted_sections;
239 unsigned free_section;
240 unsigned char free_section_entry;
241 unsigned free_sectors;
243 unsigned free_sectors_threshold;
245 struct workqueue_struct *commit_wq;
246 struct work_struct commit_work;
248 struct workqueue_struct *writer_wq;
249 struct work_struct writer_work;
251 struct workqueue_struct *recalc_wq;
252 struct work_struct recalc_work;
256 struct bio_list flush_bio_list;
258 unsigned long autocommit_jiffies;
259 struct timer_list autocommit_timer;
260 unsigned autocommit_msec;
262 wait_queue_head_t copy_to_journal_wait;
264 struct completion crypto_backoff;
266 bool journal_uptodate;
268 bool recalculate_flag;
269 bool reset_recalculate_flag;
273 bool legacy_recalculate;
275 struct alg_spec internal_hash_alg;
276 struct alg_spec journal_crypt_alg;
277 struct alg_spec journal_mac_alg;
279 atomic64_t number_of_mismatches;
281 struct notifier_block reboot_notifier;
284 struct dm_integrity_range {
285 sector_t logical_sector;
291 struct task_struct *task;
292 struct list_head wait_entry;
297 struct dm_integrity_io {
298 struct work_struct work;
300 struct dm_integrity_c *ic;
304 struct dm_integrity_range range;
306 sector_t metadata_block;
307 unsigned metadata_offset;
310 blk_status_t bi_status;
312 struct completion *completion;
314 struct dm_bio_details bio_details;
317 struct journal_completion {
318 struct dm_integrity_c *ic;
320 struct completion comp;
324 struct dm_integrity_range range;
325 struct journal_completion *comp;
328 struct bitmap_block_status {
329 struct work_struct work;
330 struct dm_integrity_c *ic;
332 unsigned long *bitmap;
333 struct bio_list bio_queue;
334 spinlock_t bio_queue_lock;
338 static struct kmem_cache *journal_io_cache;
340 #define JOURNAL_IO_MEMPOOL 32
343 #define DEBUG_print(x, ...) printk(KERN_DEBUG x, ##__VA_ARGS__)
344 static void __DEBUG_bytes(__u8 *bytes, size_t len, const char *msg, ...)
353 pr_cont(" %02x", *bytes);
359 #define DEBUG_bytes(bytes, len, msg, ...) __DEBUG_bytes(bytes, len, KERN_DEBUG msg, ##__VA_ARGS__)
361 #define DEBUG_print(x, ...) do { } while (0)
362 #define DEBUG_bytes(bytes, len, msg, ...) do { } while (0)
365 static void dm_integrity_prepare(struct request *rq)
369 static void dm_integrity_complete(struct request *rq, unsigned int nr_bytes)
374 * DM Integrity profile, protection is performed layer above (dm-crypt)
376 static const struct blk_integrity_profile dm_integrity_profile = {
377 .name = "DM-DIF-EXT-TAG",
380 .prepare_fn = dm_integrity_prepare,
381 .complete_fn = dm_integrity_complete,
384 static void dm_integrity_map_continue(struct dm_integrity_io *dio, bool from_map);
385 static void integrity_bio_wait(struct work_struct *w);
386 static void dm_integrity_dtr(struct dm_target *ti);
388 static void dm_integrity_io_error(struct dm_integrity_c *ic, const char *msg, int err)
391 atomic64_inc(&ic->number_of_mismatches);
392 if (!cmpxchg(&ic->failed, 0, err))
393 DMERR("Error on %s: %d", msg, err);
396 static int dm_integrity_failed(struct dm_integrity_c *ic)
398 return READ_ONCE(ic->failed);
401 static bool dm_integrity_disable_recalculate(struct dm_integrity_c *ic)
403 if (ic->legacy_recalculate)
405 if (!(ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) ?
406 ic->internal_hash_alg.key || ic->journal_mac_alg.key :
407 ic->internal_hash_alg.key && !ic->journal_mac_alg.key)
412 static commit_id_t dm_integrity_commit_id(struct dm_integrity_c *ic, unsigned i,
413 unsigned j, unsigned char seq)
416 * Xor the number with section and sector, so that if a piece of
417 * journal is written at wrong place, it is detected.
419 return ic->commit_ids[seq] ^ cpu_to_le64(((__u64)i << 32) ^ j);
422 static void get_area_and_offset(struct dm_integrity_c *ic, sector_t data_sector,
423 sector_t *area, sector_t *offset)
426 __u8 log2_interleave_sectors = ic->sb->log2_interleave_sectors;
427 *area = data_sector >> log2_interleave_sectors;
428 *offset = (unsigned)data_sector & ((1U << log2_interleave_sectors) - 1);
431 *offset = data_sector;
435 #define sector_to_block(ic, n) \
437 BUG_ON((n) & (unsigned)((ic)->sectors_per_block - 1)); \
438 (n) >>= (ic)->sb->log2_sectors_per_block; \
441 static __u64 get_metadata_sector_and_offset(struct dm_integrity_c *ic, sector_t area,
442 sector_t offset, unsigned *metadata_offset)
447 ms = area << ic->sb->log2_interleave_sectors;
448 if (likely(ic->log2_metadata_run >= 0))
449 ms += area << ic->log2_metadata_run;
451 ms += area * ic->metadata_run;
452 ms >>= ic->log2_buffer_sectors;
454 sector_to_block(ic, offset);
456 if (likely(ic->log2_tag_size >= 0)) {
457 ms += offset >> (SECTOR_SHIFT + ic->log2_buffer_sectors - ic->log2_tag_size);
458 mo = (offset << ic->log2_tag_size) & ((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - 1);
460 ms += (__u64)offset * ic->tag_size >> (SECTOR_SHIFT + ic->log2_buffer_sectors);
461 mo = (offset * ic->tag_size) & ((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - 1);
463 *metadata_offset = mo;
467 static sector_t get_data_sector(struct dm_integrity_c *ic, sector_t area, sector_t offset)
474 result = area << ic->sb->log2_interleave_sectors;
475 if (likely(ic->log2_metadata_run >= 0))
476 result += (area + 1) << ic->log2_metadata_run;
478 result += (area + 1) * ic->metadata_run;
480 result += (sector_t)ic->initial_sectors + offset;
486 static void wraparound_section(struct dm_integrity_c *ic, unsigned *sec_ptr)
488 if (unlikely(*sec_ptr >= ic->journal_sections))
489 *sec_ptr -= ic->journal_sections;
492 static void sb_set_version(struct dm_integrity_c *ic)
494 if (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC))
495 ic->sb->version = SB_VERSION_5;
496 else if (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING))
497 ic->sb->version = SB_VERSION_4;
498 else if (ic->mode == 'B' || ic->sb->flags & cpu_to_le32(SB_FLAG_DIRTY_BITMAP))
499 ic->sb->version = SB_VERSION_3;
500 else if (ic->meta_dev || ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
501 ic->sb->version = SB_VERSION_2;
503 ic->sb->version = SB_VERSION_1;
506 static int sb_mac(struct dm_integrity_c *ic, bool wr)
508 SHASH_DESC_ON_STACK(desc, ic->journal_mac);
510 unsigned size = crypto_shash_digestsize(ic->journal_mac);
512 if (sizeof(struct superblock) + size > 1 << SECTOR_SHIFT) {
513 dm_integrity_io_error(ic, "digest is too long", -EINVAL);
517 desc->tfm = ic->journal_mac;
519 r = crypto_shash_init(desc);
520 if (unlikely(r < 0)) {
521 dm_integrity_io_error(ic, "crypto_shash_init", r);
525 r = crypto_shash_update(desc, (__u8 *)ic->sb, (1 << SECTOR_SHIFT) - size);
526 if (unlikely(r < 0)) {
527 dm_integrity_io_error(ic, "crypto_shash_update", r);
532 r = crypto_shash_final(desc, (__u8 *)ic->sb + (1 << SECTOR_SHIFT) - size);
533 if (unlikely(r < 0)) {
534 dm_integrity_io_error(ic, "crypto_shash_final", r);
538 __u8 result[HASH_MAX_DIGESTSIZE];
539 r = crypto_shash_final(desc, result);
540 if (unlikely(r < 0)) {
541 dm_integrity_io_error(ic, "crypto_shash_final", r);
544 if (memcmp((__u8 *)ic->sb + (1 << SECTOR_SHIFT) - size, result, size)) {
545 dm_integrity_io_error(ic, "superblock mac", -EILSEQ);
546 dm_audit_log_target(DM_MSG_PREFIX, "mac-superblock", ic->ti, 0);
554 static int sync_rw_sb(struct dm_integrity_c *ic, int op, int op_flags)
556 struct dm_io_request io_req;
557 struct dm_io_region io_loc;
561 io_req.bi_op_flags = op_flags;
562 io_req.mem.type = DM_IO_KMEM;
563 io_req.mem.ptr.addr = ic->sb;
564 io_req.notify.fn = NULL;
565 io_req.client = ic->io;
566 io_loc.bdev = ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev;
567 io_loc.sector = ic->start;
568 io_loc.count = SB_SECTORS;
570 if (op == REQ_OP_WRITE) {
572 if (ic->journal_mac && ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) {
573 r = sb_mac(ic, true);
579 r = dm_io(&io_req, 1, &io_loc, NULL);
583 if (op == REQ_OP_READ) {
584 if (ic->mode != 'R' && ic->journal_mac && ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) {
585 r = sb_mac(ic, false);
594 #define BITMAP_OP_TEST_ALL_SET 0
595 #define BITMAP_OP_TEST_ALL_CLEAR 1
596 #define BITMAP_OP_SET 2
597 #define BITMAP_OP_CLEAR 3
599 static bool block_bitmap_op(struct dm_integrity_c *ic, struct page_list *bitmap,
600 sector_t sector, sector_t n_sectors, int mode)
602 unsigned long bit, end_bit, this_end_bit, page, end_page;
605 if (unlikely(((sector | n_sectors) & ((1 << ic->sb->log2_sectors_per_block) - 1)) != 0)) {
606 DMCRIT("invalid bitmap access (%llx,%llx,%d,%d,%d)",
609 ic->sb->log2_sectors_per_block,
610 ic->log2_blocks_per_bitmap_bit,
615 if (unlikely(!n_sectors))
618 bit = sector >> (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
619 end_bit = (sector + n_sectors - 1) >>
620 (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
622 page = bit / (PAGE_SIZE * 8);
623 bit %= PAGE_SIZE * 8;
625 end_page = end_bit / (PAGE_SIZE * 8);
626 end_bit %= PAGE_SIZE * 8;
629 if (page < end_page) {
630 this_end_bit = PAGE_SIZE * 8 - 1;
632 this_end_bit = end_bit;
635 data = lowmem_page_address(bitmap[page].page);
637 if (mode == BITMAP_OP_TEST_ALL_SET) {
638 while (bit <= this_end_bit) {
639 if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
641 if (data[bit / BITS_PER_LONG] != -1)
643 bit += BITS_PER_LONG;
644 } while (this_end_bit >= bit + BITS_PER_LONG - 1);
647 if (!test_bit(bit, data))
651 } else if (mode == BITMAP_OP_TEST_ALL_CLEAR) {
652 while (bit <= this_end_bit) {
653 if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
655 if (data[bit / BITS_PER_LONG] != 0)
657 bit += BITS_PER_LONG;
658 } while (this_end_bit >= bit + BITS_PER_LONG - 1);
661 if (test_bit(bit, data))
665 } else if (mode == BITMAP_OP_SET) {
666 while (bit <= this_end_bit) {
667 if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
669 data[bit / BITS_PER_LONG] = -1;
670 bit += BITS_PER_LONG;
671 } while (this_end_bit >= bit + BITS_PER_LONG - 1);
674 __set_bit(bit, data);
677 } else if (mode == BITMAP_OP_CLEAR) {
678 if (!bit && this_end_bit == PAGE_SIZE * 8 - 1)
680 else while (bit <= this_end_bit) {
681 if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
683 data[bit / BITS_PER_LONG] = 0;
684 bit += BITS_PER_LONG;
685 } while (this_end_bit >= bit + BITS_PER_LONG - 1);
688 __clear_bit(bit, data);
695 if (unlikely(page < end_page)) {
704 static void block_bitmap_copy(struct dm_integrity_c *ic, struct page_list *dst, struct page_list *src)
706 unsigned n_bitmap_pages = DIV_ROUND_UP(ic->n_bitmap_blocks, PAGE_SIZE / BITMAP_BLOCK_SIZE);
709 for (i = 0; i < n_bitmap_pages; i++) {
710 unsigned long *dst_data = lowmem_page_address(dst[i].page);
711 unsigned long *src_data = lowmem_page_address(src[i].page);
712 copy_page(dst_data, src_data);
716 static struct bitmap_block_status *sector_to_bitmap_block(struct dm_integrity_c *ic, sector_t sector)
718 unsigned bit = sector >> (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
719 unsigned bitmap_block = bit / (BITMAP_BLOCK_SIZE * 8);
721 BUG_ON(bitmap_block >= ic->n_bitmap_blocks);
722 return &ic->bbs[bitmap_block];
725 static void access_journal_check(struct dm_integrity_c *ic, unsigned section, unsigned offset,
726 bool e, const char *function)
728 #if defined(CONFIG_DM_DEBUG) || defined(INTERNAL_VERIFY)
729 unsigned limit = e ? ic->journal_section_entries : ic->journal_section_sectors;
731 if (unlikely(section >= ic->journal_sections) ||
732 unlikely(offset >= limit)) {
733 DMCRIT("%s: invalid access at (%u,%u), limit (%u,%u)",
734 function, section, offset, ic->journal_sections, limit);
740 static void page_list_location(struct dm_integrity_c *ic, unsigned section, unsigned offset,
741 unsigned *pl_index, unsigned *pl_offset)
745 access_journal_check(ic, section, offset, false, "page_list_location");
747 sector = section * ic->journal_section_sectors + offset;
749 *pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
750 *pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
753 static struct journal_sector *access_page_list(struct dm_integrity_c *ic, struct page_list *pl,
754 unsigned section, unsigned offset, unsigned *n_sectors)
756 unsigned pl_index, pl_offset;
759 page_list_location(ic, section, offset, &pl_index, &pl_offset);
762 *n_sectors = (PAGE_SIZE - pl_offset) >> SECTOR_SHIFT;
764 va = lowmem_page_address(pl[pl_index].page);
766 return (struct journal_sector *)(va + pl_offset);
769 static struct journal_sector *access_journal(struct dm_integrity_c *ic, unsigned section, unsigned offset)
771 return access_page_list(ic, ic->journal, section, offset, NULL);
774 static struct journal_entry *access_journal_entry(struct dm_integrity_c *ic, unsigned section, unsigned n)
776 unsigned rel_sector, offset;
777 struct journal_sector *js;
779 access_journal_check(ic, section, n, true, "access_journal_entry");
781 rel_sector = n % JOURNAL_BLOCK_SECTORS;
782 offset = n / JOURNAL_BLOCK_SECTORS;
784 js = access_journal(ic, section, rel_sector);
785 return (struct journal_entry *)((char *)js + offset * ic->journal_entry_size);
788 static struct journal_sector *access_journal_data(struct dm_integrity_c *ic, unsigned section, unsigned n)
790 n <<= ic->sb->log2_sectors_per_block;
792 n += JOURNAL_BLOCK_SECTORS;
794 access_journal_check(ic, section, n, false, "access_journal_data");
796 return access_journal(ic, section, n);
799 static void section_mac(struct dm_integrity_c *ic, unsigned section, __u8 result[JOURNAL_MAC_SIZE])
801 SHASH_DESC_ON_STACK(desc, ic->journal_mac);
805 desc->tfm = ic->journal_mac;
807 r = crypto_shash_init(desc);
808 if (unlikely(r < 0)) {
809 dm_integrity_io_error(ic, "crypto_shash_init", r);
813 if (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) {
816 r = crypto_shash_update(desc, (__u8 *)&ic->sb->salt, SALT_SIZE);
817 if (unlikely(r < 0)) {
818 dm_integrity_io_error(ic, "crypto_shash_update", r);
822 section_le = cpu_to_le64(section);
823 r = crypto_shash_update(desc, (__u8 *)§ion_le, sizeof section_le);
824 if (unlikely(r < 0)) {
825 dm_integrity_io_error(ic, "crypto_shash_update", r);
830 for (j = 0; j < ic->journal_section_entries; j++) {
831 struct journal_entry *je = access_journal_entry(ic, section, j);
832 r = crypto_shash_update(desc, (__u8 *)&je->u.sector, sizeof je->u.sector);
833 if (unlikely(r < 0)) {
834 dm_integrity_io_error(ic, "crypto_shash_update", r);
839 size = crypto_shash_digestsize(ic->journal_mac);
841 if (likely(size <= JOURNAL_MAC_SIZE)) {
842 r = crypto_shash_final(desc, result);
843 if (unlikely(r < 0)) {
844 dm_integrity_io_error(ic, "crypto_shash_final", r);
847 memset(result + size, 0, JOURNAL_MAC_SIZE - size);
849 __u8 digest[HASH_MAX_DIGESTSIZE];
851 if (WARN_ON(size > sizeof(digest))) {
852 dm_integrity_io_error(ic, "digest_size", -EINVAL);
855 r = crypto_shash_final(desc, digest);
856 if (unlikely(r < 0)) {
857 dm_integrity_io_error(ic, "crypto_shash_final", r);
860 memcpy(result, digest, JOURNAL_MAC_SIZE);
865 memset(result, 0, JOURNAL_MAC_SIZE);
868 static void rw_section_mac(struct dm_integrity_c *ic, unsigned section, bool wr)
870 __u8 result[JOURNAL_MAC_SIZE];
873 if (!ic->journal_mac)
876 section_mac(ic, section, result);
878 for (j = 0; j < JOURNAL_BLOCK_SECTORS; j++) {
879 struct journal_sector *js = access_journal(ic, section, j);
882 memcpy(&js->mac, result + (j * JOURNAL_MAC_PER_SECTOR), JOURNAL_MAC_PER_SECTOR);
884 if (memcmp(&js->mac, result + (j * JOURNAL_MAC_PER_SECTOR), JOURNAL_MAC_PER_SECTOR)) {
885 dm_integrity_io_error(ic, "journal mac", -EILSEQ);
886 dm_audit_log_target(DM_MSG_PREFIX, "mac-journal", ic->ti, 0);
892 static void complete_journal_op(void *context)
894 struct journal_completion *comp = context;
895 BUG_ON(!atomic_read(&comp->in_flight));
896 if (likely(atomic_dec_and_test(&comp->in_flight)))
897 complete(&comp->comp);
900 static void xor_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
901 unsigned n_sections, struct journal_completion *comp)
903 struct async_submit_ctl submit;
904 size_t n_bytes = (size_t)(n_sections * ic->journal_section_sectors) << SECTOR_SHIFT;
905 unsigned pl_index, pl_offset, section_index;
906 struct page_list *source_pl, *target_pl;
908 if (likely(encrypt)) {
909 source_pl = ic->journal;
910 target_pl = ic->journal_io;
912 source_pl = ic->journal_io;
913 target_pl = ic->journal;
916 page_list_location(ic, section, 0, &pl_index, &pl_offset);
918 atomic_add(roundup(pl_offset + n_bytes, PAGE_SIZE) >> PAGE_SHIFT, &comp->in_flight);
920 init_async_submit(&submit, ASYNC_TX_XOR_ZERO_DST, NULL, complete_journal_op, comp, NULL);
922 section_index = pl_index;
926 struct page *src_pages[2];
927 struct page *dst_page;
929 while (unlikely(pl_index == section_index)) {
932 rw_section_mac(ic, section, true);
937 page_list_location(ic, section, 0, §ion_index, &dummy);
940 this_step = min(n_bytes, (size_t)PAGE_SIZE - pl_offset);
941 dst_page = target_pl[pl_index].page;
942 src_pages[0] = source_pl[pl_index].page;
943 src_pages[1] = ic->journal_xor[pl_index].page;
945 async_xor(dst_page, src_pages, pl_offset, 2, this_step, &submit);
949 n_bytes -= this_step;
954 async_tx_issue_pending_all();
957 static void complete_journal_encrypt(struct crypto_async_request *req, int err)
959 struct journal_completion *comp = req->data;
961 if (likely(err == -EINPROGRESS)) {
962 complete(&comp->ic->crypto_backoff);
965 dm_integrity_io_error(comp->ic, "asynchronous encrypt", err);
967 complete_journal_op(comp);
970 static bool do_crypt(bool encrypt, struct skcipher_request *req, struct journal_completion *comp)
973 skcipher_request_set_callback(req, CRYPTO_TFM_REQ_MAY_BACKLOG,
974 complete_journal_encrypt, comp);
976 r = crypto_skcipher_encrypt(req);
978 r = crypto_skcipher_decrypt(req);
981 if (likely(r == -EINPROGRESS))
983 if (likely(r == -EBUSY)) {
984 wait_for_completion(&comp->ic->crypto_backoff);
985 reinit_completion(&comp->ic->crypto_backoff);
988 dm_integrity_io_error(comp->ic, "encrypt", r);
992 static void crypt_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
993 unsigned n_sections, struct journal_completion *comp)
995 struct scatterlist **source_sg;
996 struct scatterlist **target_sg;
998 atomic_add(2, &comp->in_flight);
1000 if (likely(encrypt)) {
1001 source_sg = ic->journal_scatterlist;
1002 target_sg = ic->journal_io_scatterlist;
1004 source_sg = ic->journal_io_scatterlist;
1005 target_sg = ic->journal_scatterlist;
1009 struct skcipher_request *req;
1013 if (likely(encrypt))
1014 rw_section_mac(ic, section, true);
1016 req = ic->sk_requests[section];
1017 ivsize = crypto_skcipher_ivsize(ic->journal_crypt);
1020 memcpy(iv, iv + ivsize, ivsize);
1022 req->src = source_sg[section];
1023 req->dst = target_sg[section];
1025 if (unlikely(do_crypt(encrypt, req, comp)))
1026 atomic_inc(&comp->in_flight);
1030 } while (n_sections);
1032 atomic_dec(&comp->in_flight);
1033 complete_journal_op(comp);
1036 static void encrypt_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
1037 unsigned n_sections, struct journal_completion *comp)
1039 if (ic->journal_xor)
1040 return xor_journal(ic, encrypt, section, n_sections, comp);
1042 return crypt_journal(ic, encrypt, section, n_sections, comp);
1045 static void complete_journal_io(unsigned long error, void *context)
1047 struct journal_completion *comp = context;
1048 if (unlikely(error != 0))
1049 dm_integrity_io_error(comp->ic, "writing journal", -EIO);
1050 complete_journal_op(comp);
1053 static void rw_journal_sectors(struct dm_integrity_c *ic, int op, int op_flags,
1054 unsigned sector, unsigned n_sectors, struct journal_completion *comp)
1056 struct dm_io_request io_req;
1057 struct dm_io_region io_loc;
1058 unsigned pl_index, pl_offset;
1061 if (unlikely(dm_integrity_failed(ic))) {
1063 complete_journal_io(-1UL, comp);
1067 pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
1068 pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
1071 io_req.bi_op_flags = op_flags;
1072 io_req.mem.type = DM_IO_PAGE_LIST;
1074 io_req.mem.ptr.pl = &ic->journal_io[pl_index];
1076 io_req.mem.ptr.pl = &ic->journal[pl_index];
1077 io_req.mem.offset = pl_offset;
1078 if (likely(comp != NULL)) {
1079 io_req.notify.fn = complete_journal_io;
1080 io_req.notify.context = comp;
1082 io_req.notify.fn = NULL;
1084 io_req.client = ic->io;
1085 io_loc.bdev = ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev;
1086 io_loc.sector = ic->start + SB_SECTORS + sector;
1087 io_loc.count = n_sectors;
1089 r = dm_io(&io_req, 1, &io_loc, NULL);
1091 dm_integrity_io_error(ic, op == REQ_OP_READ ? "reading journal" : "writing journal", r);
1093 WARN_ONCE(1, "asynchronous dm_io failed: %d", r);
1094 complete_journal_io(-1UL, comp);
1099 static void rw_journal(struct dm_integrity_c *ic, int op, int op_flags, unsigned section,
1100 unsigned n_sections, struct journal_completion *comp)
1102 unsigned sector, n_sectors;
1104 sector = section * ic->journal_section_sectors;
1105 n_sectors = n_sections * ic->journal_section_sectors;
1107 rw_journal_sectors(ic, op, op_flags, sector, n_sectors, comp);
1110 static void write_journal(struct dm_integrity_c *ic, unsigned commit_start, unsigned commit_sections)
1112 struct journal_completion io_comp;
1113 struct journal_completion crypt_comp_1;
1114 struct journal_completion crypt_comp_2;
1118 init_completion(&io_comp.comp);
1120 if (commit_start + commit_sections <= ic->journal_sections) {
1121 io_comp.in_flight = (atomic_t)ATOMIC_INIT(1);
1122 if (ic->journal_io) {
1123 crypt_comp_1.ic = ic;
1124 init_completion(&crypt_comp_1.comp);
1125 crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
1126 encrypt_journal(ic, true, commit_start, commit_sections, &crypt_comp_1);
1127 wait_for_completion_io(&crypt_comp_1.comp);
1129 for (i = 0; i < commit_sections; i++)
1130 rw_section_mac(ic, commit_start + i, true);
1132 rw_journal(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, commit_start,
1133 commit_sections, &io_comp);
1136 io_comp.in_flight = (atomic_t)ATOMIC_INIT(2);
1137 to_end = ic->journal_sections - commit_start;
1138 if (ic->journal_io) {
1139 crypt_comp_1.ic = ic;
1140 init_completion(&crypt_comp_1.comp);
1141 crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
1142 encrypt_journal(ic, true, commit_start, to_end, &crypt_comp_1);
1143 if (try_wait_for_completion(&crypt_comp_1.comp)) {
1144 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
1145 reinit_completion(&crypt_comp_1.comp);
1146 crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
1147 encrypt_journal(ic, true, 0, commit_sections - to_end, &crypt_comp_1);
1148 wait_for_completion_io(&crypt_comp_1.comp);
1150 crypt_comp_2.ic = ic;
1151 init_completion(&crypt_comp_2.comp);
1152 crypt_comp_2.in_flight = (atomic_t)ATOMIC_INIT(0);
1153 encrypt_journal(ic, true, 0, commit_sections - to_end, &crypt_comp_2);
1154 wait_for_completion_io(&crypt_comp_1.comp);
1155 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
1156 wait_for_completion_io(&crypt_comp_2.comp);
1159 for (i = 0; i < to_end; i++)
1160 rw_section_mac(ic, commit_start + i, true);
1161 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
1162 for (i = 0; i < commit_sections - to_end; i++)
1163 rw_section_mac(ic, i, true);
1165 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, 0, commit_sections - to_end, &io_comp);
1168 wait_for_completion_io(&io_comp.comp);
1171 static void copy_from_journal(struct dm_integrity_c *ic, unsigned section, unsigned offset,
1172 unsigned n_sectors, sector_t target, io_notify_fn fn, void *data)
1174 struct dm_io_request io_req;
1175 struct dm_io_region io_loc;
1177 unsigned sector, pl_index, pl_offset;
1179 BUG_ON((target | n_sectors | offset) & (unsigned)(ic->sectors_per_block - 1));
1181 if (unlikely(dm_integrity_failed(ic))) {
1186 sector = section * ic->journal_section_sectors + JOURNAL_BLOCK_SECTORS + offset;
1188 pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
1189 pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
1191 io_req.bi_op = REQ_OP_WRITE;
1192 io_req.bi_op_flags = 0;
1193 io_req.mem.type = DM_IO_PAGE_LIST;
1194 io_req.mem.ptr.pl = &ic->journal[pl_index];
1195 io_req.mem.offset = pl_offset;
1196 io_req.notify.fn = fn;
1197 io_req.notify.context = data;
1198 io_req.client = ic->io;
1199 io_loc.bdev = ic->dev->bdev;
1200 io_loc.sector = target;
1201 io_loc.count = n_sectors;
1203 r = dm_io(&io_req, 1, &io_loc, NULL);
1205 WARN_ONCE(1, "asynchronous dm_io failed: %d", r);
1210 static bool ranges_overlap(struct dm_integrity_range *range1, struct dm_integrity_range *range2)
1212 return range1->logical_sector < range2->logical_sector + range2->n_sectors &&
1213 range1->logical_sector + range1->n_sectors > range2->logical_sector;
1216 static bool add_new_range(struct dm_integrity_c *ic, struct dm_integrity_range *new_range, bool check_waiting)
1218 struct rb_node **n = &ic->in_progress.rb_node;
1219 struct rb_node *parent;
1221 BUG_ON((new_range->logical_sector | new_range->n_sectors) & (unsigned)(ic->sectors_per_block - 1));
1223 if (likely(check_waiting)) {
1224 struct dm_integrity_range *range;
1225 list_for_each_entry(range, &ic->wait_list, wait_entry) {
1226 if (unlikely(ranges_overlap(range, new_range)))
1234 struct dm_integrity_range *range = container_of(*n, struct dm_integrity_range, node);
1237 if (new_range->logical_sector + new_range->n_sectors <= range->logical_sector) {
1238 n = &range->node.rb_left;
1239 } else if (new_range->logical_sector >= range->logical_sector + range->n_sectors) {
1240 n = &range->node.rb_right;
1246 rb_link_node(&new_range->node, parent, n);
1247 rb_insert_color(&new_range->node, &ic->in_progress);
1252 static void remove_range_unlocked(struct dm_integrity_c *ic, struct dm_integrity_range *range)
1254 rb_erase(&range->node, &ic->in_progress);
1255 while (unlikely(!list_empty(&ic->wait_list))) {
1256 struct dm_integrity_range *last_range =
1257 list_first_entry(&ic->wait_list, struct dm_integrity_range, wait_entry);
1258 struct task_struct *last_range_task;
1259 last_range_task = last_range->task;
1260 list_del(&last_range->wait_entry);
1261 if (!add_new_range(ic, last_range, false)) {
1262 last_range->task = last_range_task;
1263 list_add(&last_range->wait_entry, &ic->wait_list);
1266 last_range->waiting = false;
1267 wake_up_process(last_range_task);
1271 static void remove_range(struct dm_integrity_c *ic, struct dm_integrity_range *range)
1273 unsigned long flags;
1275 spin_lock_irqsave(&ic->endio_wait.lock, flags);
1276 remove_range_unlocked(ic, range);
1277 spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
1280 static void wait_and_add_new_range(struct dm_integrity_c *ic, struct dm_integrity_range *new_range)
1282 new_range->waiting = true;
1283 list_add_tail(&new_range->wait_entry, &ic->wait_list);
1284 new_range->task = current;
1286 __set_current_state(TASK_UNINTERRUPTIBLE);
1287 spin_unlock_irq(&ic->endio_wait.lock);
1289 spin_lock_irq(&ic->endio_wait.lock);
1290 } while (unlikely(new_range->waiting));
1293 static void add_new_range_and_wait(struct dm_integrity_c *ic, struct dm_integrity_range *new_range)
1295 if (unlikely(!add_new_range(ic, new_range, true)))
1296 wait_and_add_new_range(ic, new_range);
1299 static void init_journal_node(struct journal_node *node)
1301 RB_CLEAR_NODE(&node->node);
1302 node->sector = (sector_t)-1;
1305 static void add_journal_node(struct dm_integrity_c *ic, struct journal_node *node, sector_t sector)
1307 struct rb_node **link;
1308 struct rb_node *parent;
1310 node->sector = sector;
1311 BUG_ON(!RB_EMPTY_NODE(&node->node));
1313 link = &ic->journal_tree_root.rb_node;
1317 struct journal_node *j;
1319 j = container_of(parent, struct journal_node, node);
1320 if (sector < j->sector)
1321 link = &j->node.rb_left;
1323 link = &j->node.rb_right;
1326 rb_link_node(&node->node, parent, link);
1327 rb_insert_color(&node->node, &ic->journal_tree_root);
1330 static void remove_journal_node(struct dm_integrity_c *ic, struct journal_node *node)
1332 BUG_ON(RB_EMPTY_NODE(&node->node));
1333 rb_erase(&node->node, &ic->journal_tree_root);
1334 init_journal_node(node);
1337 #define NOT_FOUND (-1U)
1339 static unsigned find_journal_node(struct dm_integrity_c *ic, sector_t sector, sector_t *next_sector)
1341 struct rb_node *n = ic->journal_tree_root.rb_node;
1342 unsigned found = NOT_FOUND;
1343 *next_sector = (sector_t)-1;
1345 struct journal_node *j = container_of(n, struct journal_node, node);
1346 if (sector == j->sector) {
1347 found = j - ic->journal_tree;
1349 if (sector < j->sector) {
1350 *next_sector = j->sector;
1351 n = j->node.rb_left;
1353 n = j->node.rb_right;
1360 static bool test_journal_node(struct dm_integrity_c *ic, unsigned pos, sector_t sector)
1362 struct journal_node *node, *next_node;
1363 struct rb_node *next;
1365 if (unlikely(pos >= ic->journal_entries))
1367 node = &ic->journal_tree[pos];
1368 if (unlikely(RB_EMPTY_NODE(&node->node)))
1370 if (unlikely(node->sector != sector))
1373 next = rb_next(&node->node);
1374 if (unlikely(!next))
1377 next_node = container_of(next, struct journal_node, node);
1378 return next_node->sector != sector;
1381 static bool find_newer_committed_node(struct dm_integrity_c *ic, struct journal_node *node)
1383 struct rb_node *next;
1384 struct journal_node *next_node;
1385 unsigned next_section;
1387 BUG_ON(RB_EMPTY_NODE(&node->node));
1389 next = rb_next(&node->node);
1390 if (unlikely(!next))
1393 next_node = container_of(next, struct journal_node, node);
1395 if (next_node->sector != node->sector)
1398 next_section = (unsigned)(next_node - ic->journal_tree) / ic->journal_section_entries;
1399 if (next_section >= ic->committed_section &&
1400 next_section < ic->committed_section + ic->n_committed_sections)
1402 if (next_section + ic->journal_sections < ic->committed_section + ic->n_committed_sections)
1412 static int dm_integrity_rw_tag(struct dm_integrity_c *ic, unsigned char *tag, sector_t *metadata_block,
1413 unsigned *metadata_offset, unsigned total_size, int op)
1415 #define MAY_BE_FILLER 1
1416 #define MAY_BE_HASH 2
1417 unsigned hash_offset = 0;
1418 unsigned may_be = MAY_BE_HASH | (ic->discard ? MAY_BE_FILLER : 0);
1421 unsigned char *data, *dp;
1422 struct dm_buffer *b;
1426 r = dm_integrity_failed(ic);
1430 data = dm_bufio_read(ic->bufio, *metadata_block, &b);
1432 return PTR_ERR(data);
1434 to_copy = min((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - *metadata_offset, total_size);
1435 dp = data + *metadata_offset;
1436 if (op == TAG_READ) {
1437 memcpy(tag, dp, to_copy);
1438 } else if (op == TAG_WRITE) {
1439 if (memcmp(dp, tag, to_copy)) {
1440 memcpy(dp, tag, to_copy);
1441 dm_bufio_mark_partial_buffer_dirty(b, *metadata_offset, *metadata_offset + to_copy);
1444 /* e.g.: op == TAG_CMP */
1446 if (likely(is_power_of_2(ic->tag_size))) {
1447 if (unlikely(memcmp(dp, tag, to_copy)))
1448 if (unlikely(!ic->discard) ||
1449 unlikely(memchr_inv(dp, DISCARD_FILLER, to_copy) != NULL)) {
1457 for (i = 0; i < to_copy; i++, ts--) {
1458 if (unlikely(dp[i] != tag[i]))
1459 may_be &= ~MAY_BE_HASH;
1460 if (likely(dp[i] != DISCARD_FILLER))
1461 may_be &= ~MAY_BE_FILLER;
1463 if (unlikely(hash_offset == ic->tag_size)) {
1464 if (unlikely(!may_be)) {
1465 dm_bufio_release(b);
1469 may_be = MAY_BE_HASH | (ic->discard ? MAY_BE_FILLER : 0);
1474 dm_bufio_release(b);
1477 *metadata_offset += to_copy;
1478 if (unlikely(*metadata_offset == 1U << SECTOR_SHIFT << ic->log2_buffer_sectors)) {
1479 (*metadata_block)++;
1480 *metadata_offset = 0;
1483 if (unlikely(!is_power_of_2(ic->tag_size))) {
1484 hash_offset = (hash_offset + to_copy) % ic->tag_size;
1487 total_size -= to_copy;
1488 } while (unlikely(total_size));
1491 #undef MAY_BE_FILLER
1495 struct flush_request {
1496 struct dm_io_request io_req;
1497 struct dm_io_region io_reg;
1498 struct dm_integrity_c *ic;
1499 struct completion comp;
1502 static void flush_notify(unsigned long error, void *fr_)
1504 struct flush_request *fr = fr_;
1505 if (unlikely(error != 0))
1506 dm_integrity_io_error(fr->ic, "flushing disk cache", -EIO);
1507 complete(&fr->comp);
1510 static void dm_integrity_flush_buffers(struct dm_integrity_c *ic, bool flush_data)
1514 struct flush_request fr;
1519 fr.io_req.bi_op = REQ_OP_WRITE,
1520 fr.io_req.bi_op_flags = REQ_PREFLUSH | REQ_SYNC,
1521 fr.io_req.mem.type = DM_IO_KMEM,
1522 fr.io_req.mem.ptr.addr = NULL,
1523 fr.io_req.notify.fn = flush_notify,
1524 fr.io_req.notify.context = &fr;
1525 fr.io_req.client = dm_bufio_get_dm_io_client(ic->bufio),
1526 fr.io_reg.bdev = ic->dev->bdev,
1527 fr.io_reg.sector = 0,
1528 fr.io_reg.count = 0,
1530 init_completion(&fr.comp);
1531 r = dm_io(&fr.io_req, 1, &fr.io_reg, NULL);
1535 r = dm_bufio_write_dirty_buffers(ic->bufio);
1537 dm_integrity_io_error(ic, "writing tags", r);
1540 wait_for_completion(&fr.comp);
1543 static void sleep_on_endio_wait(struct dm_integrity_c *ic)
1545 DECLARE_WAITQUEUE(wait, current);
1546 __add_wait_queue(&ic->endio_wait, &wait);
1547 __set_current_state(TASK_UNINTERRUPTIBLE);
1548 spin_unlock_irq(&ic->endio_wait.lock);
1550 spin_lock_irq(&ic->endio_wait.lock);
1551 __remove_wait_queue(&ic->endio_wait, &wait);
1554 static void autocommit_fn(struct timer_list *t)
1556 struct dm_integrity_c *ic = from_timer(ic, t, autocommit_timer);
1558 if (likely(!dm_integrity_failed(ic)))
1559 queue_work(ic->commit_wq, &ic->commit_work);
1562 static void schedule_autocommit(struct dm_integrity_c *ic)
1564 if (!timer_pending(&ic->autocommit_timer))
1565 mod_timer(&ic->autocommit_timer, jiffies + ic->autocommit_jiffies);
1568 static void submit_flush_bio(struct dm_integrity_c *ic, struct dm_integrity_io *dio)
1571 unsigned long flags;
1573 spin_lock_irqsave(&ic->endio_wait.lock, flags);
1574 bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1575 bio_list_add(&ic->flush_bio_list, bio);
1576 spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
1578 queue_work(ic->commit_wq, &ic->commit_work);
1581 static void do_endio(struct dm_integrity_c *ic, struct bio *bio)
1583 int r = dm_integrity_failed(ic);
1584 if (unlikely(r) && !bio->bi_status)
1585 bio->bi_status = errno_to_blk_status(r);
1586 if (unlikely(ic->synchronous_mode) && bio_op(bio) == REQ_OP_WRITE) {
1587 unsigned long flags;
1588 spin_lock_irqsave(&ic->endio_wait.lock, flags);
1589 bio_list_add(&ic->synchronous_bios, bio);
1590 queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, 0);
1591 spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
1597 static void do_endio_flush(struct dm_integrity_c *ic, struct dm_integrity_io *dio)
1599 struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1601 if (unlikely(dio->fua) && likely(!bio->bi_status) && likely(!dm_integrity_failed(ic)))
1602 submit_flush_bio(ic, dio);
1607 static void dec_in_flight(struct dm_integrity_io *dio)
1609 if (atomic_dec_and_test(&dio->in_flight)) {
1610 struct dm_integrity_c *ic = dio->ic;
1613 remove_range(ic, &dio->range);
1615 if (dio->op == REQ_OP_WRITE || unlikely(dio->op == REQ_OP_DISCARD))
1616 schedule_autocommit(ic);
1618 bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1620 if (unlikely(dio->bi_status) && !bio->bi_status)
1621 bio->bi_status = dio->bi_status;
1622 if (likely(!bio->bi_status) && unlikely(bio_sectors(bio) != dio->range.n_sectors)) {
1623 dio->range.logical_sector += dio->range.n_sectors;
1624 bio_advance(bio, dio->range.n_sectors << SECTOR_SHIFT);
1625 INIT_WORK(&dio->work, integrity_bio_wait);
1626 queue_work(ic->offload_wq, &dio->work);
1629 do_endio_flush(ic, dio);
1633 static void integrity_end_io(struct bio *bio)
1635 struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
1637 dm_bio_restore(&dio->bio_details, bio);
1638 if (bio->bi_integrity)
1639 bio->bi_opf |= REQ_INTEGRITY;
1641 if (dio->completion)
1642 complete(dio->completion);
1647 static void integrity_sector_checksum(struct dm_integrity_c *ic, sector_t sector,
1648 const char *data, char *result)
1650 __le64 sector_le = cpu_to_le64(sector);
1651 SHASH_DESC_ON_STACK(req, ic->internal_hash);
1653 unsigned digest_size;
1655 req->tfm = ic->internal_hash;
1657 r = crypto_shash_init(req);
1658 if (unlikely(r < 0)) {
1659 dm_integrity_io_error(ic, "crypto_shash_init", r);
1663 if (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) {
1664 r = crypto_shash_update(req, (__u8 *)&ic->sb->salt, SALT_SIZE);
1665 if (unlikely(r < 0)) {
1666 dm_integrity_io_error(ic, "crypto_shash_update", r);
1671 r = crypto_shash_update(req, (const __u8 *)§or_le, sizeof sector_le);
1672 if (unlikely(r < 0)) {
1673 dm_integrity_io_error(ic, "crypto_shash_update", r);
1677 r = crypto_shash_update(req, data, ic->sectors_per_block << SECTOR_SHIFT);
1678 if (unlikely(r < 0)) {
1679 dm_integrity_io_error(ic, "crypto_shash_update", r);
1683 r = crypto_shash_final(req, result);
1684 if (unlikely(r < 0)) {
1685 dm_integrity_io_error(ic, "crypto_shash_final", r);
1689 digest_size = crypto_shash_digestsize(ic->internal_hash);
1690 if (unlikely(digest_size < ic->tag_size))
1691 memset(result + digest_size, 0, ic->tag_size - digest_size);
1696 /* this shouldn't happen anyway, the hash functions have no reason to fail */
1697 get_random_bytes(result, ic->tag_size);
1700 static void integrity_metadata(struct work_struct *w)
1702 struct dm_integrity_io *dio = container_of(w, struct dm_integrity_io, work);
1703 struct dm_integrity_c *ic = dio->ic;
1707 if (ic->internal_hash) {
1708 struct bvec_iter iter;
1710 unsigned digest_size = crypto_shash_digestsize(ic->internal_hash);
1711 struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1713 unsigned extra_space = unlikely(digest_size > ic->tag_size) ? digest_size - ic->tag_size : 0;
1714 char checksums_onstack[max((size_t)HASH_MAX_DIGESTSIZE, MAX_TAG_SIZE)];
1716 unsigned sectors_to_process;
1718 if (unlikely(ic->mode == 'R'))
1721 if (likely(dio->op != REQ_OP_DISCARD))
1722 checksums = kmalloc((PAGE_SIZE >> SECTOR_SHIFT >> ic->sb->log2_sectors_per_block) * ic->tag_size + extra_space,
1723 GFP_NOIO | __GFP_NORETRY | __GFP_NOWARN);
1725 checksums = kmalloc(PAGE_SIZE, GFP_NOIO | __GFP_NORETRY | __GFP_NOWARN);
1727 checksums = checksums_onstack;
1728 if (WARN_ON(extra_space &&
1729 digest_size > sizeof(checksums_onstack))) {
1735 if (unlikely(dio->op == REQ_OP_DISCARD)) {
1736 sector_t bi_sector = dio->bio_details.bi_iter.bi_sector;
1737 unsigned bi_size = dio->bio_details.bi_iter.bi_size;
1738 unsigned max_size = likely(checksums != checksums_onstack) ? PAGE_SIZE : HASH_MAX_DIGESTSIZE;
1739 unsigned max_blocks = max_size / ic->tag_size;
1740 memset(checksums, DISCARD_FILLER, max_size);
1743 unsigned this_step_blocks = bi_size >> (SECTOR_SHIFT + ic->sb->log2_sectors_per_block);
1744 this_step_blocks = min(this_step_blocks, max_blocks);
1745 r = dm_integrity_rw_tag(ic, checksums, &dio->metadata_block, &dio->metadata_offset,
1746 this_step_blocks * ic->tag_size, TAG_WRITE);
1748 if (likely(checksums != checksums_onstack))
1753 /*if (bi_size < this_step_blocks << (SECTOR_SHIFT + ic->sb->log2_sectors_per_block)) {
1754 printk("BUGG: bi_sector: %llx, bi_size: %u\n", bi_sector, bi_size);
1755 printk("BUGG: this_step_blocks: %u\n", this_step_blocks);
1758 bi_size -= this_step_blocks << (SECTOR_SHIFT + ic->sb->log2_sectors_per_block);
1759 bi_sector += this_step_blocks << ic->sb->log2_sectors_per_block;
1762 if (likely(checksums != checksums_onstack))
1767 sector = dio->range.logical_sector;
1768 sectors_to_process = dio->range.n_sectors;
1770 __bio_for_each_segment(bv, bio, iter, dio->bio_details.bi_iter) {
1772 char *mem, *checksums_ptr;
1775 mem = bvec_kmap_local(&bv);
1777 checksums_ptr = checksums;
1779 integrity_sector_checksum(ic, sector, mem + pos, checksums_ptr);
1780 checksums_ptr += ic->tag_size;
1781 sectors_to_process -= ic->sectors_per_block;
1782 pos += ic->sectors_per_block << SECTOR_SHIFT;
1783 sector += ic->sectors_per_block;
1784 } while (pos < bv.bv_len && sectors_to_process && checksums != checksums_onstack);
1787 r = dm_integrity_rw_tag(ic, checksums, &dio->metadata_block, &dio->metadata_offset,
1788 checksums_ptr - checksums, dio->op == REQ_OP_READ ? TAG_CMP : TAG_WRITE);
1791 char b[BDEVNAME_SIZE];
1794 s = sector - ((r + ic->tag_size - 1) / ic->tag_size);
1795 DMERR_LIMIT("%s: Checksum failed at sector 0x%llx",
1796 bio_devname(bio, b), s);
1798 atomic64_inc(&ic->number_of_mismatches);
1799 dm_audit_log_bio(DM_MSG_PREFIX, "integrity-checksum",
1802 if (likely(checksums != checksums_onstack))
1807 if (!sectors_to_process)
1810 if (unlikely(pos < bv.bv_len)) {
1811 bv.bv_offset += pos;
1817 if (likely(checksums != checksums_onstack))
1820 struct bio_integrity_payload *bip = dio->bio_details.bi_integrity;
1824 struct bvec_iter iter;
1825 unsigned data_to_process = dio->range.n_sectors;
1826 sector_to_block(ic, data_to_process);
1827 data_to_process *= ic->tag_size;
1829 bip_for_each_vec(biv, bip, iter) {
1833 BUG_ON(PageHighMem(biv.bv_page));
1834 tag = bvec_virt(&biv);
1835 this_len = min(biv.bv_len, data_to_process);
1836 r = dm_integrity_rw_tag(ic, tag, &dio->metadata_block, &dio->metadata_offset,
1837 this_len, dio->op == REQ_OP_READ ? TAG_READ : TAG_WRITE);
1840 data_to_process -= this_len;
1841 if (!data_to_process)
1850 dio->bi_status = errno_to_blk_status(r);
1854 static int dm_integrity_map(struct dm_target *ti, struct bio *bio)
1856 struct dm_integrity_c *ic = ti->private;
1857 struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
1858 struct bio_integrity_payload *bip;
1860 sector_t area, offset;
1864 dio->op = bio_op(bio);
1866 if (unlikely(dio->op == REQ_OP_DISCARD)) {
1867 if (ti->max_io_len) {
1868 sector_t sec = dm_target_offset(ti, bio->bi_iter.bi_sector);
1869 unsigned log2_max_io_len = __fls(ti->max_io_len);
1870 sector_t start_boundary = sec >> log2_max_io_len;
1871 sector_t end_boundary = (sec + bio_sectors(bio) - 1) >> log2_max_io_len;
1872 if (start_boundary < end_boundary) {
1873 sector_t len = ti->max_io_len - (sec & (ti->max_io_len - 1));
1874 dm_accept_partial_bio(bio, len);
1879 if (unlikely(bio->bi_opf & REQ_PREFLUSH)) {
1880 submit_flush_bio(ic, dio);
1881 return DM_MAPIO_SUBMITTED;
1884 dio->range.logical_sector = dm_target_offset(ti, bio->bi_iter.bi_sector);
1885 dio->fua = dio->op == REQ_OP_WRITE && bio->bi_opf & REQ_FUA;
1886 if (unlikely(dio->fua)) {
1888 * Don't pass down the FUA flag because we have to flush
1889 * disk cache anyway.
1891 bio->bi_opf &= ~REQ_FUA;
1893 if (unlikely(dio->range.logical_sector + bio_sectors(bio) > ic->provided_data_sectors)) {
1894 DMERR("Too big sector number: 0x%llx + 0x%x > 0x%llx",
1895 dio->range.logical_sector, bio_sectors(bio),
1896 ic->provided_data_sectors);
1897 return DM_MAPIO_KILL;
1899 if (unlikely((dio->range.logical_sector | bio_sectors(bio)) & (unsigned)(ic->sectors_per_block - 1))) {
1900 DMERR("Bio not aligned on %u sectors: 0x%llx, 0x%x",
1901 ic->sectors_per_block,
1902 dio->range.logical_sector, bio_sectors(bio));
1903 return DM_MAPIO_KILL;
1906 if (ic->sectors_per_block > 1 && likely(dio->op != REQ_OP_DISCARD)) {
1907 struct bvec_iter iter;
1909 bio_for_each_segment(bv, bio, iter) {
1910 if (unlikely(bv.bv_len & ((ic->sectors_per_block << SECTOR_SHIFT) - 1))) {
1911 DMERR("Bio vector (%u,%u) is not aligned on %u-sector boundary",
1912 bv.bv_offset, bv.bv_len, ic->sectors_per_block);
1913 return DM_MAPIO_KILL;
1918 bip = bio_integrity(bio);
1919 if (!ic->internal_hash) {
1921 unsigned wanted_tag_size = bio_sectors(bio) >> ic->sb->log2_sectors_per_block;
1922 if (ic->log2_tag_size >= 0)
1923 wanted_tag_size <<= ic->log2_tag_size;
1925 wanted_tag_size *= ic->tag_size;
1926 if (unlikely(wanted_tag_size != bip->bip_iter.bi_size)) {
1927 DMERR("Invalid integrity data size %u, expected %u",
1928 bip->bip_iter.bi_size, wanted_tag_size);
1929 return DM_MAPIO_KILL;
1933 if (unlikely(bip != NULL)) {
1934 DMERR("Unexpected integrity data when using internal hash");
1935 return DM_MAPIO_KILL;
1939 if (unlikely(ic->mode == 'R') && unlikely(dio->op != REQ_OP_READ))
1940 return DM_MAPIO_KILL;
1942 get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
1943 dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset, &dio->metadata_offset);
1944 bio->bi_iter.bi_sector = get_data_sector(ic, area, offset);
1946 dm_integrity_map_continue(dio, true);
1947 return DM_MAPIO_SUBMITTED;
1950 static bool __journal_read_write(struct dm_integrity_io *dio, struct bio *bio,
1951 unsigned journal_section, unsigned journal_entry)
1953 struct dm_integrity_c *ic = dio->ic;
1954 sector_t logical_sector;
1957 logical_sector = dio->range.logical_sector;
1958 n_sectors = dio->range.n_sectors;
1960 struct bio_vec bv = bio_iovec(bio);
1963 if (unlikely(bv.bv_len >> SECTOR_SHIFT > n_sectors))
1964 bv.bv_len = n_sectors << SECTOR_SHIFT;
1965 n_sectors -= bv.bv_len >> SECTOR_SHIFT;
1966 bio_advance_iter(bio, &bio->bi_iter, bv.bv_len);
1968 mem = kmap_local_page(bv.bv_page);
1969 if (likely(dio->op == REQ_OP_WRITE))
1970 flush_dcache_page(bv.bv_page);
1973 struct journal_entry *je = access_journal_entry(ic, journal_section, journal_entry);
1975 if (unlikely(dio->op == REQ_OP_READ)) {
1976 struct journal_sector *js;
1980 if (unlikely(journal_entry_is_inprogress(je))) {
1981 flush_dcache_page(bv.bv_page);
1984 __io_wait_event(ic->copy_to_journal_wait, !journal_entry_is_inprogress(je));
1988 BUG_ON(journal_entry_get_sector(je) != logical_sector);
1989 js = access_journal_data(ic, journal_section, journal_entry);
1990 mem_ptr = mem + bv.bv_offset;
1993 memcpy(mem_ptr, js, JOURNAL_SECTOR_DATA);
1994 *(commit_id_t *)(mem_ptr + JOURNAL_SECTOR_DATA) = je->last_bytes[s];
1996 mem_ptr += 1 << SECTOR_SHIFT;
1997 } while (++s < ic->sectors_per_block);
1998 #ifdef INTERNAL_VERIFY
1999 if (ic->internal_hash) {
2000 char checksums_onstack[max((size_t)HASH_MAX_DIGESTSIZE, MAX_TAG_SIZE)];
2002 integrity_sector_checksum(ic, logical_sector, mem + bv.bv_offset, checksums_onstack);
2003 if (unlikely(memcmp(checksums_onstack, journal_entry_tag(ic, je), ic->tag_size))) {
2004 DMERR_LIMIT("Checksum failed when reading from journal, at sector 0x%llx",
2006 dm_audit_log_bio(DM_MSG_PREFIX, "journal-checksum",
2007 bio, logical_sector, 0);
2013 if (!ic->internal_hash) {
2014 struct bio_integrity_payload *bip = bio_integrity(bio);
2015 unsigned tag_todo = ic->tag_size;
2016 char *tag_ptr = journal_entry_tag(ic, je);
2019 struct bio_vec biv = bvec_iter_bvec(bip->bip_vec, bip->bip_iter);
2020 unsigned tag_now = min(biv.bv_len, tag_todo);
2022 BUG_ON(PageHighMem(biv.bv_page));
2023 tag_addr = bvec_virt(&biv);
2024 if (likely(dio->op == REQ_OP_WRITE))
2025 memcpy(tag_ptr, tag_addr, tag_now);
2027 memcpy(tag_addr, tag_ptr, tag_now);
2028 bvec_iter_advance(bip->bip_vec, &bip->bip_iter, tag_now);
2030 tag_todo -= tag_now;
2031 } while (unlikely(tag_todo)); else {
2032 if (likely(dio->op == REQ_OP_WRITE))
2033 memset(tag_ptr, 0, tag_todo);
2037 if (likely(dio->op == REQ_OP_WRITE)) {
2038 struct journal_sector *js;
2041 js = access_journal_data(ic, journal_section, journal_entry);
2042 memcpy(js, mem + bv.bv_offset, ic->sectors_per_block << SECTOR_SHIFT);
2046 je->last_bytes[s] = js[s].commit_id;
2047 } while (++s < ic->sectors_per_block);
2049 if (ic->internal_hash) {
2050 unsigned digest_size = crypto_shash_digestsize(ic->internal_hash);
2051 if (unlikely(digest_size > ic->tag_size)) {
2052 char checksums_onstack[HASH_MAX_DIGESTSIZE];
2053 integrity_sector_checksum(ic, logical_sector, (char *)js, checksums_onstack);
2054 memcpy(journal_entry_tag(ic, je), checksums_onstack, ic->tag_size);
2056 integrity_sector_checksum(ic, logical_sector, (char *)js, journal_entry_tag(ic, je));
2059 journal_entry_set_sector(je, logical_sector);
2061 logical_sector += ic->sectors_per_block;
2064 if (unlikely(journal_entry == ic->journal_section_entries)) {
2067 wraparound_section(ic, &journal_section);
2070 bv.bv_offset += ic->sectors_per_block << SECTOR_SHIFT;
2071 } while (bv.bv_len -= ic->sectors_per_block << SECTOR_SHIFT);
2073 if (unlikely(dio->op == REQ_OP_READ))
2074 flush_dcache_page(bv.bv_page);
2076 } while (n_sectors);
2078 if (likely(dio->op == REQ_OP_WRITE)) {
2080 if (unlikely(waitqueue_active(&ic->copy_to_journal_wait)))
2081 wake_up(&ic->copy_to_journal_wait);
2082 if (READ_ONCE(ic->free_sectors) <= ic->free_sectors_threshold) {
2083 queue_work(ic->commit_wq, &ic->commit_work);
2085 schedule_autocommit(ic);
2088 remove_range(ic, &dio->range);
2091 if (unlikely(bio->bi_iter.bi_size)) {
2092 sector_t area, offset;
2094 dio->range.logical_sector = logical_sector;
2095 get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
2096 dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset, &dio->metadata_offset);
2103 static void dm_integrity_map_continue(struct dm_integrity_io *dio, bool from_map)
2105 struct dm_integrity_c *ic = dio->ic;
2106 struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
2107 unsigned journal_section, journal_entry;
2108 unsigned journal_read_pos;
2109 struct completion read_comp;
2110 bool discard_retried = false;
2111 bool need_sync_io = ic->internal_hash && dio->op == REQ_OP_READ;
2112 if (unlikely(dio->op == REQ_OP_DISCARD) && ic->mode != 'D')
2113 need_sync_io = true;
2115 if (need_sync_io && from_map) {
2116 INIT_WORK(&dio->work, integrity_bio_wait);
2117 queue_work(ic->offload_wq, &dio->work);
2122 spin_lock_irq(&ic->endio_wait.lock);
2124 if (unlikely(dm_integrity_failed(ic))) {
2125 spin_unlock_irq(&ic->endio_wait.lock);
2129 dio->range.n_sectors = bio_sectors(bio);
2130 journal_read_pos = NOT_FOUND;
2131 if (ic->mode == 'J' && likely(dio->op != REQ_OP_DISCARD)) {
2132 if (dio->op == REQ_OP_WRITE) {
2133 unsigned next_entry, i, pos;
2134 unsigned ws, we, range_sectors;
2136 dio->range.n_sectors = min(dio->range.n_sectors,
2137 (sector_t)ic->free_sectors << ic->sb->log2_sectors_per_block);
2138 if (unlikely(!dio->range.n_sectors)) {
2140 goto offload_to_thread;
2141 sleep_on_endio_wait(ic);
2144 range_sectors = dio->range.n_sectors >> ic->sb->log2_sectors_per_block;
2145 ic->free_sectors -= range_sectors;
2146 journal_section = ic->free_section;
2147 journal_entry = ic->free_section_entry;
2149 next_entry = ic->free_section_entry + range_sectors;
2150 ic->free_section_entry = next_entry % ic->journal_section_entries;
2151 ic->free_section += next_entry / ic->journal_section_entries;
2152 ic->n_uncommitted_sections += next_entry / ic->journal_section_entries;
2153 wraparound_section(ic, &ic->free_section);
2155 pos = journal_section * ic->journal_section_entries + journal_entry;
2156 ws = journal_section;
2160 struct journal_entry *je;
2162 add_journal_node(ic, &ic->journal_tree[pos], dio->range.logical_sector + i);
2164 if (unlikely(pos >= ic->journal_entries))
2167 je = access_journal_entry(ic, ws, we);
2168 BUG_ON(!journal_entry_is_unused(je));
2169 journal_entry_set_inprogress(je);
2171 if (unlikely(we == ic->journal_section_entries)) {
2174 wraparound_section(ic, &ws);
2176 } while ((i += ic->sectors_per_block) < dio->range.n_sectors);
2178 spin_unlock_irq(&ic->endio_wait.lock);
2179 goto journal_read_write;
2181 sector_t next_sector;
2182 journal_read_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
2183 if (likely(journal_read_pos == NOT_FOUND)) {
2184 if (unlikely(dio->range.n_sectors > next_sector - dio->range.logical_sector))
2185 dio->range.n_sectors = next_sector - dio->range.logical_sector;
2188 unsigned jp = journal_read_pos + 1;
2189 for (i = ic->sectors_per_block; i < dio->range.n_sectors; i += ic->sectors_per_block, jp++) {
2190 if (!test_journal_node(ic, jp, dio->range.logical_sector + i))
2193 dio->range.n_sectors = i;
2197 if (unlikely(!add_new_range(ic, &dio->range, true))) {
2199 * We must not sleep in the request routine because it could
2200 * stall bios on current->bio_list.
2201 * So, we offload the bio to a workqueue if we have to sleep.
2205 spin_unlock_irq(&ic->endio_wait.lock);
2206 INIT_WORK(&dio->work, integrity_bio_wait);
2207 queue_work(ic->wait_wq, &dio->work);
2210 if (journal_read_pos != NOT_FOUND)
2211 dio->range.n_sectors = ic->sectors_per_block;
2212 wait_and_add_new_range(ic, &dio->range);
2214 * wait_and_add_new_range drops the spinlock, so the journal
2215 * may have been changed arbitrarily. We need to recheck.
2216 * To simplify the code, we restrict I/O size to just one block.
2218 if (journal_read_pos != NOT_FOUND) {
2219 sector_t next_sector;
2220 unsigned new_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
2221 if (unlikely(new_pos != journal_read_pos)) {
2222 remove_range_unlocked(ic, &dio->range);
2227 if (ic->mode == 'J' && likely(dio->op == REQ_OP_DISCARD) && !discard_retried) {
2228 sector_t next_sector;
2229 unsigned new_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
2230 if (unlikely(new_pos != NOT_FOUND) ||
2231 unlikely(next_sector < dio->range.logical_sector - dio->range.n_sectors)) {
2232 remove_range_unlocked(ic, &dio->range);
2233 spin_unlock_irq(&ic->endio_wait.lock);
2234 queue_work(ic->commit_wq, &ic->commit_work);
2235 flush_workqueue(ic->commit_wq);
2236 queue_work(ic->writer_wq, &ic->writer_work);
2237 flush_workqueue(ic->writer_wq);
2238 discard_retried = true;
2242 spin_unlock_irq(&ic->endio_wait.lock);
2244 if (unlikely(journal_read_pos != NOT_FOUND)) {
2245 journal_section = journal_read_pos / ic->journal_section_entries;
2246 journal_entry = journal_read_pos % ic->journal_section_entries;
2247 goto journal_read_write;
2250 if (ic->mode == 'B' && (dio->op == REQ_OP_WRITE || unlikely(dio->op == REQ_OP_DISCARD))) {
2251 if (!block_bitmap_op(ic, ic->may_write_bitmap, dio->range.logical_sector,
2252 dio->range.n_sectors, BITMAP_OP_TEST_ALL_SET)) {
2253 struct bitmap_block_status *bbs;
2255 bbs = sector_to_bitmap_block(ic, dio->range.logical_sector);
2256 spin_lock(&bbs->bio_queue_lock);
2257 bio_list_add(&bbs->bio_queue, bio);
2258 spin_unlock(&bbs->bio_queue_lock);
2259 queue_work(ic->writer_wq, &bbs->work);
2264 dio->in_flight = (atomic_t)ATOMIC_INIT(2);
2267 init_completion(&read_comp);
2268 dio->completion = &read_comp;
2270 dio->completion = NULL;
2272 dm_bio_record(&dio->bio_details, bio);
2273 bio_set_dev(bio, ic->dev->bdev);
2274 bio->bi_integrity = NULL;
2275 bio->bi_opf &= ~REQ_INTEGRITY;
2276 bio->bi_end_io = integrity_end_io;
2277 bio->bi_iter.bi_size = dio->range.n_sectors << SECTOR_SHIFT;
2279 if (unlikely(dio->op == REQ_OP_DISCARD) && likely(ic->mode != 'D')) {
2280 integrity_metadata(&dio->work);
2281 dm_integrity_flush_buffers(ic, false);
2283 dio->in_flight = (atomic_t)ATOMIC_INIT(1);
2284 dio->completion = NULL;
2286 submit_bio_noacct(bio);
2291 submit_bio_noacct(bio);
2294 wait_for_completion_io(&read_comp);
2295 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING) &&
2296 dio->range.logical_sector + dio->range.n_sectors > le64_to_cpu(ic->sb->recalc_sector))
2298 if (ic->mode == 'B') {
2299 if (!block_bitmap_op(ic, ic->recalc_bitmap, dio->range.logical_sector,
2300 dio->range.n_sectors, BITMAP_OP_TEST_ALL_CLEAR))
2304 if (likely(!bio->bi_status))
2305 integrity_metadata(&dio->work);
2311 INIT_WORK(&dio->work, integrity_metadata);
2312 queue_work(ic->metadata_wq, &dio->work);
2318 if (unlikely(__journal_read_write(dio, bio, journal_section, journal_entry)))
2321 do_endio_flush(ic, dio);
2325 static void integrity_bio_wait(struct work_struct *w)
2327 struct dm_integrity_io *dio = container_of(w, struct dm_integrity_io, work);
2329 dm_integrity_map_continue(dio, false);
2332 static void pad_uncommitted(struct dm_integrity_c *ic)
2334 if (ic->free_section_entry) {
2335 ic->free_sectors -= ic->journal_section_entries - ic->free_section_entry;
2336 ic->free_section_entry = 0;
2338 wraparound_section(ic, &ic->free_section);
2339 ic->n_uncommitted_sections++;
2341 if (WARN_ON(ic->journal_sections * ic->journal_section_entries !=
2342 (ic->n_uncommitted_sections + ic->n_committed_sections) *
2343 ic->journal_section_entries + ic->free_sectors)) {
2344 DMCRIT("journal_sections %u, journal_section_entries %u, "
2345 "n_uncommitted_sections %u, n_committed_sections %u, "
2346 "journal_section_entries %u, free_sectors %u",
2347 ic->journal_sections, ic->journal_section_entries,
2348 ic->n_uncommitted_sections, ic->n_committed_sections,
2349 ic->journal_section_entries, ic->free_sectors);
2353 static void integrity_commit(struct work_struct *w)
2355 struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, commit_work);
2356 unsigned commit_start, commit_sections;
2358 struct bio *flushes;
2360 del_timer(&ic->autocommit_timer);
2362 spin_lock_irq(&ic->endio_wait.lock);
2363 flushes = bio_list_get(&ic->flush_bio_list);
2364 if (unlikely(ic->mode != 'J')) {
2365 spin_unlock_irq(&ic->endio_wait.lock);
2366 dm_integrity_flush_buffers(ic, true);
2367 goto release_flush_bios;
2370 pad_uncommitted(ic);
2371 commit_start = ic->uncommitted_section;
2372 commit_sections = ic->n_uncommitted_sections;
2373 spin_unlock_irq(&ic->endio_wait.lock);
2375 if (!commit_sections)
2376 goto release_flush_bios;
2379 for (n = 0; n < commit_sections; n++) {
2380 for (j = 0; j < ic->journal_section_entries; j++) {
2381 struct journal_entry *je;
2382 je = access_journal_entry(ic, i, j);
2383 io_wait_event(ic->copy_to_journal_wait, !journal_entry_is_inprogress(je));
2385 for (j = 0; j < ic->journal_section_sectors; j++) {
2386 struct journal_sector *js;
2387 js = access_journal(ic, i, j);
2388 js->commit_id = dm_integrity_commit_id(ic, i, j, ic->commit_seq);
2391 if (unlikely(i >= ic->journal_sections))
2392 ic->commit_seq = next_commit_seq(ic->commit_seq);
2393 wraparound_section(ic, &i);
2397 write_journal(ic, commit_start, commit_sections);
2399 spin_lock_irq(&ic->endio_wait.lock);
2400 ic->uncommitted_section += commit_sections;
2401 wraparound_section(ic, &ic->uncommitted_section);
2402 ic->n_uncommitted_sections -= commit_sections;
2403 ic->n_committed_sections += commit_sections;
2404 spin_unlock_irq(&ic->endio_wait.lock);
2406 if (READ_ONCE(ic->free_sectors) <= ic->free_sectors_threshold)
2407 queue_work(ic->writer_wq, &ic->writer_work);
2411 struct bio *next = flushes->bi_next;
2412 flushes->bi_next = NULL;
2413 do_endio(ic, flushes);
2418 static void complete_copy_from_journal(unsigned long error, void *context)
2420 struct journal_io *io = context;
2421 struct journal_completion *comp = io->comp;
2422 struct dm_integrity_c *ic = comp->ic;
2423 remove_range(ic, &io->range);
2424 mempool_free(io, &ic->journal_io_mempool);
2425 if (unlikely(error != 0))
2426 dm_integrity_io_error(ic, "copying from journal", -EIO);
2427 complete_journal_op(comp);
2430 static void restore_last_bytes(struct dm_integrity_c *ic, struct journal_sector *js,
2431 struct journal_entry *je)
2435 js->commit_id = je->last_bytes[s];
2437 } while (++s < ic->sectors_per_block);
2440 static void do_journal_write(struct dm_integrity_c *ic, unsigned write_start,
2441 unsigned write_sections, bool from_replay)
2444 struct journal_completion comp;
2445 struct blk_plug plug;
2447 blk_start_plug(&plug);
2450 comp.in_flight = (atomic_t)ATOMIC_INIT(1);
2451 init_completion(&comp.comp);
2454 for (n = 0; n < write_sections; n++, i++, wraparound_section(ic, &i)) {
2455 #ifndef INTERNAL_VERIFY
2456 if (unlikely(from_replay))
2458 rw_section_mac(ic, i, false);
2459 for (j = 0; j < ic->journal_section_entries; j++) {
2460 struct journal_entry *je = access_journal_entry(ic, i, j);
2461 sector_t sec, area, offset;
2462 unsigned k, l, next_loop;
2463 sector_t metadata_block;
2464 unsigned metadata_offset;
2465 struct journal_io *io;
2467 if (journal_entry_is_unused(je))
2469 BUG_ON(unlikely(journal_entry_is_inprogress(je)) && !from_replay);
2470 sec = journal_entry_get_sector(je);
2471 if (unlikely(from_replay)) {
2472 if (unlikely(sec & (unsigned)(ic->sectors_per_block - 1))) {
2473 dm_integrity_io_error(ic, "invalid sector in journal", -EIO);
2474 sec &= ~(sector_t)(ic->sectors_per_block - 1);
2477 if (unlikely(sec >= ic->provided_data_sectors))
2479 get_area_and_offset(ic, sec, &area, &offset);
2480 restore_last_bytes(ic, access_journal_data(ic, i, j), je);
2481 for (k = j + 1; k < ic->journal_section_entries; k++) {
2482 struct journal_entry *je2 = access_journal_entry(ic, i, k);
2483 sector_t sec2, area2, offset2;
2484 if (journal_entry_is_unused(je2))
2486 BUG_ON(unlikely(journal_entry_is_inprogress(je2)) && !from_replay);
2487 sec2 = journal_entry_get_sector(je2);
2488 if (unlikely(sec2 >= ic->provided_data_sectors))
2490 get_area_and_offset(ic, sec2, &area2, &offset2);
2491 if (area2 != area || offset2 != offset + ((k - j) << ic->sb->log2_sectors_per_block))
2493 restore_last_bytes(ic, access_journal_data(ic, i, k), je2);
2497 io = mempool_alloc(&ic->journal_io_mempool, GFP_NOIO);
2499 io->range.logical_sector = sec;
2500 io->range.n_sectors = (k - j) << ic->sb->log2_sectors_per_block;
2502 spin_lock_irq(&ic->endio_wait.lock);
2503 add_new_range_and_wait(ic, &io->range);
2505 if (likely(!from_replay)) {
2506 struct journal_node *section_node = &ic->journal_tree[i * ic->journal_section_entries];
2508 /* don't write if there is newer committed sector */
2509 while (j < k && find_newer_committed_node(ic, §ion_node[j])) {
2510 struct journal_entry *je2 = access_journal_entry(ic, i, j);
2512 journal_entry_set_unused(je2);
2513 remove_journal_node(ic, §ion_node[j]);
2515 sec += ic->sectors_per_block;
2516 offset += ic->sectors_per_block;
2518 while (j < k && find_newer_committed_node(ic, §ion_node[k - 1])) {
2519 struct journal_entry *je2 = access_journal_entry(ic, i, k - 1);
2521 journal_entry_set_unused(je2);
2522 remove_journal_node(ic, §ion_node[k - 1]);
2526 remove_range_unlocked(ic, &io->range);
2527 spin_unlock_irq(&ic->endio_wait.lock);
2528 mempool_free(io, &ic->journal_io_mempool);
2531 for (l = j; l < k; l++) {
2532 remove_journal_node(ic, §ion_node[l]);
2535 spin_unlock_irq(&ic->endio_wait.lock);
2537 metadata_block = get_metadata_sector_and_offset(ic, area, offset, &metadata_offset);
2538 for (l = j; l < k; l++) {
2540 struct journal_entry *je2 = access_journal_entry(ic, i, l);
2543 #ifndef INTERNAL_VERIFY
2544 unlikely(from_replay) &&
2546 ic->internal_hash) {
2547 char test_tag[max_t(size_t, HASH_MAX_DIGESTSIZE, MAX_TAG_SIZE)];
2549 integrity_sector_checksum(ic, sec + ((l - j) << ic->sb->log2_sectors_per_block),
2550 (char *)access_journal_data(ic, i, l), test_tag);
2551 if (unlikely(memcmp(test_tag, journal_entry_tag(ic, je2), ic->tag_size))) {
2552 dm_integrity_io_error(ic, "tag mismatch when replaying journal", -EILSEQ);
2553 dm_audit_log_target(DM_MSG_PREFIX, "integrity-replay-journal", ic->ti, 0);
2557 journal_entry_set_unused(je2);
2558 r = dm_integrity_rw_tag(ic, journal_entry_tag(ic, je2), &metadata_block, &metadata_offset,
2559 ic->tag_size, TAG_WRITE);
2561 dm_integrity_io_error(ic, "reading tags", r);
2565 atomic_inc(&comp.in_flight);
2566 copy_from_journal(ic, i, j << ic->sb->log2_sectors_per_block,
2567 (k - j) << ic->sb->log2_sectors_per_block,
2568 get_data_sector(ic, area, offset),
2569 complete_copy_from_journal, io);
2575 dm_bufio_write_dirty_buffers_async(ic->bufio);
2577 blk_finish_plug(&plug);
2579 complete_journal_op(&comp);
2580 wait_for_completion_io(&comp.comp);
2582 dm_integrity_flush_buffers(ic, true);
2585 static void integrity_writer(struct work_struct *w)
2587 struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, writer_work);
2588 unsigned write_start, write_sections;
2590 unsigned prev_free_sectors;
2592 /* the following test is not needed, but it tests the replay code */
2593 if (unlikely(dm_post_suspending(ic->ti)) && !ic->meta_dev)
2596 spin_lock_irq(&ic->endio_wait.lock);
2597 write_start = ic->committed_section;
2598 write_sections = ic->n_committed_sections;
2599 spin_unlock_irq(&ic->endio_wait.lock);
2601 if (!write_sections)
2604 do_journal_write(ic, write_start, write_sections, false);
2606 spin_lock_irq(&ic->endio_wait.lock);
2608 ic->committed_section += write_sections;
2609 wraparound_section(ic, &ic->committed_section);
2610 ic->n_committed_sections -= write_sections;
2612 prev_free_sectors = ic->free_sectors;
2613 ic->free_sectors += write_sections * ic->journal_section_entries;
2614 if (unlikely(!prev_free_sectors))
2615 wake_up_locked(&ic->endio_wait);
2617 spin_unlock_irq(&ic->endio_wait.lock);
2620 static void recalc_write_super(struct dm_integrity_c *ic)
2624 dm_integrity_flush_buffers(ic, false);
2625 if (dm_integrity_failed(ic))
2628 r = sync_rw_sb(ic, REQ_OP_WRITE, 0);
2630 dm_integrity_io_error(ic, "writing superblock", r);
2633 static void integrity_recalc(struct work_struct *w)
2635 struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, recalc_work);
2636 struct dm_integrity_range range;
2637 struct dm_io_request io_req;
2638 struct dm_io_region io_loc;
2639 sector_t area, offset;
2640 sector_t metadata_block;
2641 unsigned metadata_offset;
2642 sector_t logical_sector, n_sectors;
2646 unsigned super_counter = 0;
2648 DEBUG_print("start recalculation... (position %llx)\n", le64_to_cpu(ic->sb->recalc_sector));
2650 spin_lock_irq(&ic->endio_wait.lock);
2654 if (unlikely(dm_post_suspending(ic->ti)))
2657 range.logical_sector = le64_to_cpu(ic->sb->recalc_sector);
2658 if (unlikely(range.logical_sector >= ic->provided_data_sectors)) {
2659 if (ic->mode == 'B') {
2660 block_bitmap_op(ic, ic->recalc_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_CLEAR);
2661 DEBUG_print("queue_delayed_work: bitmap_flush_work\n");
2662 queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, 0);
2667 get_area_and_offset(ic, range.logical_sector, &area, &offset);
2668 range.n_sectors = min((sector_t)RECALC_SECTORS, ic->provided_data_sectors - range.logical_sector);
2670 range.n_sectors = min(range.n_sectors, ((sector_t)1U << ic->sb->log2_interleave_sectors) - (unsigned)offset);
2672 add_new_range_and_wait(ic, &range);
2673 spin_unlock_irq(&ic->endio_wait.lock);
2674 logical_sector = range.logical_sector;
2675 n_sectors = range.n_sectors;
2677 if (ic->mode == 'B') {
2678 if (block_bitmap_op(ic, ic->recalc_bitmap, logical_sector, n_sectors, BITMAP_OP_TEST_ALL_CLEAR)) {
2679 goto advance_and_next;
2681 while (block_bitmap_op(ic, ic->recalc_bitmap, logical_sector,
2682 ic->sectors_per_block, BITMAP_OP_TEST_ALL_CLEAR)) {
2683 logical_sector += ic->sectors_per_block;
2684 n_sectors -= ic->sectors_per_block;
2687 while (block_bitmap_op(ic, ic->recalc_bitmap, logical_sector + n_sectors - ic->sectors_per_block,
2688 ic->sectors_per_block, BITMAP_OP_TEST_ALL_CLEAR)) {
2689 n_sectors -= ic->sectors_per_block;
2692 get_area_and_offset(ic, logical_sector, &area, &offset);
2695 DEBUG_print("recalculating: %llx, %llx\n", logical_sector, n_sectors);
2697 if (unlikely(++super_counter == RECALC_WRITE_SUPER)) {
2698 recalc_write_super(ic);
2699 if (ic->mode == 'B') {
2700 queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, ic->bitmap_flush_interval);
2705 if (unlikely(dm_integrity_failed(ic)))
2708 io_req.bi_op = REQ_OP_READ;
2709 io_req.bi_op_flags = 0;
2710 io_req.mem.type = DM_IO_VMA;
2711 io_req.mem.ptr.addr = ic->recalc_buffer;
2712 io_req.notify.fn = NULL;
2713 io_req.client = ic->io;
2714 io_loc.bdev = ic->dev->bdev;
2715 io_loc.sector = get_data_sector(ic, area, offset);
2716 io_loc.count = n_sectors;
2718 r = dm_io(&io_req, 1, &io_loc, NULL);
2720 dm_integrity_io_error(ic, "reading data", r);
2724 t = ic->recalc_tags;
2725 for (i = 0; i < n_sectors; i += ic->sectors_per_block) {
2726 integrity_sector_checksum(ic, logical_sector + i, ic->recalc_buffer + (i << SECTOR_SHIFT), t);
2730 metadata_block = get_metadata_sector_and_offset(ic, area, offset, &metadata_offset);
2732 r = dm_integrity_rw_tag(ic, ic->recalc_tags, &metadata_block, &metadata_offset, t - ic->recalc_tags, TAG_WRITE);
2734 dm_integrity_io_error(ic, "writing tags", r);
2738 if (ic->mode == 'B') {
2739 sector_t start, end;
2740 start = (range.logical_sector >>
2741 (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit)) <<
2742 (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
2743 end = ((range.logical_sector + range.n_sectors) >>
2744 (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit)) <<
2745 (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
2746 block_bitmap_op(ic, ic->recalc_bitmap, start, end - start, BITMAP_OP_CLEAR);
2752 spin_lock_irq(&ic->endio_wait.lock);
2753 remove_range_unlocked(ic, &range);
2754 ic->sb->recalc_sector = cpu_to_le64(range.logical_sector + range.n_sectors);
2758 remove_range(ic, &range);
2762 spin_unlock_irq(&ic->endio_wait.lock);
2764 recalc_write_super(ic);
2767 static void bitmap_block_work(struct work_struct *w)
2769 struct bitmap_block_status *bbs = container_of(w, struct bitmap_block_status, work);
2770 struct dm_integrity_c *ic = bbs->ic;
2772 struct bio_list bio_queue;
2773 struct bio_list waiting;
2775 bio_list_init(&waiting);
2777 spin_lock(&bbs->bio_queue_lock);
2778 bio_queue = bbs->bio_queue;
2779 bio_list_init(&bbs->bio_queue);
2780 spin_unlock(&bbs->bio_queue_lock);
2782 while ((bio = bio_list_pop(&bio_queue))) {
2783 struct dm_integrity_io *dio;
2785 dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
2787 if (block_bitmap_op(ic, ic->may_write_bitmap, dio->range.logical_sector,
2788 dio->range.n_sectors, BITMAP_OP_TEST_ALL_SET)) {
2789 remove_range(ic, &dio->range);
2790 INIT_WORK(&dio->work, integrity_bio_wait);
2791 queue_work(ic->offload_wq, &dio->work);
2793 block_bitmap_op(ic, ic->journal, dio->range.logical_sector,
2794 dio->range.n_sectors, BITMAP_OP_SET);
2795 bio_list_add(&waiting, bio);
2799 if (bio_list_empty(&waiting))
2802 rw_journal_sectors(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC,
2803 bbs->idx * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT),
2804 BITMAP_BLOCK_SIZE >> SECTOR_SHIFT, NULL);
2806 while ((bio = bio_list_pop(&waiting))) {
2807 struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
2809 block_bitmap_op(ic, ic->may_write_bitmap, dio->range.logical_sector,
2810 dio->range.n_sectors, BITMAP_OP_SET);
2812 remove_range(ic, &dio->range);
2813 INIT_WORK(&dio->work, integrity_bio_wait);
2814 queue_work(ic->offload_wq, &dio->work);
2817 queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, ic->bitmap_flush_interval);
2820 static void bitmap_flush_work(struct work_struct *work)
2822 struct dm_integrity_c *ic = container_of(work, struct dm_integrity_c, bitmap_flush_work.work);
2823 struct dm_integrity_range range;
2824 unsigned long limit;
2827 dm_integrity_flush_buffers(ic, false);
2829 range.logical_sector = 0;
2830 range.n_sectors = ic->provided_data_sectors;
2832 spin_lock_irq(&ic->endio_wait.lock);
2833 add_new_range_and_wait(ic, &range);
2834 spin_unlock_irq(&ic->endio_wait.lock);
2836 dm_integrity_flush_buffers(ic, true);
2838 limit = ic->provided_data_sectors;
2839 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
2840 limit = le64_to_cpu(ic->sb->recalc_sector)
2841 >> (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit)
2842 << (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
2844 /*DEBUG_print("zeroing journal\n");*/
2845 block_bitmap_op(ic, ic->journal, 0, limit, BITMAP_OP_CLEAR);
2846 block_bitmap_op(ic, ic->may_write_bitmap, 0, limit, BITMAP_OP_CLEAR);
2848 rw_journal_sectors(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, 0,
2849 ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
2851 spin_lock_irq(&ic->endio_wait.lock);
2852 remove_range_unlocked(ic, &range);
2853 while (unlikely((bio = bio_list_pop(&ic->synchronous_bios)) != NULL)) {
2855 spin_unlock_irq(&ic->endio_wait.lock);
2856 spin_lock_irq(&ic->endio_wait.lock);
2858 spin_unlock_irq(&ic->endio_wait.lock);
2862 static void init_journal(struct dm_integrity_c *ic, unsigned start_section,
2863 unsigned n_sections, unsigned char commit_seq)
2870 for (n = 0; n < n_sections; n++) {
2871 i = start_section + n;
2872 wraparound_section(ic, &i);
2873 for (j = 0; j < ic->journal_section_sectors; j++) {
2874 struct journal_sector *js = access_journal(ic, i, j);
2875 BUILD_BUG_ON(sizeof(js->sectors) != JOURNAL_SECTOR_DATA);
2876 memset(&js->sectors, 0, sizeof(js->sectors));
2877 js->commit_id = dm_integrity_commit_id(ic, i, j, commit_seq);
2879 for (j = 0; j < ic->journal_section_entries; j++) {
2880 struct journal_entry *je = access_journal_entry(ic, i, j);
2881 journal_entry_set_unused(je);
2885 write_journal(ic, start_section, n_sections);
2888 static int find_commit_seq(struct dm_integrity_c *ic, unsigned i, unsigned j, commit_id_t id)
2891 for (k = 0; k < N_COMMIT_IDS; k++) {
2892 if (dm_integrity_commit_id(ic, i, j, k) == id)
2895 dm_integrity_io_error(ic, "journal commit id", -EIO);
2899 static void replay_journal(struct dm_integrity_c *ic)
2902 bool used_commit_ids[N_COMMIT_IDS];
2903 unsigned max_commit_id_sections[N_COMMIT_IDS];
2904 unsigned write_start, write_sections;
2905 unsigned continue_section;
2907 unsigned char unused, last_used, want_commit_seq;
2909 if (ic->mode == 'R')
2912 if (ic->journal_uptodate)
2918 if (!ic->just_formatted) {
2919 DEBUG_print("reading journal\n");
2920 rw_journal(ic, REQ_OP_READ, 0, 0, ic->journal_sections, NULL);
2922 DEBUG_bytes(lowmem_page_address(ic->journal_io[0].page), 64, "read journal");
2923 if (ic->journal_io) {
2924 struct journal_completion crypt_comp;
2926 init_completion(&crypt_comp.comp);
2927 crypt_comp.in_flight = (atomic_t)ATOMIC_INIT(0);
2928 encrypt_journal(ic, false, 0, ic->journal_sections, &crypt_comp);
2929 wait_for_completion(&crypt_comp.comp);
2931 DEBUG_bytes(lowmem_page_address(ic->journal[0].page), 64, "decrypted journal");
2934 if (dm_integrity_failed(ic))
2937 journal_empty = true;
2938 memset(used_commit_ids, 0, sizeof used_commit_ids);
2939 memset(max_commit_id_sections, 0, sizeof max_commit_id_sections);
2940 for (i = 0; i < ic->journal_sections; i++) {
2941 for (j = 0; j < ic->journal_section_sectors; j++) {
2943 struct journal_sector *js = access_journal(ic, i, j);
2944 k = find_commit_seq(ic, i, j, js->commit_id);
2947 used_commit_ids[k] = true;
2948 max_commit_id_sections[k] = i;
2950 if (journal_empty) {
2951 for (j = 0; j < ic->journal_section_entries; j++) {
2952 struct journal_entry *je = access_journal_entry(ic, i, j);
2953 if (!journal_entry_is_unused(je)) {
2954 journal_empty = false;
2961 if (!used_commit_ids[N_COMMIT_IDS - 1]) {
2962 unused = N_COMMIT_IDS - 1;
2963 while (unused && !used_commit_ids[unused - 1])
2966 for (unused = 0; unused < N_COMMIT_IDS; unused++)
2967 if (!used_commit_ids[unused])
2969 if (unused == N_COMMIT_IDS) {
2970 dm_integrity_io_error(ic, "journal commit ids", -EIO);
2974 DEBUG_print("first unused commit seq %d [%d,%d,%d,%d]\n",
2975 unused, used_commit_ids[0], used_commit_ids[1],
2976 used_commit_ids[2], used_commit_ids[3]);
2978 last_used = prev_commit_seq(unused);
2979 want_commit_seq = prev_commit_seq(last_used);
2981 if (!used_commit_ids[want_commit_seq] && used_commit_ids[prev_commit_seq(want_commit_seq)])
2982 journal_empty = true;
2984 write_start = max_commit_id_sections[last_used] + 1;
2985 if (unlikely(write_start >= ic->journal_sections))
2986 want_commit_seq = next_commit_seq(want_commit_seq);
2987 wraparound_section(ic, &write_start);
2990 for (write_sections = 0; write_sections < ic->journal_sections; write_sections++) {
2991 for (j = 0; j < ic->journal_section_sectors; j++) {
2992 struct journal_sector *js = access_journal(ic, i, j);
2994 if (js->commit_id != dm_integrity_commit_id(ic, i, j, want_commit_seq)) {
2996 * This could be caused by crash during writing.
2997 * We won't replay the inconsistent part of the
3000 DEBUG_print("commit id mismatch at position (%u, %u): %d != %d\n",
3001 i, j, find_commit_seq(ic, i, j, js->commit_id), want_commit_seq);
3006 if (unlikely(i >= ic->journal_sections))
3007 want_commit_seq = next_commit_seq(want_commit_seq);
3008 wraparound_section(ic, &i);
3012 if (!journal_empty) {
3013 DEBUG_print("replaying %u sections, starting at %u, commit seq %d\n",
3014 write_sections, write_start, want_commit_seq);
3015 do_journal_write(ic, write_start, write_sections, true);
3018 if (write_sections == ic->journal_sections && (ic->mode == 'J' || journal_empty)) {
3019 continue_section = write_start;
3020 ic->commit_seq = want_commit_seq;
3021 DEBUG_print("continuing from section %u, commit seq %d\n", write_start, ic->commit_seq);
3024 unsigned char erase_seq;
3026 DEBUG_print("clearing journal\n");
3028 erase_seq = prev_commit_seq(prev_commit_seq(last_used));
3030 init_journal(ic, s, 1, erase_seq);
3032 wraparound_section(ic, &s);
3033 if (ic->journal_sections >= 2) {
3034 init_journal(ic, s, ic->journal_sections - 2, erase_seq);
3035 s += ic->journal_sections - 2;
3036 wraparound_section(ic, &s);
3037 init_journal(ic, s, 1, erase_seq);
3040 continue_section = 0;
3041 ic->commit_seq = next_commit_seq(erase_seq);
3044 ic->committed_section = continue_section;
3045 ic->n_committed_sections = 0;
3047 ic->uncommitted_section = continue_section;
3048 ic->n_uncommitted_sections = 0;
3050 ic->free_section = continue_section;
3051 ic->free_section_entry = 0;
3052 ic->free_sectors = ic->journal_entries;
3054 ic->journal_tree_root = RB_ROOT;
3055 for (i = 0; i < ic->journal_entries; i++)
3056 init_journal_node(&ic->journal_tree[i]);
3059 static void dm_integrity_enter_synchronous_mode(struct dm_integrity_c *ic)
3061 DEBUG_print("dm_integrity_enter_synchronous_mode\n");
3063 if (ic->mode == 'B') {
3064 ic->bitmap_flush_interval = msecs_to_jiffies(10) + 1;
3065 ic->synchronous_mode = 1;
3067 cancel_delayed_work_sync(&ic->bitmap_flush_work);
3068 queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, 0);
3069 flush_workqueue(ic->commit_wq);
3073 static int dm_integrity_reboot(struct notifier_block *n, unsigned long code, void *x)
3075 struct dm_integrity_c *ic = container_of(n, struct dm_integrity_c, reboot_notifier);
3077 DEBUG_print("dm_integrity_reboot\n");
3079 dm_integrity_enter_synchronous_mode(ic);
3084 static void dm_integrity_postsuspend(struct dm_target *ti)
3086 struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
3089 WARN_ON(unregister_reboot_notifier(&ic->reboot_notifier));
3091 del_timer_sync(&ic->autocommit_timer);
3094 drain_workqueue(ic->recalc_wq);
3096 if (ic->mode == 'B')
3097 cancel_delayed_work_sync(&ic->bitmap_flush_work);
3099 queue_work(ic->commit_wq, &ic->commit_work);
3100 drain_workqueue(ic->commit_wq);
3102 if (ic->mode == 'J') {
3104 queue_work(ic->writer_wq, &ic->writer_work);
3105 drain_workqueue(ic->writer_wq);
3106 dm_integrity_flush_buffers(ic, true);
3109 if (ic->mode == 'B') {
3110 dm_integrity_flush_buffers(ic, true);
3112 /* set to 0 to test bitmap replay code */
3113 init_journal(ic, 0, ic->journal_sections, 0);
3114 ic->sb->flags &= ~cpu_to_le32(SB_FLAG_DIRTY_BITMAP);
3115 r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
3117 dm_integrity_io_error(ic, "writing superblock", r);
3121 BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
3123 ic->journal_uptodate = true;
3126 static void dm_integrity_resume(struct dm_target *ti)
3128 struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
3129 __u64 old_provided_data_sectors = le64_to_cpu(ic->sb->provided_data_sectors);
3132 DEBUG_print("resume\n");
3134 if (ic->provided_data_sectors != old_provided_data_sectors) {
3135 if (ic->provided_data_sectors > old_provided_data_sectors &&
3137 ic->sb->log2_blocks_per_bitmap_bit == ic->log2_blocks_per_bitmap_bit) {
3138 rw_journal_sectors(ic, REQ_OP_READ, 0, 0,
3139 ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3140 block_bitmap_op(ic, ic->journal, old_provided_data_sectors,
3141 ic->provided_data_sectors - old_provided_data_sectors, BITMAP_OP_SET);
3142 rw_journal_sectors(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, 0,
3143 ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3146 ic->sb->provided_data_sectors = cpu_to_le64(ic->provided_data_sectors);
3147 r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
3149 dm_integrity_io_error(ic, "writing superblock", r);
3152 if (ic->sb->flags & cpu_to_le32(SB_FLAG_DIRTY_BITMAP)) {
3153 DEBUG_print("resume dirty_bitmap\n");
3154 rw_journal_sectors(ic, REQ_OP_READ, 0, 0,
3155 ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3156 if (ic->mode == 'B') {
3157 if (ic->sb->log2_blocks_per_bitmap_bit == ic->log2_blocks_per_bitmap_bit &&
3158 !ic->reset_recalculate_flag) {
3159 block_bitmap_copy(ic, ic->recalc_bitmap, ic->journal);
3160 block_bitmap_copy(ic, ic->may_write_bitmap, ic->journal);
3161 if (!block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors,
3162 BITMAP_OP_TEST_ALL_CLEAR)) {
3163 ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
3164 ic->sb->recalc_sector = cpu_to_le64(0);
3167 DEBUG_print("non-matching blocks_per_bitmap_bit: %u, %u\n",
3168 ic->sb->log2_blocks_per_bitmap_bit, ic->log2_blocks_per_bitmap_bit);
3169 ic->sb->log2_blocks_per_bitmap_bit = ic->log2_blocks_per_bitmap_bit;
3170 block_bitmap_op(ic, ic->recalc_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_SET);
3171 block_bitmap_op(ic, ic->may_write_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_SET);
3172 block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors, BITMAP_OP_SET);
3173 rw_journal_sectors(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, 0,
3174 ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3175 ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
3176 ic->sb->recalc_sector = cpu_to_le64(0);
3179 if (!(ic->sb->log2_blocks_per_bitmap_bit == ic->log2_blocks_per_bitmap_bit &&
3180 block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors, BITMAP_OP_TEST_ALL_CLEAR)) ||
3181 ic->reset_recalculate_flag) {
3182 ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
3183 ic->sb->recalc_sector = cpu_to_le64(0);
3185 init_journal(ic, 0, ic->journal_sections, 0);
3187 ic->sb->flags &= ~cpu_to_le32(SB_FLAG_DIRTY_BITMAP);
3189 r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
3191 dm_integrity_io_error(ic, "writing superblock", r);
3194 if (ic->reset_recalculate_flag) {
3195 ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
3196 ic->sb->recalc_sector = cpu_to_le64(0);
3198 if (ic->mode == 'B') {
3199 ic->sb->flags |= cpu_to_le32(SB_FLAG_DIRTY_BITMAP);
3200 ic->sb->log2_blocks_per_bitmap_bit = ic->log2_blocks_per_bitmap_bit;
3201 r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
3203 dm_integrity_io_error(ic, "writing superblock", r);
3205 block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors, BITMAP_OP_CLEAR);
3206 block_bitmap_op(ic, ic->recalc_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_CLEAR);
3207 block_bitmap_op(ic, ic->may_write_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_CLEAR);
3208 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING) &&
3209 le64_to_cpu(ic->sb->recalc_sector) < ic->provided_data_sectors) {
3210 block_bitmap_op(ic, ic->journal, le64_to_cpu(ic->sb->recalc_sector),
3211 ic->provided_data_sectors - le64_to_cpu(ic->sb->recalc_sector), BITMAP_OP_SET);
3212 block_bitmap_op(ic, ic->recalc_bitmap, le64_to_cpu(ic->sb->recalc_sector),
3213 ic->provided_data_sectors - le64_to_cpu(ic->sb->recalc_sector), BITMAP_OP_SET);
3214 block_bitmap_op(ic, ic->may_write_bitmap, le64_to_cpu(ic->sb->recalc_sector),
3215 ic->provided_data_sectors - le64_to_cpu(ic->sb->recalc_sector), BITMAP_OP_SET);
3217 rw_journal_sectors(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, 0,
3218 ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3222 DEBUG_print("testing recalc: %x\n", ic->sb->flags);
3223 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
3224 __u64 recalc_pos = le64_to_cpu(ic->sb->recalc_sector);
3225 DEBUG_print("recalc pos: %llx / %llx\n", recalc_pos, ic->provided_data_sectors);
3226 if (recalc_pos < ic->provided_data_sectors) {
3227 queue_work(ic->recalc_wq, &ic->recalc_work);
3228 } else if (recalc_pos > ic->provided_data_sectors) {
3229 ic->sb->recalc_sector = cpu_to_le64(ic->provided_data_sectors);
3230 recalc_write_super(ic);
3234 ic->reboot_notifier.notifier_call = dm_integrity_reboot;
3235 ic->reboot_notifier.next = NULL;
3236 ic->reboot_notifier.priority = INT_MAX - 1; /* be notified after md and before hardware drivers */
3237 WARN_ON(register_reboot_notifier(&ic->reboot_notifier));
3240 /* set to 1 to stress test synchronous mode */
3241 dm_integrity_enter_synchronous_mode(ic);
3245 static void dm_integrity_status(struct dm_target *ti, status_type_t type,
3246 unsigned status_flags, char *result, unsigned maxlen)
3248 struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
3253 case STATUSTYPE_INFO:
3255 (unsigned long long)atomic64_read(&ic->number_of_mismatches),
3256 ic->provided_data_sectors);
3257 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
3258 DMEMIT(" %llu", le64_to_cpu(ic->sb->recalc_sector));
3263 case STATUSTYPE_TABLE: {
3264 __u64 watermark_percentage = (__u64)(ic->journal_entries - ic->free_sectors_threshold) * 100;
3265 watermark_percentage += ic->journal_entries / 2;
3266 do_div(watermark_percentage, ic->journal_entries);
3268 arg_count += !!ic->meta_dev;
3269 arg_count += ic->sectors_per_block != 1;
3270 arg_count += !!(ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING));
3271 arg_count += ic->reset_recalculate_flag;
3272 arg_count += ic->discard;
3273 arg_count += ic->mode == 'J';
3274 arg_count += ic->mode == 'J';
3275 arg_count += ic->mode == 'B';
3276 arg_count += ic->mode == 'B';
3277 arg_count += !!ic->internal_hash_alg.alg_string;
3278 arg_count += !!ic->journal_crypt_alg.alg_string;
3279 arg_count += !!ic->journal_mac_alg.alg_string;
3280 arg_count += (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING)) != 0;
3281 arg_count += (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) != 0;
3282 arg_count += ic->legacy_recalculate;
3283 DMEMIT("%s %llu %u %c %u", ic->dev->name, ic->start,
3284 ic->tag_size, ic->mode, arg_count);
3286 DMEMIT(" meta_device:%s", ic->meta_dev->name);
3287 if (ic->sectors_per_block != 1)
3288 DMEMIT(" block_size:%u", ic->sectors_per_block << SECTOR_SHIFT);
3289 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
3290 DMEMIT(" recalculate");
3291 if (ic->reset_recalculate_flag)
3292 DMEMIT(" reset_recalculate");
3294 DMEMIT(" allow_discards");
3295 DMEMIT(" journal_sectors:%u", ic->initial_sectors - SB_SECTORS);
3296 DMEMIT(" interleave_sectors:%u", 1U << ic->sb->log2_interleave_sectors);
3297 DMEMIT(" buffer_sectors:%u", 1U << ic->log2_buffer_sectors);
3298 if (ic->mode == 'J') {
3299 DMEMIT(" journal_watermark:%u", (unsigned)watermark_percentage);
3300 DMEMIT(" commit_time:%u", ic->autocommit_msec);
3302 if (ic->mode == 'B') {
3303 DMEMIT(" sectors_per_bit:%llu", (sector_t)ic->sectors_per_block << ic->log2_blocks_per_bitmap_bit);
3304 DMEMIT(" bitmap_flush_interval:%u", jiffies_to_msecs(ic->bitmap_flush_interval));
3306 if ((ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING)) != 0)
3307 DMEMIT(" fix_padding");
3308 if ((ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) != 0)
3309 DMEMIT(" fix_hmac");
3310 if (ic->legacy_recalculate)
3311 DMEMIT(" legacy_recalculate");
3313 #define EMIT_ALG(a, n) \
3315 if (ic->a.alg_string) { \
3316 DMEMIT(" %s:%s", n, ic->a.alg_string); \
3317 if (ic->a.key_string) \
3318 DMEMIT(":%s", ic->a.key_string);\
3321 EMIT_ALG(internal_hash_alg, "internal_hash");
3322 EMIT_ALG(journal_crypt_alg, "journal_crypt");
3323 EMIT_ALG(journal_mac_alg, "journal_mac");
3326 case STATUSTYPE_IMA:
3327 DMEMIT_TARGET_NAME_VERSION(ti->type);
3328 DMEMIT(",dev_name=%s,start=%llu,tag_size=%u,mode=%c",
3329 ic->dev->name, ic->start, ic->tag_size, ic->mode);
3332 DMEMIT(",meta_device=%s", ic->meta_dev->name);
3333 if (ic->sectors_per_block != 1)
3334 DMEMIT(",block_size=%u", ic->sectors_per_block << SECTOR_SHIFT);
3336 DMEMIT(",recalculate=%c", (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) ?
3338 DMEMIT(",allow_discards=%c", ic->discard ? 'y' : 'n');
3339 DMEMIT(",fix_padding=%c",
3340 ((ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING)) != 0) ? 'y' : 'n');
3341 DMEMIT(",fix_hmac=%c",
3342 ((ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) != 0) ? 'y' : 'n');
3343 DMEMIT(",legacy_recalculate=%c", ic->legacy_recalculate ? 'y' : 'n');
3345 DMEMIT(",journal_sectors=%u", ic->initial_sectors - SB_SECTORS);
3346 DMEMIT(",interleave_sectors=%u", 1U << ic->sb->log2_interleave_sectors);
3347 DMEMIT(",buffer_sectors=%u", 1U << ic->log2_buffer_sectors);
3353 static int dm_integrity_iterate_devices(struct dm_target *ti,
3354 iterate_devices_callout_fn fn, void *data)
3356 struct dm_integrity_c *ic = ti->private;
3359 return fn(ti, ic->dev, ic->start + ic->initial_sectors + ic->metadata_run, ti->len, data);
3361 return fn(ti, ic->dev, 0, ti->len, data);
3364 static void dm_integrity_io_hints(struct dm_target *ti, struct queue_limits *limits)
3366 struct dm_integrity_c *ic = ti->private;
3368 if (ic->sectors_per_block > 1) {
3369 limits->logical_block_size = ic->sectors_per_block << SECTOR_SHIFT;
3370 limits->physical_block_size = ic->sectors_per_block << SECTOR_SHIFT;
3371 blk_limits_io_min(limits, ic->sectors_per_block << SECTOR_SHIFT);
3375 static void calculate_journal_section_size(struct dm_integrity_c *ic)
3377 unsigned sector_space = JOURNAL_SECTOR_DATA;
3379 ic->journal_sections = le32_to_cpu(ic->sb->journal_sections);
3380 ic->journal_entry_size = roundup(offsetof(struct journal_entry, last_bytes[ic->sectors_per_block]) + ic->tag_size,
3381 JOURNAL_ENTRY_ROUNDUP);
3383 if (ic->sb->flags & cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC))
3384 sector_space -= JOURNAL_MAC_PER_SECTOR;
3385 ic->journal_entries_per_sector = sector_space / ic->journal_entry_size;
3386 ic->journal_section_entries = ic->journal_entries_per_sector * JOURNAL_BLOCK_SECTORS;
3387 ic->journal_section_sectors = (ic->journal_section_entries << ic->sb->log2_sectors_per_block) + JOURNAL_BLOCK_SECTORS;
3388 ic->journal_entries = ic->journal_section_entries * ic->journal_sections;
3391 static int calculate_device_limits(struct dm_integrity_c *ic)
3393 __u64 initial_sectors;
3395 calculate_journal_section_size(ic);
3396 initial_sectors = SB_SECTORS + (__u64)ic->journal_section_sectors * ic->journal_sections;
3397 if (initial_sectors + METADATA_PADDING_SECTORS >= ic->meta_device_sectors || initial_sectors > UINT_MAX)
3399 ic->initial_sectors = initial_sectors;
3401 if (!ic->meta_dev) {
3402 sector_t last_sector, last_area, last_offset;
3404 /* we have to maintain excessive padding for compatibility with existing volumes */
3405 __u64 metadata_run_padding =
3406 ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING) ?
3407 (__u64)(METADATA_PADDING_SECTORS << SECTOR_SHIFT) :
3408 (__u64)(1 << SECTOR_SHIFT << METADATA_PADDING_SECTORS);
3410 ic->metadata_run = round_up((__u64)ic->tag_size << (ic->sb->log2_interleave_sectors - ic->sb->log2_sectors_per_block),
3411 metadata_run_padding) >> SECTOR_SHIFT;
3412 if (!(ic->metadata_run & (ic->metadata_run - 1)))
3413 ic->log2_metadata_run = __ffs(ic->metadata_run);
3415 ic->log2_metadata_run = -1;
3417 get_area_and_offset(ic, ic->provided_data_sectors - 1, &last_area, &last_offset);
3418 last_sector = get_data_sector(ic, last_area, last_offset);
3419 if (last_sector < ic->start || last_sector >= ic->meta_device_sectors)
3422 __u64 meta_size = (ic->provided_data_sectors >> ic->sb->log2_sectors_per_block) * ic->tag_size;
3423 meta_size = (meta_size + ((1U << (ic->log2_buffer_sectors + SECTOR_SHIFT)) - 1))
3424 >> (ic->log2_buffer_sectors + SECTOR_SHIFT);
3425 meta_size <<= ic->log2_buffer_sectors;
3426 if (ic->initial_sectors + meta_size < ic->initial_sectors ||
3427 ic->initial_sectors + meta_size > ic->meta_device_sectors)
3429 ic->metadata_run = 1;
3430 ic->log2_metadata_run = 0;
3436 static void get_provided_data_sectors(struct dm_integrity_c *ic)
3438 if (!ic->meta_dev) {
3440 ic->provided_data_sectors = 0;
3441 for (test_bit = fls64(ic->meta_device_sectors) - 1; test_bit >= 3; test_bit--) {
3442 __u64 prev_data_sectors = ic->provided_data_sectors;
3444 ic->provided_data_sectors |= (sector_t)1 << test_bit;
3445 if (calculate_device_limits(ic))
3446 ic->provided_data_sectors = prev_data_sectors;
3449 ic->provided_data_sectors = ic->data_device_sectors;
3450 ic->provided_data_sectors &= ~(sector_t)(ic->sectors_per_block - 1);
3454 static int initialize_superblock(struct dm_integrity_c *ic, unsigned journal_sectors, unsigned interleave_sectors)
3456 unsigned journal_sections;
3459 memset(ic->sb, 0, SB_SECTORS << SECTOR_SHIFT);
3460 memcpy(ic->sb->magic, SB_MAGIC, 8);
3461 ic->sb->integrity_tag_size = cpu_to_le16(ic->tag_size);
3462 ic->sb->log2_sectors_per_block = __ffs(ic->sectors_per_block);
3463 if (ic->journal_mac_alg.alg_string)
3464 ic->sb->flags |= cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC);
3466 calculate_journal_section_size(ic);
3467 journal_sections = journal_sectors / ic->journal_section_sectors;
3468 if (!journal_sections)
3469 journal_sections = 1;
3471 if (ic->fix_hmac && (ic->internal_hash_alg.alg_string || ic->journal_mac_alg.alg_string)) {
3472 ic->sb->flags |= cpu_to_le32(SB_FLAG_FIXED_HMAC);
3473 get_random_bytes(ic->sb->salt, SALT_SIZE);
3476 if (!ic->meta_dev) {
3477 if (ic->fix_padding)
3478 ic->sb->flags |= cpu_to_le32(SB_FLAG_FIXED_PADDING);
3479 ic->sb->journal_sections = cpu_to_le32(journal_sections);
3480 if (!interleave_sectors)
3481 interleave_sectors = DEFAULT_INTERLEAVE_SECTORS;
3482 ic->sb->log2_interleave_sectors = __fls(interleave_sectors);
3483 ic->sb->log2_interleave_sectors = max((__u8)MIN_LOG2_INTERLEAVE_SECTORS, ic->sb->log2_interleave_sectors);
3484 ic->sb->log2_interleave_sectors = min((__u8)MAX_LOG2_INTERLEAVE_SECTORS, ic->sb->log2_interleave_sectors);
3486 get_provided_data_sectors(ic);
3487 if (!ic->provided_data_sectors)
3490 ic->sb->log2_interleave_sectors = 0;
3492 get_provided_data_sectors(ic);
3493 if (!ic->provided_data_sectors)
3497 ic->sb->journal_sections = cpu_to_le32(0);
3498 for (test_bit = fls(journal_sections) - 1; test_bit >= 0; test_bit--) {
3499 __u32 prev_journal_sections = le32_to_cpu(ic->sb->journal_sections);
3500 __u32 test_journal_sections = prev_journal_sections | (1U << test_bit);
3501 if (test_journal_sections > journal_sections)
3503 ic->sb->journal_sections = cpu_to_le32(test_journal_sections);
3504 if (calculate_device_limits(ic))
3505 ic->sb->journal_sections = cpu_to_le32(prev_journal_sections);
3508 if (!le32_to_cpu(ic->sb->journal_sections)) {
3509 if (ic->log2_buffer_sectors > 3) {
3510 ic->log2_buffer_sectors--;
3511 goto try_smaller_buffer;
3517 ic->sb->provided_data_sectors = cpu_to_le64(ic->provided_data_sectors);
3524 static void dm_integrity_set(struct dm_target *ti, struct dm_integrity_c *ic)
3526 struct gendisk *disk = dm_disk(dm_table_get_md(ti->table));
3527 struct blk_integrity bi;
3529 memset(&bi, 0, sizeof(bi));
3530 bi.profile = &dm_integrity_profile;
3531 bi.tuple_size = ic->tag_size;
3532 bi.tag_size = bi.tuple_size;
3533 bi.interval_exp = ic->sb->log2_sectors_per_block + SECTOR_SHIFT;
3535 blk_integrity_register(disk, &bi);
3536 blk_queue_max_integrity_segments(disk->queue, UINT_MAX);
3539 static void dm_integrity_free_page_list(struct page_list *pl)
3545 for (i = 0; pl[i].page; i++)
3546 __free_page(pl[i].page);
3550 static struct page_list *dm_integrity_alloc_page_list(unsigned n_pages)
3552 struct page_list *pl;
3555 pl = kvmalloc_array(n_pages + 1, sizeof(struct page_list), GFP_KERNEL | __GFP_ZERO);
3559 for (i = 0; i < n_pages; i++) {
3560 pl[i].page = alloc_page(GFP_KERNEL);
3562 dm_integrity_free_page_list(pl);
3566 pl[i - 1].next = &pl[i];
3574 static void dm_integrity_free_journal_scatterlist(struct dm_integrity_c *ic, struct scatterlist **sl)
3577 for (i = 0; i < ic->journal_sections; i++)
3582 static struct scatterlist **dm_integrity_alloc_journal_scatterlist(struct dm_integrity_c *ic,
3583 struct page_list *pl)
3585 struct scatterlist **sl;
3588 sl = kvmalloc_array(ic->journal_sections,
3589 sizeof(struct scatterlist *),
3590 GFP_KERNEL | __GFP_ZERO);
3594 for (i = 0; i < ic->journal_sections; i++) {
3595 struct scatterlist *s;
3596 unsigned start_index, start_offset;
3597 unsigned end_index, end_offset;
3601 page_list_location(ic, i, 0, &start_index, &start_offset);
3602 page_list_location(ic, i, ic->journal_section_sectors - 1,
3603 &end_index, &end_offset);
3605 n_pages = (end_index - start_index + 1);
3607 s = kvmalloc_array(n_pages, sizeof(struct scatterlist),
3610 dm_integrity_free_journal_scatterlist(ic, sl);
3614 sg_init_table(s, n_pages);
3615 for (idx = start_index; idx <= end_index; idx++) {
3616 char *va = lowmem_page_address(pl[idx].page);
3617 unsigned start = 0, end = PAGE_SIZE;
3618 if (idx == start_index)
3619 start = start_offset;
3620 if (idx == end_index)
3621 end = end_offset + (1 << SECTOR_SHIFT);
3622 sg_set_buf(&s[idx - start_index], va + start, end - start);
3631 static void free_alg(struct alg_spec *a)
3633 kfree_sensitive(a->alg_string);
3634 kfree_sensitive(a->key);
3635 memset(a, 0, sizeof *a);
3638 static int get_alg_and_key(const char *arg, struct alg_spec *a, char **error, char *error_inval)
3644 a->alg_string = kstrdup(strchr(arg, ':') + 1, GFP_KERNEL);
3648 k = strchr(a->alg_string, ':');
3651 a->key_string = k + 1;
3652 if (strlen(a->key_string) & 1)
3655 a->key_size = strlen(a->key_string) / 2;
3656 a->key = kmalloc(a->key_size, GFP_KERNEL);
3659 if (hex2bin(a->key, a->key_string, a->key_size))
3665 *error = error_inval;
3668 *error = "Out of memory for an argument";
3672 static int get_mac(struct crypto_shash **hash, struct alg_spec *a, char **error,
3673 char *error_alg, char *error_key)
3677 if (a->alg_string) {
3678 *hash = crypto_alloc_shash(a->alg_string, 0, CRYPTO_ALG_ALLOCATES_MEMORY);
3679 if (IS_ERR(*hash)) {
3687 r = crypto_shash_setkey(*hash, a->key, a->key_size);
3692 } else if (crypto_shash_get_flags(*hash) & CRYPTO_TFM_NEED_KEY) {
3701 static int create_journal(struct dm_integrity_c *ic, char **error)
3705 __u64 journal_pages, journal_desc_size, journal_tree_size;
3706 unsigned char *crypt_data = NULL, *crypt_iv = NULL;
3707 struct skcipher_request *req = NULL;
3709 ic->commit_ids[0] = cpu_to_le64(0x1111111111111111ULL);
3710 ic->commit_ids[1] = cpu_to_le64(0x2222222222222222ULL);
3711 ic->commit_ids[2] = cpu_to_le64(0x3333333333333333ULL);
3712 ic->commit_ids[3] = cpu_to_le64(0x4444444444444444ULL);
3714 journal_pages = roundup((__u64)ic->journal_sections * ic->journal_section_sectors,
3715 PAGE_SIZE >> SECTOR_SHIFT) >> (PAGE_SHIFT - SECTOR_SHIFT);
3716 journal_desc_size = journal_pages * sizeof(struct page_list);
3717 if (journal_pages >= totalram_pages() - totalhigh_pages() || journal_desc_size > ULONG_MAX) {
3718 *error = "Journal doesn't fit into memory";
3722 ic->journal_pages = journal_pages;
3724 ic->journal = dm_integrity_alloc_page_list(ic->journal_pages);
3726 *error = "Could not allocate memory for journal";
3730 if (ic->journal_crypt_alg.alg_string) {
3731 unsigned ivsize, blocksize;
3732 struct journal_completion comp;
3735 ic->journal_crypt = crypto_alloc_skcipher(ic->journal_crypt_alg.alg_string, 0, CRYPTO_ALG_ALLOCATES_MEMORY);
3736 if (IS_ERR(ic->journal_crypt)) {
3737 *error = "Invalid journal cipher";
3738 r = PTR_ERR(ic->journal_crypt);
3739 ic->journal_crypt = NULL;
3742 ivsize = crypto_skcipher_ivsize(ic->journal_crypt);
3743 blocksize = crypto_skcipher_blocksize(ic->journal_crypt);
3745 if (ic->journal_crypt_alg.key) {
3746 r = crypto_skcipher_setkey(ic->journal_crypt, ic->journal_crypt_alg.key,
3747 ic->journal_crypt_alg.key_size);
3749 *error = "Error setting encryption key";
3753 DEBUG_print("cipher %s, block size %u iv size %u\n",
3754 ic->journal_crypt_alg.alg_string, blocksize, ivsize);
3756 ic->journal_io = dm_integrity_alloc_page_list(ic->journal_pages);
3757 if (!ic->journal_io) {
3758 *error = "Could not allocate memory for journal io";
3763 if (blocksize == 1) {
3764 struct scatterlist *sg;
3766 req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
3768 *error = "Could not allocate crypt request";
3773 crypt_iv = kzalloc(ivsize, GFP_KERNEL);
3775 *error = "Could not allocate iv";
3780 ic->journal_xor = dm_integrity_alloc_page_list(ic->journal_pages);
3781 if (!ic->journal_xor) {
3782 *error = "Could not allocate memory for journal xor";
3787 sg = kvmalloc_array(ic->journal_pages + 1,
3788 sizeof(struct scatterlist),
3791 *error = "Unable to allocate sg list";
3795 sg_init_table(sg, ic->journal_pages + 1);
3796 for (i = 0; i < ic->journal_pages; i++) {
3797 char *va = lowmem_page_address(ic->journal_xor[i].page);
3799 sg_set_buf(&sg[i], va, PAGE_SIZE);
3801 sg_set_buf(&sg[i], &ic->commit_ids, sizeof ic->commit_ids);
3803 skcipher_request_set_crypt(req, sg, sg,
3804 PAGE_SIZE * ic->journal_pages + sizeof ic->commit_ids, crypt_iv);
3805 init_completion(&comp.comp);
3806 comp.in_flight = (atomic_t)ATOMIC_INIT(1);
3807 if (do_crypt(true, req, &comp))
3808 wait_for_completion(&comp.comp);
3810 r = dm_integrity_failed(ic);
3812 *error = "Unable to encrypt journal";
3815 DEBUG_bytes(lowmem_page_address(ic->journal_xor[0].page), 64, "xor data");
3817 crypto_free_skcipher(ic->journal_crypt);
3818 ic->journal_crypt = NULL;
3820 unsigned crypt_len = roundup(ivsize, blocksize);
3822 req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
3824 *error = "Could not allocate crypt request";
3829 crypt_iv = kmalloc(ivsize, GFP_KERNEL);
3831 *error = "Could not allocate iv";
3836 crypt_data = kmalloc(crypt_len, GFP_KERNEL);
3838 *error = "Unable to allocate crypt data";
3843 ic->journal_scatterlist = dm_integrity_alloc_journal_scatterlist(ic, ic->journal);
3844 if (!ic->journal_scatterlist) {
3845 *error = "Unable to allocate sg list";
3849 ic->journal_io_scatterlist = dm_integrity_alloc_journal_scatterlist(ic, ic->journal_io);
3850 if (!ic->journal_io_scatterlist) {
3851 *error = "Unable to allocate sg list";
3855 ic->sk_requests = kvmalloc_array(ic->journal_sections,
3856 sizeof(struct skcipher_request *),
3857 GFP_KERNEL | __GFP_ZERO);
3858 if (!ic->sk_requests) {
3859 *error = "Unable to allocate sk requests";
3863 for (i = 0; i < ic->journal_sections; i++) {
3864 struct scatterlist sg;
3865 struct skcipher_request *section_req;
3866 __le32 section_le = cpu_to_le32(i);
3868 memset(crypt_iv, 0x00, ivsize);
3869 memset(crypt_data, 0x00, crypt_len);
3870 memcpy(crypt_data, §ion_le, min((size_t)crypt_len, sizeof(section_le)));
3872 sg_init_one(&sg, crypt_data, crypt_len);
3873 skcipher_request_set_crypt(req, &sg, &sg, crypt_len, crypt_iv);
3874 init_completion(&comp.comp);
3875 comp.in_flight = (atomic_t)ATOMIC_INIT(1);
3876 if (do_crypt(true, req, &comp))
3877 wait_for_completion(&comp.comp);
3879 r = dm_integrity_failed(ic);
3881 *error = "Unable to generate iv";
3885 section_req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
3887 *error = "Unable to allocate crypt request";
3891 section_req->iv = kmalloc_array(ivsize, 2,
3893 if (!section_req->iv) {
3894 skcipher_request_free(section_req);
3895 *error = "Unable to allocate iv";
3899 memcpy(section_req->iv + ivsize, crypt_data, ivsize);
3900 section_req->cryptlen = (size_t)ic->journal_section_sectors << SECTOR_SHIFT;
3901 ic->sk_requests[i] = section_req;
3902 DEBUG_bytes(crypt_data, ivsize, "iv(%u)", i);
3907 for (i = 0; i < N_COMMIT_IDS; i++) {
3910 for (j = 0; j < i; j++) {
3911 if (ic->commit_ids[j] == ic->commit_ids[i]) {
3912 ic->commit_ids[i] = cpu_to_le64(le64_to_cpu(ic->commit_ids[i]) + 1);
3913 goto retest_commit_id;
3916 DEBUG_print("commit id %u: %016llx\n", i, ic->commit_ids[i]);
3919 journal_tree_size = (__u64)ic->journal_entries * sizeof(struct journal_node);
3920 if (journal_tree_size > ULONG_MAX) {
3921 *error = "Journal doesn't fit into memory";
3925 ic->journal_tree = kvmalloc(journal_tree_size, GFP_KERNEL);
3926 if (!ic->journal_tree) {
3927 *error = "Could not allocate memory for journal tree";
3933 skcipher_request_free(req);
3939 * Construct a integrity mapping
3943 * offset from the start of the device
3945 * D - direct writes, J - journal writes, B - bitmap mode, R - recovery mode
3946 * number of optional arguments
3947 * optional arguments:
3949 * interleave_sectors
3956 * bitmap_flush_interval
3962 static int dm_integrity_ctr(struct dm_target *ti, unsigned argc, char **argv)
3964 struct dm_integrity_c *ic;
3967 unsigned extra_args;
3968 struct dm_arg_set as;
3969 static const struct dm_arg _args[] = {
3970 {0, 18, "Invalid number of feature args"},
3972 unsigned journal_sectors, interleave_sectors, buffer_sectors, journal_watermark, sync_msec;
3973 bool should_write_sb;
3975 unsigned long long start;
3976 __s8 log2_sectors_per_bitmap_bit = -1;
3977 __s8 log2_blocks_per_bitmap_bit;
3978 __u64 bits_in_journal;
3979 __u64 n_bitmap_bits;
3981 #define DIRECT_ARGUMENTS 4
3983 if (argc <= DIRECT_ARGUMENTS) {
3984 ti->error = "Invalid argument count";
3988 ic = kzalloc(sizeof(struct dm_integrity_c), GFP_KERNEL);
3990 ti->error = "Cannot allocate integrity context";
3994 ti->per_io_data_size = sizeof(struct dm_integrity_io);
3997 ic->in_progress = RB_ROOT;
3998 INIT_LIST_HEAD(&ic->wait_list);
3999 init_waitqueue_head(&ic->endio_wait);
4000 bio_list_init(&ic->flush_bio_list);
4001 init_waitqueue_head(&ic->copy_to_journal_wait);
4002 init_completion(&ic->crypto_backoff);
4003 atomic64_set(&ic->number_of_mismatches, 0);
4004 ic->bitmap_flush_interval = BITMAP_FLUSH_INTERVAL;
4006 r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &ic->dev);
4008 ti->error = "Device lookup failed";
4012 if (sscanf(argv[1], "%llu%c", &start, &dummy) != 1 || start != (sector_t)start) {
4013 ti->error = "Invalid starting offset";
4019 if (strcmp(argv[2], "-")) {
4020 if (sscanf(argv[2], "%u%c", &ic->tag_size, &dummy) != 1 || !ic->tag_size) {
4021 ti->error = "Invalid tag size";
4027 if (!strcmp(argv[3], "J") || !strcmp(argv[3], "B") ||
4028 !strcmp(argv[3], "D") || !strcmp(argv[3], "R")) {
4029 ic->mode = argv[3][0];
4031 ti->error = "Invalid mode (expecting J, B, D, R)";
4036 journal_sectors = 0;
4037 interleave_sectors = DEFAULT_INTERLEAVE_SECTORS;
4038 buffer_sectors = DEFAULT_BUFFER_SECTORS;
4039 journal_watermark = DEFAULT_JOURNAL_WATERMARK;
4040 sync_msec = DEFAULT_SYNC_MSEC;
4041 ic->sectors_per_block = 1;
4043 as.argc = argc - DIRECT_ARGUMENTS;
4044 as.argv = argv + DIRECT_ARGUMENTS;
4045 r = dm_read_arg_group(_args, &as, &extra_args, &ti->error);
4049 while (extra_args--) {
4050 const char *opt_string;
4052 unsigned long long llval;
4053 opt_string = dm_shift_arg(&as);
4056 ti->error = "Not enough feature arguments";
4059 if (sscanf(opt_string, "journal_sectors:%u%c", &val, &dummy) == 1)
4060 journal_sectors = val ? val : 1;
4061 else if (sscanf(opt_string, "interleave_sectors:%u%c", &val, &dummy) == 1)
4062 interleave_sectors = val;
4063 else if (sscanf(opt_string, "buffer_sectors:%u%c", &val, &dummy) == 1)
4064 buffer_sectors = val;
4065 else if (sscanf(opt_string, "journal_watermark:%u%c", &val, &dummy) == 1 && val <= 100)
4066 journal_watermark = val;
4067 else if (sscanf(opt_string, "commit_time:%u%c", &val, &dummy) == 1)
4069 else if (!strncmp(opt_string, "meta_device:", strlen("meta_device:"))) {
4071 dm_put_device(ti, ic->meta_dev);
4072 ic->meta_dev = NULL;
4074 r = dm_get_device(ti, strchr(opt_string, ':') + 1,
4075 dm_table_get_mode(ti->table), &ic->meta_dev);
4077 ti->error = "Device lookup failed";
4080 } else if (sscanf(opt_string, "block_size:%u%c", &val, &dummy) == 1) {
4081 if (val < 1 << SECTOR_SHIFT ||
4082 val > MAX_SECTORS_PER_BLOCK << SECTOR_SHIFT ||
4085 ti->error = "Invalid block_size argument";
4088 ic->sectors_per_block = val >> SECTOR_SHIFT;
4089 } else if (sscanf(opt_string, "sectors_per_bit:%llu%c", &llval, &dummy) == 1) {
4090 log2_sectors_per_bitmap_bit = !llval ? 0 : __ilog2_u64(llval);
4091 } else if (sscanf(opt_string, "bitmap_flush_interval:%u%c", &val, &dummy) == 1) {
4092 if (val >= (uint64_t)UINT_MAX * 1000 / HZ) {
4094 ti->error = "Invalid bitmap_flush_interval argument";
4097 ic->bitmap_flush_interval = msecs_to_jiffies(val);
4098 } else if (!strncmp(opt_string, "internal_hash:", strlen("internal_hash:"))) {
4099 r = get_alg_and_key(opt_string, &ic->internal_hash_alg, &ti->error,
4100 "Invalid internal_hash argument");
4103 } else if (!strncmp(opt_string, "journal_crypt:", strlen("journal_crypt:"))) {
4104 r = get_alg_and_key(opt_string, &ic->journal_crypt_alg, &ti->error,
4105 "Invalid journal_crypt argument");
4108 } else if (!strncmp(opt_string, "journal_mac:", strlen("journal_mac:"))) {
4109 r = get_alg_and_key(opt_string, &ic->journal_mac_alg, &ti->error,
4110 "Invalid journal_mac argument");
4113 } else if (!strcmp(opt_string, "recalculate")) {
4114 ic->recalculate_flag = true;
4115 } else if (!strcmp(opt_string, "reset_recalculate")) {
4116 ic->recalculate_flag = true;
4117 ic->reset_recalculate_flag = true;
4118 } else if (!strcmp(opt_string, "allow_discards")) {
4120 } else if (!strcmp(opt_string, "fix_padding")) {
4121 ic->fix_padding = true;
4122 } else if (!strcmp(opt_string, "fix_hmac")) {
4123 ic->fix_hmac = true;
4124 } else if (!strcmp(opt_string, "legacy_recalculate")) {
4125 ic->legacy_recalculate = true;
4128 ti->error = "Invalid argument";
4133 ic->data_device_sectors = bdev_nr_sectors(ic->dev->bdev);
4135 ic->meta_device_sectors = ic->data_device_sectors;
4137 ic->meta_device_sectors = bdev_nr_sectors(ic->meta_dev->bdev);
4139 if (!journal_sectors) {
4140 journal_sectors = min((sector_t)DEFAULT_MAX_JOURNAL_SECTORS,
4141 ic->data_device_sectors >> DEFAULT_JOURNAL_SIZE_FACTOR);
4144 if (!buffer_sectors)
4146 ic->log2_buffer_sectors = min((int)__fls(buffer_sectors), 31 - SECTOR_SHIFT);
4148 r = get_mac(&ic->internal_hash, &ic->internal_hash_alg, &ti->error,
4149 "Invalid internal hash", "Error setting internal hash key");
4153 r = get_mac(&ic->journal_mac, &ic->journal_mac_alg, &ti->error,
4154 "Invalid journal mac", "Error setting journal mac key");
4158 if (!ic->tag_size) {
4159 if (!ic->internal_hash) {
4160 ti->error = "Unknown tag size";
4164 ic->tag_size = crypto_shash_digestsize(ic->internal_hash);
4166 if (ic->tag_size > MAX_TAG_SIZE) {
4167 ti->error = "Too big tag size";
4171 if (!(ic->tag_size & (ic->tag_size - 1)))
4172 ic->log2_tag_size = __ffs(ic->tag_size);
4174 ic->log2_tag_size = -1;
4176 if (ic->mode == 'B' && !ic->internal_hash) {
4178 ti->error = "Bitmap mode can be only used with internal hash";
4182 if (ic->discard && !ic->internal_hash) {
4184 ti->error = "Discard can be only used with internal hash";
4188 ic->autocommit_jiffies = msecs_to_jiffies(sync_msec);
4189 ic->autocommit_msec = sync_msec;
4190 timer_setup(&ic->autocommit_timer, autocommit_fn, 0);
4192 ic->io = dm_io_client_create();
4193 if (IS_ERR(ic->io)) {
4194 r = PTR_ERR(ic->io);
4196 ti->error = "Cannot allocate dm io";
4200 r = mempool_init_slab_pool(&ic->journal_io_mempool, JOURNAL_IO_MEMPOOL, journal_io_cache);
4202 ti->error = "Cannot allocate mempool";
4206 ic->metadata_wq = alloc_workqueue("dm-integrity-metadata",
4207 WQ_MEM_RECLAIM, METADATA_WORKQUEUE_MAX_ACTIVE);
4208 if (!ic->metadata_wq) {
4209 ti->error = "Cannot allocate workqueue";
4215 * If this workqueue were percpu, it would cause bio reordering
4216 * and reduced performance.
4218 ic->wait_wq = alloc_workqueue("dm-integrity-wait", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
4220 ti->error = "Cannot allocate workqueue";
4225 ic->offload_wq = alloc_workqueue("dm-integrity-offload", WQ_MEM_RECLAIM,
4226 METADATA_WORKQUEUE_MAX_ACTIVE);
4227 if (!ic->offload_wq) {
4228 ti->error = "Cannot allocate workqueue";
4233 ic->commit_wq = alloc_workqueue("dm-integrity-commit", WQ_MEM_RECLAIM, 1);
4234 if (!ic->commit_wq) {
4235 ti->error = "Cannot allocate workqueue";
4239 INIT_WORK(&ic->commit_work, integrity_commit);
4241 if (ic->mode == 'J' || ic->mode == 'B') {
4242 ic->writer_wq = alloc_workqueue("dm-integrity-writer", WQ_MEM_RECLAIM, 1);
4243 if (!ic->writer_wq) {
4244 ti->error = "Cannot allocate workqueue";
4248 INIT_WORK(&ic->writer_work, integrity_writer);
4251 ic->sb = alloc_pages_exact(SB_SECTORS << SECTOR_SHIFT, GFP_KERNEL);
4254 ti->error = "Cannot allocate superblock area";
4258 r = sync_rw_sb(ic, REQ_OP_READ, 0);
4260 ti->error = "Error reading superblock";
4263 should_write_sb = false;
4264 if (memcmp(ic->sb->magic, SB_MAGIC, 8)) {
4265 if (ic->mode != 'R') {
4266 if (memchr_inv(ic->sb, 0, SB_SECTORS << SECTOR_SHIFT)) {
4268 ti->error = "The device is not initialized";
4273 r = initialize_superblock(ic, journal_sectors, interleave_sectors);
4275 ti->error = "Could not initialize superblock";
4278 if (ic->mode != 'R')
4279 should_write_sb = true;
4282 if (!ic->sb->version || ic->sb->version > SB_VERSION_5) {
4284 ti->error = "Unknown version";
4287 if (le16_to_cpu(ic->sb->integrity_tag_size) != ic->tag_size) {
4289 ti->error = "Tag size doesn't match the information in superblock";
4292 if (ic->sb->log2_sectors_per_block != __ffs(ic->sectors_per_block)) {
4294 ti->error = "Block size doesn't match the information in superblock";
4297 if (!le32_to_cpu(ic->sb->journal_sections)) {
4299 ti->error = "Corrupted superblock, journal_sections is 0";
4302 /* make sure that ti->max_io_len doesn't overflow */
4303 if (!ic->meta_dev) {
4304 if (ic->sb->log2_interleave_sectors < MIN_LOG2_INTERLEAVE_SECTORS ||
4305 ic->sb->log2_interleave_sectors > MAX_LOG2_INTERLEAVE_SECTORS) {
4307 ti->error = "Invalid interleave_sectors in the superblock";
4311 if (ic->sb->log2_interleave_sectors) {
4313 ti->error = "Invalid interleave_sectors in the superblock";
4317 if (!!(ic->sb->flags & cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC)) != !!ic->journal_mac_alg.alg_string) {
4319 ti->error = "Journal mac mismatch";
4323 get_provided_data_sectors(ic);
4324 if (!ic->provided_data_sectors) {
4326 ti->error = "The device is too small";
4331 r = calculate_device_limits(ic);
4334 if (ic->log2_buffer_sectors > 3) {
4335 ic->log2_buffer_sectors--;
4336 goto try_smaller_buffer;
4339 ti->error = "The device is too small";
4343 if (log2_sectors_per_bitmap_bit < 0)
4344 log2_sectors_per_bitmap_bit = __fls(DEFAULT_SECTORS_PER_BITMAP_BIT);
4345 if (log2_sectors_per_bitmap_bit < ic->sb->log2_sectors_per_block)
4346 log2_sectors_per_bitmap_bit = ic->sb->log2_sectors_per_block;
4348 bits_in_journal = ((__u64)ic->journal_section_sectors * ic->journal_sections) << (SECTOR_SHIFT + 3);
4349 if (bits_in_journal > UINT_MAX)
4350 bits_in_journal = UINT_MAX;
4351 while (bits_in_journal < (ic->provided_data_sectors + ((sector_t)1 << log2_sectors_per_bitmap_bit) - 1) >> log2_sectors_per_bitmap_bit)
4352 log2_sectors_per_bitmap_bit++;
4354 log2_blocks_per_bitmap_bit = log2_sectors_per_bitmap_bit - ic->sb->log2_sectors_per_block;
4355 ic->log2_blocks_per_bitmap_bit = log2_blocks_per_bitmap_bit;
4356 if (should_write_sb) {
4357 ic->sb->log2_blocks_per_bitmap_bit = log2_blocks_per_bitmap_bit;
4359 n_bitmap_bits = ((ic->provided_data_sectors >> ic->sb->log2_sectors_per_block)
4360 + (((sector_t)1 << log2_blocks_per_bitmap_bit) - 1)) >> log2_blocks_per_bitmap_bit;
4361 ic->n_bitmap_blocks = DIV_ROUND_UP(n_bitmap_bits, BITMAP_BLOCK_SIZE * 8);
4364 ic->log2_buffer_sectors = min(ic->log2_buffer_sectors, (__u8)__ffs(ic->metadata_run));
4366 if (ti->len > ic->provided_data_sectors) {
4368 ti->error = "Not enough provided sectors for requested mapping size";
4373 threshold = (__u64)ic->journal_entries * (100 - journal_watermark);
4375 do_div(threshold, 100);
4376 ic->free_sectors_threshold = threshold;
4378 DEBUG_print("initialized:\n");
4379 DEBUG_print(" integrity_tag_size %u\n", le16_to_cpu(ic->sb->integrity_tag_size));
4380 DEBUG_print(" journal_entry_size %u\n", ic->journal_entry_size);
4381 DEBUG_print(" journal_entries_per_sector %u\n", ic->journal_entries_per_sector);
4382 DEBUG_print(" journal_section_entries %u\n", ic->journal_section_entries);
4383 DEBUG_print(" journal_section_sectors %u\n", ic->journal_section_sectors);
4384 DEBUG_print(" journal_sections %u\n", (unsigned)le32_to_cpu(ic->sb->journal_sections));
4385 DEBUG_print(" journal_entries %u\n", ic->journal_entries);
4386 DEBUG_print(" log2_interleave_sectors %d\n", ic->sb->log2_interleave_sectors);
4387 DEBUG_print(" data_device_sectors 0x%llx\n", bdev_nr_sectors(ic->dev->bdev));
4388 DEBUG_print(" initial_sectors 0x%x\n", ic->initial_sectors);
4389 DEBUG_print(" metadata_run 0x%x\n", ic->metadata_run);
4390 DEBUG_print(" log2_metadata_run %d\n", ic->log2_metadata_run);
4391 DEBUG_print(" provided_data_sectors 0x%llx (%llu)\n", ic->provided_data_sectors, ic->provided_data_sectors);
4392 DEBUG_print(" log2_buffer_sectors %u\n", ic->log2_buffer_sectors);
4393 DEBUG_print(" bits_in_journal %llu\n", bits_in_journal);
4395 if (ic->recalculate_flag && !(ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))) {
4396 ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
4397 ic->sb->recalc_sector = cpu_to_le64(0);
4400 if (ic->internal_hash) {
4401 ic->recalc_wq = alloc_workqueue("dm-integrity-recalc", WQ_MEM_RECLAIM, 1);
4402 if (!ic->recalc_wq ) {
4403 ti->error = "Cannot allocate workqueue";
4407 INIT_WORK(&ic->recalc_work, integrity_recalc);
4408 ic->recalc_buffer = vmalloc(RECALC_SECTORS << SECTOR_SHIFT);
4409 if (!ic->recalc_buffer) {
4410 ti->error = "Cannot allocate buffer for recalculating";
4414 ic->recalc_tags = kvmalloc_array(RECALC_SECTORS >> ic->sb->log2_sectors_per_block,
4415 ic->tag_size, GFP_KERNEL);
4416 if (!ic->recalc_tags) {
4417 ti->error = "Cannot allocate tags for recalculating";
4422 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
4423 ti->error = "Recalculate can only be specified with internal_hash";
4429 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING) &&
4430 le64_to_cpu(ic->sb->recalc_sector) < ic->provided_data_sectors &&
4431 dm_integrity_disable_recalculate(ic)) {
4432 ti->error = "Recalculating with HMAC is disabled for security reasons - if you really need it, use the argument \"legacy_recalculate\"";
4437 ic->bufio = dm_bufio_client_create(ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev,
4438 1U << (SECTOR_SHIFT + ic->log2_buffer_sectors), 1, 0, NULL, NULL);
4439 if (IS_ERR(ic->bufio)) {
4440 r = PTR_ERR(ic->bufio);
4441 ti->error = "Cannot initialize dm-bufio";
4445 dm_bufio_set_sector_offset(ic->bufio, ic->start + ic->initial_sectors);
4447 if (ic->mode != 'R') {
4448 r = create_journal(ic, &ti->error);
4454 if (ic->mode == 'B') {
4456 unsigned n_bitmap_pages = DIV_ROUND_UP(ic->n_bitmap_blocks, PAGE_SIZE / BITMAP_BLOCK_SIZE);
4458 ic->recalc_bitmap = dm_integrity_alloc_page_list(n_bitmap_pages);
4459 if (!ic->recalc_bitmap) {
4463 ic->may_write_bitmap = dm_integrity_alloc_page_list(n_bitmap_pages);
4464 if (!ic->may_write_bitmap) {
4468 ic->bbs = kvmalloc_array(ic->n_bitmap_blocks, sizeof(struct bitmap_block_status), GFP_KERNEL);
4473 INIT_DELAYED_WORK(&ic->bitmap_flush_work, bitmap_flush_work);
4474 for (i = 0; i < ic->n_bitmap_blocks; i++) {
4475 struct bitmap_block_status *bbs = &ic->bbs[i];
4476 unsigned sector, pl_index, pl_offset;
4478 INIT_WORK(&bbs->work, bitmap_block_work);
4481 bio_list_init(&bbs->bio_queue);
4482 spin_lock_init(&bbs->bio_queue_lock);
4484 sector = i * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT);
4485 pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
4486 pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
4488 bbs->bitmap = lowmem_page_address(ic->journal[pl_index].page) + pl_offset;
4492 if (should_write_sb) {
4495 init_journal(ic, 0, ic->journal_sections, 0);
4496 r = dm_integrity_failed(ic);
4498 ti->error = "Error initializing journal";
4501 r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
4503 ti->error = "Error initializing superblock";
4506 ic->just_formatted = true;
4509 if (!ic->meta_dev) {
4510 r = dm_set_target_max_io_len(ti, 1U << ic->sb->log2_interleave_sectors);
4514 if (ic->mode == 'B') {
4515 unsigned max_io_len = ((sector_t)ic->sectors_per_block << ic->log2_blocks_per_bitmap_bit) * (BITMAP_BLOCK_SIZE * 8);
4517 max_io_len = 1U << 31;
4518 DEBUG_print("max_io_len: old %u, new %u\n", ti->max_io_len, max_io_len);
4519 if (!ti->max_io_len || ti->max_io_len > max_io_len) {
4520 r = dm_set_target_max_io_len(ti, max_io_len);
4526 if (!ic->internal_hash)
4527 dm_integrity_set(ti, ic);
4529 ti->num_flush_bios = 1;
4530 ti->flush_supported = true;
4532 ti->num_discard_bios = 1;
4534 dm_audit_log_ctr(DM_MSG_PREFIX, ti, 1);
4538 dm_audit_log_ctr(DM_MSG_PREFIX, ti, 0);
4539 dm_integrity_dtr(ti);
4543 static void dm_integrity_dtr(struct dm_target *ti)
4545 struct dm_integrity_c *ic = ti->private;
4547 BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
4548 BUG_ON(!list_empty(&ic->wait_list));
4550 if (ic->metadata_wq)
4551 destroy_workqueue(ic->metadata_wq);
4553 destroy_workqueue(ic->wait_wq);
4555 destroy_workqueue(ic->offload_wq);
4557 destroy_workqueue(ic->commit_wq);
4559 destroy_workqueue(ic->writer_wq);
4561 destroy_workqueue(ic->recalc_wq);
4562 vfree(ic->recalc_buffer);
4563 kvfree(ic->recalc_tags);
4566 dm_bufio_client_destroy(ic->bufio);
4567 mempool_exit(&ic->journal_io_mempool);
4569 dm_io_client_destroy(ic->io);
4571 dm_put_device(ti, ic->dev);
4573 dm_put_device(ti, ic->meta_dev);
4574 dm_integrity_free_page_list(ic->journal);
4575 dm_integrity_free_page_list(ic->journal_io);
4576 dm_integrity_free_page_list(ic->journal_xor);
4577 dm_integrity_free_page_list(ic->recalc_bitmap);
4578 dm_integrity_free_page_list(ic->may_write_bitmap);
4579 if (ic->journal_scatterlist)
4580 dm_integrity_free_journal_scatterlist(ic, ic->journal_scatterlist);
4581 if (ic->journal_io_scatterlist)
4582 dm_integrity_free_journal_scatterlist(ic, ic->journal_io_scatterlist);
4583 if (ic->sk_requests) {
4586 for (i = 0; i < ic->journal_sections; i++) {
4587 struct skcipher_request *req = ic->sk_requests[i];
4589 kfree_sensitive(req->iv);
4590 skcipher_request_free(req);
4593 kvfree(ic->sk_requests);
4595 kvfree(ic->journal_tree);
4597 free_pages_exact(ic->sb, SB_SECTORS << SECTOR_SHIFT);
4599 if (ic->internal_hash)
4600 crypto_free_shash(ic->internal_hash);
4601 free_alg(&ic->internal_hash_alg);
4603 if (ic->journal_crypt)
4604 crypto_free_skcipher(ic->journal_crypt);
4605 free_alg(&ic->journal_crypt_alg);
4607 if (ic->journal_mac)
4608 crypto_free_shash(ic->journal_mac);
4609 free_alg(&ic->journal_mac_alg);
4612 dm_audit_log_dtr(DM_MSG_PREFIX, ti, 1);
4615 static struct target_type integrity_target = {
4616 .name = "integrity",
4617 .version = {1, 10, 0},
4618 .module = THIS_MODULE,
4619 .features = DM_TARGET_SINGLETON | DM_TARGET_INTEGRITY,
4620 .ctr = dm_integrity_ctr,
4621 .dtr = dm_integrity_dtr,
4622 .map = dm_integrity_map,
4623 .postsuspend = dm_integrity_postsuspend,
4624 .resume = dm_integrity_resume,
4625 .status = dm_integrity_status,
4626 .iterate_devices = dm_integrity_iterate_devices,
4627 .io_hints = dm_integrity_io_hints,
4630 static int __init dm_integrity_init(void)
4634 journal_io_cache = kmem_cache_create("integrity_journal_io",
4635 sizeof(struct journal_io), 0, 0, NULL);
4636 if (!journal_io_cache) {
4637 DMERR("can't allocate journal io cache");
4641 r = dm_register_target(&integrity_target);
4644 DMERR("register failed %d", r);
4649 static void __exit dm_integrity_exit(void)
4651 dm_unregister_target(&integrity_target);
4652 kmem_cache_destroy(journal_io_cache);
4655 module_init(dm_integrity_init);
4656 module_exit(dm_integrity_exit);
4658 MODULE_AUTHOR("Milan Broz");
4659 MODULE_AUTHOR("Mikulas Patocka");
4660 MODULE_DESCRIPTION(DM_NAME " target for integrity tags extension");
4661 MODULE_LICENSE("GPL");