2 * Copyright (C) 2016-2017 Red Hat, Inc. All rights reserved.
3 * Copyright (C) 2016-2017 Milan Broz
4 * Copyright (C) 2016-2017 Mikulas Patocka
6 * This file is released under the GPL.
9 #include "dm-bio-record.h"
11 #include <linux/compiler.h>
12 #include <linux/module.h>
13 #include <linux/device-mapper.h>
14 #include <linux/dm-io.h>
15 #include <linux/vmalloc.h>
16 #include <linux/sort.h>
17 #include <linux/rbtree.h>
18 #include <linux/delay.h>
19 #include <linux/random.h>
20 #include <linux/reboot.h>
21 #include <crypto/hash.h>
22 #include <crypto/skcipher.h>
23 #include <linux/async_tx.h>
24 #include <linux/dm-bufio.h>
26 #define DM_MSG_PREFIX "integrity"
28 #define DEFAULT_INTERLEAVE_SECTORS 32768
29 #define DEFAULT_JOURNAL_SIZE_FACTOR 7
30 #define DEFAULT_SECTORS_PER_BITMAP_BIT 32768
31 #define DEFAULT_BUFFER_SECTORS 128
32 #define DEFAULT_JOURNAL_WATERMARK 50
33 #define DEFAULT_SYNC_MSEC 10000
34 #define DEFAULT_MAX_JOURNAL_SECTORS 131072
35 #define MIN_LOG2_INTERLEAVE_SECTORS 3
36 #define MAX_LOG2_INTERLEAVE_SECTORS 31
37 #define METADATA_WORKQUEUE_MAX_ACTIVE 16
38 #define RECALC_SECTORS 32768
39 #define RECALC_WRITE_SUPER 16
40 #define BITMAP_BLOCK_SIZE 4096 /* don't change it */
41 #define BITMAP_FLUSH_INTERVAL (10 * HZ)
42 #define DISCARD_FILLER 0xf6
46 * Warning - DEBUG_PRINT prints security-sensitive data to the log,
47 * so it should not be enabled in the official kernel
50 //#define INTERNAL_VERIFY
56 #define SB_MAGIC "integrt"
57 #define SB_VERSION_1 1
58 #define SB_VERSION_2 2
59 #define SB_VERSION_3 3
60 #define SB_VERSION_4 4
61 #define SB_VERSION_5 5
63 #define MAX_SECTORS_PER_BLOCK 8
68 __u8 log2_interleave_sectors;
69 __le16 integrity_tag_size;
70 __le32 journal_sections;
71 __le64 provided_data_sectors; /* userspace uses this value */
73 __u8 log2_sectors_per_block;
74 __u8 log2_blocks_per_bitmap_bit;
81 #define SB_FLAG_HAVE_JOURNAL_MAC 0x1
82 #define SB_FLAG_RECALCULATING 0x2
83 #define SB_FLAG_DIRTY_BITMAP 0x4
84 #define SB_FLAG_FIXED_PADDING 0x8
85 #define SB_FLAG_FIXED_HMAC 0x10
87 #define JOURNAL_ENTRY_ROUNDUP 8
89 typedef __le64 commit_id_t;
90 #define JOURNAL_MAC_PER_SECTOR 8
92 struct journal_entry {
100 commit_id_t last_bytes[];
104 #define journal_entry_tag(ic, je) ((__u8 *)&(je)->last_bytes[(ic)->sectors_per_block])
106 #if BITS_PER_LONG == 64
107 #define journal_entry_set_sector(je, x) do { smp_wmb(); WRITE_ONCE((je)->u.sector, cpu_to_le64(x)); } while (0)
109 #define journal_entry_set_sector(je, x) do { (je)->u.s.sector_lo = cpu_to_le32(x); smp_wmb(); WRITE_ONCE((je)->u.s.sector_hi, cpu_to_le32((x) >> 32)); } while (0)
111 #define journal_entry_get_sector(je) le64_to_cpu((je)->u.sector)
112 #define journal_entry_is_unused(je) ((je)->u.s.sector_hi == cpu_to_le32(-1))
113 #define journal_entry_set_unused(je) do { ((je)->u.s.sector_hi = cpu_to_le32(-1)); } while (0)
114 #define journal_entry_is_inprogress(je) ((je)->u.s.sector_hi == cpu_to_le32(-2))
115 #define journal_entry_set_inprogress(je) do { ((je)->u.s.sector_hi = cpu_to_le32(-2)); } while (0)
117 #define JOURNAL_BLOCK_SECTORS 8
118 #define JOURNAL_SECTOR_DATA ((1 << SECTOR_SHIFT) - sizeof(commit_id_t))
119 #define JOURNAL_MAC_SIZE (JOURNAL_MAC_PER_SECTOR * JOURNAL_BLOCK_SECTORS)
121 struct journal_sector {
122 __u8 entries[JOURNAL_SECTOR_DATA - JOURNAL_MAC_PER_SECTOR];
123 __u8 mac[JOURNAL_MAC_PER_SECTOR];
124 commit_id_t commit_id;
127 #define MAX_TAG_SIZE (JOURNAL_SECTOR_DATA - JOURNAL_MAC_PER_SECTOR - offsetof(struct journal_entry, last_bytes[MAX_SECTORS_PER_BLOCK]))
129 #define METADATA_PADDING_SECTORS 8
131 #define N_COMMIT_IDS 4
133 static unsigned char prev_commit_seq(unsigned char seq)
135 return (seq + N_COMMIT_IDS - 1) % N_COMMIT_IDS;
138 static unsigned char next_commit_seq(unsigned char seq)
140 return (seq + 1) % N_COMMIT_IDS;
144 * In-memory structures
147 struct journal_node {
159 struct dm_integrity_c {
161 struct dm_dev *meta_dev;
165 mempool_t journal_io_mempool;
166 struct dm_io_client *io;
167 struct dm_bufio_client *bufio;
168 struct workqueue_struct *metadata_wq;
169 struct superblock *sb;
170 unsigned journal_pages;
171 unsigned n_bitmap_blocks;
173 struct page_list *journal;
174 struct page_list *journal_io;
175 struct page_list *journal_xor;
176 struct page_list *recalc_bitmap;
177 struct page_list *may_write_bitmap;
178 struct bitmap_block_status *bbs;
179 unsigned bitmap_flush_interval;
180 int synchronous_mode;
181 struct bio_list synchronous_bios;
182 struct delayed_work bitmap_flush_work;
184 struct crypto_skcipher *journal_crypt;
185 struct scatterlist **journal_scatterlist;
186 struct scatterlist **journal_io_scatterlist;
187 struct skcipher_request **sk_requests;
189 struct crypto_shash *journal_mac;
191 struct journal_node *journal_tree;
192 struct rb_root journal_tree_root;
194 sector_t provided_data_sectors;
196 unsigned short journal_entry_size;
197 unsigned char journal_entries_per_sector;
198 unsigned char journal_section_entries;
199 unsigned short journal_section_sectors;
200 unsigned journal_sections;
201 unsigned journal_entries;
202 sector_t data_device_sectors;
203 sector_t meta_device_sectors;
204 unsigned initial_sectors;
205 unsigned metadata_run;
206 __s8 log2_metadata_run;
207 __u8 log2_buffer_sectors;
208 __u8 sectors_per_block;
209 __u8 log2_blocks_per_bitmap_bit;
215 struct crypto_shash *internal_hash;
217 struct dm_target *ti;
219 /* these variables are locked with endio_wait.lock */
220 struct rb_root in_progress;
221 struct list_head wait_list;
222 wait_queue_head_t endio_wait;
223 struct workqueue_struct *wait_wq;
224 struct workqueue_struct *offload_wq;
226 unsigned char commit_seq;
227 commit_id_t commit_ids[N_COMMIT_IDS];
229 unsigned committed_section;
230 unsigned n_committed_sections;
232 unsigned uncommitted_section;
233 unsigned n_uncommitted_sections;
235 unsigned free_section;
236 unsigned char free_section_entry;
237 unsigned free_sectors;
239 unsigned free_sectors_threshold;
241 struct workqueue_struct *commit_wq;
242 struct work_struct commit_work;
244 struct workqueue_struct *writer_wq;
245 struct work_struct writer_work;
247 struct workqueue_struct *recalc_wq;
248 struct work_struct recalc_work;
252 struct bio_list flush_bio_list;
254 unsigned long autocommit_jiffies;
255 struct timer_list autocommit_timer;
256 unsigned autocommit_msec;
258 wait_queue_head_t copy_to_journal_wait;
260 struct completion crypto_backoff;
262 bool wrote_to_journal;
263 bool journal_uptodate;
265 bool recalculate_flag;
266 bool reset_recalculate_flag;
270 bool legacy_recalculate;
272 struct alg_spec internal_hash_alg;
273 struct alg_spec journal_crypt_alg;
274 struct alg_spec journal_mac_alg;
276 atomic64_t number_of_mismatches;
278 struct notifier_block reboot_notifier;
281 struct dm_integrity_range {
282 sector_t logical_sector;
288 struct task_struct *task;
289 struct list_head wait_entry;
294 struct dm_integrity_io {
295 struct work_struct work;
297 struct dm_integrity_c *ic;
301 struct dm_integrity_range range;
303 sector_t metadata_block;
304 unsigned metadata_offset;
307 blk_status_t bi_status;
309 struct completion *completion;
311 struct dm_bio_details bio_details;
314 struct journal_completion {
315 struct dm_integrity_c *ic;
317 struct completion comp;
321 struct dm_integrity_range range;
322 struct journal_completion *comp;
325 struct bitmap_block_status {
326 struct work_struct work;
327 struct dm_integrity_c *ic;
329 unsigned long *bitmap;
330 struct bio_list bio_queue;
331 spinlock_t bio_queue_lock;
335 static struct kmem_cache *journal_io_cache;
337 #define JOURNAL_IO_MEMPOOL 32
340 #define DEBUG_print(x, ...) printk(KERN_DEBUG x, ##__VA_ARGS__)
341 static void __DEBUG_bytes(__u8 *bytes, size_t len, const char *msg, ...)
350 pr_cont(" %02x", *bytes);
356 #define DEBUG_bytes(bytes, len, msg, ...) __DEBUG_bytes(bytes, len, KERN_DEBUG msg, ##__VA_ARGS__)
358 #define DEBUG_print(x, ...) do { } while (0)
359 #define DEBUG_bytes(bytes, len, msg, ...) do { } while (0)
362 static void dm_integrity_prepare(struct request *rq)
366 static void dm_integrity_complete(struct request *rq, unsigned int nr_bytes)
371 * DM Integrity profile, protection is performed layer above (dm-crypt)
373 static const struct blk_integrity_profile dm_integrity_profile = {
374 .name = "DM-DIF-EXT-TAG",
377 .prepare_fn = dm_integrity_prepare,
378 .complete_fn = dm_integrity_complete,
381 static void dm_integrity_map_continue(struct dm_integrity_io *dio, bool from_map);
382 static void integrity_bio_wait(struct work_struct *w);
383 static void dm_integrity_dtr(struct dm_target *ti);
385 static void dm_integrity_io_error(struct dm_integrity_c *ic, const char *msg, int err)
388 atomic64_inc(&ic->number_of_mismatches);
389 if (!cmpxchg(&ic->failed, 0, err))
390 DMERR("Error on %s: %d", msg, err);
393 static int dm_integrity_failed(struct dm_integrity_c *ic)
395 return READ_ONCE(ic->failed);
398 static bool dm_integrity_disable_recalculate(struct dm_integrity_c *ic)
400 if (ic->legacy_recalculate)
402 if (!(ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) ?
403 ic->internal_hash_alg.key || ic->journal_mac_alg.key :
404 ic->internal_hash_alg.key && !ic->journal_mac_alg.key)
409 static commit_id_t dm_integrity_commit_id(struct dm_integrity_c *ic, unsigned i,
410 unsigned j, unsigned char seq)
413 * Xor the number with section and sector, so that if a piece of
414 * journal is written at wrong place, it is detected.
416 return ic->commit_ids[seq] ^ cpu_to_le64(((__u64)i << 32) ^ j);
419 static void get_area_and_offset(struct dm_integrity_c *ic, sector_t data_sector,
420 sector_t *area, sector_t *offset)
423 __u8 log2_interleave_sectors = ic->sb->log2_interleave_sectors;
424 *area = data_sector >> log2_interleave_sectors;
425 *offset = (unsigned)data_sector & ((1U << log2_interleave_sectors) - 1);
428 *offset = data_sector;
432 #define sector_to_block(ic, n) \
434 BUG_ON((n) & (unsigned)((ic)->sectors_per_block - 1)); \
435 (n) >>= (ic)->sb->log2_sectors_per_block; \
438 static __u64 get_metadata_sector_and_offset(struct dm_integrity_c *ic, sector_t area,
439 sector_t offset, unsigned *metadata_offset)
444 ms = area << ic->sb->log2_interleave_sectors;
445 if (likely(ic->log2_metadata_run >= 0))
446 ms += area << ic->log2_metadata_run;
448 ms += area * ic->metadata_run;
449 ms >>= ic->log2_buffer_sectors;
451 sector_to_block(ic, offset);
453 if (likely(ic->log2_tag_size >= 0)) {
454 ms += offset >> (SECTOR_SHIFT + ic->log2_buffer_sectors - ic->log2_tag_size);
455 mo = (offset << ic->log2_tag_size) & ((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - 1);
457 ms += (__u64)offset * ic->tag_size >> (SECTOR_SHIFT + ic->log2_buffer_sectors);
458 mo = (offset * ic->tag_size) & ((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - 1);
460 *metadata_offset = mo;
464 static sector_t get_data_sector(struct dm_integrity_c *ic, sector_t area, sector_t offset)
471 result = area << ic->sb->log2_interleave_sectors;
472 if (likely(ic->log2_metadata_run >= 0))
473 result += (area + 1) << ic->log2_metadata_run;
475 result += (area + 1) * ic->metadata_run;
477 result += (sector_t)ic->initial_sectors + offset;
483 static void wraparound_section(struct dm_integrity_c *ic, unsigned *sec_ptr)
485 if (unlikely(*sec_ptr >= ic->journal_sections))
486 *sec_ptr -= ic->journal_sections;
489 static void sb_set_version(struct dm_integrity_c *ic)
491 if (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC))
492 ic->sb->version = SB_VERSION_5;
493 else if (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING))
494 ic->sb->version = SB_VERSION_4;
495 else if (ic->mode == 'B' || ic->sb->flags & cpu_to_le32(SB_FLAG_DIRTY_BITMAP))
496 ic->sb->version = SB_VERSION_3;
497 else if (ic->meta_dev || ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
498 ic->sb->version = SB_VERSION_2;
500 ic->sb->version = SB_VERSION_1;
503 static int sb_mac(struct dm_integrity_c *ic, bool wr)
505 SHASH_DESC_ON_STACK(desc, ic->journal_mac);
507 unsigned size = crypto_shash_digestsize(ic->journal_mac);
509 if (sizeof(struct superblock) + size > 1 << SECTOR_SHIFT) {
510 dm_integrity_io_error(ic, "digest is too long", -EINVAL);
514 desc->tfm = ic->journal_mac;
516 r = crypto_shash_init(desc);
517 if (unlikely(r < 0)) {
518 dm_integrity_io_error(ic, "crypto_shash_init", r);
522 r = crypto_shash_update(desc, (__u8 *)ic->sb, (1 << SECTOR_SHIFT) - size);
523 if (unlikely(r < 0)) {
524 dm_integrity_io_error(ic, "crypto_shash_update", r);
529 r = crypto_shash_final(desc, (__u8 *)ic->sb + (1 << SECTOR_SHIFT) - size);
530 if (unlikely(r < 0)) {
531 dm_integrity_io_error(ic, "crypto_shash_final", r);
535 __u8 result[HASH_MAX_DIGESTSIZE];
536 r = crypto_shash_final(desc, result);
537 if (unlikely(r < 0)) {
538 dm_integrity_io_error(ic, "crypto_shash_final", r);
541 if (memcmp((__u8 *)ic->sb + (1 << SECTOR_SHIFT) - size, result, size)) {
542 dm_integrity_io_error(ic, "superblock mac", -EILSEQ);
550 static int sync_rw_sb(struct dm_integrity_c *ic, int op, int op_flags)
552 struct dm_io_request io_req;
553 struct dm_io_region io_loc;
557 io_req.bi_op_flags = op_flags;
558 io_req.mem.type = DM_IO_KMEM;
559 io_req.mem.ptr.addr = ic->sb;
560 io_req.notify.fn = NULL;
561 io_req.client = ic->io;
562 io_loc.bdev = ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev;
563 io_loc.sector = ic->start;
564 io_loc.count = SB_SECTORS;
566 if (op == REQ_OP_WRITE) {
568 if (ic->journal_mac && ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) {
569 r = sb_mac(ic, true);
575 r = dm_io(&io_req, 1, &io_loc, NULL);
579 if (op == REQ_OP_READ) {
580 if (ic->mode != 'R' && ic->journal_mac && ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) {
581 r = sb_mac(ic, false);
590 #define BITMAP_OP_TEST_ALL_SET 0
591 #define BITMAP_OP_TEST_ALL_CLEAR 1
592 #define BITMAP_OP_SET 2
593 #define BITMAP_OP_CLEAR 3
595 static bool block_bitmap_op(struct dm_integrity_c *ic, struct page_list *bitmap,
596 sector_t sector, sector_t n_sectors, int mode)
598 unsigned long bit, end_bit, this_end_bit, page, end_page;
601 if (unlikely(((sector | n_sectors) & ((1 << ic->sb->log2_sectors_per_block) - 1)) != 0)) {
602 DMCRIT("invalid bitmap access (%llx,%llx,%d,%d,%d)",
605 ic->sb->log2_sectors_per_block,
606 ic->log2_blocks_per_bitmap_bit,
611 if (unlikely(!n_sectors))
614 bit = sector >> (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
615 end_bit = (sector + n_sectors - 1) >>
616 (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
618 page = bit / (PAGE_SIZE * 8);
619 bit %= PAGE_SIZE * 8;
621 end_page = end_bit / (PAGE_SIZE * 8);
622 end_bit %= PAGE_SIZE * 8;
625 if (page < end_page) {
626 this_end_bit = PAGE_SIZE * 8 - 1;
628 this_end_bit = end_bit;
631 data = lowmem_page_address(bitmap[page].page);
633 if (mode == BITMAP_OP_TEST_ALL_SET) {
634 while (bit <= this_end_bit) {
635 if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
637 if (data[bit / BITS_PER_LONG] != -1)
639 bit += BITS_PER_LONG;
640 } while (this_end_bit >= bit + BITS_PER_LONG - 1);
643 if (!test_bit(bit, data))
647 } else if (mode == BITMAP_OP_TEST_ALL_CLEAR) {
648 while (bit <= this_end_bit) {
649 if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
651 if (data[bit / BITS_PER_LONG] != 0)
653 bit += BITS_PER_LONG;
654 } while (this_end_bit >= bit + BITS_PER_LONG - 1);
657 if (test_bit(bit, data))
661 } else if (mode == BITMAP_OP_SET) {
662 while (bit <= this_end_bit) {
663 if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
665 data[bit / BITS_PER_LONG] = -1;
666 bit += BITS_PER_LONG;
667 } while (this_end_bit >= bit + BITS_PER_LONG - 1);
670 __set_bit(bit, data);
673 } else if (mode == BITMAP_OP_CLEAR) {
674 if (!bit && this_end_bit == PAGE_SIZE * 8 - 1)
676 else while (bit <= this_end_bit) {
677 if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
679 data[bit / BITS_PER_LONG] = 0;
680 bit += BITS_PER_LONG;
681 } while (this_end_bit >= bit + BITS_PER_LONG - 1);
684 __clear_bit(bit, data);
691 if (unlikely(page < end_page)) {
700 static void block_bitmap_copy(struct dm_integrity_c *ic, struct page_list *dst, struct page_list *src)
702 unsigned n_bitmap_pages = DIV_ROUND_UP(ic->n_bitmap_blocks, PAGE_SIZE / BITMAP_BLOCK_SIZE);
705 for (i = 0; i < n_bitmap_pages; i++) {
706 unsigned long *dst_data = lowmem_page_address(dst[i].page);
707 unsigned long *src_data = lowmem_page_address(src[i].page);
708 copy_page(dst_data, src_data);
712 static struct bitmap_block_status *sector_to_bitmap_block(struct dm_integrity_c *ic, sector_t sector)
714 unsigned bit = sector >> (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
715 unsigned bitmap_block = bit / (BITMAP_BLOCK_SIZE * 8);
717 BUG_ON(bitmap_block >= ic->n_bitmap_blocks);
718 return &ic->bbs[bitmap_block];
721 static void access_journal_check(struct dm_integrity_c *ic, unsigned section, unsigned offset,
722 bool e, const char *function)
724 #if defined(CONFIG_DM_DEBUG) || defined(INTERNAL_VERIFY)
725 unsigned limit = e ? ic->journal_section_entries : ic->journal_section_sectors;
727 if (unlikely(section >= ic->journal_sections) ||
728 unlikely(offset >= limit)) {
729 DMCRIT("%s: invalid access at (%u,%u), limit (%u,%u)",
730 function, section, offset, ic->journal_sections, limit);
736 static void page_list_location(struct dm_integrity_c *ic, unsigned section, unsigned offset,
737 unsigned *pl_index, unsigned *pl_offset)
741 access_journal_check(ic, section, offset, false, "page_list_location");
743 sector = section * ic->journal_section_sectors + offset;
745 *pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
746 *pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
749 static struct journal_sector *access_page_list(struct dm_integrity_c *ic, struct page_list *pl,
750 unsigned section, unsigned offset, unsigned *n_sectors)
752 unsigned pl_index, pl_offset;
755 page_list_location(ic, section, offset, &pl_index, &pl_offset);
758 *n_sectors = (PAGE_SIZE - pl_offset) >> SECTOR_SHIFT;
760 va = lowmem_page_address(pl[pl_index].page);
762 return (struct journal_sector *)(va + pl_offset);
765 static struct journal_sector *access_journal(struct dm_integrity_c *ic, unsigned section, unsigned offset)
767 return access_page_list(ic, ic->journal, section, offset, NULL);
770 static struct journal_entry *access_journal_entry(struct dm_integrity_c *ic, unsigned section, unsigned n)
772 unsigned rel_sector, offset;
773 struct journal_sector *js;
775 access_journal_check(ic, section, n, true, "access_journal_entry");
777 rel_sector = n % JOURNAL_BLOCK_SECTORS;
778 offset = n / JOURNAL_BLOCK_SECTORS;
780 js = access_journal(ic, section, rel_sector);
781 return (struct journal_entry *)((char *)js + offset * ic->journal_entry_size);
784 static struct journal_sector *access_journal_data(struct dm_integrity_c *ic, unsigned section, unsigned n)
786 n <<= ic->sb->log2_sectors_per_block;
788 n += JOURNAL_BLOCK_SECTORS;
790 access_journal_check(ic, section, n, false, "access_journal_data");
792 return access_journal(ic, section, n);
795 static void section_mac(struct dm_integrity_c *ic, unsigned section, __u8 result[JOURNAL_MAC_SIZE])
797 SHASH_DESC_ON_STACK(desc, ic->journal_mac);
801 desc->tfm = ic->journal_mac;
803 r = crypto_shash_init(desc);
804 if (unlikely(r < 0)) {
805 dm_integrity_io_error(ic, "crypto_shash_init", r);
809 if (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) {
812 r = crypto_shash_update(desc, (__u8 *)&ic->sb->salt, SALT_SIZE);
813 if (unlikely(r < 0)) {
814 dm_integrity_io_error(ic, "crypto_shash_update", r);
818 section_le = cpu_to_le64(section);
819 r = crypto_shash_update(desc, (__u8 *)§ion_le, sizeof section_le);
820 if (unlikely(r < 0)) {
821 dm_integrity_io_error(ic, "crypto_shash_update", r);
826 for (j = 0; j < ic->journal_section_entries; j++) {
827 struct journal_entry *je = access_journal_entry(ic, section, j);
828 r = crypto_shash_update(desc, (__u8 *)&je->u.sector, sizeof je->u.sector);
829 if (unlikely(r < 0)) {
830 dm_integrity_io_error(ic, "crypto_shash_update", r);
835 size = crypto_shash_digestsize(ic->journal_mac);
837 if (likely(size <= JOURNAL_MAC_SIZE)) {
838 r = crypto_shash_final(desc, result);
839 if (unlikely(r < 0)) {
840 dm_integrity_io_error(ic, "crypto_shash_final", r);
843 memset(result + size, 0, JOURNAL_MAC_SIZE - size);
845 __u8 digest[HASH_MAX_DIGESTSIZE];
847 if (WARN_ON(size > sizeof(digest))) {
848 dm_integrity_io_error(ic, "digest_size", -EINVAL);
851 r = crypto_shash_final(desc, digest);
852 if (unlikely(r < 0)) {
853 dm_integrity_io_error(ic, "crypto_shash_final", r);
856 memcpy(result, digest, JOURNAL_MAC_SIZE);
861 memset(result, 0, JOURNAL_MAC_SIZE);
864 static void rw_section_mac(struct dm_integrity_c *ic, unsigned section, bool wr)
866 __u8 result[JOURNAL_MAC_SIZE];
869 if (!ic->journal_mac)
872 section_mac(ic, section, result);
874 for (j = 0; j < JOURNAL_BLOCK_SECTORS; j++) {
875 struct journal_sector *js = access_journal(ic, section, j);
878 memcpy(&js->mac, result + (j * JOURNAL_MAC_PER_SECTOR), JOURNAL_MAC_PER_SECTOR);
880 if (memcmp(&js->mac, result + (j * JOURNAL_MAC_PER_SECTOR), JOURNAL_MAC_PER_SECTOR))
881 dm_integrity_io_error(ic, "journal mac", -EILSEQ);
886 static void complete_journal_op(void *context)
888 struct journal_completion *comp = context;
889 BUG_ON(!atomic_read(&comp->in_flight));
890 if (likely(atomic_dec_and_test(&comp->in_flight)))
891 complete(&comp->comp);
894 static void xor_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
895 unsigned n_sections, struct journal_completion *comp)
897 struct async_submit_ctl submit;
898 size_t n_bytes = (size_t)(n_sections * ic->journal_section_sectors) << SECTOR_SHIFT;
899 unsigned pl_index, pl_offset, section_index;
900 struct page_list *source_pl, *target_pl;
902 if (likely(encrypt)) {
903 source_pl = ic->journal;
904 target_pl = ic->journal_io;
906 source_pl = ic->journal_io;
907 target_pl = ic->journal;
910 page_list_location(ic, section, 0, &pl_index, &pl_offset);
912 atomic_add(roundup(pl_offset + n_bytes, PAGE_SIZE) >> PAGE_SHIFT, &comp->in_flight);
914 init_async_submit(&submit, ASYNC_TX_XOR_ZERO_DST, NULL, complete_journal_op, comp, NULL);
916 section_index = pl_index;
920 struct page *src_pages[2];
921 struct page *dst_page;
923 while (unlikely(pl_index == section_index)) {
926 rw_section_mac(ic, section, true);
931 page_list_location(ic, section, 0, §ion_index, &dummy);
934 this_step = min(n_bytes, (size_t)PAGE_SIZE - pl_offset);
935 dst_page = target_pl[pl_index].page;
936 src_pages[0] = source_pl[pl_index].page;
937 src_pages[1] = ic->journal_xor[pl_index].page;
939 async_xor(dst_page, src_pages, pl_offset, 2, this_step, &submit);
943 n_bytes -= this_step;
948 async_tx_issue_pending_all();
951 static void complete_journal_encrypt(struct crypto_async_request *req, int err)
953 struct journal_completion *comp = req->data;
955 if (likely(err == -EINPROGRESS)) {
956 complete(&comp->ic->crypto_backoff);
959 dm_integrity_io_error(comp->ic, "asynchronous encrypt", err);
961 complete_journal_op(comp);
964 static bool do_crypt(bool encrypt, struct skcipher_request *req, struct journal_completion *comp)
967 skcipher_request_set_callback(req, CRYPTO_TFM_REQ_MAY_BACKLOG,
968 complete_journal_encrypt, comp);
970 r = crypto_skcipher_encrypt(req);
972 r = crypto_skcipher_decrypt(req);
975 if (likely(r == -EINPROGRESS))
977 if (likely(r == -EBUSY)) {
978 wait_for_completion(&comp->ic->crypto_backoff);
979 reinit_completion(&comp->ic->crypto_backoff);
982 dm_integrity_io_error(comp->ic, "encrypt", r);
986 static void crypt_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
987 unsigned n_sections, struct journal_completion *comp)
989 struct scatterlist **source_sg;
990 struct scatterlist **target_sg;
992 atomic_add(2, &comp->in_flight);
994 if (likely(encrypt)) {
995 source_sg = ic->journal_scatterlist;
996 target_sg = ic->journal_io_scatterlist;
998 source_sg = ic->journal_io_scatterlist;
999 target_sg = ic->journal_scatterlist;
1003 struct skcipher_request *req;
1007 if (likely(encrypt))
1008 rw_section_mac(ic, section, true);
1010 req = ic->sk_requests[section];
1011 ivsize = crypto_skcipher_ivsize(ic->journal_crypt);
1014 memcpy(iv, iv + ivsize, ivsize);
1016 req->src = source_sg[section];
1017 req->dst = target_sg[section];
1019 if (unlikely(do_crypt(encrypt, req, comp)))
1020 atomic_inc(&comp->in_flight);
1024 } while (n_sections);
1026 atomic_dec(&comp->in_flight);
1027 complete_journal_op(comp);
1030 static void encrypt_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
1031 unsigned n_sections, struct journal_completion *comp)
1033 if (ic->journal_xor)
1034 return xor_journal(ic, encrypt, section, n_sections, comp);
1036 return crypt_journal(ic, encrypt, section, n_sections, comp);
1039 static void complete_journal_io(unsigned long error, void *context)
1041 struct journal_completion *comp = context;
1042 if (unlikely(error != 0))
1043 dm_integrity_io_error(comp->ic, "writing journal", -EIO);
1044 complete_journal_op(comp);
1047 static void rw_journal_sectors(struct dm_integrity_c *ic, int op, int op_flags,
1048 unsigned sector, unsigned n_sectors, struct journal_completion *comp)
1050 struct dm_io_request io_req;
1051 struct dm_io_region io_loc;
1052 unsigned pl_index, pl_offset;
1055 if (unlikely(dm_integrity_failed(ic))) {
1057 complete_journal_io(-1UL, comp);
1061 pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
1062 pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
1065 io_req.bi_op_flags = op_flags;
1066 io_req.mem.type = DM_IO_PAGE_LIST;
1068 io_req.mem.ptr.pl = &ic->journal_io[pl_index];
1070 io_req.mem.ptr.pl = &ic->journal[pl_index];
1071 io_req.mem.offset = pl_offset;
1072 if (likely(comp != NULL)) {
1073 io_req.notify.fn = complete_journal_io;
1074 io_req.notify.context = comp;
1076 io_req.notify.fn = NULL;
1078 io_req.client = ic->io;
1079 io_loc.bdev = ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev;
1080 io_loc.sector = ic->start + SB_SECTORS + sector;
1081 io_loc.count = n_sectors;
1083 r = dm_io(&io_req, 1, &io_loc, NULL);
1085 dm_integrity_io_error(ic, op == REQ_OP_READ ? "reading journal" : "writing journal", r);
1087 WARN_ONCE(1, "asynchronous dm_io failed: %d", r);
1088 complete_journal_io(-1UL, comp);
1093 static void rw_journal(struct dm_integrity_c *ic, int op, int op_flags, unsigned section,
1094 unsigned n_sections, struct journal_completion *comp)
1096 unsigned sector, n_sectors;
1098 sector = section * ic->journal_section_sectors;
1099 n_sectors = n_sections * ic->journal_section_sectors;
1101 rw_journal_sectors(ic, op, op_flags, sector, n_sectors, comp);
1104 static void write_journal(struct dm_integrity_c *ic, unsigned commit_start, unsigned commit_sections)
1106 struct journal_completion io_comp;
1107 struct journal_completion crypt_comp_1;
1108 struct journal_completion crypt_comp_2;
1112 init_completion(&io_comp.comp);
1114 if (commit_start + commit_sections <= ic->journal_sections) {
1115 io_comp.in_flight = (atomic_t)ATOMIC_INIT(1);
1116 if (ic->journal_io) {
1117 crypt_comp_1.ic = ic;
1118 init_completion(&crypt_comp_1.comp);
1119 crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
1120 encrypt_journal(ic, true, commit_start, commit_sections, &crypt_comp_1);
1121 wait_for_completion_io(&crypt_comp_1.comp);
1123 for (i = 0; i < commit_sections; i++)
1124 rw_section_mac(ic, commit_start + i, true);
1126 rw_journal(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, commit_start,
1127 commit_sections, &io_comp);
1130 io_comp.in_flight = (atomic_t)ATOMIC_INIT(2);
1131 to_end = ic->journal_sections - commit_start;
1132 if (ic->journal_io) {
1133 crypt_comp_1.ic = ic;
1134 init_completion(&crypt_comp_1.comp);
1135 crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
1136 encrypt_journal(ic, true, commit_start, to_end, &crypt_comp_1);
1137 if (try_wait_for_completion(&crypt_comp_1.comp)) {
1138 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
1139 reinit_completion(&crypt_comp_1.comp);
1140 crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
1141 encrypt_journal(ic, true, 0, commit_sections - to_end, &crypt_comp_1);
1142 wait_for_completion_io(&crypt_comp_1.comp);
1144 crypt_comp_2.ic = ic;
1145 init_completion(&crypt_comp_2.comp);
1146 crypt_comp_2.in_flight = (atomic_t)ATOMIC_INIT(0);
1147 encrypt_journal(ic, true, 0, commit_sections - to_end, &crypt_comp_2);
1148 wait_for_completion_io(&crypt_comp_1.comp);
1149 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
1150 wait_for_completion_io(&crypt_comp_2.comp);
1153 for (i = 0; i < to_end; i++)
1154 rw_section_mac(ic, commit_start + i, true);
1155 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
1156 for (i = 0; i < commit_sections - to_end; i++)
1157 rw_section_mac(ic, i, true);
1159 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, 0, commit_sections - to_end, &io_comp);
1162 wait_for_completion_io(&io_comp.comp);
1165 static void copy_from_journal(struct dm_integrity_c *ic, unsigned section, unsigned offset,
1166 unsigned n_sectors, sector_t target, io_notify_fn fn, void *data)
1168 struct dm_io_request io_req;
1169 struct dm_io_region io_loc;
1171 unsigned sector, pl_index, pl_offset;
1173 BUG_ON((target | n_sectors | offset) & (unsigned)(ic->sectors_per_block - 1));
1175 if (unlikely(dm_integrity_failed(ic))) {
1180 sector = section * ic->journal_section_sectors + JOURNAL_BLOCK_SECTORS + offset;
1182 pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
1183 pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
1185 io_req.bi_op = REQ_OP_WRITE;
1186 io_req.bi_op_flags = 0;
1187 io_req.mem.type = DM_IO_PAGE_LIST;
1188 io_req.mem.ptr.pl = &ic->journal[pl_index];
1189 io_req.mem.offset = pl_offset;
1190 io_req.notify.fn = fn;
1191 io_req.notify.context = data;
1192 io_req.client = ic->io;
1193 io_loc.bdev = ic->dev->bdev;
1194 io_loc.sector = target;
1195 io_loc.count = n_sectors;
1197 r = dm_io(&io_req, 1, &io_loc, NULL);
1199 WARN_ONCE(1, "asynchronous dm_io failed: %d", r);
1204 static bool ranges_overlap(struct dm_integrity_range *range1, struct dm_integrity_range *range2)
1206 return range1->logical_sector < range2->logical_sector + range2->n_sectors &&
1207 range1->logical_sector + range1->n_sectors > range2->logical_sector;
1210 static bool add_new_range(struct dm_integrity_c *ic, struct dm_integrity_range *new_range, bool check_waiting)
1212 struct rb_node **n = &ic->in_progress.rb_node;
1213 struct rb_node *parent;
1215 BUG_ON((new_range->logical_sector | new_range->n_sectors) & (unsigned)(ic->sectors_per_block - 1));
1217 if (likely(check_waiting)) {
1218 struct dm_integrity_range *range;
1219 list_for_each_entry(range, &ic->wait_list, wait_entry) {
1220 if (unlikely(ranges_overlap(range, new_range)))
1228 struct dm_integrity_range *range = container_of(*n, struct dm_integrity_range, node);
1231 if (new_range->logical_sector + new_range->n_sectors <= range->logical_sector) {
1232 n = &range->node.rb_left;
1233 } else if (new_range->logical_sector >= range->logical_sector + range->n_sectors) {
1234 n = &range->node.rb_right;
1240 rb_link_node(&new_range->node, parent, n);
1241 rb_insert_color(&new_range->node, &ic->in_progress);
1246 static void remove_range_unlocked(struct dm_integrity_c *ic, struct dm_integrity_range *range)
1248 rb_erase(&range->node, &ic->in_progress);
1249 while (unlikely(!list_empty(&ic->wait_list))) {
1250 struct dm_integrity_range *last_range =
1251 list_first_entry(&ic->wait_list, struct dm_integrity_range, wait_entry);
1252 struct task_struct *last_range_task;
1253 last_range_task = last_range->task;
1254 list_del(&last_range->wait_entry);
1255 if (!add_new_range(ic, last_range, false)) {
1256 last_range->task = last_range_task;
1257 list_add(&last_range->wait_entry, &ic->wait_list);
1260 last_range->waiting = false;
1261 wake_up_process(last_range_task);
1265 static void remove_range(struct dm_integrity_c *ic, struct dm_integrity_range *range)
1267 unsigned long flags;
1269 spin_lock_irqsave(&ic->endio_wait.lock, flags);
1270 remove_range_unlocked(ic, range);
1271 spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
1274 static void wait_and_add_new_range(struct dm_integrity_c *ic, struct dm_integrity_range *new_range)
1276 new_range->waiting = true;
1277 list_add_tail(&new_range->wait_entry, &ic->wait_list);
1278 new_range->task = current;
1280 __set_current_state(TASK_UNINTERRUPTIBLE);
1281 spin_unlock_irq(&ic->endio_wait.lock);
1283 spin_lock_irq(&ic->endio_wait.lock);
1284 } while (unlikely(new_range->waiting));
1287 static void add_new_range_and_wait(struct dm_integrity_c *ic, struct dm_integrity_range *new_range)
1289 if (unlikely(!add_new_range(ic, new_range, true)))
1290 wait_and_add_new_range(ic, new_range);
1293 static void init_journal_node(struct journal_node *node)
1295 RB_CLEAR_NODE(&node->node);
1296 node->sector = (sector_t)-1;
1299 static void add_journal_node(struct dm_integrity_c *ic, struct journal_node *node, sector_t sector)
1301 struct rb_node **link;
1302 struct rb_node *parent;
1304 node->sector = sector;
1305 BUG_ON(!RB_EMPTY_NODE(&node->node));
1307 link = &ic->journal_tree_root.rb_node;
1311 struct journal_node *j;
1313 j = container_of(parent, struct journal_node, node);
1314 if (sector < j->sector)
1315 link = &j->node.rb_left;
1317 link = &j->node.rb_right;
1320 rb_link_node(&node->node, parent, link);
1321 rb_insert_color(&node->node, &ic->journal_tree_root);
1324 static void remove_journal_node(struct dm_integrity_c *ic, struct journal_node *node)
1326 BUG_ON(RB_EMPTY_NODE(&node->node));
1327 rb_erase(&node->node, &ic->journal_tree_root);
1328 init_journal_node(node);
1331 #define NOT_FOUND (-1U)
1333 static unsigned find_journal_node(struct dm_integrity_c *ic, sector_t sector, sector_t *next_sector)
1335 struct rb_node *n = ic->journal_tree_root.rb_node;
1336 unsigned found = NOT_FOUND;
1337 *next_sector = (sector_t)-1;
1339 struct journal_node *j = container_of(n, struct journal_node, node);
1340 if (sector == j->sector) {
1341 found = j - ic->journal_tree;
1343 if (sector < j->sector) {
1344 *next_sector = j->sector;
1345 n = j->node.rb_left;
1347 n = j->node.rb_right;
1354 static bool test_journal_node(struct dm_integrity_c *ic, unsigned pos, sector_t sector)
1356 struct journal_node *node, *next_node;
1357 struct rb_node *next;
1359 if (unlikely(pos >= ic->journal_entries))
1361 node = &ic->journal_tree[pos];
1362 if (unlikely(RB_EMPTY_NODE(&node->node)))
1364 if (unlikely(node->sector != sector))
1367 next = rb_next(&node->node);
1368 if (unlikely(!next))
1371 next_node = container_of(next, struct journal_node, node);
1372 return next_node->sector != sector;
1375 static bool find_newer_committed_node(struct dm_integrity_c *ic, struct journal_node *node)
1377 struct rb_node *next;
1378 struct journal_node *next_node;
1379 unsigned next_section;
1381 BUG_ON(RB_EMPTY_NODE(&node->node));
1383 next = rb_next(&node->node);
1384 if (unlikely(!next))
1387 next_node = container_of(next, struct journal_node, node);
1389 if (next_node->sector != node->sector)
1392 next_section = (unsigned)(next_node - ic->journal_tree) / ic->journal_section_entries;
1393 if (next_section >= ic->committed_section &&
1394 next_section < ic->committed_section + ic->n_committed_sections)
1396 if (next_section + ic->journal_sections < ic->committed_section + ic->n_committed_sections)
1406 static int dm_integrity_rw_tag(struct dm_integrity_c *ic, unsigned char *tag, sector_t *metadata_block,
1407 unsigned *metadata_offset, unsigned total_size, int op)
1409 #define MAY_BE_FILLER 1
1410 #define MAY_BE_HASH 2
1411 unsigned hash_offset = 0;
1412 unsigned may_be = MAY_BE_HASH | (ic->discard ? MAY_BE_FILLER : 0);
1415 unsigned char *data, *dp;
1416 struct dm_buffer *b;
1420 r = dm_integrity_failed(ic);
1424 data = dm_bufio_read(ic->bufio, *metadata_block, &b);
1426 return PTR_ERR(data);
1428 to_copy = min((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - *metadata_offset, total_size);
1429 dp = data + *metadata_offset;
1430 if (op == TAG_READ) {
1431 memcpy(tag, dp, to_copy);
1432 } else if (op == TAG_WRITE) {
1433 if (memcmp(dp, tag, to_copy)) {
1434 memcpy(dp, tag, to_copy);
1435 dm_bufio_mark_partial_buffer_dirty(b, *metadata_offset, *metadata_offset + to_copy);
1438 /* e.g.: op == TAG_CMP */
1440 if (likely(is_power_of_2(ic->tag_size))) {
1441 if (unlikely(memcmp(dp, tag, to_copy)))
1442 if (unlikely(!ic->discard) ||
1443 unlikely(memchr_inv(dp, DISCARD_FILLER, to_copy) != NULL)) {
1451 for (i = 0; i < to_copy; i++, ts--) {
1452 if (unlikely(dp[i] != tag[i]))
1453 may_be &= ~MAY_BE_HASH;
1454 if (likely(dp[i] != DISCARD_FILLER))
1455 may_be &= ~MAY_BE_FILLER;
1457 if (unlikely(hash_offset == ic->tag_size)) {
1458 if (unlikely(!may_be)) {
1459 dm_bufio_release(b);
1463 may_be = MAY_BE_HASH | (ic->discard ? MAY_BE_FILLER : 0);
1468 dm_bufio_release(b);
1471 *metadata_offset += to_copy;
1472 if (unlikely(*metadata_offset == 1U << SECTOR_SHIFT << ic->log2_buffer_sectors)) {
1473 (*metadata_block)++;
1474 *metadata_offset = 0;
1477 if (unlikely(!is_power_of_2(ic->tag_size))) {
1478 hash_offset = (hash_offset + to_copy) % ic->tag_size;
1481 total_size -= to_copy;
1482 } while (unlikely(total_size));
1485 #undef MAY_BE_FILLER
1489 struct flush_request {
1490 struct dm_io_request io_req;
1491 struct dm_io_region io_reg;
1492 struct dm_integrity_c *ic;
1493 struct completion comp;
1496 static void flush_notify(unsigned long error, void *fr_)
1498 struct flush_request *fr = fr_;
1499 if (unlikely(error != 0))
1500 dm_integrity_io_error(fr->ic, "flushing disk cache", -EIO);
1501 complete(&fr->comp);
1504 static void dm_integrity_flush_buffers(struct dm_integrity_c *ic, bool flush_data)
1508 struct flush_request fr;
1513 fr.io_req.bi_op = REQ_OP_WRITE,
1514 fr.io_req.bi_op_flags = REQ_PREFLUSH | REQ_SYNC,
1515 fr.io_req.mem.type = DM_IO_KMEM,
1516 fr.io_req.mem.ptr.addr = NULL,
1517 fr.io_req.notify.fn = flush_notify,
1518 fr.io_req.notify.context = &fr;
1519 fr.io_req.client = dm_bufio_get_dm_io_client(ic->bufio),
1520 fr.io_reg.bdev = ic->dev->bdev,
1521 fr.io_reg.sector = 0,
1522 fr.io_reg.count = 0,
1524 init_completion(&fr.comp);
1525 r = dm_io(&fr.io_req, 1, &fr.io_reg, NULL);
1529 r = dm_bufio_write_dirty_buffers(ic->bufio);
1531 dm_integrity_io_error(ic, "writing tags", r);
1534 wait_for_completion(&fr.comp);
1537 static void sleep_on_endio_wait(struct dm_integrity_c *ic)
1539 DECLARE_WAITQUEUE(wait, current);
1540 __add_wait_queue(&ic->endio_wait, &wait);
1541 __set_current_state(TASK_UNINTERRUPTIBLE);
1542 spin_unlock_irq(&ic->endio_wait.lock);
1544 spin_lock_irq(&ic->endio_wait.lock);
1545 __remove_wait_queue(&ic->endio_wait, &wait);
1548 static void autocommit_fn(struct timer_list *t)
1550 struct dm_integrity_c *ic = from_timer(ic, t, autocommit_timer);
1552 if (likely(!dm_integrity_failed(ic)))
1553 queue_work(ic->commit_wq, &ic->commit_work);
1556 static void schedule_autocommit(struct dm_integrity_c *ic)
1558 if (!timer_pending(&ic->autocommit_timer))
1559 mod_timer(&ic->autocommit_timer, jiffies + ic->autocommit_jiffies);
1562 static void submit_flush_bio(struct dm_integrity_c *ic, struct dm_integrity_io *dio)
1565 unsigned long flags;
1567 spin_lock_irqsave(&ic->endio_wait.lock, flags);
1568 bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1569 bio_list_add(&ic->flush_bio_list, bio);
1570 spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
1572 queue_work(ic->commit_wq, &ic->commit_work);
1575 static void do_endio(struct dm_integrity_c *ic, struct bio *bio)
1577 int r = dm_integrity_failed(ic);
1578 if (unlikely(r) && !bio->bi_status)
1579 bio->bi_status = errno_to_blk_status(r);
1580 if (unlikely(ic->synchronous_mode) && bio_op(bio) == REQ_OP_WRITE) {
1581 unsigned long flags;
1582 spin_lock_irqsave(&ic->endio_wait.lock, flags);
1583 bio_list_add(&ic->synchronous_bios, bio);
1584 queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, 0);
1585 spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
1591 static void do_endio_flush(struct dm_integrity_c *ic, struct dm_integrity_io *dio)
1593 struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1595 if (unlikely(dio->fua) && likely(!bio->bi_status) && likely(!dm_integrity_failed(ic)))
1596 submit_flush_bio(ic, dio);
1601 static void dec_in_flight(struct dm_integrity_io *dio)
1603 if (atomic_dec_and_test(&dio->in_flight)) {
1604 struct dm_integrity_c *ic = dio->ic;
1607 remove_range(ic, &dio->range);
1609 if (dio->op == REQ_OP_WRITE || unlikely(dio->op == REQ_OP_DISCARD))
1610 schedule_autocommit(ic);
1612 bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1614 if (unlikely(dio->bi_status) && !bio->bi_status)
1615 bio->bi_status = dio->bi_status;
1616 if (likely(!bio->bi_status) && unlikely(bio_sectors(bio) != dio->range.n_sectors)) {
1617 dio->range.logical_sector += dio->range.n_sectors;
1618 bio_advance(bio, dio->range.n_sectors << SECTOR_SHIFT);
1619 INIT_WORK(&dio->work, integrity_bio_wait);
1620 queue_work(ic->offload_wq, &dio->work);
1623 do_endio_flush(ic, dio);
1627 static void integrity_end_io(struct bio *bio)
1629 struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
1631 dm_bio_restore(&dio->bio_details, bio);
1632 if (bio->bi_integrity)
1633 bio->bi_opf |= REQ_INTEGRITY;
1635 if (dio->completion)
1636 complete(dio->completion);
1641 static void integrity_sector_checksum(struct dm_integrity_c *ic, sector_t sector,
1642 const char *data, char *result)
1644 __le64 sector_le = cpu_to_le64(sector);
1645 SHASH_DESC_ON_STACK(req, ic->internal_hash);
1647 unsigned digest_size;
1649 req->tfm = ic->internal_hash;
1651 r = crypto_shash_init(req);
1652 if (unlikely(r < 0)) {
1653 dm_integrity_io_error(ic, "crypto_shash_init", r);
1657 if (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) {
1658 r = crypto_shash_update(req, (__u8 *)&ic->sb->salt, SALT_SIZE);
1659 if (unlikely(r < 0)) {
1660 dm_integrity_io_error(ic, "crypto_shash_update", r);
1665 r = crypto_shash_update(req, (const __u8 *)§or_le, sizeof sector_le);
1666 if (unlikely(r < 0)) {
1667 dm_integrity_io_error(ic, "crypto_shash_update", r);
1671 r = crypto_shash_update(req, data, ic->sectors_per_block << SECTOR_SHIFT);
1672 if (unlikely(r < 0)) {
1673 dm_integrity_io_error(ic, "crypto_shash_update", r);
1677 r = crypto_shash_final(req, result);
1678 if (unlikely(r < 0)) {
1679 dm_integrity_io_error(ic, "crypto_shash_final", r);
1683 digest_size = crypto_shash_digestsize(ic->internal_hash);
1684 if (unlikely(digest_size < ic->tag_size))
1685 memset(result + digest_size, 0, ic->tag_size - digest_size);
1690 /* this shouldn't happen anyway, the hash functions have no reason to fail */
1691 get_random_bytes(result, ic->tag_size);
1694 static void integrity_metadata(struct work_struct *w)
1696 struct dm_integrity_io *dio = container_of(w, struct dm_integrity_io, work);
1697 struct dm_integrity_c *ic = dio->ic;
1701 if (ic->internal_hash) {
1702 struct bvec_iter iter;
1704 unsigned digest_size = crypto_shash_digestsize(ic->internal_hash);
1705 struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1707 unsigned extra_space = unlikely(digest_size > ic->tag_size) ? digest_size - ic->tag_size : 0;
1708 char checksums_onstack[max((size_t)HASH_MAX_DIGESTSIZE, MAX_TAG_SIZE)];
1710 unsigned sectors_to_process;
1712 if (unlikely(ic->mode == 'R'))
1715 if (likely(dio->op != REQ_OP_DISCARD))
1716 checksums = kmalloc((PAGE_SIZE >> SECTOR_SHIFT >> ic->sb->log2_sectors_per_block) * ic->tag_size + extra_space,
1717 GFP_NOIO | __GFP_NORETRY | __GFP_NOWARN);
1719 checksums = kmalloc(PAGE_SIZE, GFP_NOIO | __GFP_NORETRY | __GFP_NOWARN);
1721 checksums = checksums_onstack;
1722 if (WARN_ON(extra_space &&
1723 digest_size > sizeof(checksums_onstack))) {
1729 if (unlikely(dio->op == REQ_OP_DISCARD)) {
1730 sector_t bi_sector = dio->bio_details.bi_iter.bi_sector;
1731 unsigned bi_size = dio->bio_details.bi_iter.bi_size;
1732 unsigned max_size = likely(checksums != checksums_onstack) ? PAGE_SIZE : HASH_MAX_DIGESTSIZE;
1733 unsigned max_blocks = max_size / ic->tag_size;
1734 memset(checksums, DISCARD_FILLER, max_size);
1737 unsigned this_step_blocks = bi_size >> (SECTOR_SHIFT + ic->sb->log2_sectors_per_block);
1738 this_step_blocks = min(this_step_blocks, max_blocks);
1739 r = dm_integrity_rw_tag(ic, checksums, &dio->metadata_block, &dio->metadata_offset,
1740 this_step_blocks * ic->tag_size, TAG_WRITE);
1742 if (likely(checksums != checksums_onstack))
1747 /*if (bi_size < this_step_blocks << (SECTOR_SHIFT + ic->sb->log2_sectors_per_block)) {
1748 printk("BUGG: bi_sector: %llx, bi_size: %u\n", bi_sector, bi_size);
1749 printk("BUGG: this_step_blocks: %u\n", this_step_blocks);
1752 bi_size -= this_step_blocks << (SECTOR_SHIFT + ic->sb->log2_sectors_per_block);
1753 bi_sector += this_step_blocks << ic->sb->log2_sectors_per_block;
1756 if (likely(checksums != checksums_onstack))
1761 sector = dio->range.logical_sector;
1762 sectors_to_process = dio->range.n_sectors;
1764 __bio_for_each_segment(bv, bio, iter, dio->bio_details.bi_iter) {
1766 char *mem, *checksums_ptr;
1769 mem = (char *)kmap_atomic(bv.bv_page) + bv.bv_offset;
1771 checksums_ptr = checksums;
1773 integrity_sector_checksum(ic, sector, mem + pos, checksums_ptr);
1774 checksums_ptr += ic->tag_size;
1775 sectors_to_process -= ic->sectors_per_block;
1776 pos += ic->sectors_per_block << SECTOR_SHIFT;
1777 sector += ic->sectors_per_block;
1778 } while (pos < bv.bv_len && sectors_to_process && checksums != checksums_onstack);
1781 r = dm_integrity_rw_tag(ic, checksums, &dio->metadata_block, &dio->metadata_offset,
1782 checksums_ptr - checksums, dio->op == REQ_OP_READ ? TAG_CMP : TAG_WRITE);
1785 char b[BDEVNAME_SIZE];
1786 DMERR_LIMIT("%s: Checksum failed at sector 0x%llx", bio_devname(bio, b),
1787 (sector - ((r + ic->tag_size - 1) / ic->tag_size)));
1789 atomic64_inc(&ic->number_of_mismatches);
1791 if (likely(checksums != checksums_onstack))
1796 if (!sectors_to_process)
1799 if (unlikely(pos < bv.bv_len)) {
1800 bv.bv_offset += pos;
1806 if (likely(checksums != checksums_onstack))
1809 struct bio_integrity_payload *bip = dio->bio_details.bi_integrity;
1813 struct bvec_iter iter;
1814 unsigned data_to_process = dio->range.n_sectors;
1815 sector_to_block(ic, data_to_process);
1816 data_to_process *= ic->tag_size;
1818 bip_for_each_vec(biv, bip, iter) {
1822 BUG_ON(PageHighMem(biv.bv_page));
1823 tag = bvec_virt(&biv);
1824 this_len = min(biv.bv_len, data_to_process);
1825 r = dm_integrity_rw_tag(ic, tag, &dio->metadata_block, &dio->metadata_offset,
1826 this_len, dio->op == REQ_OP_READ ? TAG_READ : TAG_WRITE);
1829 data_to_process -= this_len;
1830 if (!data_to_process)
1839 dio->bi_status = errno_to_blk_status(r);
1843 static int dm_integrity_map(struct dm_target *ti, struct bio *bio)
1845 struct dm_integrity_c *ic = ti->private;
1846 struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
1847 struct bio_integrity_payload *bip;
1849 sector_t area, offset;
1853 dio->op = bio_op(bio);
1855 if (unlikely(dio->op == REQ_OP_DISCARD)) {
1856 if (ti->max_io_len) {
1857 sector_t sec = dm_target_offset(ti, bio->bi_iter.bi_sector);
1858 unsigned log2_max_io_len = __fls(ti->max_io_len);
1859 sector_t start_boundary = sec >> log2_max_io_len;
1860 sector_t end_boundary = (sec + bio_sectors(bio) - 1) >> log2_max_io_len;
1861 if (start_boundary < end_boundary) {
1862 sector_t len = ti->max_io_len - (sec & (ti->max_io_len - 1));
1863 dm_accept_partial_bio(bio, len);
1868 if (unlikely(bio->bi_opf & REQ_PREFLUSH)) {
1869 submit_flush_bio(ic, dio);
1870 return DM_MAPIO_SUBMITTED;
1873 dio->range.logical_sector = dm_target_offset(ti, bio->bi_iter.bi_sector);
1874 dio->fua = dio->op == REQ_OP_WRITE && bio->bi_opf & REQ_FUA;
1875 if (unlikely(dio->fua)) {
1877 * Don't pass down the FUA flag because we have to flush
1878 * disk cache anyway.
1880 bio->bi_opf &= ~REQ_FUA;
1882 if (unlikely(dio->range.logical_sector + bio_sectors(bio) > ic->provided_data_sectors)) {
1883 DMERR("Too big sector number: 0x%llx + 0x%x > 0x%llx",
1884 dio->range.logical_sector, bio_sectors(bio),
1885 ic->provided_data_sectors);
1886 return DM_MAPIO_KILL;
1888 if (unlikely((dio->range.logical_sector | bio_sectors(bio)) & (unsigned)(ic->sectors_per_block - 1))) {
1889 DMERR("Bio not aligned on %u sectors: 0x%llx, 0x%x",
1890 ic->sectors_per_block,
1891 dio->range.logical_sector, bio_sectors(bio));
1892 return DM_MAPIO_KILL;
1895 if (ic->sectors_per_block > 1 && likely(dio->op != REQ_OP_DISCARD)) {
1896 struct bvec_iter iter;
1898 bio_for_each_segment(bv, bio, iter) {
1899 if (unlikely(bv.bv_len & ((ic->sectors_per_block << SECTOR_SHIFT) - 1))) {
1900 DMERR("Bio vector (%u,%u) is not aligned on %u-sector boundary",
1901 bv.bv_offset, bv.bv_len, ic->sectors_per_block);
1902 return DM_MAPIO_KILL;
1907 bip = bio_integrity(bio);
1908 if (!ic->internal_hash) {
1910 unsigned wanted_tag_size = bio_sectors(bio) >> ic->sb->log2_sectors_per_block;
1911 if (ic->log2_tag_size >= 0)
1912 wanted_tag_size <<= ic->log2_tag_size;
1914 wanted_tag_size *= ic->tag_size;
1915 if (unlikely(wanted_tag_size != bip->bip_iter.bi_size)) {
1916 DMERR("Invalid integrity data size %u, expected %u",
1917 bip->bip_iter.bi_size, wanted_tag_size);
1918 return DM_MAPIO_KILL;
1922 if (unlikely(bip != NULL)) {
1923 DMERR("Unexpected integrity data when using internal hash");
1924 return DM_MAPIO_KILL;
1928 if (unlikely(ic->mode == 'R') && unlikely(dio->op != REQ_OP_READ))
1929 return DM_MAPIO_KILL;
1931 get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
1932 dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset, &dio->metadata_offset);
1933 bio->bi_iter.bi_sector = get_data_sector(ic, area, offset);
1935 dm_integrity_map_continue(dio, true);
1936 return DM_MAPIO_SUBMITTED;
1939 static bool __journal_read_write(struct dm_integrity_io *dio, struct bio *bio,
1940 unsigned journal_section, unsigned journal_entry)
1942 struct dm_integrity_c *ic = dio->ic;
1943 sector_t logical_sector;
1946 logical_sector = dio->range.logical_sector;
1947 n_sectors = dio->range.n_sectors;
1949 struct bio_vec bv = bio_iovec(bio);
1952 if (unlikely(bv.bv_len >> SECTOR_SHIFT > n_sectors))
1953 bv.bv_len = n_sectors << SECTOR_SHIFT;
1954 n_sectors -= bv.bv_len >> SECTOR_SHIFT;
1955 bio_advance_iter(bio, &bio->bi_iter, bv.bv_len);
1957 mem = kmap_atomic(bv.bv_page);
1958 if (likely(dio->op == REQ_OP_WRITE))
1959 flush_dcache_page(bv.bv_page);
1962 struct journal_entry *je = access_journal_entry(ic, journal_section, journal_entry);
1964 if (unlikely(dio->op == REQ_OP_READ)) {
1965 struct journal_sector *js;
1969 if (unlikely(journal_entry_is_inprogress(je))) {
1970 flush_dcache_page(bv.bv_page);
1973 __io_wait_event(ic->copy_to_journal_wait, !journal_entry_is_inprogress(je));
1977 BUG_ON(journal_entry_get_sector(je) != logical_sector);
1978 js = access_journal_data(ic, journal_section, journal_entry);
1979 mem_ptr = mem + bv.bv_offset;
1982 memcpy(mem_ptr, js, JOURNAL_SECTOR_DATA);
1983 *(commit_id_t *)(mem_ptr + JOURNAL_SECTOR_DATA) = je->last_bytes[s];
1985 mem_ptr += 1 << SECTOR_SHIFT;
1986 } while (++s < ic->sectors_per_block);
1987 #ifdef INTERNAL_VERIFY
1988 if (ic->internal_hash) {
1989 char checksums_onstack[max((size_t)HASH_MAX_DIGESTSIZE, MAX_TAG_SIZE)];
1991 integrity_sector_checksum(ic, logical_sector, mem + bv.bv_offset, checksums_onstack);
1992 if (unlikely(memcmp(checksums_onstack, journal_entry_tag(ic, je), ic->tag_size))) {
1993 DMERR_LIMIT("Checksum failed when reading from journal, at sector 0x%llx",
2000 if (!ic->internal_hash) {
2001 struct bio_integrity_payload *bip = bio_integrity(bio);
2002 unsigned tag_todo = ic->tag_size;
2003 char *tag_ptr = journal_entry_tag(ic, je);
2006 struct bio_vec biv = bvec_iter_bvec(bip->bip_vec, bip->bip_iter);
2007 unsigned tag_now = min(biv.bv_len, tag_todo);
2009 BUG_ON(PageHighMem(biv.bv_page));
2010 tag_addr = bvec_virt(&biv);
2011 if (likely(dio->op == REQ_OP_WRITE))
2012 memcpy(tag_ptr, tag_addr, tag_now);
2014 memcpy(tag_addr, tag_ptr, tag_now);
2015 bvec_iter_advance(bip->bip_vec, &bip->bip_iter, tag_now);
2017 tag_todo -= tag_now;
2018 } while (unlikely(tag_todo)); else {
2019 if (likely(dio->op == REQ_OP_WRITE))
2020 memset(tag_ptr, 0, tag_todo);
2024 if (likely(dio->op == REQ_OP_WRITE)) {
2025 struct journal_sector *js;
2028 js = access_journal_data(ic, journal_section, journal_entry);
2029 memcpy(js, mem + bv.bv_offset, ic->sectors_per_block << SECTOR_SHIFT);
2033 je->last_bytes[s] = js[s].commit_id;
2034 } while (++s < ic->sectors_per_block);
2036 if (ic->internal_hash) {
2037 unsigned digest_size = crypto_shash_digestsize(ic->internal_hash);
2038 if (unlikely(digest_size > ic->tag_size)) {
2039 char checksums_onstack[HASH_MAX_DIGESTSIZE];
2040 integrity_sector_checksum(ic, logical_sector, (char *)js, checksums_onstack);
2041 memcpy(journal_entry_tag(ic, je), checksums_onstack, ic->tag_size);
2043 integrity_sector_checksum(ic, logical_sector, (char *)js, journal_entry_tag(ic, je));
2046 journal_entry_set_sector(je, logical_sector);
2048 logical_sector += ic->sectors_per_block;
2051 if (unlikely(journal_entry == ic->journal_section_entries)) {
2054 wraparound_section(ic, &journal_section);
2057 bv.bv_offset += ic->sectors_per_block << SECTOR_SHIFT;
2058 } while (bv.bv_len -= ic->sectors_per_block << SECTOR_SHIFT);
2060 if (unlikely(dio->op == REQ_OP_READ))
2061 flush_dcache_page(bv.bv_page);
2063 } while (n_sectors);
2065 if (likely(dio->op == REQ_OP_WRITE)) {
2067 if (unlikely(waitqueue_active(&ic->copy_to_journal_wait)))
2068 wake_up(&ic->copy_to_journal_wait);
2069 if (READ_ONCE(ic->free_sectors) <= ic->free_sectors_threshold) {
2070 queue_work(ic->commit_wq, &ic->commit_work);
2072 schedule_autocommit(ic);
2075 remove_range(ic, &dio->range);
2078 if (unlikely(bio->bi_iter.bi_size)) {
2079 sector_t area, offset;
2081 dio->range.logical_sector = logical_sector;
2082 get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
2083 dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset, &dio->metadata_offset);
2090 static void dm_integrity_map_continue(struct dm_integrity_io *dio, bool from_map)
2092 struct dm_integrity_c *ic = dio->ic;
2093 struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
2094 unsigned journal_section, journal_entry;
2095 unsigned journal_read_pos;
2096 struct completion read_comp;
2097 bool discard_retried = false;
2098 bool need_sync_io = ic->internal_hash && dio->op == REQ_OP_READ;
2099 if (unlikely(dio->op == REQ_OP_DISCARD) && ic->mode != 'D')
2100 need_sync_io = true;
2102 if (need_sync_io && from_map) {
2103 INIT_WORK(&dio->work, integrity_bio_wait);
2104 queue_work(ic->offload_wq, &dio->work);
2109 spin_lock_irq(&ic->endio_wait.lock);
2111 if (unlikely(dm_integrity_failed(ic))) {
2112 spin_unlock_irq(&ic->endio_wait.lock);
2116 dio->range.n_sectors = bio_sectors(bio);
2117 journal_read_pos = NOT_FOUND;
2118 if (ic->mode == 'J' && likely(dio->op != REQ_OP_DISCARD)) {
2119 if (dio->op == REQ_OP_WRITE) {
2120 unsigned next_entry, i, pos;
2121 unsigned ws, we, range_sectors;
2123 dio->range.n_sectors = min(dio->range.n_sectors,
2124 (sector_t)ic->free_sectors << ic->sb->log2_sectors_per_block);
2125 if (unlikely(!dio->range.n_sectors)) {
2127 goto offload_to_thread;
2128 sleep_on_endio_wait(ic);
2131 range_sectors = dio->range.n_sectors >> ic->sb->log2_sectors_per_block;
2132 ic->free_sectors -= range_sectors;
2133 journal_section = ic->free_section;
2134 journal_entry = ic->free_section_entry;
2136 next_entry = ic->free_section_entry + range_sectors;
2137 ic->free_section_entry = next_entry % ic->journal_section_entries;
2138 ic->free_section += next_entry / ic->journal_section_entries;
2139 ic->n_uncommitted_sections += next_entry / ic->journal_section_entries;
2140 wraparound_section(ic, &ic->free_section);
2142 pos = journal_section * ic->journal_section_entries + journal_entry;
2143 ws = journal_section;
2147 struct journal_entry *je;
2149 add_journal_node(ic, &ic->journal_tree[pos], dio->range.logical_sector + i);
2151 if (unlikely(pos >= ic->journal_entries))
2154 je = access_journal_entry(ic, ws, we);
2155 BUG_ON(!journal_entry_is_unused(je));
2156 journal_entry_set_inprogress(je);
2158 if (unlikely(we == ic->journal_section_entries)) {
2161 wraparound_section(ic, &ws);
2163 } while ((i += ic->sectors_per_block) < dio->range.n_sectors);
2165 spin_unlock_irq(&ic->endio_wait.lock);
2166 goto journal_read_write;
2168 sector_t next_sector;
2169 journal_read_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
2170 if (likely(journal_read_pos == NOT_FOUND)) {
2171 if (unlikely(dio->range.n_sectors > next_sector - dio->range.logical_sector))
2172 dio->range.n_sectors = next_sector - dio->range.logical_sector;
2175 unsigned jp = journal_read_pos + 1;
2176 for (i = ic->sectors_per_block; i < dio->range.n_sectors; i += ic->sectors_per_block, jp++) {
2177 if (!test_journal_node(ic, jp, dio->range.logical_sector + i))
2180 dio->range.n_sectors = i;
2184 if (unlikely(!add_new_range(ic, &dio->range, true))) {
2186 * We must not sleep in the request routine because it could
2187 * stall bios on current->bio_list.
2188 * So, we offload the bio to a workqueue if we have to sleep.
2192 spin_unlock_irq(&ic->endio_wait.lock);
2193 INIT_WORK(&dio->work, integrity_bio_wait);
2194 queue_work(ic->wait_wq, &dio->work);
2197 if (journal_read_pos != NOT_FOUND)
2198 dio->range.n_sectors = ic->sectors_per_block;
2199 wait_and_add_new_range(ic, &dio->range);
2201 * wait_and_add_new_range drops the spinlock, so the journal
2202 * may have been changed arbitrarily. We need to recheck.
2203 * To simplify the code, we restrict I/O size to just one block.
2205 if (journal_read_pos != NOT_FOUND) {
2206 sector_t next_sector;
2207 unsigned new_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
2208 if (unlikely(new_pos != journal_read_pos)) {
2209 remove_range_unlocked(ic, &dio->range);
2214 if (ic->mode == 'J' && likely(dio->op == REQ_OP_DISCARD) && !discard_retried) {
2215 sector_t next_sector;
2216 unsigned new_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
2217 if (unlikely(new_pos != NOT_FOUND) ||
2218 unlikely(next_sector < dio->range.logical_sector - dio->range.n_sectors)) {
2219 remove_range_unlocked(ic, &dio->range);
2220 spin_unlock_irq(&ic->endio_wait.lock);
2221 queue_work(ic->commit_wq, &ic->commit_work);
2222 flush_workqueue(ic->commit_wq);
2223 queue_work(ic->writer_wq, &ic->writer_work);
2224 flush_workqueue(ic->writer_wq);
2225 discard_retried = true;
2229 spin_unlock_irq(&ic->endio_wait.lock);
2231 if (unlikely(journal_read_pos != NOT_FOUND)) {
2232 journal_section = journal_read_pos / ic->journal_section_entries;
2233 journal_entry = journal_read_pos % ic->journal_section_entries;
2234 goto journal_read_write;
2237 if (ic->mode == 'B' && (dio->op == REQ_OP_WRITE || unlikely(dio->op == REQ_OP_DISCARD))) {
2238 if (!block_bitmap_op(ic, ic->may_write_bitmap, dio->range.logical_sector,
2239 dio->range.n_sectors, BITMAP_OP_TEST_ALL_SET)) {
2240 struct bitmap_block_status *bbs;
2242 bbs = sector_to_bitmap_block(ic, dio->range.logical_sector);
2243 spin_lock(&bbs->bio_queue_lock);
2244 bio_list_add(&bbs->bio_queue, bio);
2245 spin_unlock(&bbs->bio_queue_lock);
2246 queue_work(ic->writer_wq, &bbs->work);
2251 dio->in_flight = (atomic_t)ATOMIC_INIT(2);
2254 init_completion(&read_comp);
2255 dio->completion = &read_comp;
2257 dio->completion = NULL;
2259 dm_bio_record(&dio->bio_details, bio);
2260 bio_set_dev(bio, ic->dev->bdev);
2261 bio->bi_integrity = NULL;
2262 bio->bi_opf &= ~REQ_INTEGRITY;
2263 bio->bi_end_io = integrity_end_io;
2264 bio->bi_iter.bi_size = dio->range.n_sectors << SECTOR_SHIFT;
2266 if (unlikely(dio->op == REQ_OP_DISCARD) && likely(ic->mode != 'D')) {
2267 integrity_metadata(&dio->work);
2268 dm_integrity_flush_buffers(ic, false);
2270 dio->in_flight = (atomic_t)ATOMIC_INIT(1);
2271 dio->completion = NULL;
2273 submit_bio_noacct(bio);
2278 submit_bio_noacct(bio);
2281 wait_for_completion_io(&read_comp);
2282 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING) &&
2283 dio->range.logical_sector + dio->range.n_sectors > le64_to_cpu(ic->sb->recalc_sector))
2285 if (ic->mode == 'B') {
2286 if (!block_bitmap_op(ic, ic->recalc_bitmap, dio->range.logical_sector,
2287 dio->range.n_sectors, BITMAP_OP_TEST_ALL_CLEAR))
2291 if (likely(!bio->bi_status))
2292 integrity_metadata(&dio->work);
2298 INIT_WORK(&dio->work, integrity_metadata);
2299 queue_work(ic->metadata_wq, &dio->work);
2305 if (unlikely(__journal_read_write(dio, bio, journal_section, journal_entry)))
2308 do_endio_flush(ic, dio);
2312 static void integrity_bio_wait(struct work_struct *w)
2314 struct dm_integrity_io *dio = container_of(w, struct dm_integrity_io, work);
2316 dm_integrity_map_continue(dio, false);
2319 static void pad_uncommitted(struct dm_integrity_c *ic)
2321 if (ic->free_section_entry) {
2322 ic->free_sectors -= ic->journal_section_entries - ic->free_section_entry;
2323 ic->free_section_entry = 0;
2325 wraparound_section(ic, &ic->free_section);
2326 ic->n_uncommitted_sections++;
2328 if (WARN_ON(ic->journal_sections * ic->journal_section_entries !=
2329 (ic->n_uncommitted_sections + ic->n_committed_sections) *
2330 ic->journal_section_entries + ic->free_sectors)) {
2331 DMCRIT("journal_sections %u, journal_section_entries %u, "
2332 "n_uncommitted_sections %u, n_committed_sections %u, "
2333 "journal_section_entries %u, free_sectors %u",
2334 ic->journal_sections, ic->journal_section_entries,
2335 ic->n_uncommitted_sections, ic->n_committed_sections,
2336 ic->journal_section_entries, ic->free_sectors);
2340 static void integrity_commit(struct work_struct *w)
2342 struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, commit_work);
2343 unsigned commit_start, commit_sections;
2345 struct bio *flushes;
2347 del_timer(&ic->autocommit_timer);
2349 spin_lock_irq(&ic->endio_wait.lock);
2350 flushes = bio_list_get(&ic->flush_bio_list);
2351 if (unlikely(ic->mode != 'J')) {
2352 spin_unlock_irq(&ic->endio_wait.lock);
2353 dm_integrity_flush_buffers(ic, true);
2354 goto release_flush_bios;
2357 pad_uncommitted(ic);
2358 commit_start = ic->uncommitted_section;
2359 commit_sections = ic->n_uncommitted_sections;
2360 spin_unlock_irq(&ic->endio_wait.lock);
2362 if (!commit_sections)
2363 goto release_flush_bios;
2365 ic->wrote_to_journal = true;
2368 for (n = 0; n < commit_sections; n++) {
2369 for (j = 0; j < ic->journal_section_entries; j++) {
2370 struct journal_entry *je;
2371 je = access_journal_entry(ic, i, j);
2372 io_wait_event(ic->copy_to_journal_wait, !journal_entry_is_inprogress(je));
2374 for (j = 0; j < ic->journal_section_sectors; j++) {
2375 struct journal_sector *js;
2376 js = access_journal(ic, i, j);
2377 js->commit_id = dm_integrity_commit_id(ic, i, j, ic->commit_seq);
2380 if (unlikely(i >= ic->journal_sections))
2381 ic->commit_seq = next_commit_seq(ic->commit_seq);
2382 wraparound_section(ic, &i);
2386 write_journal(ic, commit_start, commit_sections);
2388 spin_lock_irq(&ic->endio_wait.lock);
2389 ic->uncommitted_section += commit_sections;
2390 wraparound_section(ic, &ic->uncommitted_section);
2391 ic->n_uncommitted_sections -= commit_sections;
2392 ic->n_committed_sections += commit_sections;
2393 spin_unlock_irq(&ic->endio_wait.lock);
2395 if (READ_ONCE(ic->free_sectors) <= ic->free_sectors_threshold)
2396 queue_work(ic->writer_wq, &ic->writer_work);
2400 struct bio *next = flushes->bi_next;
2401 flushes->bi_next = NULL;
2402 do_endio(ic, flushes);
2407 static void complete_copy_from_journal(unsigned long error, void *context)
2409 struct journal_io *io = context;
2410 struct journal_completion *comp = io->comp;
2411 struct dm_integrity_c *ic = comp->ic;
2412 remove_range(ic, &io->range);
2413 mempool_free(io, &ic->journal_io_mempool);
2414 if (unlikely(error != 0))
2415 dm_integrity_io_error(ic, "copying from journal", -EIO);
2416 complete_journal_op(comp);
2419 static void restore_last_bytes(struct dm_integrity_c *ic, struct journal_sector *js,
2420 struct journal_entry *je)
2424 js->commit_id = je->last_bytes[s];
2426 } while (++s < ic->sectors_per_block);
2429 static void do_journal_write(struct dm_integrity_c *ic, unsigned write_start,
2430 unsigned write_sections, bool from_replay)
2433 struct journal_completion comp;
2434 struct blk_plug plug;
2436 blk_start_plug(&plug);
2439 comp.in_flight = (atomic_t)ATOMIC_INIT(1);
2440 init_completion(&comp.comp);
2443 for (n = 0; n < write_sections; n++, i++, wraparound_section(ic, &i)) {
2444 #ifndef INTERNAL_VERIFY
2445 if (unlikely(from_replay))
2447 rw_section_mac(ic, i, false);
2448 for (j = 0; j < ic->journal_section_entries; j++) {
2449 struct journal_entry *je = access_journal_entry(ic, i, j);
2450 sector_t sec, area, offset;
2451 unsigned k, l, next_loop;
2452 sector_t metadata_block;
2453 unsigned metadata_offset;
2454 struct journal_io *io;
2456 if (journal_entry_is_unused(je))
2458 BUG_ON(unlikely(journal_entry_is_inprogress(je)) && !from_replay);
2459 sec = journal_entry_get_sector(je);
2460 if (unlikely(from_replay)) {
2461 if (unlikely(sec & (unsigned)(ic->sectors_per_block - 1))) {
2462 dm_integrity_io_error(ic, "invalid sector in journal", -EIO);
2463 sec &= ~(sector_t)(ic->sectors_per_block - 1);
2465 if (unlikely(sec >= ic->provided_data_sectors)) {
2466 journal_entry_set_unused(je);
2470 get_area_and_offset(ic, sec, &area, &offset);
2471 restore_last_bytes(ic, access_journal_data(ic, i, j), je);
2472 for (k = j + 1; k < ic->journal_section_entries; k++) {
2473 struct journal_entry *je2 = access_journal_entry(ic, i, k);
2474 sector_t sec2, area2, offset2;
2475 if (journal_entry_is_unused(je2))
2477 BUG_ON(unlikely(journal_entry_is_inprogress(je2)) && !from_replay);
2478 sec2 = journal_entry_get_sector(je2);
2479 if (unlikely(sec2 >= ic->provided_data_sectors))
2481 get_area_and_offset(ic, sec2, &area2, &offset2);
2482 if (area2 != area || offset2 != offset + ((k - j) << ic->sb->log2_sectors_per_block))
2484 restore_last_bytes(ic, access_journal_data(ic, i, k), je2);
2488 io = mempool_alloc(&ic->journal_io_mempool, GFP_NOIO);
2490 io->range.logical_sector = sec;
2491 io->range.n_sectors = (k - j) << ic->sb->log2_sectors_per_block;
2493 spin_lock_irq(&ic->endio_wait.lock);
2494 add_new_range_and_wait(ic, &io->range);
2496 if (likely(!from_replay)) {
2497 struct journal_node *section_node = &ic->journal_tree[i * ic->journal_section_entries];
2499 /* don't write if there is newer committed sector */
2500 while (j < k && find_newer_committed_node(ic, §ion_node[j])) {
2501 struct journal_entry *je2 = access_journal_entry(ic, i, j);
2503 journal_entry_set_unused(je2);
2504 remove_journal_node(ic, §ion_node[j]);
2506 sec += ic->sectors_per_block;
2507 offset += ic->sectors_per_block;
2509 while (j < k && find_newer_committed_node(ic, §ion_node[k - 1])) {
2510 struct journal_entry *je2 = access_journal_entry(ic, i, k - 1);
2512 journal_entry_set_unused(je2);
2513 remove_journal_node(ic, §ion_node[k - 1]);
2517 remove_range_unlocked(ic, &io->range);
2518 spin_unlock_irq(&ic->endio_wait.lock);
2519 mempool_free(io, &ic->journal_io_mempool);
2522 for (l = j; l < k; l++) {
2523 remove_journal_node(ic, §ion_node[l]);
2526 spin_unlock_irq(&ic->endio_wait.lock);
2528 metadata_block = get_metadata_sector_and_offset(ic, area, offset, &metadata_offset);
2529 for (l = j; l < k; l++) {
2531 struct journal_entry *je2 = access_journal_entry(ic, i, l);
2534 #ifndef INTERNAL_VERIFY
2535 unlikely(from_replay) &&
2537 ic->internal_hash) {
2538 char test_tag[max_t(size_t, HASH_MAX_DIGESTSIZE, MAX_TAG_SIZE)];
2540 integrity_sector_checksum(ic, sec + ((l - j) << ic->sb->log2_sectors_per_block),
2541 (char *)access_journal_data(ic, i, l), test_tag);
2542 if (unlikely(memcmp(test_tag, journal_entry_tag(ic, je2), ic->tag_size)))
2543 dm_integrity_io_error(ic, "tag mismatch when replaying journal", -EILSEQ);
2546 journal_entry_set_unused(je2);
2547 r = dm_integrity_rw_tag(ic, journal_entry_tag(ic, je2), &metadata_block, &metadata_offset,
2548 ic->tag_size, TAG_WRITE);
2550 dm_integrity_io_error(ic, "reading tags", r);
2554 atomic_inc(&comp.in_flight);
2555 copy_from_journal(ic, i, j << ic->sb->log2_sectors_per_block,
2556 (k - j) << ic->sb->log2_sectors_per_block,
2557 get_data_sector(ic, area, offset),
2558 complete_copy_from_journal, io);
2564 dm_bufio_write_dirty_buffers_async(ic->bufio);
2566 blk_finish_plug(&plug);
2568 complete_journal_op(&comp);
2569 wait_for_completion_io(&comp.comp);
2571 dm_integrity_flush_buffers(ic, true);
2574 static void integrity_writer(struct work_struct *w)
2576 struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, writer_work);
2577 unsigned write_start, write_sections;
2579 unsigned prev_free_sectors;
2581 spin_lock_irq(&ic->endio_wait.lock);
2582 write_start = ic->committed_section;
2583 write_sections = ic->n_committed_sections;
2584 spin_unlock_irq(&ic->endio_wait.lock);
2586 if (!write_sections)
2589 do_journal_write(ic, write_start, write_sections, false);
2591 spin_lock_irq(&ic->endio_wait.lock);
2593 ic->committed_section += write_sections;
2594 wraparound_section(ic, &ic->committed_section);
2595 ic->n_committed_sections -= write_sections;
2597 prev_free_sectors = ic->free_sectors;
2598 ic->free_sectors += write_sections * ic->journal_section_entries;
2599 if (unlikely(!prev_free_sectors))
2600 wake_up_locked(&ic->endio_wait);
2602 spin_unlock_irq(&ic->endio_wait.lock);
2605 static void recalc_write_super(struct dm_integrity_c *ic)
2609 dm_integrity_flush_buffers(ic, false);
2610 if (dm_integrity_failed(ic))
2613 r = sync_rw_sb(ic, REQ_OP_WRITE, 0);
2615 dm_integrity_io_error(ic, "writing superblock", r);
2618 static void integrity_recalc(struct work_struct *w)
2620 struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, recalc_work);
2621 struct dm_integrity_range range;
2622 struct dm_io_request io_req;
2623 struct dm_io_region io_loc;
2624 sector_t area, offset;
2625 sector_t metadata_block;
2626 unsigned metadata_offset;
2627 sector_t logical_sector, n_sectors;
2631 unsigned super_counter = 0;
2633 DEBUG_print("start recalculation... (position %llx)\n", le64_to_cpu(ic->sb->recalc_sector));
2635 spin_lock_irq(&ic->endio_wait.lock);
2639 if (unlikely(dm_post_suspending(ic->ti)))
2642 range.logical_sector = le64_to_cpu(ic->sb->recalc_sector);
2643 if (unlikely(range.logical_sector >= ic->provided_data_sectors)) {
2644 if (ic->mode == 'B') {
2645 block_bitmap_op(ic, ic->recalc_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_CLEAR);
2646 DEBUG_print("queue_delayed_work: bitmap_flush_work\n");
2647 queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, 0);
2652 get_area_and_offset(ic, range.logical_sector, &area, &offset);
2653 range.n_sectors = min((sector_t)RECALC_SECTORS, ic->provided_data_sectors - range.logical_sector);
2655 range.n_sectors = min(range.n_sectors, ((sector_t)1U << ic->sb->log2_interleave_sectors) - (unsigned)offset);
2657 add_new_range_and_wait(ic, &range);
2658 spin_unlock_irq(&ic->endio_wait.lock);
2659 logical_sector = range.logical_sector;
2660 n_sectors = range.n_sectors;
2662 if (ic->mode == 'B') {
2663 if (block_bitmap_op(ic, ic->recalc_bitmap, logical_sector, n_sectors, BITMAP_OP_TEST_ALL_CLEAR)) {
2664 goto advance_and_next;
2666 while (block_bitmap_op(ic, ic->recalc_bitmap, logical_sector,
2667 ic->sectors_per_block, BITMAP_OP_TEST_ALL_CLEAR)) {
2668 logical_sector += ic->sectors_per_block;
2669 n_sectors -= ic->sectors_per_block;
2672 while (block_bitmap_op(ic, ic->recalc_bitmap, logical_sector + n_sectors - ic->sectors_per_block,
2673 ic->sectors_per_block, BITMAP_OP_TEST_ALL_CLEAR)) {
2674 n_sectors -= ic->sectors_per_block;
2677 get_area_and_offset(ic, logical_sector, &area, &offset);
2680 DEBUG_print("recalculating: %llx, %llx\n", logical_sector, n_sectors);
2682 if (unlikely(++super_counter == RECALC_WRITE_SUPER)) {
2683 recalc_write_super(ic);
2684 if (ic->mode == 'B') {
2685 queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, ic->bitmap_flush_interval);
2690 if (unlikely(dm_integrity_failed(ic)))
2693 io_req.bi_op = REQ_OP_READ;
2694 io_req.bi_op_flags = 0;
2695 io_req.mem.type = DM_IO_VMA;
2696 io_req.mem.ptr.addr = ic->recalc_buffer;
2697 io_req.notify.fn = NULL;
2698 io_req.client = ic->io;
2699 io_loc.bdev = ic->dev->bdev;
2700 io_loc.sector = get_data_sector(ic, area, offset);
2701 io_loc.count = n_sectors;
2703 r = dm_io(&io_req, 1, &io_loc, NULL);
2705 dm_integrity_io_error(ic, "reading data", r);
2709 t = ic->recalc_tags;
2710 for (i = 0; i < n_sectors; i += ic->sectors_per_block) {
2711 integrity_sector_checksum(ic, logical_sector + i, ic->recalc_buffer + (i << SECTOR_SHIFT), t);
2715 metadata_block = get_metadata_sector_and_offset(ic, area, offset, &metadata_offset);
2717 r = dm_integrity_rw_tag(ic, ic->recalc_tags, &metadata_block, &metadata_offset, t - ic->recalc_tags, TAG_WRITE);
2719 dm_integrity_io_error(ic, "writing tags", r);
2723 if (ic->mode == 'B') {
2724 sector_t start, end;
2725 start = (range.logical_sector >>
2726 (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit)) <<
2727 (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
2728 end = ((range.logical_sector + range.n_sectors) >>
2729 (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit)) <<
2730 (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
2731 block_bitmap_op(ic, ic->recalc_bitmap, start, end - start, BITMAP_OP_CLEAR);
2737 spin_lock_irq(&ic->endio_wait.lock);
2738 remove_range_unlocked(ic, &range);
2739 ic->sb->recalc_sector = cpu_to_le64(range.logical_sector + range.n_sectors);
2743 remove_range(ic, &range);
2747 spin_unlock_irq(&ic->endio_wait.lock);
2749 recalc_write_super(ic);
2752 static void bitmap_block_work(struct work_struct *w)
2754 struct bitmap_block_status *bbs = container_of(w, struct bitmap_block_status, work);
2755 struct dm_integrity_c *ic = bbs->ic;
2757 struct bio_list bio_queue;
2758 struct bio_list waiting;
2760 bio_list_init(&waiting);
2762 spin_lock(&bbs->bio_queue_lock);
2763 bio_queue = bbs->bio_queue;
2764 bio_list_init(&bbs->bio_queue);
2765 spin_unlock(&bbs->bio_queue_lock);
2767 while ((bio = bio_list_pop(&bio_queue))) {
2768 struct dm_integrity_io *dio;
2770 dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
2772 if (block_bitmap_op(ic, ic->may_write_bitmap, dio->range.logical_sector,
2773 dio->range.n_sectors, BITMAP_OP_TEST_ALL_SET)) {
2774 remove_range(ic, &dio->range);
2775 INIT_WORK(&dio->work, integrity_bio_wait);
2776 queue_work(ic->offload_wq, &dio->work);
2778 block_bitmap_op(ic, ic->journal, dio->range.logical_sector,
2779 dio->range.n_sectors, BITMAP_OP_SET);
2780 bio_list_add(&waiting, bio);
2784 if (bio_list_empty(&waiting))
2787 rw_journal_sectors(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC,
2788 bbs->idx * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT),
2789 BITMAP_BLOCK_SIZE >> SECTOR_SHIFT, NULL);
2791 while ((bio = bio_list_pop(&waiting))) {
2792 struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
2794 block_bitmap_op(ic, ic->may_write_bitmap, dio->range.logical_sector,
2795 dio->range.n_sectors, BITMAP_OP_SET);
2797 remove_range(ic, &dio->range);
2798 INIT_WORK(&dio->work, integrity_bio_wait);
2799 queue_work(ic->offload_wq, &dio->work);
2802 queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, ic->bitmap_flush_interval);
2805 static void bitmap_flush_work(struct work_struct *work)
2807 struct dm_integrity_c *ic = container_of(work, struct dm_integrity_c, bitmap_flush_work.work);
2808 struct dm_integrity_range range;
2809 unsigned long limit;
2812 dm_integrity_flush_buffers(ic, false);
2814 range.logical_sector = 0;
2815 range.n_sectors = ic->provided_data_sectors;
2817 spin_lock_irq(&ic->endio_wait.lock);
2818 add_new_range_and_wait(ic, &range);
2819 spin_unlock_irq(&ic->endio_wait.lock);
2821 dm_integrity_flush_buffers(ic, true);
2823 limit = ic->provided_data_sectors;
2824 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
2825 limit = le64_to_cpu(ic->sb->recalc_sector)
2826 >> (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit)
2827 << (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
2829 /*DEBUG_print("zeroing journal\n");*/
2830 block_bitmap_op(ic, ic->journal, 0, limit, BITMAP_OP_CLEAR);
2831 block_bitmap_op(ic, ic->may_write_bitmap, 0, limit, BITMAP_OP_CLEAR);
2833 rw_journal_sectors(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, 0,
2834 ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
2836 spin_lock_irq(&ic->endio_wait.lock);
2837 remove_range_unlocked(ic, &range);
2838 while (unlikely((bio = bio_list_pop(&ic->synchronous_bios)) != NULL)) {
2840 spin_unlock_irq(&ic->endio_wait.lock);
2841 spin_lock_irq(&ic->endio_wait.lock);
2843 spin_unlock_irq(&ic->endio_wait.lock);
2847 static void init_journal(struct dm_integrity_c *ic, unsigned start_section,
2848 unsigned n_sections, unsigned char commit_seq)
2855 for (n = 0; n < n_sections; n++) {
2856 i = start_section + n;
2857 wraparound_section(ic, &i);
2858 for (j = 0; j < ic->journal_section_sectors; j++) {
2859 struct journal_sector *js = access_journal(ic, i, j);
2860 memset(&js->entries, 0, JOURNAL_SECTOR_DATA);
2861 js->commit_id = dm_integrity_commit_id(ic, i, j, commit_seq);
2863 for (j = 0; j < ic->journal_section_entries; j++) {
2864 struct journal_entry *je = access_journal_entry(ic, i, j);
2865 journal_entry_set_unused(je);
2869 write_journal(ic, start_section, n_sections);
2872 static int find_commit_seq(struct dm_integrity_c *ic, unsigned i, unsigned j, commit_id_t id)
2875 for (k = 0; k < N_COMMIT_IDS; k++) {
2876 if (dm_integrity_commit_id(ic, i, j, k) == id)
2879 dm_integrity_io_error(ic, "journal commit id", -EIO);
2883 static void replay_journal(struct dm_integrity_c *ic)
2886 bool used_commit_ids[N_COMMIT_IDS];
2887 unsigned max_commit_id_sections[N_COMMIT_IDS];
2888 unsigned write_start, write_sections;
2889 unsigned continue_section;
2891 unsigned char unused, last_used, want_commit_seq;
2893 if (ic->mode == 'R')
2896 if (ic->journal_uptodate)
2902 if (!ic->just_formatted) {
2903 DEBUG_print("reading journal\n");
2904 rw_journal(ic, REQ_OP_READ, 0, 0, ic->journal_sections, NULL);
2906 DEBUG_bytes(lowmem_page_address(ic->journal_io[0].page), 64, "read journal");
2907 if (ic->journal_io) {
2908 struct journal_completion crypt_comp;
2910 init_completion(&crypt_comp.comp);
2911 crypt_comp.in_flight = (atomic_t)ATOMIC_INIT(0);
2912 encrypt_journal(ic, false, 0, ic->journal_sections, &crypt_comp);
2913 wait_for_completion(&crypt_comp.comp);
2915 DEBUG_bytes(lowmem_page_address(ic->journal[0].page), 64, "decrypted journal");
2918 if (dm_integrity_failed(ic))
2921 journal_empty = true;
2922 memset(used_commit_ids, 0, sizeof used_commit_ids);
2923 memset(max_commit_id_sections, 0, sizeof max_commit_id_sections);
2924 for (i = 0; i < ic->journal_sections; i++) {
2925 for (j = 0; j < ic->journal_section_sectors; j++) {
2927 struct journal_sector *js = access_journal(ic, i, j);
2928 k = find_commit_seq(ic, i, j, js->commit_id);
2931 used_commit_ids[k] = true;
2932 max_commit_id_sections[k] = i;
2934 if (journal_empty) {
2935 for (j = 0; j < ic->journal_section_entries; j++) {
2936 struct journal_entry *je = access_journal_entry(ic, i, j);
2937 if (!journal_entry_is_unused(je)) {
2938 journal_empty = false;
2945 if (!used_commit_ids[N_COMMIT_IDS - 1]) {
2946 unused = N_COMMIT_IDS - 1;
2947 while (unused && !used_commit_ids[unused - 1])
2950 for (unused = 0; unused < N_COMMIT_IDS; unused++)
2951 if (!used_commit_ids[unused])
2953 if (unused == N_COMMIT_IDS) {
2954 dm_integrity_io_error(ic, "journal commit ids", -EIO);
2958 DEBUG_print("first unused commit seq %d [%d,%d,%d,%d]\n",
2959 unused, used_commit_ids[0], used_commit_ids[1],
2960 used_commit_ids[2], used_commit_ids[3]);
2962 last_used = prev_commit_seq(unused);
2963 want_commit_seq = prev_commit_seq(last_used);
2965 if (!used_commit_ids[want_commit_seq] && used_commit_ids[prev_commit_seq(want_commit_seq)])
2966 journal_empty = true;
2968 write_start = max_commit_id_sections[last_used] + 1;
2969 if (unlikely(write_start >= ic->journal_sections))
2970 want_commit_seq = next_commit_seq(want_commit_seq);
2971 wraparound_section(ic, &write_start);
2974 for (write_sections = 0; write_sections < ic->journal_sections; write_sections++) {
2975 for (j = 0; j < ic->journal_section_sectors; j++) {
2976 struct journal_sector *js = access_journal(ic, i, j);
2978 if (js->commit_id != dm_integrity_commit_id(ic, i, j, want_commit_seq)) {
2980 * This could be caused by crash during writing.
2981 * We won't replay the inconsistent part of the
2984 DEBUG_print("commit id mismatch at position (%u, %u): %d != %d\n",
2985 i, j, find_commit_seq(ic, i, j, js->commit_id), want_commit_seq);
2990 if (unlikely(i >= ic->journal_sections))
2991 want_commit_seq = next_commit_seq(want_commit_seq);
2992 wraparound_section(ic, &i);
2996 if (!journal_empty) {
2997 DEBUG_print("replaying %u sections, starting at %u, commit seq %d\n",
2998 write_sections, write_start, want_commit_seq);
2999 do_journal_write(ic, write_start, write_sections, true);
3002 if (write_sections == ic->journal_sections && (ic->mode == 'J' || journal_empty)) {
3003 continue_section = write_start;
3004 ic->commit_seq = want_commit_seq;
3005 DEBUG_print("continuing from section %u, commit seq %d\n", write_start, ic->commit_seq);
3008 unsigned char erase_seq;
3010 DEBUG_print("clearing journal\n");
3012 erase_seq = prev_commit_seq(prev_commit_seq(last_used));
3014 init_journal(ic, s, 1, erase_seq);
3016 wraparound_section(ic, &s);
3017 if (ic->journal_sections >= 2) {
3018 init_journal(ic, s, ic->journal_sections - 2, erase_seq);
3019 s += ic->journal_sections - 2;
3020 wraparound_section(ic, &s);
3021 init_journal(ic, s, 1, erase_seq);
3024 continue_section = 0;
3025 ic->commit_seq = next_commit_seq(erase_seq);
3028 ic->committed_section = continue_section;
3029 ic->n_committed_sections = 0;
3031 ic->uncommitted_section = continue_section;
3032 ic->n_uncommitted_sections = 0;
3034 ic->free_section = continue_section;
3035 ic->free_section_entry = 0;
3036 ic->free_sectors = ic->journal_entries;
3038 ic->journal_tree_root = RB_ROOT;
3039 for (i = 0; i < ic->journal_entries; i++)
3040 init_journal_node(&ic->journal_tree[i]);
3043 static void dm_integrity_enter_synchronous_mode(struct dm_integrity_c *ic)
3045 DEBUG_print("dm_integrity_enter_synchronous_mode\n");
3047 if (ic->mode == 'B') {
3048 ic->bitmap_flush_interval = msecs_to_jiffies(10) + 1;
3049 ic->synchronous_mode = 1;
3051 cancel_delayed_work_sync(&ic->bitmap_flush_work);
3052 queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, 0);
3053 flush_workqueue(ic->commit_wq);
3057 static int dm_integrity_reboot(struct notifier_block *n, unsigned long code, void *x)
3059 struct dm_integrity_c *ic = container_of(n, struct dm_integrity_c, reboot_notifier);
3061 DEBUG_print("dm_integrity_reboot\n");
3063 dm_integrity_enter_synchronous_mode(ic);
3068 static void dm_integrity_postsuspend(struct dm_target *ti)
3070 struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
3073 WARN_ON(unregister_reboot_notifier(&ic->reboot_notifier));
3075 del_timer_sync(&ic->autocommit_timer);
3078 drain_workqueue(ic->recalc_wq);
3080 if (ic->mode == 'B')
3081 cancel_delayed_work_sync(&ic->bitmap_flush_work);
3083 queue_work(ic->commit_wq, &ic->commit_work);
3084 drain_workqueue(ic->commit_wq);
3086 if (ic->mode == 'J') {
3087 queue_work(ic->writer_wq, &ic->writer_work);
3088 drain_workqueue(ic->writer_wq);
3089 dm_integrity_flush_buffers(ic, true);
3090 if (ic->wrote_to_journal) {
3091 init_journal(ic, ic->free_section,
3092 ic->journal_sections - ic->free_section, ic->commit_seq);
3093 if (ic->free_section) {
3094 init_journal(ic, 0, ic->free_section,
3095 next_commit_seq(ic->commit_seq));
3100 if (ic->mode == 'B') {
3101 dm_integrity_flush_buffers(ic, true);
3103 /* set to 0 to test bitmap replay code */
3104 init_journal(ic, 0, ic->journal_sections, 0);
3105 ic->sb->flags &= ~cpu_to_le32(SB_FLAG_DIRTY_BITMAP);
3106 r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
3108 dm_integrity_io_error(ic, "writing superblock", r);
3112 BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
3114 ic->journal_uptodate = true;
3117 static void dm_integrity_resume(struct dm_target *ti)
3119 struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
3120 __u64 old_provided_data_sectors = le64_to_cpu(ic->sb->provided_data_sectors);
3123 DEBUG_print("resume\n");
3125 ic->wrote_to_journal = false;
3127 if (ic->provided_data_sectors != old_provided_data_sectors) {
3128 if (ic->provided_data_sectors > old_provided_data_sectors &&
3130 ic->sb->log2_blocks_per_bitmap_bit == ic->log2_blocks_per_bitmap_bit) {
3131 rw_journal_sectors(ic, REQ_OP_READ, 0, 0,
3132 ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3133 block_bitmap_op(ic, ic->journal, old_provided_data_sectors,
3134 ic->provided_data_sectors - old_provided_data_sectors, BITMAP_OP_SET);
3135 rw_journal_sectors(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, 0,
3136 ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3139 ic->sb->provided_data_sectors = cpu_to_le64(ic->provided_data_sectors);
3140 r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
3142 dm_integrity_io_error(ic, "writing superblock", r);
3145 if (ic->sb->flags & cpu_to_le32(SB_FLAG_DIRTY_BITMAP)) {
3146 DEBUG_print("resume dirty_bitmap\n");
3147 rw_journal_sectors(ic, REQ_OP_READ, 0, 0,
3148 ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3149 if (ic->mode == 'B') {
3150 if (ic->sb->log2_blocks_per_bitmap_bit == ic->log2_blocks_per_bitmap_bit &&
3151 !ic->reset_recalculate_flag) {
3152 block_bitmap_copy(ic, ic->recalc_bitmap, ic->journal);
3153 block_bitmap_copy(ic, ic->may_write_bitmap, ic->journal);
3154 if (!block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors,
3155 BITMAP_OP_TEST_ALL_CLEAR)) {
3156 ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
3157 ic->sb->recalc_sector = cpu_to_le64(0);
3160 DEBUG_print("non-matching blocks_per_bitmap_bit: %u, %u\n",
3161 ic->sb->log2_blocks_per_bitmap_bit, ic->log2_blocks_per_bitmap_bit);
3162 ic->sb->log2_blocks_per_bitmap_bit = ic->log2_blocks_per_bitmap_bit;
3163 block_bitmap_op(ic, ic->recalc_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_SET);
3164 block_bitmap_op(ic, ic->may_write_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_SET);
3165 block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors, BITMAP_OP_SET);
3166 rw_journal_sectors(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, 0,
3167 ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3168 ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
3169 ic->sb->recalc_sector = cpu_to_le64(0);
3172 if (!(ic->sb->log2_blocks_per_bitmap_bit == ic->log2_blocks_per_bitmap_bit &&
3173 block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors, BITMAP_OP_TEST_ALL_CLEAR)) ||
3174 ic->reset_recalculate_flag) {
3175 ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
3176 ic->sb->recalc_sector = cpu_to_le64(0);
3178 init_journal(ic, 0, ic->journal_sections, 0);
3180 ic->sb->flags &= ~cpu_to_le32(SB_FLAG_DIRTY_BITMAP);
3182 r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
3184 dm_integrity_io_error(ic, "writing superblock", r);
3187 if (ic->reset_recalculate_flag) {
3188 ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
3189 ic->sb->recalc_sector = cpu_to_le64(0);
3191 if (ic->mode == 'B') {
3192 ic->sb->flags |= cpu_to_le32(SB_FLAG_DIRTY_BITMAP);
3193 ic->sb->log2_blocks_per_bitmap_bit = ic->log2_blocks_per_bitmap_bit;
3194 r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
3196 dm_integrity_io_error(ic, "writing superblock", r);
3198 block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors, BITMAP_OP_CLEAR);
3199 block_bitmap_op(ic, ic->recalc_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_CLEAR);
3200 block_bitmap_op(ic, ic->may_write_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_CLEAR);
3201 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING) &&
3202 le64_to_cpu(ic->sb->recalc_sector) < ic->provided_data_sectors) {
3203 block_bitmap_op(ic, ic->journal, le64_to_cpu(ic->sb->recalc_sector),
3204 ic->provided_data_sectors - le64_to_cpu(ic->sb->recalc_sector), BITMAP_OP_SET);
3205 block_bitmap_op(ic, ic->recalc_bitmap, le64_to_cpu(ic->sb->recalc_sector),
3206 ic->provided_data_sectors - le64_to_cpu(ic->sb->recalc_sector), BITMAP_OP_SET);
3207 block_bitmap_op(ic, ic->may_write_bitmap, le64_to_cpu(ic->sb->recalc_sector),
3208 ic->provided_data_sectors - le64_to_cpu(ic->sb->recalc_sector), BITMAP_OP_SET);
3210 rw_journal_sectors(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, 0,
3211 ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3215 DEBUG_print("testing recalc: %x\n", ic->sb->flags);
3216 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
3217 __u64 recalc_pos = le64_to_cpu(ic->sb->recalc_sector);
3218 DEBUG_print("recalc pos: %llx / %llx\n", recalc_pos, ic->provided_data_sectors);
3219 if (recalc_pos < ic->provided_data_sectors) {
3220 queue_work(ic->recalc_wq, &ic->recalc_work);
3221 } else if (recalc_pos > ic->provided_data_sectors) {
3222 ic->sb->recalc_sector = cpu_to_le64(ic->provided_data_sectors);
3223 recalc_write_super(ic);
3227 ic->reboot_notifier.notifier_call = dm_integrity_reboot;
3228 ic->reboot_notifier.next = NULL;
3229 ic->reboot_notifier.priority = INT_MAX - 1; /* be notified after md and before hardware drivers */
3230 WARN_ON(register_reboot_notifier(&ic->reboot_notifier));
3233 /* set to 1 to stress test synchronous mode */
3234 dm_integrity_enter_synchronous_mode(ic);
3238 static void dm_integrity_status(struct dm_target *ti, status_type_t type,
3239 unsigned status_flags, char *result, unsigned maxlen)
3241 struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
3246 case STATUSTYPE_INFO:
3248 (unsigned long long)atomic64_read(&ic->number_of_mismatches),
3249 ic->provided_data_sectors);
3250 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
3251 DMEMIT(" %llu", le64_to_cpu(ic->sb->recalc_sector));
3256 case STATUSTYPE_TABLE: {
3257 __u64 watermark_percentage = (__u64)(ic->journal_entries - ic->free_sectors_threshold) * 100;
3258 watermark_percentage += ic->journal_entries / 2;
3259 do_div(watermark_percentage, ic->journal_entries);
3261 arg_count += !!ic->meta_dev;
3262 arg_count += ic->sectors_per_block != 1;
3263 arg_count += !!(ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING));
3264 arg_count += ic->reset_recalculate_flag;
3265 arg_count += ic->discard;
3266 arg_count += ic->mode == 'J';
3267 arg_count += ic->mode == 'J';
3268 arg_count += ic->mode == 'B';
3269 arg_count += ic->mode == 'B';
3270 arg_count += !!ic->internal_hash_alg.alg_string;
3271 arg_count += !!ic->journal_crypt_alg.alg_string;
3272 arg_count += !!ic->journal_mac_alg.alg_string;
3273 arg_count += (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING)) != 0;
3274 arg_count += (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) != 0;
3275 arg_count += ic->legacy_recalculate;
3276 DMEMIT("%s %llu %u %c %u", ic->dev->name, ic->start,
3277 ic->tag_size, ic->mode, arg_count);
3279 DMEMIT(" meta_device:%s", ic->meta_dev->name);
3280 if (ic->sectors_per_block != 1)
3281 DMEMIT(" block_size:%u", ic->sectors_per_block << SECTOR_SHIFT);
3282 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
3283 DMEMIT(" recalculate");
3284 if (ic->reset_recalculate_flag)
3285 DMEMIT(" reset_recalculate");
3287 DMEMIT(" allow_discards");
3288 DMEMIT(" journal_sectors:%u", ic->initial_sectors - SB_SECTORS);
3289 DMEMIT(" interleave_sectors:%u", 1U << ic->sb->log2_interleave_sectors);
3290 DMEMIT(" buffer_sectors:%u", 1U << ic->log2_buffer_sectors);
3291 if (ic->mode == 'J') {
3292 DMEMIT(" journal_watermark:%u", (unsigned)watermark_percentage);
3293 DMEMIT(" commit_time:%u", ic->autocommit_msec);
3295 if (ic->mode == 'B') {
3296 DMEMIT(" sectors_per_bit:%llu", (sector_t)ic->sectors_per_block << ic->log2_blocks_per_bitmap_bit);
3297 DMEMIT(" bitmap_flush_interval:%u", jiffies_to_msecs(ic->bitmap_flush_interval));
3299 if ((ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING)) != 0)
3300 DMEMIT(" fix_padding");
3301 if ((ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) != 0)
3302 DMEMIT(" fix_hmac");
3303 if (ic->legacy_recalculate)
3304 DMEMIT(" legacy_recalculate");
3306 #define EMIT_ALG(a, n) \
3308 if (ic->a.alg_string) { \
3309 DMEMIT(" %s:%s", n, ic->a.alg_string); \
3310 if (ic->a.key_string) \
3311 DMEMIT(":%s", ic->a.key_string);\
3314 EMIT_ALG(internal_hash_alg, "internal_hash");
3315 EMIT_ALG(journal_crypt_alg, "journal_crypt");
3316 EMIT_ALG(journal_mac_alg, "journal_mac");
3319 case STATUSTYPE_IMA:
3320 DMEMIT_TARGET_NAME_VERSION(ti->type);
3321 DMEMIT(",dev_name=%s,start=%llu,tag_size=%u,mode=%c",
3322 ic->dev->name, ic->start, ic->tag_size, ic->mode);
3325 DMEMIT(",meta_device=%s", ic->meta_dev->name);
3326 if (ic->sectors_per_block != 1)
3327 DMEMIT(",block_size=%u", ic->sectors_per_block << SECTOR_SHIFT);
3329 DMEMIT(",recalculate=%c", (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) ?
3331 DMEMIT(",allow_discards=%c", ic->discard ? 'y' : 'n');
3332 DMEMIT(",fix_padding=%c",
3333 ((ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING)) != 0) ? 'y' : 'n');
3334 DMEMIT(",fix_hmac=%c",
3335 ((ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) != 0) ? 'y' : 'n');
3336 DMEMIT(",legacy_recalculate=%c", ic->legacy_recalculate ? 'y' : 'n');
3338 DMEMIT(",journal_sectors=%u", ic->initial_sectors - SB_SECTORS);
3339 DMEMIT(",interleave_sectors=%u", 1U << ic->sb->log2_interleave_sectors);
3340 DMEMIT(",buffer_sectors=%u", 1U << ic->log2_buffer_sectors);
3346 static int dm_integrity_iterate_devices(struct dm_target *ti,
3347 iterate_devices_callout_fn fn, void *data)
3349 struct dm_integrity_c *ic = ti->private;
3352 return fn(ti, ic->dev, ic->start + ic->initial_sectors + ic->metadata_run, ti->len, data);
3354 return fn(ti, ic->dev, 0, ti->len, data);
3357 static void dm_integrity_io_hints(struct dm_target *ti, struct queue_limits *limits)
3359 struct dm_integrity_c *ic = ti->private;
3361 if (ic->sectors_per_block > 1) {
3362 limits->logical_block_size = ic->sectors_per_block << SECTOR_SHIFT;
3363 limits->physical_block_size = ic->sectors_per_block << SECTOR_SHIFT;
3364 blk_limits_io_min(limits, ic->sectors_per_block << SECTOR_SHIFT);
3368 static void calculate_journal_section_size(struct dm_integrity_c *ic)
3370 unsigned sector_space = JOURNAL_SECTOR_DATA;
3372 ic->journal_sections = le32_to_cpu(ic->sb->journal_sections);
3373 ic->journal_entry_size = roundup(offsetof(struct journal_entry, last_bytes[ic->sectors_per_block]) + ic->tag_size,
3374 JOURNAL_ENTRY_ROUNDUP);
3376 if (ic->sb->flags & cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC))
3377 sector_space -= JOURNAL_MAC_PER_SECTOR;
3378 ic->journal_entries_per_sector = sector_space / ic->journal_entry_size;
3379 ic->journal_section_entries = ic->journal_entries_per_sector * JOURNAL_BLOCK_SECTORS;
3380 ic->journal_section_sectors = (ic->journal_section_entries << ic->sb->log2_sectors_per_block) + JOURNAL_BLOCK_SECTORS;
3381 ic->journal_entries = ic->journal_section_entries * ic->journal_sections;
3384 static int calculate_device_limits(struct dm_integrity_c *ic)
3386 __u64 initial_sectors;
3388 calculate_journal_section_size(ic);
3389 initial_sectors = SB_SECTORS + (__u64)ic->journal_section_sectors * ic->journal_sections;
3390 if (initial_sectors + METADATA_PADDING_SECTORS >= ic->meta_device_sectors || initial_sectors > UINT_MAX)
3392 ic->initial_sectors = initial_sectors;
3394 if (!ic->meta_dev) {
3395 sector_t last_sector, last_area, last_offset;
3397 /* we have to maintain excessive padding for compatibility with existing volumes */
3398 __u64 metadata_run_padding =
3399 ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING) ?
3400 (__u64)(METADATA_PADDING_SECTORS << SECTOR_SHIFT) :
3401 (__u64)(1 << SECTOR_SHIFT << METADATA_PADDING_SECTORS);
3403 ic->metadata_run = round_up((__u64)ic->tag_size << (ic->sb->log2_interleave_sectors - ic->sb->log2_sectors_per_block),
3404 metadata_run_padding) >> SECTOR_SHIFT;
3405 if (!(ic->metadata_run & (ic->metadata_run - 1)))
3406 ic->log2_metadata_run = __ffs(ic->metadata_run);
3408 ic->log2_metadata_run = -1;
3410 get_area_and_offset(ic, ic->provided_data_sectors - 1, &last_area, &last_offset);
3411 last_sector = get_data_sector(ic, last_area, last_offset);
3412 if (last_sector < ic->start || last_sector >= ic->meta_device_sectors)
3415 __u64 meta_size = (ic->provided_data_sectors >> ic->sb->log2_sectors_per_block) * ic->tag_size;
3416 meta_size = (meta_size + ((1U << (ic->log2_buffer_sectors + SECTOR_SHIFT)) - 1))
3417 >> (ic->log2_buffer_sectors + SECTOR_SHIFT);
3418 meta_size <<= ic->log2_buffer_sectors;
3419 if (ic->initial_sectors + meta_size < ic->initial_sectors ||
3420 ic->initial_sectors + meta_size > ic->meta_device_sectors)
3422 ic->metadata_run = 1;
3423 ic->log2_metadata_run = 0;
3429 static void get_provided_data_sectors(struct dm_integrity_c *ic)
3431 if (!ic->meta_dev) {
3433 ic->provided_data_sectors = 0;
3434 for (test_bit = fls64(ic->meta_device_sectors) - 1; test_bit >= 3; test_bit--) {
3435 __u64 prev_data_sectors = ic->provided_data_sectors;
3437 ic->provided_data_sectors |= (sector_t)1 << test_bit;
3438 if (calculate_device_limits(ic))
3439 ic->provided_data_sectors = prev_data_sectors;
3442 ic->provided_data_sectors = ic->data_device_sectors;
3443 ic->provided_data_sectors &= ~(sector_t)(ic->sectors_per_block - 1);
3447 static int initialize_superblock(struct dm_integrity_c *ic, unsigned journal_sectors, unsigned interleave_sectors)
3449 unsigned journal_sections;
3452 memset(ic->sb, 0, SB_SECTORS << SECTOR_SHIFT);
3453 memcpy(ic->sb->magic, SB_MAGIC, 8);
3454 ic->sb->integrity_tag_size = cpu_to_le16(ic->tag_size);
3455 ic->sb->log2_sectors_per_block = __ffs(ic->sectors_per_block);
3456 if (ic->journal_mac_alg.alg_string)
3457 ic->sb->flags |= cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC);
3459 calculate_journal_section_size(ic);
3460 journal_sections = journal_sectors / ic->journal_section_sectors;
3461 if (!journal_sections)
3462 journal_sections = 1;
3464 if (ic->fix_hmac && (ic->internal_hash_alg.alg_string || ic->journal_mac_alg.alg_string)) {
3465 ic->sb->flags |= cpu_to_le32(SB_FLAG_FIXED_HMAC);
3466 get_random_bytes(ic->sb->salt, SALT_SIZE);
3469 if (!ic->meta_dev) {
3470 if (ic->fix_padding)
3471 ic->sb->flags |= cpu_to_le32(SB_FLAG_FIXED_PADDING);
3472 ic->sb->journal_sections = cpu_to_le32(journal_sections);
3473 if (!interleave_sectors)
3474 interleave_sectors = DEFAULT_INTERLEAVE_SECTORS;
3475 ic->sb->log2_interleave_sectors = __fls(interleave_sectors);
3476 ic->sb->log2_interleave_sectors = max((__u8)MIN_LOG2_INTERLEAVE_SECTORS, ic->sb->log2_interleave_sectors);
3477 ic->sb->log2_interleave_sectors = min((__u8)MAX_LOG2_INTERLEAVE_SECTORS, ic->sb->log2_interleave_sectors);
3479 get_provided_data_sectors(ic);
3480 if (!ic->provided_data_sectors)
3483 ic->sb->log2_interleave_sectors = 0;
3485 get_provided_data_sectors(ic);
3486 if (!ic->provided_data_sectors)
3490 ic->sb->journal_sections = cpu_to_le32(0);
3491 for (test_bit = fls(journal_sections) - 1; test_bit >= 0; test_bit--) {
3492 __u32 prev_journal_sections = le32_to_cpu(ic->sb->journal_sections);
3493 __u32 test_journal_sections = prev_journal_sections | (1U << test_bit);
3494 if (test_journal_sections > journal_sections)
3496 ic->sb->journal_sections = cpu_to_le32(test_journal_sections);
3497 if (calculate_device_limits(ic))
3498 ic->sb->journal_sections = cpu_to_le32(prev_journal_sections);
3501 if (!le32_to_cpu(ic->sb->journal_sections)) {
3502 if (ic->log2_buffer_sectors > 3) {
3503 ic->log2_buffer_sectors--;
3504 goto try_smaller_buffer;
3510 ic->sb->provided_data_sectors = cpu_to_le64(ic->provided_data_sectors);
3517 static void dm_integrity_set(struct dm_target *ti, struct dm_integrity_c *ic)
3519 struct gendisk *disk = dm_disk(dm_table_get_md(ti->table));
3520 struct blk_integrity bi;
3522 memset(&bi, 0, sizeof(bi));
3523 bi.profile = &dm_integrity_profile;
3524 bi.tuple_size = ic->tag_size;
3525 bi.tag_size = bi.tuple_size;
3526 bi.interval_exp = ic->sb->log2_sectors_per_block + SECTOR_SHIFT;
3528 blk_integrity_register(disk, &bi);
3529 blk_queue_max_integrity_segments(disk->queue, UINT_MAX);
3532 static void dm_integrity_free_page_list(struct page_list *pl)
3538 for (i = 0; pl[i].page; i++)
3539 __free_page(pl[i].page);
3543 static struct page_list *dm_integrity_alloc_page_list(unsigned n_pages)
3545 struct page_list *pl;
3548 pl = kvmalloc_array(n_pages + 1, sizeof(struct page_list), GFP_KERNEL | __GFP_ZERO);
3552 for (i = 0; i < n_pages; i++) {
3553 pl[i].page = alloc_page(GFP_KERNEL);
3555 dm_integrity_free_page_list(pl);
3559 pl[i - 1].next = &pl[i];
3567 static void dm_integrity_free_journal_scatterlist(struct dm_integrity_c *ic, struct scatterlist **sl)
3570 for (i = 0; i < ic->journal_sections; i++)
3575 static struct scatterlist **dm_integrity_alloc_journal_scatterlist(struct dm_integrity_c *ic,
3576 struct page_list *pl)
3578 struct scatterlist **sl;
3581 sl = kvmalloc_array(ic->journal_sections,
3582 sizeof(struct scatterlist *),
3583 GFP_KERNEL | __GFP_ZERO);
3587 for (i = 0; i < ic->journal_sections; i++) {
3588 struct scatterlist *s;
3589 unsigned start_index, start_offset;
3590 unsigned end_index, end_offset;
3594 page_list_location(ic, i, 0, &start_index, &start_offset);
3595 page_list_location(ic, i, ic->journal_section_sectors - 1,
3596 &end_index, &end_offset);
3598 n_pages = (end_index - start_index + 1);
3600 s = kvmalloc_array(n_pages, sizeof(struct scatterlist),
3603 dm_integrity_free_journal_scatterlist(ic, sl);
3607 sg_init_table(s, n_pages);
3608 for (idx = start_index; idx <= end_index; idx++) {
3609 char *va = lowmem_page_address(pl[idx].page);
3610 unsigned start = 0, end = PAGE_SIZE;
3611 if (idx == start_index)
3612 start = start_offset;
3613 if (idx == end_index)
3614 end = end_offset + (1 << SECTOR_SHIFT);
3615 sg_set_buf(&s[idx - start_index], va + start, end - start);
3624 static void free_alg(struct alg_spec *a)
3626 kfree_sensitive(a->alg_string);
3627 kfree_sensitive(a->key);
3628 memset(a, 0, sizeof *a);
3631 static int get_alg_and_key(const char *arg, struct alg_spec *a, char **error, char *error_inval)
3637 a->alg_string = kstrdup(strchr(arg, ':') + 1, GFP_KERNEL);
3641 k = strchr(a->alg_string, ':');
3644 a->key_string = k + 1;
3645 if (strlen(a->key_string) & 1)
3648 a->key_size = strlen(a->key_string) / 2;
3649 a->key = kmalloc(a->key_size, GFP_KERNEL);
3652 if (hex2bin(a->key, a->key_string, a->key_size))
3658 *error = error_inval;
3661 *error = "Out of memory for an argument";
3665 static int get_mac(struct crypto_shash **hash, struct alg_spec *a, char **error,
3666 char *error_alg, char *error_key)
3670 if (a->alg_string) {
3671 *hash = crypto_alloc_shash(a->alg_string, 0, CRYPTO_ALG_ALLOCATES_MEMORY);
3672 if (IS_ERR(*hash)) {
3680 r = crypto_shash_setkey(*hash, a->key, a->key_size);
3685 } else if (crypto_shash_get_flags(*hash) & CRYPTO_TFM_NEED_KEY) {
3694 static int create_journal(struct dm_integrity_c *ic, char **error)
3698 __u64 journal_pages, journal_desc_size, journal_tree_size;
3699 unsigned char *crypt_data = NULL, *crypt_iv = NULL;
3700 struct skcipher_request *req = NULL;
3702 ic->commit_ids[0] = cpu_to_le64(0x1111111111111111ULL);
3703 ic->commit_ids[1] = cpu_to_le64(0x2222222222222222ULL);
3704 ic->commit_ids[2] = cpu_to_le64(0x3333333333333333ULL);
3705 ic->commit_ids[3] = cpu_to_le64(0x4444444444444444ULL);
3707 journal_pages = roundup((__u64)ic->journal_sections * ic->journal_section_sectors,
3708 PAGE_SIZE >> SECTOR_SHIFT) >> (PAGE_SHIFT - SECTOR_SHIFT);
3709 journal_desc_size = journal_pages * sizeof(struct page_list);
3710 if (journal_pages >= totalram_pages() - totalhigh_pages() || journal_desc_size > ULONG_MAX) {
3711 *error = "Journal doesn't fit into memory";
3715 ic->journal_pages = journal_pages;
3717 ic->journal = dm_integrity_alloc_page_list(ic->journal_pages);
3719 *error = "Could not allocate memory for journal";
3723 if (ic->journal_crypt_alg.alg_string) {
3724 unsigned ivsize, blocksize;
3725 struct journal_completion comp;
3728 ic->journal_crypt = crypto_alloc_skcipher(ic->journal_crypt_alg.alg_string, 0, CRYPTO_ALG_ALLOCATES_MEMORY);
3729 if (IS_ERR(ic->journal_crypt)) {
3730 *error = "Invalid journal cipher";
3731 r = PTR_ERR(ic->journal_crypt);
3732 ic->journal_crypt = NULL;
3735 ivsize = crypto_skcipher_ivsize(ic->journal_crypt);
3736 blocksize = crypto_skcipher_blocksize(ic->journal_crypt);
3738 if (ic->journal_crypt_alg.key) {
3739 r = crypto_skcipher_setkey(ic->journal_crypt, ic->journal_crypt_alg.key,
3740 ic->journal_crypt_alg.key_size);
3742 *error = "Error setting encryption key";
3746 DEBUG_print("cipher %s, block size %u iv size %u\n",
3747 ic->journal_crypt_alg.alg_string, blocksize, ivsize);
3749 ic->journal_io = dm_integrity_alloc_page_list(ic->journal_pages);
3750 if (!ic->journal_io) {
3751 *error = "Could not allocate memory for journal io";
3756 if (blocksize == 1) {
3757 struct scatterlist *sg;
3759 req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
3761 *error = "Could not allocate crypt request";
3766 crypt_iv = kzalloc(ivsize, GFP_KERNEL);
3768 *error = "Could not allocate iv";
3773 ic->journal_xor = dm_integrity_alloc_page_list(ic->journal_pages);
3774 if (!ic->journal_xor) {
3775 *error = "Could not allocate memory for journal xor";
3780 sg = kvmalloc_array(ic->journal_pages + 1,
3781 sizeof(struct scatterlist),
3784 *error = "Unable to allocate sg list";
3788 sg_init_table(sg, ic->journal_pages + 1);
3789 for (i = 0; i < ic->journal_pages; i++) {
3790 char *va = lowmem_page_address(ic->journal_xor[i].page);
3792 sg_set_buf(&sg[i], va, PAGE_SIZE);
3794 sg_set_buf(&sg[i], &ic->commit_ids, sizeof ic->commit_ids);
3796 skcipher_request_set_crypt(req, sg, sg,
3797 PAGE_SIZE * ic->journal_pages + sizeof ic->commit_ids, crypt_iv);
3798 init_completion(&comp.comp);
3799 comp.in_flight = (atomic_t)ATOMIC_INIT(1);
3800 if (do_crypt(true, req, &comp))
3801 wait_for_completion(&comp.comp);
3803 r = dm_integrity_failed(ic);
3805 *error = "Unable to encrypt journal";
3808 DEBUG_bytes(lowmem_page_address(ic->journal_xor[0].page), 64, "xor data");
3810 crypto_free_skcipher(ic->journal_crypt);
3811 ic->journal_crypt = NULL;
3813 unsigned crypt_len = roundup(ivsize, blocksize);
3815 req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
3817 *error = "Could not allocate crypt request";
3822 crypt_iv = kmalloc(ivsize, GFP_KERNEL);
3824 *error = "Could not allocate iv";
3829 crypt_data = kmalloc(crypt_len, GFP_KERNEL);
3831 *error = "Unable to allocate crypt data";
3836 ic->journal_scatterlist = dm_integrity_alloc_journal_scatterlist(ic, ic->journal);
3837 if (!ic->journal_scatterlist) {
3838 *error = "Unable to allocate sg list";
3842 ic->journal_io_scatterlist = dm_integrity_alloc_journal_scatterlist(ic, ic->journal_io);
3843 if (!ic->journal_io_scatterlist) {
3844 *error = "Unable to allocate sg list";
3848 ic->sk_requests = kvmalloc_array(ic->journal_sections,
3849 sizeof(struct skcipher_request *),
3850 GFP_KERNEL | __GFP_ZERO);
3851 if (!ic->sk_requests) {
3852 *error = "Unable to allocate sk requests";
3856 for (i = 0; i < ic->journal_sections; i++) {
3857 struct scatterlist sg;
3858 struct skcipher_request *section_req;
3859 __le32 section_le = cpu_to_le32(i);
3861 memset(crypt_iv, 0x00, ivsize);
3862 memset(crypt_data, 0x00, crypt_len);
3863 memcpy(crypt_data, §ion_le, min((size_t)crypt_len, sizeof(section_le)));
3865 sg_init_one(&sg, crypt_data, crypt_len);
3866 skcipher_request_set_crypt(req, &sg, &sg, crypt_len, crypt_iv);
3867 init_completion(&comp.comp);
3868 comp.in_flight = (atomic_t)ATOMIC_INIT(1);
3869 if (do_crypt(true, req, &comp))
3870 wait_for_completion(&comp.comp);
3872 r = dm_integrity_failed(ic);
3874 *error = "Unable to generate iv";
3878 section_req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
3880 *error = "Unable to allocate crypt request";
3884 section_req->iv = kmalloc_array(ivsize, 2,
3886 if (!section_req->iv) {
3887 skcipher_request_free(section_req);
3888 *error = "Unable to allocate iv";
3892 memcpy(section_req->iv + ivsize, crypt_data, ivsize);
3893 section_req->cryptlen = (size_t)ic->journal_section_sectors << SECTOR_SHIFT;
3894 ic->sk_requests[i] = section_req;
3895 DEBUG_bytes(crypt_data, ivsize, "iv(%u)", i);
3900 for (i = 0; i < N_COMMIT_IDS; i++) {
3903 for (j = 0; j < i; j++) {
3904 if (ic->commit_ids[j] == ic->commit_ids[i]) {
3905 ic->commit_ids[i] = cpu_to_le64(le64_to_cpu(ic->commit_ids[i]) + 1);
3906 goto retest_commit_id;
3909 DEBUG_print("commit id %u: %016llx\n", i, ic->commit_ids[i]);
3912 journal_tree_size = (__u64)ic->journal_entries * sizeof(struct journal_node);
3913 if (journal_tree_size > ULONG_MAX) {
3914 *error = "Journal doesn't fit into memory";
3918 ic->journal_tree = kvmalloc(journal_tree_size, GFP_KERNEL);
3919 if (!ic->journal_tree) {
3920 *error = "Could not allocate memory for journal tree";
3926 skcipher_request_free(req);
3932 * Construct a integrity mapping
3936 * offset from the start of the device
3938 * D - direct writes, J - journal writes, B - bitmap mode, R - recovery mode
3939 * number of optional arguments
3940 * optional arguments:
3942 * interleave_sectors
3949 * bitmap_flush_interval
3955 static int dm_integrity_ctr(struct dm_target *ti, unsigned argc, char **argv)
3957 struct dm_integrity_c *ic;
3960 unsigned extra_args;
3961 struct dm_arg_set as;
3962 static const struct dm_arg _args[] = {
3963 {0, 18, "Invalid number of feature args"},
3965 unsigned journal_sectors, interleave_sectors, buffer_sectors, journal_watermark, sync_msec;
3966 bool should_write_sb;
3968 unsigned long long start;
3969 __s8 log2_sectors_per_bitmap_bit = -1;
3970 __s8 log2_blocks_per_bitmap_bit;
3971 __u64 bits_in_journal;
3972 __u64 n_bitmap_bits;
3974 #define DIRECT_ARGUMENTS 4
3976 if (argc <= DIRECT_ARGUMENTS) {
3977 ti->error = "Invalid argument count";
3981 ic = kzalloc(sizeof(struct dm_integrity_c), GFP_KERNEL);
3983 ti->error = "Cannot allocate integrity context";
3987 ti->per_io_data_size = sizeof(struct dm_integrity_io);
3990 ic->in_progress = RB_ROOT;
3991 INIT_LIST_HEAD(&ic->wait_list);
3992 init_waitqueue_head(&ic->endio_wait);
3993 bio_list_init(&ic->flush_bio_list);
3994 init_waitqueue_head(&ic->copy_to_journal_wait);
3995 init_completion(&ic->crypto_backoff);
3996 atomic64_set(&ic->number_of_mismatches, 0);
3997 ic->bitmap_flush_interval = BITMAP_FLUSH_INTERVAL;
3999 r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &ic->dev);
4001 ti->error = "Device lookup failed";
4005 if (sscanf(argv[1], "%llu%c", &start, &dummy) != 1 || start != (sector_t)start) {
4006 ti->error = "Invalid starting offset";
4012 if (strcmp(argv[2], "-")) {
4013 if (sscanf(argv[2], "%u%c", &ic->tag_size, &dummy) != 1 || !ic->tag_size) {
4014 ti->error = "Invalid tag size";
4020 if (!strcmp(argv[3], "J") || !strcmp(argv[3], "B") ||
4021 !strcmp(argv[3], "D") || !strcmp(argv[3], "R")) {
4022 ic->mode = argv[3][0];
4024 ti->error = "Invalid mode (expecting J, B, D, R)";
4029 journal_sectors = 0;
4030 interleave_sectors = DEFAULT_INTERLEAVE_SECTORS;
4031 buffer_sectors = DEFAULT_BUFFER_SECTORS;
4032 journal_watermark = DEFAULT_JOURNAL_WATERMARK;
4033 sync_msec = DEFAULT_SYNC_MSEC;
4034 ic->sectors_per_block = 1;
4036 as.argc = argc - DIRECT_ARGUMENTS;
4037 as.argv = argv + DIRECT_ARGUMENTS;
4038 r = dm_read_arg_group(_args, &as, &extra_args, &ti->error);
4042 while (extra_args--) {
4043 const char *opt_string;
4045 unsigned long long llval;
4046 opt_string = dm_shift_arg(&as);
4049 ti->error = "Not enough feature arguments";
4052 if (sscanf(opt_string, "journal_sectors:%u%c", &val, &dummy) == 1)
4053 journal_sectors = val ? val : 1;
4054 else if (sscanf(opt_string, "interleave_sectors:%u%c", &val, &dummy) == 1)
4055 interleave_sectors = val;
4056 else if (sscanf(opt_string, "buffer_sectors:%u%c", &val, &dummy) == 1)
4057 buffer_sectors = val;
4058 else if (sscanf(opt_string, "journal_watermark:%u%c", &val, &dummy) == 1 && val <= 100)
4059 journal_watermark = val;
4060 else if (sscanf(opt_string, "commit_time:%u%c", &val, &dummy) == 1)
4062 else if (!strncmp(opt_string, "meta_device:", strlen("meta_device:"))) {
4064 dm_put_device(ti, ic->meta_dev);
4065 ic->meta_dev = NULL;
4067 r = dm_get_device(ti, strchr(opt_string, ':') + 1,
4068 dm_table_get_mode(ti->table), &ic->meta_dev);
4070 ti->error = "Device lookup failed";
4073 } else if (sscanf(opt_string, "block_size:%u%c", &val, &dummy) == 1) {
4074 if (val < 1 << SECTOR_SHIFT ||
4075 val > MAX_SECTORS_PER_BLOCK << SECTOR_SHIFT ||
4078 ti->error = "Invalid block_size argument";
4081 ic->sectors_per_block = val >> SECTOR_SHIFT;
4082 } else if (sscanf(opt_string, "sectors_per_bit:%llu%c", &llval, &dummy) == 1) {
4083 log2_sectors_per_bitmap_bit = !llval ? 0 : __ilog2_u64(llval);
4084 } else if (sscanf(opt_string, "bitmap_flush_interval:%u%c", &val, &dummy) == 1) {
4085 if (val >= (uint64_t)UINT_MAX * 1000 / HZ) {
4087 ti->error = "Invalid bitmap_flush_interval argument";
4090 ic->bitmap_flush_interval = msecs_to_jiffies(val);
4091 } else if (!strncmp(opt_string, "internal_hash:", strlen("internal_hash:"))) {
4092 r = get_alg_and_key(opt_string, &ic->internal_hash_alg, &ti->error,
4093 "Invalid internal_hash argument");
4096 } else if (!strncmp(opt_string, "journal_crypt:", strlen("journal_crypt:"))) {
4097 r = get_alg_and_key(opt_string, &ic->journal_crypt_alg, &ti->error,
4098 "Invalid journal_crypt argument");
4101 } else if (!strncmp(opt_string, "journal_mac:", strlen("journal_mac:"))) {
4102 r = get_alg_and_key(opt_string, &ic->journal_mac_alg, &ti->error,
4103 "Invalid journal_mac argument");
4106 } else if (!strcmp(opt_string, "recalculate")) {
4107 ic->recalculate_flag = true;
4108 } else if (!strcmp(opt_string, "reset_recalculate")) {
4109 ic->recalculate_flag = true;
4110 ic->reset_recalculate_flag = true;
4111 } else if (!strcmp(opt_string, "allow_discards")) {
4113 } else if (!strcmp(opt_string, "fix_padding")) {
4114 ic->fix_padding = true;
4115 } else if (!strcmp(opt_string, "fix_hmac")) {
4116 ic->fix_hmac = true;
4117 } else if (!strcmp(opt_string, "legacy_recalculate")) {
4118 ic->legacy_recalculate = true;
4121 ti->error = "Invalid argument";
4126 ic->data_device_sectors = i_size_read(ic->dev->bdev->bd_inode) >> SECTOR_SHIFT;
4128 ic->meta_device_sectors = ic->data_device_sectors;
4130 ic->meta_device_sectors = i_size_read(ic->meta_dev->bdev->bd_inode) >> SECTOR_SHIFT;
4132 if (!journal_sectors) {
4133 journal_sectors = min((sector_t)DEFAULT_MAX_JOURNAL_SECTORS,
4134 ic->data_device_sectors >> DEFAULT_JOURNAL_SIZE_FACTOR);
4137 if (!buffer_sectors)
4139 ic->log2_buffer_sectors = min((int)__fls(buffer_sectors), 31 - SECTOR_SHIFT);
4141 r = get_mac(&ic->internal_hash, &ic->internal_hash_alg, &ti->error,
4142 "Invalid internal hash", "Error setting internal hash key");
4146 r = get_mac(&ic->journal_mac, &ic->journal_mac_alg, &ti->error,
4147 "Invalid journal mac", "Error setting journal mac key");
4151 if (!ic->tag_size) {
4152 if (!ic->internal_hash) {
4153 ti->error = "Unknown tag size";
4157 ic->tag_size = crypto_shash_digestsize(ic->internal_hash);
4159 if (ic->tag_size > MAX_TAG_SIZE) {
4160 ti->error = "Too big tag size";
4164 if (!(ic->tag_size & (ic->tag_size - 1)))
4165 ic->log2_tag_size = __ffs(ic->tag_size);
4167 ic->log2_tag_size = -1;
4169 if (ic->mode == 'B' && !ic->internal_hash) {
4171 ti->error = "Bitmap mode can be only used with internal hash";
4175 if (ic->discard && !ic->internal_hash) {
4177 ti->error = "Discard can be only used with internal hash";
4181 ic->autocommit_jiffies = msecs_to_jiffies(sync_msec);
4182 ic->autocommit_msec = sync_msec;
4183 timer_setup(&ic->autocommit_timer, autocommit_fn, 0);
4185 ic->io = dm_io_client_create();
4186 if (IS_ERR(ic->io)) {
4187 r = PTR_ERR(ic->io);
4189 ti->error = "Cannot allocate dm io";
4193 r = mempool_init_slab_pool(&ic->journal_io_mempool, JOURNAL_IO_MEMPOOL, journal_io_cache);
4195 ti->error = "Cannot allocate mempool";
4199 ic->metadata_wq = alloc_workqueue("dm-integrity-metadata",
4200 WQ_MEM_RECLAIM, METADATA_WORKQUEUE_MAX_ACTIVE);
4201 if (!ic->metadata_wq) {
4202 ti->error = "Cannot allocate workqueue";
4208 * If this workqueue were percpu, it would cause bio reordering
4209 * and reduced performance.
4211 ic->wait_wq = alloc_workqueue("dm-integrity-wait", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
4213 ti->error = "Cannot allocate workqueue";
4218 ic->offload_wq = alloc_workqueue("dm-integrity-offload", WQ_MEM_RECLAIM,
4219 METADATA_WORKQUEUE_MAX_ACTIVE);
4220 if (!ic->offload_wq) {
4221 ti->error = "Cannot allocate workqueue";
4226 ic->commit_wq = alloc_workqueue("dm-integrity-commit", WQ_MEM_RECLAIM, 1);
4227 if (!ic->commit_wq) {
4228 ti->error = "Cannot allocate workqueue";
4232 INIT_WORK(&ic->commit_work, integrity_commit);
4234 if (ic->mode == 'J' || ic->mode == 'B') {
4235 ic->writer_wq = alloc_workqueue("dm-integrity-writer", WQ_MEM_RECLAIM, 1);
4236 if (!ic->writer_wq) {
4237 ti->error = "Cannot allocate workqueue";
4241 INIT_WORK(&ic->writer_work, integrity_writer);
4244 ic->sb = alloc_pages_exact(SB_SECTORS << SECTOR_SHIFT, GFP_KERNEL);
4247 ti->error = "Cannot allocate superblock area";
4251 r = sync_rw_sb(ic, REQ_OP_READ, 0);
4253 ti->error = "Error reading superblock";
4256 should_write_sb = false;
4257 if (memcmp(ic->sb->magic, SB_MAGIC, 8)) {
4258 if (ic->mode != 'R') {
4259 if (memchr_inv(ic->sb, 0, SB_SECTORS << SECTOR_SHIFT)) {
4261 ti->error = "The device is not initialized";
4266 r = initialize_superblock(ic, journal_sectors, interleave_sectors);
4268 ti->error = "Could not initialize superblock";
4271 if (ic->mode != 'R')
4272 should_write_sb = true;
4275 if (!ic->sb->version || ic->sb->version > SB_VERSION_5) {
4277 ti->error = "Unknown version";
4280 if (le16_to_cpu(ic->sb->integrity_tag_size) != ic->tag_size) {
4282 ti->error = "Tag size doesn't match the information in superblock";
4285 if (ic->sb->log2_sectors_per_block != __ffs(ic->sectors_per_block)) {
4287 ti->error = "Block size doesn't match the information in superblock";
4290 if (!le32_to_cpu(ic->sb->journal_sections)) {
4292 ti->error = "Corrupted superblock, journal_sections is 0";
4295 /* make sure that ti->max_io_len doesn't overflow */
4296 if (!ic->meta_dev) {
4297 if (ic->sb->log2_interleave_sectors < MIN_LOG2_INTERLEAVE_SECTORS ||
4298 ic->sb->log2_interleave_sectors > MAX_LOG2_INTERLEAVE_SECTORS) {
4300 ti->error = "Invalid interleave_sectors in the superblock";
4304 if (ic->sb->log2_interleave_sectors) {
4306 ti->error = "Invalid interleave_sectors in the superblock";
4310 if (!!(ic->sb->flags & cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC)) != !!ic->journal_mac_alg.alg_string) {
4312 ti->error = "Journal mac mismatch";
4316 get_provided_data_sectors(ic);
4317 if (!ic->provided_data_sectors) {
4319 ti->error = "The device is too small";
4324 r = calculate_device_limits(ic);
4327 if (ic->log2_buffer_sectors > 3) {
4328 ic->log2_buffer_sectors--;
4329 goto try_smaller_buffer;
4332 ti->error = "The device is too small";
4336 if (log2_sectors_per_bitmap_bit < 0)
4337 log2_sectors_per_bitmap_bit = __fls(DEFAULT_SECTORS_PER_BITMAP_BIT);
4338 if (log2_sectors_per_bitmap_bit < ic->sb->log2_sectors_per_block)
4339 log2_sectors_per_bitmap_bit = ic->sb->log2_sectors_per_block;
4341 bits_in_journal = ((__u64)ic->journal_section_sectors * ic->journal_sections) << (SECTOR_SHIFT + 3);
4342 if (bits_in_journal > UINT_MAX)
4343 bits_in_journal = UINT_MAX;
4344 while (bits_in_journal < (ic->provided_data_sectors + ((sector_t)1 << log2_sectors_per_bitmap_bit) - 1) >> log2_sectors_per_bitmap_bit)
4345 log2_sectors_per_bitmap_bit++;
4347 log2_blocks_per_bitmap_bit = log2_sectors_per_bitmap_bit - ic->sb->log2_sectors_per_block;
4348 ic->log2_blocks_per_bitmap_bit = log2_blocks_per_bitmap_bit;
4349 if (should_write_sb) {
4350 ic->sb->log2_blocks_per_bitmap_bit = log2_blocks_per_bitmap_bit;
4352 n_bitmap_bits = ((ic->provided_data_sectors >> ic->sb->log2_sectors_per_block)
4353 + (((sector_t)1 << log2_blocks_per_bitmap_bit) - 1)) >> log2_blocks_per_bitmap_bit;
4354 ic->n_bitmap_blocks = DIV_ROUND_UP(n_bitmap_bits, BITMAP_BLOCK_SIZE * 8);
4357 ic->log2_buffer_sectors = min(ic->log2_buffer_sectors, (__u8)__ffs(ic->metadata_run));
4359 if (ti->len > ic->provided_data_sectors) {
4361 ti->error = "Not enough provided sectors for requested mapping size";
4366 threshold = (__u64)ic->journal_entries * (100 - journal_watermark);
4368 do_div(threshold, 100);
4369 ic->free_sectors_threshold = threshold;
4371 DEBUG_print("initialized:\n");
4372 DEBUG_print(" integrity_tag_size %u\n", le16_to_cpu(ic->sb->integrity_tag_size));
4373 DEBUG_print(" journal_entry_size %u\n", ic->journal_entry_size);
4374 DEBUG_print(" journal_entries_per_sector %u\n", ic->journal_entries_per_sector);
4375 DEBUG_print(" journal_section_entries %u\n", ic->journal_section_entries);
4376 DEBUG_print(" journal_section_sectors %u\n", ic->journal_section_sectors);
4377 DEBUG_print(" journal_sections %u\n", (unsigned)le32_to_cpu(ic->sb->journal_sections));
4378 DEBUG_print(" journal_entries %u\n", ic->journal_entries);
4379 DEBUG_print(" log2_interleave_sectors %d\n", ic->sb->log2_interleave_sectors);
4380 DEBUG_print(" data_device_sectors 0x%llx\n", i_size_read(ic->dev->bdev->bd_inode) >> SECTOR_SHIFT);
4381 DEBUG_print(" initial_sectors 0x%x\n", ic->initial_sectors);
4382 DEBUG_print(" metadata_run 0x%x\n", ic->metadata_run);
4383 DEBUG_print(" log2_metadata_run %d\n", ic->log2_metadata_run);
4384 DEBUG_print(" provided_data_sectors 0x%llx (%llu)\n", ic->provided_data_sectors, ic->provided_data_sectors);
4385 DEBUG_print(" log2_buffer_sectors %u\n", ic->log2_buffer_sectors);
4386 DEBUG_print(" bits_in_journal %llu\n", bits_in_journal);
4388 if (ic->recalculate_flag && !(ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))) {
4389 ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
4390 ic->sb->recalc_sector = cpu_to_le64(0);
4393 if (ic->internal_hash) {
4394 size_t recalc_tags_size;
4395 ic->recalc_wq = alloc_workqueue("dm-integrity-recalc", WQ_MEM_RECLAIM, 1);
4396 if (!ic->recalc_wq ) {
4397 ti->error = "Cannot allocate workqueue";
4401 INIT_WORK(&ic->recalc_work, integrity_recalc);
4402 ic->recalc_buffer = vmalloc(RECALC_SECTORS << SECTOR_SHIFT);
4403 if (!ic->recalc_buffer) {
4404 ti->error = "Cannot allocate buffer for recalculating";
4408 recalc_tags_size = (RECALC_SECTORS >> ic->sb->log2_sectors_per_block) * ic->tag_size;
4409 if (crypto_shash_digestsize(ic->internal_hash) > ic->tag_size)
4410 recalc_tags_size += crypto_shash_digestsize(ic->internal_hash) - ic->tag_size;
4411 ic->recalc_tags = kvmalloc(recalc_tags_size, GFP_KERNEL);
4412 if (!ic->recalc_tags) {
4413 ti->error = "Cannot allocate tags for recalculating";
4418 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
4419 ti->error = "Recalculate can only be specified with internal_hash";
4425 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING) &&
4426 le64_to_cpu(ic->sb->recalc_sector) < ic->provided_data_sectors &&
4427 dm_integrity_disable_recalculate(ic)) {
4428 ti->error = "Recalculating with HMAC is disabled for security reasons - if you really need it, use the argument \"legacy_recalculate\"";
4433 ic->bufio = dm_bufio_client_create(ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev,
4434 1U << (SECTOR_SHIFT + ic->log2_buffer_sectors), 1, 0, NULL, NULL);
4435 if (IS_ERR(ic->bufio)) {
4436 r = PTR_ERR(ic->bufio);
4437 ti->error = "Cannot initialize dm-bufio";
4441 dm_bufio_set_sector_offset(ic->bufio, ic->start + ic->initial_sectors);
4443 if (ic->mode != 'R') {
4444 r = create_journal(ic, &ti->error);
4450 if (ic->mode == 'B') {
4452 unsigned n_bitmap_pages = DIV_ROUND_UP(ic->n_bitmap_blocks, PAGE_SIZE / BITMAP_BLOCK_SIZE);
4454 ic->recalc_bitmap = dm_integrity_alloc_page_list(n_bitmap_pages);
4455 if (!ic->recalc_bitmap) {
4459 ic->may_write_bitmap = dm_integrity_alloc_page_list(n_bitmap_pages);
4460 if (!ic->may_write_bitmap) {
4464 ic->bbs = kvmalloc_array(ic->n_bitmap_blocks, sizeof(struct bitmap_block_status), GFP_KERNEL);
4469 INIT_DELAYED_WORK(&ic->bitmap_flush_work, bitmap_flush_work);
4470 for (i = 0; i < ic->n_bitmap_blocks; i++) {
4471 struct bitmap_block_status *bbs = &ic->bbs[i];
4472 unsigned sector, pl_index, pl_offset;
4474 INIT_WORK(&bbs->work, bitmap_block_work);
4477 bio_list_init(&bbs->bio_queue);
4478 spin_lock_init(&bbs->bio_queue_lock);
4480 sector = i * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT);
4481 pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
4482 pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
4484 bbs->bitmap = lowmem_page_address(ic->journal[pl_index].page) + pl_offset;
4488 if (should_write_sb) {
4489 init_journal(ic, 0, ic->journal_sections, 0);
4490 r = dm_integrity_failed(ic);
4492 ti->error = "Error initializing journal";
4495 r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
4497 ti->error = "Error initializing superblock";
4500 ic->just_formatted = true;
4503 if (!ic->meta_dev) {
4504 r = dm_set_target_max_io_len(ti, 1U << ic->sb->log2_interleave_sectors);
4508 if (ic->mode == 'B') {
4509 unsigned max_io_len = ((sector_t)ic->sectors_per_block << ic->log2_blocks_per_bitmap_bit) * (BITMAP_BLOCK_SIZE * 8);
4511 max_io_len = 1U << 31;
4512 DEBUG_print("max_io_len: old %u, new %u\n", ti->max_io_len, max_io_len);
4513 if (!ti->max_io_len || ti->max_io_len > max_io_len) {
4514 r = dm_set_target_max_io_len(ti, max_io_len);
4520 if (!ic->internal_hash)
4521 dm_integrity_set(ti, ic);
4523 ti->num_flush_bios = 1;
4524 ti->flush_supported = true;
4526 ti->num_discard_bios = 1;
4531 dm_integrity_dtr(ti);
4535 static void dm_integrity_dtr(struct dm_target *ti)
4537 struct dm_integrity_c *ic = ti->private;
4539 BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
4540 BUG_ON(!list_empty(&ic->wait_list));
4542 if (ic->mode == 'B')
4543 cancel_delayed_work_sync(&ic->bitmap_flush_work);
4544 if (ic->metadata_wq)
4545 destroy_workqueue(ic->metadata_wq);
4547 destroy_workqueue(ic->wait_wq);
4549 destroy_workqueue(ic->offload_wq);
4551 destroy_workqueue(ic->commit_wq);
4553 destroy_workqueue(ic->writer_wq);
4555 destroy_workqueue(ic->recalc_wq);
4556 vfree(ic->recalc_buffer);
4557 kvfree(ic->recalc_tags);
4560 dm_bufio_client_destroy(ic->bufio);
4561 mempool_exit(&ic->journal_io_mempool);
4563 dm_io_client_destroy(ic->io);
4565 dm_put_device(ti, ic->dev);
4567 dm_put_device(ti, ic->meta_dev);
4568 dm_integrity_free_page_list(ic->journal);
4569 dm_integrity_free_page_list(ic->journal_io);
4570 dm_integrity_free_page_list(ic->journal_xor);
4571 dm_integrity_free_page_list(ic->recalc_bitmap);
4572 dm_integrity_free_page_list(ic->may_write_bitmap);
4573 if (ic->journal_scatterlist)
4574 dm_integrity_free_journal_scatterlist(ic, ic->journal_scatterlist);
4575 if (ic->journal_io_scatterlist)
4576 dm_integrity_free_journal_scatterlist(ic, ic->journal_io_scatterlist);
4577 if (ic->sk_requests) {
4580 for (i = 0; i < ic->journal_sections; i++) {
4581 struct skcipher_request *req = ic->sk_requests[i];
4583 kfree_sensitive(req->iv);
4584 skcipher_request_free(req);
4587 kvfree(ic->sk_requests);
4589 kvfree(ic->journal_tree);
4591 free_pages_exact(ic->sb, SB_SECTORS << SECTOR_SHIFT);
4593 if (ic->internal_hash)
4594 crypto_free_shash(ic->internal_hash);
4595 free_alg(&ic->internal_hash_alg);
4597 if (ic->journal_crypt)
4598 crypto_free_skcipher(ic->journal_crypt);
4599 free_alg(&ic->journal_crypt_alg);
4601 if (ic->journal_mac)
4602 crypto_free_shash(ic->journal_mac);
4603 free_alg(&ic->journal_mac_alg);
4608 static struct target_type integrity_target = {
4609 .name = "integrity",
4610 .version = {1, 10, 0},
4611 .module = THIS_MODULE,
4612 .features = DM_TARGET_SINGLETON | DM_TARGET_INTEGRITY,
4613 .ctr = dm_integrity_ctr,
4614 .dtr = dm_integrity_dtr,
4615 .map = dm_integrity_map,
4616 .postsuspend = dm_integrity_postsuspend,
4617 .resume = dm_integrity_resume,
4618 .status = dm_integrity_status,
4619 .iterate_devices = dm_integrity_iterate_devices,
4620 .io_hints = dm_integrity_io_hints,
4623 static int __init dm_integrity_init(void)
4627 journal_io_cache = kmem_cache_create("integrity_journal_io",
4628 sizeof(struct journal_io), 0, 0, NULL);
4629 if (!journal_io_cache) {
4630 DMERR("can't allocate journal io cache");
4634 r = dm_register_target(&integrity_target);
4637 DMERR("register failed %d", r);
4642 static void __exit dm_integrity_exit(void)
4644 dm_unregister_target(&integrity_target);
4645 kmem_cache_destroy(journal_io_cache);
4648 module_init(dm_integrity_init);
4649 module_exit(dm_integrity_exit);
4651 MODULE_AUTHOR("Milan Broz");
4652 MODULE_AUTHOR("Mikulas Patocka");
4653 MODULE_DESCRIPTION(DM_NAME " target for integrity tags extension");
4654 MODULE_LICENSE("GPL");