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 __u16 integrity_tag_size;
70 __u32 journal_sections;
71 __u64 provided_data_sectors; /* userspace uses this value */
73 __u8 log2_sectors_per_block;
74 __u8 log2_blocks_per_bitmap_bit;
81 #define SB_FLAG_HAVE_JOURNAL_MAC 0x1
82 #define SB_FLAG_RECALCULATING 0x2
83 #define SB_FLAG_DIRTY_BITMAP 0x4
84 #define SB_FLAG_FIXED_PADDING 0x8
85 #define SB_FLAG_FIXED_HMAC 0x10
87 #define JOURNAL_ENTRY_ROUNDUP 8
89 typedef __u64 commit_id_t;
90 #define JOURNAL_MAC_PER_SECTOR 8
92 struct journal_entry {
100 commit_id_t last_bytes[];
104 #define journal_entry_tag(ic, je) ((__u8 *)&(je)->last_bytes[(ic)->sectors_per_block])
106 #if BITS_PER_LONG == 64
107 #define journal_entry_set_sector(je, x) do { smp_wmb(); WRITE_ONCE((je)->u.sector, cpu_to_le64(x)); } while (0)
109 #define journal_entry_set_sector(je, x) do { (je)->u.s.sector_lo = cpu_to_le32(x); smp_wmb(); WRITE_ONCE((je)->u.s.sector_hi, cpu_to_le32((x) >> 32)); } while (0)
111 #define journal_entry_get_sector(je) le64_to_cpu((je)->u.sector)
112 #define journal_entry_is_unused(je) ((je)->u.s.sector_hi == cpu_to_le32(-1))
113 #define journal_entry_set_unused(je) do { ((je)->u.s.sector_hi = cpu_to_le32(-1)); } while (0)
114 #define journal_entry_is_inprogress(je) ((je)->u.s.sector_hi == cpu_to_le32(-2))
115 #define journal_entry_set_inprogress(je) do { ((je)->u.s.sector_hi = cpu_to_le32(-2)); } while (0)
117 #define JOURNAL_BLOCK_SECTORS 8
118 #define JOURNAL_SECTOR_DATA ((1 << SECTOR_SHIFT) - sizeof(commit_id_t))
119 #define JOURNAL_MAC_SIZE (JOURNAL_MAC_PER_SECTOR * JOURNAL_BLOCK_SECTORS)
121 struct journal_sector {
122 __u8 entries[JOURNAL_SECTOR_DATA - JOURNAL_MAC_PER_SECTOR];
123 __u8 mac[JOURNAL_MAC_PER_SECTOR];
124 commit_id_t commit_id;
127 #define MAX_TAG_SIZE (JOURNAL_SECTOR_DATA - JOURNAL_MAC_PER_SECTOR - offsetof(struct journal_entry, last_bytes[MAX_SECTORS_PER_BLOCK]))
129 #define METADATA_PADDING_SECTORS 8
131 #define N_COMMIT_IDS 4
133 static unsigned char prev_commit_seq(unsigned char seq)
135 return (seq + N_COMMIT_IDS - 1) % N_COMMIT_IDS;
138 static unsigned char next_commit_seq(unsigned char seq)
140 return (seq + 1) % N_COMMIT_IDS;
144 * In-memory structures
147 struct journal_node {
159 struct dm_integrity_c {
161 struct dm_dev *meta_dev;
165 mempool_t journal_io_mempool;
166 struct dm_io_client *io;
167 struct dm_bufio_client *bufio;
168 struct workqueue_struct *metadata_wq;
169 struct superblock *sb;
170 unsigned journal_pages;
171 unsigned n_bitmap_blocks;
173 struct page_list *journal;
174 struct page_list *journal_io;
175 struct page_list *journal_xor;
176 struct page_list *recalc_bitmap;
177 struct page_list *may_write_bitmap;
178 struct bitmap_block_status *bbs;
179 unsigned bitmap_flush_interval;
180 int synchronous_mode;
181 struct bio_list synchronous_bios;
182 struct delayed_work bitmap_flush_work;
184 struct crypto_skcipher *journal_crypt;
185 struct scatterlist **journal_scatterlist;
186 struct scatterlist **journal_io_scatterlist;
187 struct skcipher_request **sk_requests;
189 struct crypto_shash *journal_mac;
191 struct journal_node *journal_tree;
192 struct rb_root journal_tree_root;
194 sector_t provided_data_sectors;
196 unsigned short journal_entry_size;
197 unsigned char journal_entries_per_sector;
198 unsigned char journal_section_entries;
199 unsigned short journal_section_sectors;
200 unsigned journal_sections;
201 unsigned journal_entries;
202 sector_t data_device_sectors;
203 sector_t meta_device_sectors;
204 unsigned initial_sectors;
205 unsigned metadata_run;
206 __s8 log2_metadata_run;
207 __u8 log2_buffer_sectors;
208 __u8 sectors_per_block;
209 __u8 log2_blocks_per_bitmap_bit;
215 struct crypto_shash *internal_hash;
217 struct dm_target *ti;
219 /* these variables are locked with endio_wait.lock */
220 struct rb_root in_progress;
221 struct list_head wait_list;
222 wait_queue_head_t endio_wait;
223 struct workqueue_struct *wait_wq;
224 struct workqueue_struct *offload_wq;
226 unsigned char commit_seq;
227 commit_id_t commit_ids[N_COMMIT_IDS];
229 unsigned committed_section;
230 unsigned n_committed_sections;
232 unsigned uncommitted_section;
233 unsigned n_uncommitted_sections;
235 unsigned free_section;
236 unsigned char free_section_entry;
237 unsigned free_sectors;
239 unsigned free_sectors_threshold;
241 struct workqueue_struct *commit_wq;
242 struct work_struct commit_work;
244 struct workqueue_struct *writer_wq;
245 struct work_struct writer_work;
247 struct workqueue_struct *recalc_wq;
248 struct work_struct recalc_work;
252 struct bio_list flush_bio_list;
254 unsigned long autocommit_jiffies;
255 struct timer_list autocommit_timer;
256 unsigned autocommit_msec;
258 wait_queue_head_t copy_to_journal_wait;
260 struct completion crypto_backoff;
262 bool journal_uptodate;
264 bool recalculate_flag;
265 bool reset_recalculate_flag;
269 bool legacy_recalculate;
271 struct alg_spec internal_hash_alg;
272 struct alg_spec journal_crypt_alg;
273 struct alg_spec journal_mac_alg;
275 atomic64_t number_of_mismatches;
277 struct notifier_block reboot_notifier;
280 struct dm_integrity_range {
281 sector_t logical_sector;
287 struct task_struct *task;
288 struct list_head wait_entry;
293 struct dm_integrity_io {
294 struct work_struct work;
296 struct dm_integrity_c *ic;
300 struct dm_integrity_range range;
302 sector_t metadata_block;
303 unsigned metadata_offset;
306 blk_status_t bi_status;
308 struct completion *completion;
310 struct dm_bio_details bio_details;
313 struct journal_completion {
314 struct dm_integrity_c *ic;
316 struct completion comp;
320 struct dm_integrity_range range;
321 struct journal_completion *comp;
324 struct bitmap_block_status {
325 struct work_struct work;
326 struct dm_integrity_c *ic;
328 unsigned long *bitmap;
329 struct bio_list bio_queue;
330 spinlock_t bio_queue_lock;
334 static struct kmem_cache *journal_io_cache;
336 #define JOURNAL_IO_MEMPOOL 32
339 #define DEBUG_print(x, ...) printk(KERN_DEBUG x, ##__VA_ARGS__)
340 static void __DEBUG_bytes(__u8 *bytes, size_t len, const char *msg, ...)
349 pr_cont(" %02x", *bytes);
355 #define DEBUG_bytes(bytes, len, msg, ...) __DEBUG_bytes(bytes, len, KERN_DEBUG msg, ##__VA_ARGS__)
357 #define DEBUG_print(x, ...) do { } while (0)
358 #define DEBUG_bytes(bytes, len, msg, ...) do { } while (0)
361 static void dm_integrity_prepare(struct request *rq)
365 static void dm_integrity_complete(struct request *rq, unsigned int nr_bytes)
370 * DM Integrity profile, protection is performed layer above (dm-crypt)
372 static const struct blk_integrity_profile dm_integrity_profile = {
373 .name = "DM-DIF-EXT-TAG",
376 .prepare_fn = dm_integrity_prepare,
377 .complete_fn = dm_integrity_complete,
380 static void dm_integrity_map_continue(struct dm_integrity_io *dio, bool from_map);
381 static void integrity_bio_wait(struct work_struct *w);
382 static void dm_integrity_dtr(struct dm_target *ti);
384 static void dm_integrity_io_error(struct dm_integrity_c *ic, const char *msg, int err)
387 atomic64_inc(&ic->number_of_mismatches);
388 if (!cmpxchg(&ic->failed, 0, err))
389 DMERR("Error on %s: %d", msg, err);
392 static int dm_integrity_failed(struct dm_integrity_c *ic)
394 return READ_ONCE(ic->failed);
397 static bool dm_integrity_disable_recalculate(struct dm_integrity_c *ic)
399 if (ic->legacy_recalculate)
401 if (!(ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) ?
402 ic->internal_hash_alg.key || ic->journal_mac_alg.key :
403 ic->internal_hash_alg.key && !ic->journal_mac_alg.key)
408 static commit_id_t dm_integrity_commit_id(struct dm_integrity_c *ic, unsigned i,
409 unsigned j, unsigned char seq)
412 * Xor the number with section and sector, so that if a piece of
413 * journal is written at wrong place, it is detected.
415 return ic->commit_ids[seq] ^ cpu_to_le64(((__u64)i << 32) ^ j);
418 static void get_area_and_offset(struct dm_integrity_c *ic, sector_t data_sector,
419 sector_t *area, sector_t *offset)
422 __u8 log2_interleave_sectors = ic->sb->log2_interleave_sectors;
423 *area = data_sector >> log2_interleave_sectors;
424 *offset = (unsigned)data_sector & ((1U << log2_interleave_sectors) - 1);
427 *offset = data_sector;
431 #define sector_to_block(ic, n) \
433 BUG_ON((n) & (unsigned)((ic)->sectors_per_block - 1)); \
434 (n) >>= (ic)->sb->log2_sectors_per_block; \
437 static __u64 get_metadata_sector_and_offset(struct dm_integrity_c *ic, sector_t area,
438 sector_t offset, unsigned *metadata_offset)
443 ms = area << ic->sb->log2_interleave_sectors;
444 if (likely(ic->log2_metadata_run >= 0))
445 ms += area << ic->log2_metadata_run;
447 ms += area * ic->metadata_run;
448 ms >>= ic->log2_buffer_sectors;
450 sector_to_block(ic, offset);
452 if (likely(ic->log2_tag_size >= 0)) {
453 ms += offset >> (SECTOR_SHIFT + ic->log2_buffer_sectors - ic->log2_tag_size);
454 mo = (offset << ic->log2_tag_size) & ((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - 1);
456 ms += (__u64)offset * ic->tag_size >> (SECTOR_SHIFT + ic->log2_buffer_sectors);
457 mo = (offset * ic->tag_size) & ((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - 1);
459 *metadata_offset = mo;
463 static sector_t get_data_sector(struct dm_integrity_c *ic, sector_t area, sector_t offset)
470 result = area << ic->sb->log2_interleave_sectors;
471 if (likely(ic->log2_metadata_run >= 0))
472 result += (area + 1) << ic->log2_metadata_run;
474 result += (area + 1) * ic->metadata_run;
476 result += (sector_t)ic->initial_sectors + offset;
482 static void wraparound_section(struct dm_integrity_c *ic, unsigned *sec_ptr)
484 if (unlikely(*sec_ptr >= ic->journal_sections))
485 *sec_ptr -= ic->journal_sections;
488 static void sb_set_version(struct dm_integrity_c *ic)
490 if (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC))
491 ic->sb->version = SB_VERSION_5;
492 else if (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING))
493 ic->sb->version = SB_VERSION_4;
494 else if (ic->mode == 'B' || ic->sb->flags & cpu_to_le32(SB_FLAG_DIRTY_BITMAP))
495 ic->sb->version = SB_VERSION_3;
496 else if (ic->meta_dev || ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
497 ic->sb->version = SB_VERSION_2;
499 ic->sb->version = SB_VERSION_1;
502 static int sb_mac(struct dm_integrity_c *ic, bool wr)
504 SHASH_DESC_ON_STACK(desc, ic->journal_mac);
506 unsigned size = crypto_shash_digestsize(ic->journal_mac);
508 if (sizeof(struct superblock) + size > 1 << SECTOR_SHIFT) {
509 dm_integrity_io_error(ic, "digest is too long", -EINVAL);
513 desc->tfm = ic->journal_mac;
515 r = crypto_shash_init(desc);
516 if (unlikely(r < 0)) {
517 dm_integrity_io_error(ic, "crypto_shash_init", r);
521 r = crypto_shash_update(desc, (__u8 *)ic->sb, (1 << SECTOR_SHIFT) - size);
522 if (unlikely(r < 0)) {
523 dm_integrity_io_error(ic, "crypto_shash_update", r);
528 r = crypto_shash_final(desc, (__u8 *)ic->sb + (1 << SECTOR_SHIFT) - size);
529 if (unlikely(r < 0)) {
530 dm_integrity_io_error(ic, "crypto_shash_final", r);
534 __u8 result[HASH_MAX_DIGESTSIZE];
535 r = crypto_shash_final(desc, result);
536 if (unlikely(r < 0)) {
537 dm_integrity_io_error(ic, "crypto_shash_final", r);
540 if (memcmp((__u8 *)ic->sb + (1 << SECTOR_SHIFT) - size, result, size)) {
541 dm_integrity_io_error(ic, "superblock mac", -EILSEQ);
549 static int sync_rw_sb(struct dm_integrity_c *ic, int op, int op_flags)
551 struct dm_io_request io_req;
552 struct dm_io_region io_loc;
556 io_req.bi_op_flags = op_flags;
557 io_req.mem.type = DM_IO_KMEM;
558 io_req.mem.ptr.addr = ic->sb;
559 io_req.notify.fn = NULL;
560 io_req.client = ic->io;
561 io_loc.bdev = ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev;
562 io_loc.sector = ic->start;
563 io_loc.count = SB_SECTORS;
565 if (op == REQ_OP_WRITE) {
567 if (ic->journal_mac && ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) {
568 r = sb_mac(ic, true);
574 r = dm_io(&io_req, 1, &io_loc, NULL);
578 if (op == REQ_OP_READ) {
579 if (ic->mode != 'R' && ic->journal_mac && ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) {
580 r = sb_mac(ic, false);
589 #define BITMAP_OP_TEST_ALL_SET 0
590 #define BITMAP_OP_TEST_ALL_CLEAR 1
591 #define BITMAP_OP_SET 2
592 #define BITMAP_OP_CLEAR 3
594 static bool block_bitmap_op(struct dm_integrity_c *ic, struct page_list *bitmap,
595 sector_t sector, sector_t n_sectors, int mode)
597 unsigned long bit, end_bit, this_end_bit, page, end_page;
600 if (unlikely(((sector | n_sectors) & ((1 << ic->sb->log2_sectors_per_block) - 1)) != 0)) {
601 DMCRIT("invalid bitmap access (%llx,%llx,%d,%d,%d)",
604 ic->sb->log2_sectors_per_block,
605 ic->log2_blocks_per_bitmap_bit,
610 if (unlikely(!n_sectors))
613 bit = sector >> (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
614 end_bit = (sector + n_sectors - 1) >>
615 (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
617 page = bit / (PAGE_SIZE * 8);
618 bit %= PAGE_SIZE * 8;
620 end_page = end_bit / (PAGE_SIZE * 8);
621 end_bit %= PAGE_SIZE * 8;
624 if (page < end_page) {
625 this_end_bit = PAGE_SIZE * 8 - 1;
627 this_end_bit = end_bit;
630 data = lowmem_page_address(bitmap[page].page);
632 if (mode == BITMAP_OP_TEST_ALL_SET) {
633 while (bit <= this_end_bit) {
634 if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
636 if (data[bit / BITS_PER_LONG] != -1)
638 bit += BITS_PER_LONG;
639 } while (this_end_bit >= bit + BITS_PER_LONG - 1);
642 if (!test_bit(bit, data))
646 } else if (mode == BITMAP_OP_TEST_ALL_CLEAR) {
647 while (bit <= this_end_bit) {
648 if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
650 if (data[bit / BITS_PER_LONG] != 0)
652 bit += BITS_PER_LONG;
653 } while (this_end_bit >= bit + BITS_PER_LONG - 1);
656 if (test_bit(bit, data))
660 } else if (mode == BITMAP_OP_SET) {
661 while (bit <= this_end_bit) {
662 if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
664 data[bit / BITS_PER_LONG] = -1;
665 bit += BITS_PER_LONG;
666 } while (this_end_bit >= bit + BITS_PER_LONG - 1);
669 __set_bit(bit, data);
672 } else if (mode == BITMAP_OP_CLEAR) {
673 if (!bit && this_end_bit == PAGE_SIZE * 8 - 1)
675 else while (bit <= this_end_bit) {
676 if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
678 data[bit / BITS_PER_LONG] = 0;
679 bit += BITS_PER_LONG;
680 } while (this_end_bit >= bit + BITS_PER_LONG - 1);
683 __clear_bit(bit, data);
690 if (unlikely(page < end_page)) {
699 static void block_bitmap_copy(struct dm_integrity_c *ic, struct page_list *dst, struct page_list *src)
701 unsigned n_bitmap_pages = DIV_ROUND_UP(ic->n_bitmap_blocks, PAGE_SIZE / BITMAP_BLOCK_SIZE);
704 for (i = 0; i < n_bitmap_pages; i++) {
705 unsigned long *dst_data = lowmem_page_address(dst[i].page);
706 unsigned long *src_data = lowmem_page_address(src[i].page);
707 copy_page(dst_data, src_data);
711 static struct bitmap_block_status *sector_to_bitmap_block(struct dm_integrity_c *ic, sector_t sector)
713 unsigned bit = sector >> (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
714 unsigned bitmap_block = bit / (BITMAP_BLOCK_SIZE * 8);
716 BUG_ON(bitmap_block >= ic->n_bitmap_blocks);
717 return &ic->bbs[bitmap_block];
720 static void access_journal_check(struct dm_integrity_c *ic, unsigned section, unsigned offset,
721 bool e, const char *function)
723 #if defined(CONFIG_DM_DEBUG) || defined(INTERNAL_VERIFY)
724 unsigned limit = e ? ic->journal_section_entries : ic->journal_section_sectors;
726 if (unlikely(section >= ic->journal_sections) ||
727 unlikely(offset >= limit)) {
728 DMCRIT("%s: invalid access at (%u,%u), limit (%u,%u)",
729 function, section, offset, ic->journal_sections, limit);
735 static void page_list_location(struct dm_integrity_c *ic, unsigned section, unsigned offset,
736 unsigned *pl_index, unsigned *pl_offset)
740 access_journal_check(ic, section, offset, false, "page_list_location");
742 sector = section * ic->journal_section_sectors + offset;
744 *pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
745 *pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
748 static struct journal_sector *access_page_list(struct dm_integrity_c *ic, struct page_list *pl,
749 unsigned section, unsigned offset, unsigned *n_sectors)
751 unsigned pl_index, pl_offset;
754 page_list_location(ic, section, offset, &pl_index, &pl_offset);
757 *n_sectors = (PAGE_SIZE - pl_offset) >> SECTOR_SHIFT;
759 va = lowmem_page_address(pl[pl_index].page);
761 return (struct journal_sector *)(va + pl_offset);
764 static struct journal_sector *access_journal(struct dm_integrity_c *ic, unsigned section, unsigned offset)
766 return access_page_list(ic, ic->journal, section, offset, NULL);
769 static struct journal_entry *access_journal_entry(struct dm_integrity_c *ic, unsigned section, unsigned n)
771 unsigned rel_sector, offset;
772 struct journal_sector *js;
774 access_journal_check(ic, section, n, true, "access_journal_entry");
776 rel_sector = n % JOURNAL_BLOCK_SECTORS;
777 offset = n / JOURNAL_BLOCK_SECTORS;
779 js = access_journal(ic, section, rel_sector);
780 return (struct journal_entry *)((char *)js + offset * ic->journal_entry_size);
783 static struct journal_sector *access_journal_data(struct dm_integrity_c *ic, unsigned section, unsigned n)
785 n <<= ic->sb->log2_sectors_per_block;
787 n += JOURNAL_BLOCK_SECTORS;
789 access_journal_check(ic, section, n, false, "access_journal_data");
791 return access_journal(ic, section, n);
794 static void section_mac(struct dm_integrity_c *ic, unsigned section, __u8 result[JOURNAL_MAC_SIZE])
796 SHASH_DESC_ON_STACK(desc, ic->journal_mac);
800 desc->tfm = ic->journal_mac;
802 r = crypto_shash_init(desc);
803 if (unlikely(r < 0)) {
804 dm_integrity_io_error(ic, "crypto_shash_init", r);
808 if (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) {
811 r = crypto_shash_update(desc, (__u8 *)&ic->sb->salt, SALT_SIZE);
812 if (unlikely(r < 0)) {
813 dm_integrity_io_error(ic, "crypto_shash_update", r);
817 section_le = cpu_to_le64(section);
818 r = crypto_shash_update(desc, (__u8 *)§ion_le, sizeof section_le);
819 if (unlikely(r < 0)) {
820 dm_integrity_io_error(ic, "crypto_shash_update", r);
825 for (j = 0; j < ic->journal_section_entries; j++) {
826 struct journal_entry *je = access_journal_entry(ic, section, j);
827 r = crypto_shash_update(desc, (__u8 *)&je->u.sector, sizeof je->u.sector);
828 if (unlikely(r < 0)) {
829 dm_integrity_io_error(ic, "crypto_shash_update", r);
834 size = crypto_shash_digestsize(ic->journal_mac);
836 if (likely(size <= JOURNAL_MAC_SIZE)) {
837 r = crypto_shash_final(desc, result);
838 if (unlikely(r < 0)) {
839 dm_integrity_io_error(ic, "crypto_shash_final", r);
842 memset(result + size, 0, JOURNAL_MAC_SIZE - size);
844 __u8 digest[HASH_MAX_DIGESTSIZE];
846 if (WARN_ON(size > sizeof(digest))) {
847 dm_integrity_io_error(ic, "digest_size", -EINVAL);
850 r = crypto_shash_final(desc, digest);
851 if (unlikely(r < 0)) {
852 dm_integrity_io_error(ic, "crypto_shash_final", r);
855 memcpy(result, digest, JOURNAL_MAC_SIZE);
860 memset(result, 0, JOURNAL_MAC_SIZE);
863 static void rw_section_mac(struct dm_integrity_c *ic, unsigned section, bool wr)
865 __u8 result[JOURNAL_MAC_SIZE];
868 if (!ic->journal_mac)
871 section_mac(ic, section, result);
873 for (j = 0; j < JOURNAL_BLOCK_SECTORS; j++) {
874 struct journal_sector *js = access_journal(ic, section, j);
877 memcpy(&js->mac, result + (j * JOURNAL_MAC_PER_SECTOR), JOURNAL_MAC_PER_SECTOR);
879 if (memcmp(&js->mac, result + (j * JOURNAL_MAC_PER_SECTOR), JOURNAL_MAC_PER_SECTOR))
880 dm_integrity_io_error(ic, "journal mac", -EILSEQ);
885 static void complete_journal_op(void *context)
887 struct journal_completion *comp = context;
888 BUG_ON(!atomic_read(&comp->in_flight));
889 if (likely(atomic_dec_and_test(&comp->in_flight)))
890 complete(&comp->comp);
893 static void xor_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
894 unsigned n_sections, struct journal_completion *comp)
896 struct async_submit_ctl submit;
897 size_t n_bytes = (size_t)(n_sections * ic->journal_section_sectors) << SECTOR_SHIFT;
898 unsigned pl_index, pl_offset, section_index;
899 struct page_list *source_pl, *target_pl;
901 if (likely(encrypt)) {
902 source_pl = ic->journal;
903 target_pl = ic->journal_io;
905 source_pl = ic->journal_io;
906 target_pl = ic->journal;
909 page_list_location(ic, section, 0, &pl_index, &pl_offset);
911 atomic_add(roundup(pl_offset + n_bytes, PAGE_SIZE) >> PAGE_SHIFT, &comp->in_flight);
913 init_async_submit(&submit, ASYNC_TX_XOR_ZERO_DST, NULL, complete_journal_op, comp, NULL);
915 section_index = pl_index;
919 struct page *src_pages[2];
920 struct page *dst_page;
922 while (unlikely(pl_index == section_index)) {
925 rw_section_mac(ic, section, true);
930 page_list_location(ic, section, 0, §ion_index, &dummy);
933 this_step = min(n_bytes, (size_t)PAGE_SIZE - pl_offset);
934 dst_page = target_pl[pl_index].page;
935 src_pages[0] = source_pl[pl_index].page;
936 src_pages[1] = ic->journal_xor[pl_index].page;
938 async_xor(dst_page, src_pages, pl_offset, 2, this_step, &submit);
942 n_bytes -= this_step;
947 async_tx_issue_pending_all();
950 static void complete_journal_encrypt(struct crypto_async_request *req, int err)
952 struct journal_completion *comp = req->data;
954 if (likely(err == -EINPROGRESS)) {
955 complete(&comp->ic->crypto_backoff);
958 dm_integrity_io_error(comp->ic, "asynchronous encrypt", err);
960 complete_journal_op(comp);
963 static bool do_crypt(bool encrypt, struct skcipher_request *req, struct journal_completion *comp)
966 skcipher_request_set_callback(req, CRYPTO_TFM_REQ_MAY_BACKLOG,
967 complete_journal_encrypt, comp);
969 r = crypto_skcipher_encrypt(req);
971 r = crypto_skcipher_decrypt(req);
974 if (likely(r == -EINPROGRESS))
976 if (likely(r == -EBUSY)) {
977 wait_for_completion(&comp->ic->crypto_backoff);
978 reinit_completion(&comp->ic->crypto_backoff);
981 dm_integrity_io_error(comp->ic, "encrypt", r);
985 static void crypt_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
986 unsigned n_sections, struct journal_completion *comp)
988 struct scatterlist **source_sg;
989 struct scatterlist **target_sg;
991 atomic_add(2, &comp->in_flight);
993 if (likely(encrypt)) {
994 source_sg = ic->journal_scatterlist;
995 target_sg = ic->journal_io_scatterlist;
997 source_sg = ic->journal_io_scatterlist;
998 target_sg = ic->journal_scatterlist;
1002 struct skcipher_request *req;
1006 if (likely(encrypt))
1007 rw_section_mac(ic, section, true);
1009 req = ic->sk_requests[section];
1010 ivsize = crypto_skcipher_ivsize(ic->journal_crypt);
1013 memcpy(iv, iv + ivsize, ivsize);
1015 req->src = source_sg[section];
1016 req->dst = target_sg[section];
1018 if (unlikely(do_crypt(encrypt, req, comp)))
1019 atomic_inc(&comp->in_flight);
1023 } while (n_sections);
1025 atomic_dec(&comp->in_flight);
1026 complete_journal_op(comp);
1029 static void encrypt_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
1030 unsigned n_sections, struct journal_completion *comp)
1032 if (ic->journal_xor)
1033 return xor_journal(ic, encrypt, section, n_sections, comp);
1035 return crypt_journal(ic, encrypt, section, n_sections, comp);
1038 static void complete_journal_io(unsigned long error, void *context)
1040 struct journal_completion *comp = context;
1041 if (unlikely(error != 0))
1042 dm_integrity_io_error(comp->ic, "writing journal", -EIO);
1043 complete_journal_op(comp);
1046 static void rw_journal_sectors(struct dm_integrity_c *ic, int op, int op_flags,
1047 unsigned sector, unsigned n_sectors, struct journal_completion *comp)
1049 struct dm_io_request io_req;
1050 struct dm_io_region io_loc;
1051 unsigned pl_index, pl_offset;
1054 if (unlikely(dm_integrity_failed(ic))) {
1056 complete_journal_io(-1UL, comp);
1060 pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
1061 pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
1064 io_req.bi_op_flags = op_flags;
1065 io_req.mem.type = DM_IO_PAGE_LIST;
1067 io_req.mem.ptr.pl = &ic->journal_io[pl_index];
1069 io_req.mem.ptr.pl = &ic->journal[pl_index];
1070 io_req.mem.offset = pl_offset;
1071 if (likely(comp != NULL)) {
1072 io_req.notify.fn = complete_journal_io;
1073 io_req.notify.context = comp;
1075 io_req.notify.fn = NULL;
1077 io_req.client = ic->io;
1078 io_loc.bdev = ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev;
1079 io_loc.sector = ic->start + SB_SECTORS + sector;
1080 io_loc.count = n_sectors;
1082 r = dm_io(&io_req, 1, &io_loc, NULL);
1084 dm_integrity_io_error(ic, op == REQ_OP_READ ? "reading journal" : "writing journal", r);
1086 WARN_ONCE(1, "asynchronous dm_io failed: %d", r);
1087 complete_journal_io(-1UL, comp);
1092 static void rw_journal(struct dm_integrity_c *ic, int op, int op_flags, unsigned section,
1093 unsigned n_sections, struct journal_completion *comp)
1095 unsigned sector, n_sectors;
1097 sector = section * ic->journal_section_sectors;
1098 n_sectors = n_sections * ic->journal_section_sectors;
1100 rw_journal_sectors(ic, op, op_flags, sector, n_sectors, comp);
1103 static void write_journal(struct dm_integrity_c *ic, unsigned commit_start, unsigned commit_sections)
1105 struct journal_completion io_comp;
1106 struct journal_completion crypt_comp_1;
1107 struct journal_completion crypt_comp_2;
1111 init_completion(&io_comp.comp);
1113 if (commit_start + commit_sections <= ic->journal_sections) {
1114 io_comp.in_flight = (atomic_t)ATOMIC_INIT(1);
1115 if (ic->journal_io) {
1116 crypt_comp_1.ic = ic;
1117 init_completion(&crypt_comp_1.comp);
1118 crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
1119 encrypt_journal(ic, true, commit_start, commit_sections, &crypt_comp_1);
1120 wait_for_completion_io(&crypt_comp_1.comp);
1122 for (i = 0; i < commit_sections; i++)
1123 rw_section_mac(ic, commit_start + i, true);
1125 rw_journal(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, commit_start,
1126 commit_sections, &io_comp);
1129 io_comp.in_flight = (atomic_t)ATOMIC_INIT(2);
1130 to_end = ic->journal_sections - commit_start;
1131 if (ic->journal_io) {
1132 crypt_comp_1.ic = ic;
1133 init_completion(&crypt_comp_1.comp);
1134 crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
1135 encrypt_journal(ic, true, commit_start, to_end, &crypt_comp_1);
1136 if (try_wait_for_completion(&crypt_comp_1.comp)) {
1137 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
1138 reinit_completion(&crypt_comp_1.comp);
1139 crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
1140 encrypt_journal(ic, true, 0, commit_sections - to_end, &crypt_comp_1);
1141 wait_for_completion_io(&crypt_comp_1.comp);
1143 crypt_comp_2.ic = ic;
1144 init_completion(&crypt_comp_2.comp);
1145 crypt_comp_2.in_flight = (atomic_t)ATOMIC_INIT(0);
1146 encrypt_journal(ic, true, 0, commit_sections - to_end, &crypt_comp_2);
1147 wait_for_completion_io(&crypt_comp_1.comp);
1148 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
1149 wait_for_completion_io(&crypt_comp_2.comp);
1152 for (i = 0; i < to_end; i++)
1153 rw_section_mac(ic, commit_start + i, true);
1154 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
1155 for (i = 0; i < commit_sections - to_end; i++)
1156 rw_section_mac(ic, i, true);
1158 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, 0, commit_sections - to_end, &io_comp);
1161 wait_for_completion_io(&io_comp.comp);
1164 static void copy_from_journal(struct dm_integrity_c *ic, unsigned section, unsigned offset,
1165 unsigned n_sectors, sector_t target, io_notify_fn fn, void *data)
1167 struct dm_io_request io_req;
1168 struct dm_io_region io_loc;
1170 unsigned sector, pl_index, pl_offset;
1172 BUG_ON((target | n_sectors | offset) & (unsigned)(ic->sectors_per_block - 1));
1174 if (unlikely(dm_integrity_failed(ic))) {
1179 sector = section * ic->journal_section_sectors + JOURNAL_BLOCK_SECTORS + offset;
1181 pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
1182 pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
1184 io_req.bi_op = REQ_OP_WRITE;
1185 io_req.bi_op_flags = 0;
1186 io_req.mem.type = DM_IO_PAGE_LIST;
1187 io_req.mem.ptr.pl = &ic->journal[pl_index];
1188 io_req.mem.offset = pl_offset;
1189 io_req.notify.fn = fn;
1190 io_req.notify.context = data;
1191 io_req.client = ic->io;
1192 io_loc.bdev = ic->dev->bdev;
1193 io_loc.sector = target;
1194 io_loc.count = n_sectors;
1196 r = dm_io(&io_req, 1, &io_loc, NULL);
1198 WARN_ONCE(1, "asynchronous dm_io failed: %d", r);
1203 static bool ranges_overlap(struct dm_integrity_range *range1, struct dm_integrity_range *range2)
1205 return range1->logical_sector < range2->logical_sector + range2->n_sectors &&
1206 range1->logical_sector + range1->n_sectors > range2->logical_sector;
1209 static bool add_new_range(struct dm_integrity_c *ic, struct dm_integrity_range *new_range, bool check_waiting)
1211 struct rb_node **n = &ic->in_progress.rb_node;
1212 struct rb_node *parent;
1214 BUG_ON((new_range->logical_sector | new_range->n_sectors) & (unsigned)(ic->sectors_per_block - 1));
1216 if (likely(check_waiting)) {
1217 struct dm_integrity_range *range;
1218 list_for_each_entry(range, &ic->wait_list, wait_entry) {
1219 if (unlikely(ranges_overlap(range, new_range)))
1227 struct dm_integrity_range *range = container_of(*n, struct dm_integrity_range, node);
1230 if (new_range->logical_sector + new_range->n_sectors <= range->logical_sector) {
1231 n = &range->node.rb_left;
1232 } else if (new_range->logical_sector >= range->logical_sector + range->n_sectors) {
1233 n = &range->node.rb_right;
1239 rb_link_node(&new_range->node, parent, n);
1240 rb_insert_color(&new_range->node, &ic->in_progress);
1245 static void remove_range_unlocked(struct dm_integrity_c *ic, struct dm_integrity_range *range)
1247 rb_erase(&range->node, &ic->in_progress);
1248 while (unlikely(!list_empty(&ic->wait_list))) {
1249 struct dm_integrity_range *last_range =
1250 list_first_entry(&ic->wait_list, struct dm_integrity_range, wait_entry);
1251 struct task_struct *last_range_task;
1252 last_range_task = last_range->task;
1253 list_del(&last_range->wait_entry);
1254 if (!add_new_range(ic, last_range, false)) {
1255 last_range->task = last_range_task;
1256 list_add(&last_range->wait_entry, &ic->wait_list);
1259 last_range->waiting = false;
1260 wake_up_process(last_range_task);
1264 static void remove_range(struct dm_integrity_c *ic, struct dm_integrity_range *range)
1266 unsigned long flags;
1268 spin_lock_irqsave(&ic->endio_wait.lock, flags);
1269 remove_range_unlocked(ic, range);
1270 spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
1273 static void wait_and_add_new_range(struct dm_integrity_c *ic, struct dm_integrity_range *new_range)
1275 new_range->waiting = true;
1276 list_add_tail(&new_range->wait_entry, &ic->wait_list);
1277 new_range->task = current;
1279 __set_current_state(TASK_UNINTERRUPTIBLE);
1280 spin_unlock_irq(&ic->endio_wait.lock);
1282 spin_lock_irq(&ic->endio_wait.lock);
1283 } while (unlikely(new_range->waiting));
1286 static void add_new_range_and_wait(struct dm_integrity_c *ic, struct dm_integrity_range *new_range)
1288 if (unlikely(!add_new_range(ic, new_range, true)))
1289 wait_and_add_new_range(ic, new_range);
1292 static void init_journal_node(struct journal_node *node)
1294 RB_CLEAR_NODE(&node->node);
1295 node->sector = (sector_t)-1;
1298 static void add_journal_node(struct dm_integrity_c *ic, struct journal_node *node, sector_t sector)
1300 struct rb_node **link;
1301 struct rb_node *parent;
1303 node->sector = sector;
1304 BUG_ON(!RB_EMPTY_NODE(&node->node));
1306 link = &ic->journal_tree_root.rb_node;
1310 struct journal_node *j;
1312 j = container_of(parent, struct journal_node, node);
1313 if (sector < j->sector)
1314 link = &j->node.rb_left;
1316 link = &j->node.rb_right;
1319 rb_link_node(&node->node, parent, link);
1320 rb_insert_color(&node->node, &ic->journal_tree_root);
1323 static void remove_journal_node(struct dm_integrity_c *ic, struct journal_node *node)
1325 BUG_ON(RB_EMPTY_NODE(&node->node));
1326 rb_erase(&node->node, &ic->journal_tree_root);
1327 init_journal_node(node);
1330 #define NOT_FOUND (-1U)
1332 static unsigned find_journal_node(struct dm_integrity_c *ic, sector_t sector, sector_t *next_sector)
1334 struct rb_node *n = ic->journal_tree_root.rb_node;
1335 unsigned found = NOT_FOUND;
1336 *next_sector = (sector_t)-1;
1338 struct journal_node *j = container_of(n, struct journal_node, node);
1339 if (sector == j->sector) {
1340 found = j - ic->journal_tree;
1342 if (sector < j->sector) {
1343 *next_sector = j->sector;
1344 n = j->node.rb_left;
1346 n = j->node.rb_right;
1353 static bool test_journal_node(struct dm_integrity_c *ic, unsigned pos, sector_t sector)
1355 struct journal_node *node, *next_node;
1356 struct rb_node *next;
1358 if (unlikely(pos >= ic->journal_entries))
1360 node = &ic->journal_tree[pos];
1361 if (unlikely(RB_EMPTY_NODE(&node->node)))
1363 if (unlikely(node->sector != sector))
1366 next = rb_next(&node->node);
1367 if (unlikely(!next))
1370 next_node = container_of(next, struct journal_node, node);
1371 return next_node->sector != sector;
1374 static bool find_newer_committed_node(struct dm_integrity_c *ic, struct journal_node *node)
1376 struct rb_node *next;
1377 struct journal_node *next_node;
1378 unsigned next_section;
1380 BUG_ON(RB_EMPTY_NODE(&node->node));
1382 next = rb_next(&node->node);
1383 if (unlikely(!next))
1386 next_node = container_of(next, struct journal_node, node);
1388 if (next_node->sector != node->sector)
1391 next_section = (unsigned)(next_node - ic->journal_tree) / ic->journal_section_entries;
1392 if (next_section >= ic->committed_section &&
1393 next_section < ic->committed_section + ic->n_committed_sections)
1395 if (next_section + ic->journal_sections < ic->committed_section + ic->n_committed_sections)
1405 static int dm_integrity_rw_tag(struct dm_integrity_c *ic, unsigned char *tag, sector_t *metadata_block,
1406 unsigned *metadata_offset, unsigned total_size, int op)
1408 #define MAY_BE_FILLER 1
1409 #define MAY_BE_HASH 2
1410 unsigned hash_offset = 0;
1411 unsigned may_be = MAY_BE_HASH | (ic->discard ? MAY_BE_FILLER : 0);
1414 unsigned char *data, *dp;
1415 struct dm_buffer *b;
1419 r = dm_integrity_failed(ic);
1423 data = dm_bufio_read(ic->bufio, *metadata_block, &b);
1425 return PTR_ERR(data);
1427 to_copy = min((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - *metadata_offset, total_size);
1428 dp = data + *metadata_offset;
1429 if (op == TAG_READ) {
1430 memcpy(tag, dp, to_copy);
1431 } else if (op == TAG_WRITE) {
1432 if (memcmp(dp, tag, to_copy)) {
1433 memcpy(dp, tag, to_copy);
1434 dm_bufio_mark_partial_buffer_dirty(b, *metadata_offset, *metadata_offset + to_copy);
1437 /* e.g.: op == TAG_CMP */
1439 if (likely(is_power_of_2(ic->tag_size))) {
1440 if (unlikely(memcmp(dp, tag, to_copy)))
1441 if (unlikely(!ic->discard) ||
1442 unlikely(memchr_inv(dp, DISCARD_FILLER, to_copy) != NULL)) {
1450 for (i = 0; i < to_copy; i++, ts--) {
1451 if (unlikely(dp[i] != tag[i]))
1452 may_be &= ~MAY_BE_HASH;
1453 if (likely(dp[i] != DISCARD_FILLER))
1454 may_be &= ~MAY_BE_FILLER;
1456 if (unlikely(hash_offset == ic->tag_size)) {
1457 if (unlikely(!may_be)) {
1458 dm_bufio_release(b);
1462 may_be = MAY_BE_HASH | (ic->discard ? MAY_BE_FILLER : 0);
1467 dm_bufio_release(b);
1470 *metadata_offset += to_copy;
1471 if (unlikely(*metadata_offset == 1U << SECTOR_SHIFT << ic->log2_buffer_sectors)) {
1472 (*metadata_block)++;
1473 *metadata_offset = 0;
1476 if (unlikely(!is_power_of_2(ic->tag_size))) {
1477 hash_offset = (hash_offset + to_copy) % ic->tag_size;
1480 total_size -= to_copy;
1481 } while (unlikely(total_size));
1484 #undef MAY_BE_FILLER
1488 struct flush_request {
1489 struct dm_io_request io_req;
1490 struct dm_io_region io_reg;
1491 struct dm_integrity_c *ic;
1492 struct completion comp;
1495 static void flush_notify(unsigned long error, void *fr_)
1497 struct flush_request *fr = fr_;
1498 if (unlikely(error != 0))
1499 dm_integrity_io_error(fr->ic, "flushing disk cache", -EIO);
1500 complete(&fr->comp);
1503 static void dm_integrity_flush_buffers(struct dm_integrity_c *ic, bool flush_data)
1507 struct flush_request fr;
1512 fr.io_req.bi_op = REQ_OP_WRITE,
1513 fr.io_req.bi_op_flags = REQ_PREFLUSH | REQ_SYNC,
1514 fr.io_req.mem.type = DM_IO_KMEM,
1515 fr.io_req.mem.ptr.addr = NULL,
1516 fr.io_req.notify.fn = flush_notify,
1517 fr.io_req.notify.context = &fr;
1518 fr.io_req.client = dm_bufio_get_dm_io_client(ic->bufio),
1519 fr.io_reg.bdev = ic->dev->bdev,
1520 fr.io_reg.sector = 0,
1521 fr.io_reg.count = 0,
1523 init_completion(&fr.comp);
1524 r = dm_io(&fr.io_req, 1, &fr.io_reg, NULL);
1528 r = dm_bufio_write_dirty_buffers(ic->bufio);
1530 dm_integrity_io_error(ic, "writing tags", r);
1533 wait_for_completion(&fr.comp);
1536 static void sleep_on_endio_wait(struct dm_integrity_c *ic)
1538 DECLARE_WAITQUEUE(wait, current);
1539 __add_wait_queue(&ic->endio_wait, &wait);
1540 __set_current_state(TASK_UNINTERRUPTIBLE);
1541 spin_unlock_irq(&ic->endio_wait.lock);
1543 spin_lock_irq(&ic->endio_wait.lock);
1544 __remove_wait_queue(&ic->endio_wait, &wait);
1547 static void autocommit_fn(struct timer_list *t)
1549 struct dm_integrity_c *ic = from_timer(ic, t, autocommit_timer);
1551 if (likely(!dm_integrity_failed(ic)))
1552 queue_work(ic->commit_wq, &ic->commit_work);
1555 static void schedule_autocommit(struct dm_integrity_c *ic)
1557 if (!timer_pending(&ic->autocommit_timer))
1558 mod_timer(&ic->autocommit_timer, jiffies + ic->autocommit_jiffies);
1561 static void submit_flush_bio(struct dm_integrity_c *ic, struct dm_integrity_io *dio)
1564 unsigned long flags;
1566 spin_lock_irqsave(&ic->endio_wait.lock, flags);
1567 bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1568 bio_list_add(&ic->flush_bio_list, bio);
1569 spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
1571 queue_work(ic->commit_wq, &ic->commit_work);
1574 static void do_endio(struct dm_integrity_c *ic, struct bio *bio)
1576 int r = dm_integrity_failed(ic);
1577 if (unlikely(r) && !bio->bi_status)
1578 bio->bi_status = errno_to_blk_status(r);
1579 if (unlikely(ic->synchronous_mode) && bio_op(bio) == REQ_OP_WRITE) {
1580 unsigned long flags;
1581 spin_lock_irqsave(&ic->endio_wait.lock, flags);
1582 bio_list_add(&ic->synchronous_bios, bio);
1583 queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, 0);
1584 spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
1590 static void do_endio_flush(struct dm_integrity_c *ic, struct dm_integrity_io *dio)
1592 struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1594 if (unlikely(dio->fua) && likely(!bio->bi_status) && likely(!dm_integrity_failed(ic)))
1595 submit_flush_bio(ic, dio);
1600 static void dec_in_flight(struct dm_integrity_io *dio)
1602 if (atomic_dec_and_test(&dio->in_flight)) {
1603 struct dm_integrity_c *ic = dio->ic;
1606 remove_range(ic, &dio->range);
1608 if (dio->op == REQ_OP_WRITE || unlikely(dio->op == REQ_OP_DISCARD))
1609 schedule_autocommit(ic);
1611 bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1613 if (unlikely(dio->bi_status) && !bio->bi_status)
1614 bio->bi_status = dio->bi_status;
1615 if (likely(!bio->bi_status) && unlikely(bio_sectors(bio) != dio->range.n_sectors)) {
1616 dio->range.logical_sector += dio->range.n_sectors;
1617 bio_advance(bio, dio->range.n_sectors << SECTOR_SHIFT);
1618 INIT_WORK(&dio->work, integrity_bio_wait);
1619 queue_work(ic->offload_wq, &dio->work);
1622 do_endio_flush(ic, dio);
1626 static void integrity_end_io(struct bio *bio)
1628 struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
1630 dm_bio_restore(&dio->bio_details, bio);
1631 if (bio->bi_integrity)
1632 bio->bi_opf |= REQ_INTEGRITY;
1634 if (dio->completion)
1635 complete(dio->completion);
1640 static void integrity_sector_checksum(struct dm_integrity_c *ic, sector_t sector,
1641 const char *data, char *result)
1643 __u64 sector_le = cpu_to_le64(sector);
1644 SHASH_DESC_ON_STACK(req, ic->internal_hash);
1646 unsigned digest_size;
1648 req->tfm = ic->internal_hash;
1650 r = crypto_shash_init(req);
1651 if (unlikely(r < 0)) {
1652 dm_integrity_io_error(ic, "crypto_shash_init", r);
1656 if (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) {
1657 r = crypto_shash_update(req, (__u8 *)&ic->sb->salt, SALT_SIZE);
1658 if (unlikely(r < 0)) {
1659 dm_integrity_io_error(ic, "crypto_shash_update", r);
1664 r = crypto_shash_update(req, (const __u8 *)§or_le, sizeof sector_le);
1665 if (unlikely(r < 0)) {
1666 dm_integrity_io_error(ic, "crypto_shash_update", r);
1670 r = crypto_shash_update(req, data, ic->sectors_per_block << SECTOR_SHIFT);
1671 if (unlikely(r < 0)) {
1672 dm_integrity_io_error(ic, "crypto_shash_update", r);
1676 r = crypto_shash_final(req, result);
1677 if (unlikely(r < 0)) {
1678 dm_integrity_io_error(ic, "crypto_shash_final", r);
1682 digest_size = crypto_shash_digestsize(ic->internal_hash);
1683 if (unlikely(digest_size < ic->tag_size))
1684 memset(result + digest_size, 0, ic->tag_size - digest_size);
1689 /* this shouldn't happen anyway, the hash functions have no reason to fail */
1690 get_random_bytes(result, ic->tag_size);
1693 static void integrity_metadata(struct work_struct *w)
1695 struct dm_integrity_io *dio = container_of(w, struct dm_integrity_io, work);
1696 struct dm_integrity_c *ic = dio->ic;
1700 if (ic->internal_hash) {
1701 struct bvec_iter iter;
1703 unsigned digest_size = crypto_shash_digestsize(ic->internal_hash);
1704 struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1706 unsigned extra_space = unlikely(digest_size > ic->tag_size) ? digest_size - ic->tag_size : 0;
1707 char checksums_onstack[max((size_t)HASH_MAX_DIGESTSIZE, MAX_TAG_SIZE)];
1709 unsigned sectors_to_process;
1711 if (unlikely(ic->mode == 'R'))
1714 if (likely(dio->op != REQ_OP_DISCARD))
1715 checksums = kmalloc((PAGE_SIZE >> SECTOR_SHIFT >> ic->sb->log2_sectors_per_block) * ic->tag_size + extra_space,
1716 GFP_NOIO | __GFP_NORETRY | __GFP_NOWARN);
1718 checksums = kmalloc(PAGE_SIZE, GFP_NOIO | __GFP_NORETRY | __GFP_NOWARN);
1720 checksums = checksums_onstack;
1721 if (WARN_ON(extra_space &&
1722 digest_size > sizeof(checksums_onstack))) {
1728 if (unlikely(dio->op == REQ_OP_DISCARD)) {
1729 sector_t bi_sector = dio->bio_details.bi_iter.bi_sector;
1730 unsigned bi_size = dio->bio_details.bi_iter.bi_size;
1731 unsigned max_size = likely(checksums != checksums_onstack) ? PAGE_SIZE : HASH_MAX_DIGESTSIZE;
1732 unsigned max_blocks = max_size / ic->tag_size;
1733 memset(checksums, DISCARD_FILLER, max_size);
1736 unsigned this_step_blocks = bi_size >> (SECTOR_SHIFT + ic->sb->log2_sectors_per_block);
1737 this_step_blocks = min(this_step_blocks, max_blocks);
1738 r = dm_integrity_rw_tag(ic, checksums, &dio->metadata_block, &dio->metadata_offset,
1739 this_step_blocks * ic->tag_size, TAG_WRITE);
1741 if (likely(checksums != checksums_onstack))
1746 /*if (bi_size < this_step_blocks << (SECTOR_SHIFT + ic->sb->log2_sectors_per_block)) {
1747 printk("BUGG: bi_sector: %llx, bi_size: %u\n", bi_sector, bi_size);
1748 printk("BUGG: this_step_blocks: %u\n", this_step_blocks);
1751 bi_size -= this_step_blocks << (SECTOR_SHIFT + ic->sb->log2_sectors_per_block);
1752 bi_sector += this_step_blocks << ic->sb->log2_sectors_per_block;
1755 if (likely(checksums != checksums_onstack))
1760 sector = dio->range.logical_sector;
1761 sectors_to_process = dio->range.n_sectors;
1763 __bio_for_each_segment(bv, bio, iter, dio->bio_details.bi_iter) {
1765 char *mem, *checksums_ptr;
1768 mem = (char *)kmap_atomic(bv.bv_page) + bv.bv_offset;
1770 checksums_ptr = checksums;
1772 integrity_sector_checksum(ic, sector, mem + pos, checksums_ptr);
1773 checksums_ptr += ic->tag_size;
1774 sectors_to_process -= ic->sectors_per_block;
1775 pos += ic->sectors_per_block << SECTOR_SHIFT;
1776 sector += ic->sectors_per_block;
1777 } while (pos < bv.bv_len && sectors_to_process && checksums != checksums_onstack);
1780 r = dm_integrity_rw_tag(ic, checksums, &dio->metadata_block, &dio->metadata_offset,
1781 checksums_ptr - checksums, dio->op == REQ_OP_READ ? TAG_CMP : TAG_WRITE);
1784 char b[BDEVNAME_SIZE];
1785 DMERR_LIMIT("%s: Checksum failed at sector 0x%llx", bio_devname(bio, b),
1786 (sector - ((r + ic->tag_size - 1) / ic->tag_size)));
1788 atomic64_inc(&ic->number_of_mismatches);
1790 if (likely(checksums != checksums_onstack))
1795 if (!sectors_to_process)
1798 if (unlikely(pos < bv.bv_len)) {
1799 bv.bv_offset += pos;
1805 if (likely(checksums != checksums_onstack))
1808 struct bio_integrity_payload *bip = dio->bio_details.bi_integrity;
1812 struct bvec_iter iter;
1813 unsigned data_to_process = dio->range.n_sectors;
1814 sector_to_block(ic, data_to_process);
1815 data_to_process *= ic->tag_size;
1817 bip_for_each_vec(biv, bip, iter) {
1821 BUG_ON(PageHighMem(biv.bv_page));
1822 tag = lowmem_page_address(biv.bv_page) + biv.bv_offset;
1823 this_len = min(biv.bv_len, data_to_process);
1824 r = dm_integrity_rw_tag(ic, tag, &dio->metadata_block, &dio->metadata_offset,
1825 this_len, dio->op == REQ_OP_READ ? TAG_READ : TAG_WRITE);
1828 data_to_process -= this_len;
1829 if (!data_to_process)
1838 dio->bi_status = errno_to_blk_status(r);
1842 static int dm_integrity_map(struct dm_target *ti, struct bio *bio)
1844 struct dm_integrity_c *ic = ti->private;
1845 struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
1846 struct bio_integrity_payload *bip;
1848 sector_t area, offset;
1852 dio->op = bio_op(bio);
1854 if (unlikely(dio->op == REQ_OP_DISCARD)) {
1855 if (ti->max_io_len) {
1856 sector_t sec = dm_target_offset(ti, bio->bi_iter.bi_sector);
1857 unsigned log2_max_io_len = __fls(ti->max_io_len);
1858 sector_t start_boundary = sec >> log2_max_io_len;
1859 sector_t end_boundary = (sec + bio_sectors(bio) - 1) >> log2_max_io_len;
1860 if (start_boundary < end_boundary) {
1861 sector_t len = ti->max_io_len - (sec & (ti->max_io_len - 1));
1862 dm_accept_partial_bio(bio, len);
1867 if (unlikely(bio->bi_opf & REQ_PREFLUSH)) {
1868 submit_flush_bio(ic, dio);
1869 return DM_MAPIO_SUBMITTED;
1872 dio->range.logical_sector = dm_target_offset(ti, bio->bi_iter.bi_sector);
1873 dio->fua = dio->op == REQ_OP_WRITE && bio->bi_opf & REQ_FUA;
1874 if (unlikely(dio->fua)) {
1876 * Don't pass down the FUA flag because we have to flush
1877 * disk cache anyway.
1879 bio->bi_opf &= ~REQ_FUA;
1881 if (unlikely(dio->range.logical_sector + bio_sectors(bio) > ic->provided_data_sectors)) {
1882 DMERR("Too big sector number: 0x%llx + 0x%x > 0x%llx",
1883 dio->range.logical_sector, bio_sectors(bio),
1884 ic->provided_data_sectors);
1885 return DM_MAPIO_KILL;
1887 if (unlikely((dio->range.logical_sector | bio_sectors(bio)) & (unsigned)(ic->sectors_per_block - 1))) {
1888 DMERR("Bio not aligned on %u sectors: 0x%llx, 0x%x",
1889 ic->sectors_per_block,
1890 dio->range.logical_sector, bio_sectors(bio));
1891 return DM_MAPIO_KILL;
1894 if (ic->sectors_per_block > 1 && likely(dio->op != REQ_OP_DISCARD)) {
1895 struct bvec_iter iter;
1897 bio_for_each_segment(bv, bio, iter) {
1898 if (unlikely(bv.bv_len & ((ic->sectors_per_block << SECTOR_SHIFT) - 1))) {
1899 DMERR("Bio vector (%u,%u) is not aligned on %u-sector boundary",
1900 bv.bv_offset, bv.bv_len, ic->sectors_per_block);
1901 return DM_MAPIO_KILL;
1906 bip = bio_integrity(bio);
1907 if (!ic->internal_hash) {
1909 unsigned wanted_tag_size = bio_sectors(bio) >> ic->sb->log2_sectors_per_block;
1910 if (ic->log2_tag_size >= 0)
1911 wanted_tag_size <<= ic->log2_tag_size;
1913 wanted_tag_size *= ic->tag_size;
1914 if (unlikely(wanted_tag_size != bip->bip_iter.bi_size)) {
1915 DMERR("Invalid integrity data size %u, expected %u",
1916 bip->bip_iter.bi_size, wanted_tag_size);
1917 return DM_MAPIO_KILL;
1921 if (unlikely(bip != NULL)) {
1922 DMERR("Unexpected integrity data when using internal hash");
1923 return DM_MAPIO_KILL;
1927 if (unlikely(ic->mode == 'R') && unlikely(dio->op != REQ_OP_READ))
1928 return DM_MAPIO_KILL;
1930 get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
1931 dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset, &dio->metadata_offset);
1932 bio->bi_iter.bi_sector = get_data_sector(ic, area, offset);
1934 dm_integrity_map_continue(dio, true);
1935 return DM_MAPIO_SUBMITTED;
1938 static bool __journal_read_write(struct dm_integrity_io *dio, struct bio *bio,
1939 unsigned journal_section, unsigned journal_entry)
1941 struct dm_integrity_c *ic = dio->ic;
1942 sector_t logical_sector;
1945 logical_sector = dio->range.logical_sector;
1946 n_sectors = dio->range.n_sectors;
1948 struct bio_vec bv = bio_iovec(bio);
1951 if (unlikely(bv.bv_len >> SECTOR_SHIFT > n_sectors))
1952 bv.bv_len = n_sectors << SECTOR_SHIFT;
1953 n_sectors -= bv.bv_len >> SECTOR_SHIFT;
1954 bio_advance_iter(bio, &bio->bi_iter, bv.bv_len);
1956 mem = kmap_atomic(bv.bv_page);
1957 if (likely(dio->op == REQ_OP_WRITE))
1958 flush_dcache_page(bv.bv_page);
1961 struct journal_entry *je = access_journal_entry(ic, journal_section, journal_entry);
1963 if (unlikely(dio->op == REQ_OP_READ)) {
1964 struct journal_sector *js;
1968 if (unlikely(journal_entry_is_inprogress(je))) {
1969 flush_dcache_page(bv.bv_page);
1972 __io_wait_event(ic->copy_to_journal_wait, !journal_entry_is_inprogress(je));
1976 BUG_ON(journal_entry_get_sector(je) != logical_sector);
1977 js = access_journal_data(ic, journal_section, journal_entry);
1978 mem_ptr = mem + bv.bv_offset;
1981 memcpy(mem_ptr, js, JOURNAL_SECTOR_DATA);
1982 *(commit_id_t *)(mem_ptr + JOURNAL_SECTOR_DATA) = je->last_bytes[s];
1984 mem_ptr += 1 << SECTOR_SHIFT;
1985 } while (++s < ic->sectors_per_block);
1986 #ifdef INTERNAL_VERIFY
1987 if (ic->internal_hash) {
1988 char checksums_onstack[max((size_t)HASH_MAX_DIGESTSIZE, MAX_TAG_SIZE)];
1990 integrity_sector_checksum(ic, logical_sector, mem + bv.bv_offset, checksums_onstack);
1991 if (unlikely(memcmp(checksums_onstack, journal_entry_tag(ic, je), ic->tag_size))) {
1992 DMERR_LIMIT("Checksum failed when reading from journal, at sector 0x%llx",
1999 if (!ic->internal_hash) {
2000 struct bio_integrity_payload *bip = bio_integrity(bio);
2001 unsigned tag_todo = ic->tag_size;
2002 char *tag_ptr = journal_entry_tag(ic, je);
2005 struct bio_vec biv = bvec_iter_bvec(bip->bip_vec, bip->bip_iter);
2006 unsigned tag_now = min(biv.bv_len, tag_todo);
2008 BUG_ON(PageHighMem(biv.bv_page));
2009 tag_addr = lowmem_page_address(biv.bv_page) + biv.bv_offset;
2010 if (likely(dio->op == REQ_OP_WRITE))
2011 memcpy(tag_ptr, tag_addr, tag_now);
2013 memcpy(tag_addr, tag_ptr, tag_now);
2014 bvec_iter_advance(bip->bip_vec, &bip->bip_iter, tag_now);
2016 tag_todo -= tag_now;
2017 } while (unlikely(tag_todo)); else {
2018 if (likely(dio->op == REQ_OP_WRITE))
2019 memset(tag_ptr, 0, tag_todo);
2023 if (likely(dio->op == REQ_OP_WRITE)) {
2024 struct journal_sector *js;
2027 js = access_journal_data(ic, journal_section, journal_entry);
2028 memcpy(js, mem + bv.bv_offset, ic->sectors_per_block << SECTOR_SHIFT);
2032 je->last_bytes[s] = js[s].commit_id;
2033 } while (++s < ic->sectors_per_block);
2035 if (ic->internal_hash) {
2036 unsigned digest_size = crypto_shash_digestsize(ic->internal_hash);
2037 if (unlikely(digest_size > ic->tag_size)) {
2038 char checksums_onstack[HASH_MAX_DIGESTSIZE];
2039 integrity_sector_checksum(ic, logical_sector, (char *)js, checksums_onstack);
2040 memcpy(journal_entry_tag(ic, je), checksums_onstack, ic->tag_size);
2042 integrity_sector_checksum(ic, logical_sector, (char *)js, journal_entry_tag(ic, je));
2045 journal_entry_set_sector(je, logical_sector);
2047 logical_sector += ic->sectors_per_block;
2050 if (unlikely(journal_entry == ic->journal_section_entries)) {
2053 wraparound_section(ic, &journal_section);
2056 bv.bv_offset += ic->sectors_per_block << SECTOR_SHIFT;
2057 } while (bv.bv_len -= ic->sectors_per_block << SECTOR_SHIFT);
2059 if (unlikely(dio->op == REQ_OP_READ))
2060 flush_dcache_page(bv.bv_page);
2062 } while (n_sectors);
2064 if (likely(dio->op == REQ_OP_WRITE)) {
2066 if (unlikely(waitqueue_active(&ic->copy_to_journal_wait)))
2067 wake_up(&ic->copy_to_journal_wait);
2068 if (READ_ONCE(ic->free_sectors) <= ic->free_sectors_threshold) {
2069 queue_work(ic->commit_wq, &ic->commit_work);
2071 schedule_autocommit(ic);
2074 remove_range(ic, &dio->range);
2077 if (unlikely(bio->bi_iter.bi_size)) {
2078 sector_t area, offset;
2080 dio->range.logical_sector = logical_sector;
2081 get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
2082 dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset, &dio->metadata_offset);
2089 static void dm_integrity_map_continue(struct dm_integrity_io *dio, bool from_map)
2091 struct dm_integrity_c *ic = dio->ic;
2092 struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
2093 unsigned journal_section, journal_entry;
2094 unsigned journal_read_pos;
2095 struct completion read_comp;
2096 bool discard_retried = false;
2097 bool need_sync_io = ic->internal_hash && dio->op == REQ_OP_READ;
2098 if (unlikely(dio->op == REQ_OP_DISCARD) && ic->mode != 'D')
2099 need_sync_io = true;
2101 if (need_sync_io && from_map) {
2102 INIT_WORK(&dio->work, integrity_bio_wait);
2103 queue_work(ic->offload_wq, &dio->work);
2108 spin_lock_irq(&ic->endio_wait.lock);
2110 if (unlikely(dm_integrity_failed(ic))) {
2111 spin_unlock_irq(&ic->endio_wait.lock);
2115 dio->range.n_sectors = bio_sectors(bio);
2116 journal_read_pos = NOT_FOUND;
2117 if (ic->mode == 'J' && likely(dio->op != REQ_OP_DISCARD)) {
2118 if (dio->op == REQ_OP_WRITE) {
2119 unsigned next_entry, i, pos;
2120 unsigned ws, we, range_sectors;
2122 dio->range.n_sectors = min(dio->range.n_sectors,
2123 (sector_t)ic->free_sectors << ic->sb->log2_sectors_per_block);
2124 if (unlikely(!dio->range.n_sectors)) {
2126 goto offload_to_thread;
2127 sleep_on_endio_wait(ic);
2130 range_sectors = dio->range.n_sectors >> ic->sb->log2_sectors_per_block;
2131 ic->free_sectors -= range_sectors;
2132 journal_section = ic->free_section;
2133 journal_entry = ic->free_section_entry;
2135 next_entry = ic->free_section_entry + range_sectors;
2136 ic->free_section_entry = next_entry % ic->journal_section_entries;
2137 ic->free_section += next_entry / ic->journal_section_entries;
2138 ic->n_uncommitted_sections += next_entry / ic->journal_section_entries;
2139 wraparound_section(ic, &ic->free_section);
2141 pos = journal_section * ic->journal_section_entries + journal_entry;
2142 ws = journal_section;
2146 struct journal_entry *je;
2148 add_journal_node(ic, &ic->journal_tree[pos], dio->range.logical_sector + i);
2150 if (unlikely(pos >= ic->journal_entries))
2153 je = access_journal_entry(ic, ws, we);
2154 BUG_ON(!journal_entry_is_unused(je));
2155 journal_entry_set_inprogress(je);
2157 if (unlikely(we == ic->journal_section_entries)) {
2160 wraparound_section(ic, &ws);
2162 } while ((i += ic->sectors_per_block) < dio->range.n_sectors);
2164 spin_unlock_irq(&ic->endio_wait.lock);
2165 goto journal_read_write;
2167 sector_t next_sector;
2168 journal_read_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
2169 if (likely(journal_read_pos == NOT_FOUND)) {
2170 if (unlikely(dio->range.n_sectors > next_sector - dio->range.logical_sector))
2171 dio->range.n_sectors = next_sector - dio->range.logical_sector;
2174 unsigned jp = journal_read_pos + 1;
2175 for (i = ic->sectors_per_block; i < dio->range.n_sectors; i += ic->sectors_per_block, jp++) {
2176 if (!test_journal_node(ic, jp, dio->range.logical_sector + i))
2179 dio->range.n_sectors = i;
2183 if (unlikely(!add_new_range(ic, &dio->range, true))) {
2185 * We must not sleep in the request routine because it could
2186 * stall bios on current->bio_list.
2187 * So, we offload the bio to a workqueue if we have to sleep.
2191 spin_unlock_irq(&ic->endio_wait.lock);
2192 INIT_WORK(&dio->work, integrity_bio_wait);
2193 queue_work(ic->wait_wq, &dio->work);
2196 if (journal_read_pos != NOT_FOUND)
2197 dio->range.n_sectors = ic->sectors_per_block;
2198 wait_and_add_new_range(ic, &dio->range);
2200 * wait_and_add_new_range drops the spinlock, so the journal
2201 * may have been changed arbitrarily. We need to recheck.
2202 * To simplify the code, we restrict I/O size to just one block.
2204 if (journal_read_pos != NOT_FOUND) {
2205 sector_t next_sector;
2206 unsigned new_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
2207 if (unlikely(new_pos != journal_read_pos)) {
2208 remove_range_unlocked(ic, &dio->range);
2213 if (ic->mode == 'J' && likely(dio->op == REQ_OP_DISCARD) && !discard_retried) {
2214 sector_t next_sector;
2215 unsigned new_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
2216 if (unlikely(new_pos != NOT_FOUND) ||
2217 unlikely(next_sector < dio->range.logical_sector - dio->range.n_sectors)) {
2218 remove_range_unlocked(ic, &dio->range);
2219 spin_unlock_irq(&ic->endio_wait.lock);
2220 queue_work(ic->commit_wq, &ic->commit_work);
2221 flush_workqueue(ic->commit_wq);
2222 queue_work(ic->writer_wq, &ic->writer_work);
2223 flush_workqueue(ic->writer_wq);
2224 discard_retried = true;
2228 spin_unlock_irq(&ic->endio_wait.lock);
2230 if (unlikely(journal_read_pos != NOT_FOUND)) {
2231 journal_section = journal_read_pos / ic->journal_section_entries;
2232 journal_entry = journal_read_pos % ic->journal_section_entries;
2233 goto journal_read_write;
2236 if (ic->mode == 'B' && (dio->op == REQ_OP_WRITE || unlikely(dio->op == REQ_OP_DISCARD))) {
2237 if (!block_bitmap_op(ic, ic->may_write_bitmap, dio->range.logical_sector,
2238 dio->range.n_sectors, BITMAP_OP_TEST_ALL_SET)) {
2239 struct bitmap_block_status *bbs;
2241 bbs = sector_to_bitmap_block(ic, dio->range.logical_sector);
2242 spin_lock(&bbs->bio_queue_lock);
2243 bio_list_add(&bbs->bio_queue, bio);
2244 spin_unlock(&bbs->bio_queue_lock);
2245 queue_work(ic->writer_wq, &bbs->work);
2250 dio->in_flight = (atomic_t)ATOMIC_INIT(2);
2253 init_completion(&read_comp);
2254 dio->completion = &read_comp;
2256 dio->completion = NULL;
2258 dm_bio_record(&dio->bio_details, bio);
2259 bio_set_dev(bio, ic->dev->bdev);
2260 bio->bi_integrity = NULL;
2261 bio->bi_opf &= ~REQ_INTEGRITY;
2262 bio->bi_end_io = integrity_end_io;
2263 bio->bi_iter.bi_size = dio->range.n_sectors << SECTOR_SHIFT;
2265 if (unlikely(dio->op == REQ_OP_DISCARD) && likely(ic->mode != 'D')) {
2266 integrity_metadata(&dio->work);
2267 dm_integrity_flush_buffers(ic, false);
2269 dio->in_flight = (atomic_t)ATOMIC_INIT(1);
2270 dio->completion = NULL;
2272 submit_bio_noacct(bio);
2277 submit_bio_noacct(bio);
2280 wait_for_completion_io(&read_comp);
2281 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING) &&
2282 dio->range.logical_sector + dio->range.n_sectors > le64_to_cpu(ic->sb->recalc_sector))
2284 if (ic->mode == 'B') {
2285 if (!block_bitmap_op(ic, ic->recalc_bitmap, dio->range.logical_sector,
2286 dio->range.n_sectors, BITMAP_OP_TEST_ALL_CLEAR))
2290 if (likely(!bio->bi_status))
2291 integrity_metadata(&dio->work);
2297 INIT_WORK(&dio->work, integrity_metadata);
2298 queue_work(ic->metadata_wq, &dio->work);
2304 if (unlikely(__journal_read_write(dio, bio, journal_section, journal_entry)))
2307 do_endio_flush(ic, dio);
2311 static void integrity_bio_wait(struct work_struct *w)
2313 struct dm_integrity_io *dio = container_of(w, struct dm_integrity_io, work);
2315 dm_integrity_map_continue(dio, false);
2318 static void pad_uncommitted(struct dm_integrity_c *ic)
2320 if (ic->free_section_entry) {
2321 ic->free_sectors -= ic->journal_section_entries - ic->free_section_entry;
2322 ic->free_section_entry = 0;
2324 wraparound_section(ic, &ic->free_section);
2325 ic->n_uncommitted_sections++;
2327 if (WARN_ON(ic->journal_sections * ic->journal_section_entries !=
2328 (ic->n_uncommitted_sections + ic->n_committed_sections) *
2329 ic->journal_section_entries + ic->free_sectors)) {
2330 DMCRIT("journal_sections %u, journal_section_entries %u, "
2331 "n_uncommitted_sections %u, n_committed_sections %u, "
2332 "journal_section_entries %u, free_sectors %u",
2333 ic->journal_sections, ic->journal_section_entries,
2334 ic->n_uncommitted_sections, ic->n_committed_sections,
2335 ic->journal_section_entries, ic->free_sectors);
2339 static void integrity_commit(struct work_struct *w)
2341 struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, commit_work);
2342 unsigned commit_start, commit_sections;
2344 struct bio *flushes;
2346 del_timer(&ic->autocommit_timer);
2348 spin_lock_irq(&ic->endio_wait.lock);
2349 flushes = bio_list_get(&ic->flush_bio_list);
2350 if (unlikely(ic->mode != 'J')) {
2351 spin_unlock_irq(&ic->endio_wait.lock);
2352 dm_integrity_flush_buffers(ic, true);
2353 goto release_flush_bios;
2356 pad_uncommitted(ic);
2357 commit_start = ic->uncommitted_section;
2358 commit_sections = ic->n_uncommitted_sections;
2359 spin_unlock_irq(&ic->endio_wait.lock);
2361 if (!commit_sections)
2362 goto release_flush_bios;
2365 for (n = 0; n < commit_sections; n++) {
2366 for (j = 0; j < ic->journal_section_entries; j++) {
2367 struct journal_entry *je;
2368 je = access_journal_entry(ic, i, j);
2369 io_wait_event(ic->copy_to_journal_wait, !journal_entry_is_inprogress(je));
2371 for (j = 0; j < ic->journal_section_sectors; j++) {
2372 struct journal_sector *js;
2373 js = access_journal(ic, i, j);
2374 js->commit_id = dm_integrity_commit_id(ic, i, j, ic->commit_seq);
2377 if (unlikely(i >= ic->journal_sections))
2378 ic->commit_seq = next_commit_seq(ic->commit_seq);
2379 wraparound_section(ic, &i);
2383 write_journal(ic, commit_start, commit_sections);
2385 spin_lock_irq(&ic->endio_wait.lock);
2386 ic->uncommitted_section += commit_sections;
2387 wraparound_section(ic, &ic->uncommitted_section);
2388 ic->n_uncommitted_sections -= commit_sections;
2389 ic->n_committed_sections += commit_sections;
2390 spin_unlock_irq(&ic->endio_wait.lock);
2392 if (READ_ONCE(ic->free_sectors) <= ic->free_sectors_threshold)
2393 queue_work(ic->writer_wq, &ic->writer_work);
2397 struct bio *next = flushes->bi_next;
2398 flushes->bi_next = NULL;
2399 do_endio(ic, flushes);
2404 static void complete_copy_from_journal(unsigned long error, void *context)
2406 struct journal_io *io = context;
2407 struct journal_completion *comp = io->comp;
2408 struct dm_integrity_c *ic = comp->ic;
2409 remove_range(ic, &io->range);
2410 mempool_free(io, &ic->journal_io_mempool);
2411 if (unlikely(error != 0))
2412 dm_integrity_io_error(ic, "copying from journal", -EIO);
2413 complete_journal_op(comp);
2416 static void restore_last_bytes(struct dm_integrity_c *ic, struct journal_sector *js,
2417 struct journal_entry *je)
2421 js->commit_id = je->last_bytes[s];
2423 } while (++s < ic->sectors_per_block);
2426 static void do_journal_write(struct dm_integrity_c *ic, unsigned write_start,
2427 unsigned write_sections, bool from_replay)
2430 struct journal_completion comp;
2431 struct blk_plug plug;
2433 blk_start_plug(&plug);
2436 comp.in_flight = (atomic_t)ATOMIC_INIT(1);
2437 init_completion(&comp.comp);
2440 for (n = 0; n < write_sections; n++, i++, wraparound_section(ic, &i)) {
2441 #ifndef INTERNAL_VERIFY
2442 if (unlikely(from_replay))
2444 rw_section_mac(ic, i, false);
2445 for (j = 0; j < ic->journal_section_entries; j++) {
2446 struct journal_entry *je = access_journal_entry(ic, i, j);
2447 sector_t sec, area, offset;
2448 unsigned k, l, next_loop;
2449 sector_t metadata_block;
2450 unsigned metadata_offset;
2451 struct journal_io *io;
2453 if (journal_entry_is_unused(je))
2455 BUG_ON(unlikely(journal_entry_is_inprogress(je)) && !from_replay);
2456 sec = journal_entry_get_sector(je);
2457 if (unlikely(from_replay)) {
2458 if (unlikely(sec & (unsigned)(ic->sectors_per_block - 1))) {
2459 dm_integrity_io_error(ic, "invalid sector in journal", -EIO);
2460 sec &= ~(sector_t)(ic->sectors_per_block - 1);
2463 if (unlikely(sec >= ic->provided_data_sectors))
2465 get_area_and_offset(ic, sec, &area, &offset);
2466 restore_last_bytes(ic, access_journal_data(ic, i, j), je);
2467 for (k = j + 1; k < ic->journal_section_entries; k++) {
2468 struct journal_entry *je2 = access_journal_entry(ic, i, k);
2469 sector_t sec2, area2, offset2;
2470 if (journal_entry_is_unused(je2))
2472 BUG_ON(unlikely(journal_entry_is_inprogress(je2)) && !from_replay);
2473 sec2 = journal_entry_get_sector(je2);
2474 if (unlikely(sec2 >= ic->provided_data_sectors))
2476 get_area_and_offset(ic, sec2, &area2, &offset2);
2477 if (area2 != area || offset2 != offset + ((k - j) << ic->sb->log2_sectors_per_block))
2479 restore_last_bytes(ic, access_journal_data(ic, i, k), je2);
2483 io = mempool_alloc(&ic->journal_io_mempool, GFP_NOIO);
2485 io->range.logical_sector = sec;
2486 io->range.n_sectors = (k - j) << ic->sb->log2_sectors_per_block;
2488 spin_lock_irq(&ic->endio_wait.lock);
2489 add_new_range_and_wait(ic, &io->range);
2491 if (likely(!from_replay)) {
2492 struct journal_node *section_node = &ic->journal_tree[i * ic->journal_section_entries];
2494 /* don't write if there is newer committed sector */
2495 while (j < k && find_newer_committed_node(ic, §ion_node[j])) {
2496 struct journal_entry *je2 = access_journal_entry(ic, i, j);
2498 journal_entry_set_unused(je2);
2499 remove_journal_node(ic, §ion_node[j]);
2501 sec += ic->sectors_per_block;
2502 offset += ic->sectors_per_block;
2504 while (j < k && find_newer_committed_node(ic, §ion_node[k - 1])) {
2505 struct journal_entry *je2 = access_journal_entry(ic, i, k - 1);
2507 journal_entry_set_unused(je2);
2508 remove_journal_node(ic, §ion_node[k - 1]);
2512 remove_range_unlocked(ic, &io->range);
2513 spin_unlock_irq(&ic->endio_wait.lock);
2514 mempool_free(io, &ic->journal_io_mempool);
2517 for (l = j; l < k; l++) {
2518 remove_journal_node(ic, §ion_node[l]);
2521 spin_unlock_irq(&ic->endio_wait.lock);
2523 metadata_block = get_metadata_sector_and_offset(ic, area, offset, &metadata_offset);
2524 for (l = j; l < k; l++) {
2526 struct journal_entry *je2 = access_journal_entry(ic, i, l);
2529 #ifndef INTERNAL_VERIFY
2530 unlikely(from_replay) &&
2532 ic->internal_hash) {
2533 char test_tag[max_t(size_t, HASH_MAX_DIGESTSIZE, MAX_TAG_SIZE)];
2535 integrity_sector_checksum(ic, sec + ((l - j) << ic->sb->log2_sectors_per_block),
2536 (char *)access_journal_data(ic, i, l), test_tag);
2537 if (unlikely(memcmp(test_tag, journal_entry_tag(ic, je2), ic->tag_size)))
2538 dm_integrity_io_error(ic, "tag mismatch when replaying journal", -EILSEQ);
2541 journal_entry_set_unused(je2);
2542 r = dm_integrity_rw_tag(ic, journal_entry_tag(ic, je2), &metadata_block, &metadata_offset,
2543 ic->tag_size, TAG_WRITE);
2545 dm_integrity_io_error(ic, "reading tags", r);
2549 atomic_inc(&comp.in_flight);
2550 copy_from_journal(ic, i, j << ic->sb->log2_sectors_per_block,
2551 (k - j) << ic->sb->log2_sectors_per_block,
2552 get_data_sector(ic, area, offset),
2553 complete_copy_from_journal, io);
2559 dm_bufio_write_dirty_buffers_async(ic->bufio);
2561 blk_finish_plug(&plug);
2563 complete_journal_op(&comp);
2564 wait_for_completion_io(&comp.comp);
2566 dm_integrity_flush_buffers(ic, true);
2569 static void integrity_writer(struct work_struct *w)
2571 struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, writer_work);
2572 unsigned write_start, write_sections;
2574 unsigned prev_free_sectors;
2576 /* the following test is not needed, but it tests the replay code */
2577 if (unlikely(dm_post_suspending(ic->ti)) && !ic->meta_dev)
2580 spin_lock_irq(&ic->endio_wait.lock);
2581 write_start = ic->committed_section;
2582 write_sections = ic->n_committed_sections;
2583 spin_unlock_irq(&ic->endio_wait.lock);
2585 if (!write_sections)
2588 do_journal_write(ic, write_start, write_sections, false);
2590 spin_lock_irq(&ic->endio_wait.lock);
2592 ic->committed_section += write_sections;
2593 wraparound_section(ic, &ic->committed_section);
2594 ic->n_committed_sections -= write_sections;
2596 prev_free_sectors = ic->free_sectors;
2597 ic->free_sectors += write_sections * ic->journal_section_entries;
2598 if (unlikely(!prev_free_sectors))
2599 wake_up_locked(&ic->endio_wait);
2601 spin_unlock_irq(&ic->endio_wait.lock);
2604 static void recalc_write_super(struct dm_integrity_c *ic)
2608 dm_integrity_flush_buffers(ic, false);
2609 if (dm_integrity_failed(ic))
2612 r = sync_rw_sb(ic, REQ_OP_WRITE, 0);
2614 dm_integrity_io_error(ic, "writing superblock", r);
2617 static void integrity_recalc(struct work_struct *w)
2619 struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, recalc_work);
2620 struct dm_integrity_range range;
2621 struct dm_io_request io_req;
2622 struct dm_io_region io_loc;
2623 sector_t area, offset;
2624 sector_t metadata_block;
2625 unsigned metadata_offset;
2626 sector_t logical_sector, n_sectors;
2630 unsigned super_counter = 0;
2632 DEBUG_print("start recalculation... (position %llx)\n", le64_to_cpu(ic->sb->recalc_sector));
2634 spin_lock_irq(&ic->endio_wait.lock);
2638 if (unlikely(dm_post_suspending(ic->ti)))
2641 range.logical_sector = le64_to_cpu(ic->sb->recalc_sector);
2642 if (unlikely(range.logical_sector >= ic->provided_data_sectors)) {
2643 if (ic->mode == 'B') {
2644 block_bitmap_op(ic, ic->recalc_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_CLEAR);
2645 DEBUG_print("queue_delayed_work: bitmap_flush_work\n");
2646 queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, 0);
2651 get_area_and_offset(ic, range.logical_sector, &area, &offset);
2652 range.n_sectors = min((sector_t)RECALC_SECTORS, ic->provided_data_sectors - range.logical_sector);
2654 range.n_sectors = min(range.n_sectors, ((sector_t)1U << ic->sb->log2_interleave_sectors) - (unsigned)offset);
2656 add_new_range_and_wait(ic, &range);
2657 spin_unlock_irq(&ic->endio_wait.lock);
2658 logical_sector = range.logical_sector;
2659 n_sectors = range.n_sectors;
2661 if (ic->mode == 'B') {
2662 if (block_bitmap_op(ic, ic->recalc_bitmap, logical_sector, n_sectors, BITMAP_OP_TEST_ALL_CLEAR)) {
2663 goto advance_and_next;
2665 while (block_bitmap_op(ic, ic->recalc_bitmap, logical_sector,
2666 ic->sectors_per_block, BITMAP_OP_TEST_ALL_CLEAR)) {
2667 logical_sector += ic->sectors_per_block;
2668 n_sectors -= ic->sectors_per_block;
2671 while (block_bitmap_op(ic, ic->recalc_bitmap, logical_sector + n_sectors - ic->sectors_per_block,
2672 ic->sectors_per_block, BITMAP_OP_TEST_ALL_CLEAR)) {
2673 n_sectors -= ic->sectors_per_block;
2676 get_area_and_offset(ic, logical_sector, &area, &offset);
2679 DEBUG_print("recalculating: %llx, %llx\n", logical_sector, n_sectors);
2681 if (unlikely(++super_counter == RECALC_WRITE_SUPER)) {
2682 recalc_write_super(ic);
2683 if (ic->mode == 'B') {
2684 queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, ic->bitmap_flush_interval);
2689 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 t = ic->recalc_tags + (n_sectors >> ic->sb->log2_sectors_per_block) * ic->tag_size;
2718 metadata_block = get_metadata_sector_and_offset(ic, area, offset, &metadata_offset);
2720 r = dm_integrity_rw_tag(ic, ic->recalc_tags, &metadata_block, &metadata_offset, t - ic->recalc_tags, TAG_WRITE);
2722 dm_integrity_io_error(ic, "writing tags", r);
2726 if (ic->mode == 'B') {
2727 sector_t start, end;
2728 start = (range.logical_sector >>
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 end = ((range.logical_sector + range.n_sectors) >>
2732 (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit)) <<
2733 (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
2734 block_bitmap_op(ic, ic->recalc_bitmap, start, end - start, BITMAP_OP_CLEAR);
2740 spin_lock_irq(&ic->endio_wait.lock);
2741 remove_range_unlocked(ic, &range);
2742 ic->sb->recalc_sector = cpu_to_le64(range.logical_sector + range.n_sectors);
2746 remove_range(ic, &range);
2750 spin_unlock_irq(&ic->endio_wait.lock);
2752 recalc_write_super(ic);
2755 static void bitmap_block_work(struct work_struct *w)
2757 struct bitmap_block_status *bbs = container_of(w, struct bitmap_block_status, work);
2758 struct dm_integrity_c *ic = bbs->ic;
2760 struct bio_list bio_queue;
2761 struct bio_list waiting;
2763 bio_list_init(&waiting);
2765 spin_lock(&bbs->bio_queue_lock);
2766 bio_queue = bbs->bio_queue;
2767 bio_list_init(&bbs->bio_queue);
2768 spin_unlock(&bbs->bio_queue_lock);
2770 while ((bio = bio_list_pop(&bio_queue))) {
2771 struct dm_integrity_io *dio;
2773 dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
2775 if (block_bitmap_op(ic, ic->may_write_bitmap, dio->range.logical_sector,
2776 dio->range.n_sectors, BITMAP_OP_TEST_ALL_SET)) {
2777 remove_range(ic, &dio->range);
2778 INIT_WORK(&dio->work, integrity_bio_wait);
2779 queue_work(ic->offload_wq, &dio->work);
2781 block_bitmap_op(ic, ic->journal, dio->range.logical_sector,
2782 dio->range.n_sectors, BITMAP_OP_SET);
2783 bio_list_add(&waiting, bio);
2787 if (bio_list_empty(&waiting))
2790 rw_journal_sectors(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC,
2791 bbs->idx * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT),
2792 BITMAP_BLOCK_SIZE >> SECTOR_SHIFT, NULL);
2794 while ((bio = bio_list_pop(&waiting))) {
2795 struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
2797 block_bitmap_op(ic, ic->may_write_bitmap, dio->range.logical_sector,
2798 dio->range.n_sectors, BITMAP_OP_SET);
2800 remove_range(ic, &dio->range);
2801 INIT_WORK(&dio->work, integrity_bio_wait);
2802 queue_work(ic->offload_wq, &dio->work);
2805 queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, ic->bitmap_flush_interval);
2808 static void bitmap_flush_work(struct work_struct *work)
2810 struct dm_integrity_c *ic = container_of(work, struct dm_integrity_c, bitmap_flush_work.work);
2811 struct dm_integrity_range range;
2812 unsigned long limit;
2815 dm_integrity_flush_buffers(ic, false);
2817 range.logical_sector = 0;
2818 range.n_sectors = ic->provided_data_sectors;
2820 spin_lock_irq(&ic->endio_wait.lock);
2821 add_new_range_and_wait(ic, &range);
2822 spin_unlock_irq(&ic->endio_wait.lock);
2824 dm_integrity_flush_buffers(ic, true);
2826 limit = ic->provided_data_sectors;
2827 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
2828 limit = le64_to_cpu(ic->sb->recalc_sector)
2829 >> (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit)
2830 << (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
2832 /*DEBUG_print("zeroing journal\n");*/
2833 block_bitmap_op(ic, ic->journal, 0, limit, BITMAP_OP_CLEAR);
2834 block_bitmap_op(ic, ic->may_write_bitmap, 0, limit, BITMAP_OP_CLEAR);
2836 rw_journal_sectors(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, 0,
2837 ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
2839 spin_lock_irq(&ic->endio_wait.lock);
2840 remove_range_unlocked(ic, &range);
2841 while (unlikely((bio = bio_list_pop(&ic->synchronous_bios)) != NULL)) {
2843 spin_unlock_irq(&ic->endio_wait.lock);
2844 spin_lock_irq(&ic->endio_wait.lock);
2846 spin_unlock_irq(&ic->endio_wait.lock);
2850 static void init_journal(struct dm_integrity_c *ic, unsigned start_section,
2851 unsigned n_sections, unsigned char commit_seq)
2858 for (n = 0; n < n_sections; n++) {
2859 i = start_section + n;
2860 wraparound_section(ic, &i);
2861 for (j = 0; j < ic->journal_section_sectors; j++) {
2862 struct journal_sector *js = access_journal(ic, i, j);
2863 memset(&js->entries, 0, JOURNAL_SECTOR_DATA);
2864 js->commit_id = dm_integrity_commit_id(ic, i, j, commit_seq);
2866 for (j = 0; j < ic->journal_section_entries; j++) {
2867 struct journal_entry *je = access_journal_entry(ic, i, j);
2868 journal_entry_set_unused(je);
2872 write_journal(ic, start_section, n_sections);
2875 static int find_commit_seq(struct dm_integrity_c *ic, unsigned i, unsigned j, commit_id_t id)
2878 for (k = 0; k < N_COMMIT_IDS; k++) {
2879 if (dm_integrity_commit_id(ic, i, j, k) == id)
2882 dm_integrity_io_error(ic, "journal commit id", -EIO);
2886 static void replay_journal(struct dm_integrity_c *ic)
2889 bool used_commit_ids[N_COMMIT_IDS];
2890 unsigned max_commit_id_sections[N_COMMIT_IDS];
2891 unsigned write_start, write_sections;
2892 unsigned continue_section;
2894 unsigned char unused, last_used, want_commit_seq;
2896 if (ic->mode == 'R')
2899 if (ic->journal_uptodate)
2905 if (!ic->just_formatted) {
2906 DEBUG_print("reading journal\n");
2907 rw_journal(ic, REQ_OP_READ, 0, 0, ic->journal_sections, NULL);
2909 DEBUG_bytes(lowmem_page_address(ic->journal_io[0].page), 64, "read journal");
2910 if (ic->journal_io) {
2911 struct journal_completion crypt_comp;
2913 init_completion(&crypt_comp.comp);
2914 crypt_comp.in_flight = (atomic_t)ATOMIC_INIT(0);
2915 encrypt_journal(ic, false, 0, ic->journal_sections, &crypt_comp);
2916 wait_for_completion(&crypt_comp.comp);
2918 DEBUG_bytes(lowmem_page_address(ic->journal[0].page), 64, "decrypted journal");
2921 if (dm_integrity_failed(ic))
2924 journal_empty = true;
2925 memset(used_commit_ids, 0, sizeof used_commit_ids);
2926 memset(max_commit_id_sections, 0, sizeof max_commit_id_sections);
2927 for (i = 0; i < ic->journal_sections; i++) {
2928 for (j = 0; j < ic->journal_section_sectors; j++) {
2930 struct journal_sector *js = access_journal(ic, i, j);
2931 k = find_commit_seq(ic, i, j, js->commit_id);
2934 used_commit_ids[k] = true;
2935 max_commit_id_sections[k] = i;
2937 if (journal_empty) {
2938 for (j = 0; j < ic->journal_section_entries; j++) {
2939 struct journal_entry *je = access_journal_entry(ic, i, j);
2940 if (!journal_entry_is_unused(je)) {
2941 journal_empty = false;
2948 if (!used_commit_ids[N_COMMIT_IDS - 1]) {
2949 unused = N_COMMIT_IDS - 1;
2950 while (unused && !used_commit_ids[unused - 1])
2953 for (unused = 0; unused < N_COMMIT_IDS; unused++)
2954 if (!used_commit_ids[unused])
2956 if (unused == N_COMMIT_IDS) {
2957 dm_integrity_io_error(ic, "journal commit ids", -EIO);
2961 DEBUG_print("first unused commit seq %d [%d,%d,%d,%d]\n",
2962 unused, used_commit_ids[0], used_commit_ids[1],
2963 used_commit_ids[2], used_commit_ids[3]);
2965 last_used = prev_commit_seq(unused);
2966 want_commit_seq = prev_commit_seq(last_used);
2968 if (!used_commit_ids[want_commit_seq] && used_commit_ids[prev_commit_seq(want_commit_seq)])
2969 journal_empty = true;
2971 write_start = max_commit_id_sections[last_used] + 1;
2972 if (unlikely(write_start >= ic->journal_sections))
2973 want_commit_seq = next_commit_seq(want_commit_seq);
2974 wraparound_section(ic, &write_start);
2977 for (write_sections = 0; write_sections < ic->journal_sections; write_sections++) {
2978 for (j = 0; j < ic->journal_section_sectors; j++) {
2979 struct journal_sector *js = access_journal(ic, i, j);
2981 if (js->commit_id != dm_integrity_commit_id(ic, i, j, want_commit_seq)) {
2983 * This could be caused by crash during writing.
2984 * We won't replay the inconsistent part of the
2987 DEBUG_print("commit id mismatch at position (%u, %u): %d != %d\n",
2988 i, j, find_commit_seq(ic, i, j, js->commit_id), want_commit_seq);
2993 if (unlikely(i >= ic->journal_sections))
2994 want_commit_seq = next_commit_seq(want_commit_seq);
2995 wraparound_section(ic, &i);
2999 if (!journal_empty) {
3000 DEBUG_print("replaying %u sections, starting at %u, commit seq %d\n",
3001 write_sections, write_start, want_commit_seq);
3002 do_journal_write(ic, write_start, write_sections, true);
3005 if (write_sections == ic->journal_sections && (ic->mode == 'J' || journal_empty)) {
3006 continue_section = write_start;
3007 ic->commit_seq = want_commit_seq;
3008 DEBUG_print("continuing from section %u, commit seq %d\n", write_start, ic->commit_seq);
3011 unsigned char erase_seq;
3013 DEBUG_print("clearing journal\n");
3015 erase_seq = prev_commit_seq(prev_commit_seq(last_used));
3017 init_journal(ic, s, 1, erase_seq);
3019 wraparound_section(ic, &s);
3020 if (ic->journal_sections >= 2) {
3021 init_journal(ic, s, ic->journal_sections - 2, erase_seq);
3022 s += ic->journal_sections - 2;
3023 wraparound_section(ic, &s);
3024 init_journal(ic, s, 1, erase_seq);
3027 continue_section = 0;
3028 ic->commit_seq = next_commit_seq(erase_seq);
3031 ic->committed_section = continue_section;
3032 ic->n_committed_sections = 0;
3034 ic->uncommitted_section = continue_section;
3035 ic->n_uncommitted_sections = 0;
3037 ic->free_section = continue_section;
3038 ic->free_section_entry = 0;
3039 ic->free_sectors = ic->journal_entries;
3041 ic->journal_tree_root = RB_ROOT;
3042 for (i = 0; i < ic->journal_entries; i++)
3043 init_journal_node(&ic->journal_tree[i]);
3046 static void dm_integrity_enter_synchronous_mode(struct dm_integrity_c *ic)
3048 DEBUG_print("dm_integrity_enter_synchronous_mode\n");
3050 if (ic->mode == 'B') {
3051 ic->bitmap_flush_interval = msecs_to_jiffies(10) + 1;
3052 ic->synchronous_mode = 1;
3054 cancel_delayed_work_sync(&ic->bitmap_flush_work);
3055 queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, 0);
3056 flush_workqueue(ic->commit_wq);
3060 static int dm_integrity_reboot(struct notifier_block *n, unsigned long code, void *x)
3062 struct dm_integrity_c *ic = container_of(n, struct dm_integrity_c, reboot_notifier);
3064 DEBUG_print("dm_integrity_reboot\n");
3066 dm_integrity_enter_synchronous_mode(ic);
3071 static void dm_integrity_postsuspend(struct dm_target *ti)
3073 struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
3076 WARN_ON(unregister_reboot_notifier(&ic->reboot_notifier));
3078 del_timer_sync(&ic->autocommit_timer);
3081 drain_workqueue(ic->recalc_wq);
3083 if (ic->mode == 'B')
3084 cancel_delayed_work_sync(&ic->bitmap_flush_work);
3086 queue_work(ic->commit_wq, &ic->commit_work);
3087 drain_workqueue(ic->commit_wq);
3089 if (ic->mode == 'J') {
3091 queue_work(ic->writer_wq, &ic->writer_work);
3092 drain_workqueue(ic->writer_wq);
3093 dm_integrity_flush_buffers(ic, true);
3096 if (ic->mode == 'B') {
3097 dm_integrity_flush_buffers(ic, true);
3099 /* set to 0 to test bitmap replay code */
3100 init_journal(ic, 0, ic->journal_sections, 0);
3101 ic->sb->flags &= ~cpu_to_le32(SB_FLAG_DIRTY_BITMAP);
3102 r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
3104 dm_integrity_io_error(ic, "writing superblock", r);
3108 BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
3110 ic->journal_uptodate = true;
3113 static void dm_integrity_resume(struct dm_target *ti)
3115 struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
3116 __u64 old_provided_data_sectors = le64_to_cpu(ic->sb->provided_data_sectors);
3119 DEBUG_print("resume\n");
3121 if (ic->provided_data_sectors != old_provided_data_sectors) {
3122 if (ic->provided_data_sectors > old_provided_data_sectors &&
3124 ic->sb->log2_blocks_per_bitmap_bit == ic->log2_blocks_per_bitmap_bit) {
3125 rw_journal_sectors(ic, REQ_OP_READ, 0, 0,
3126 ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3127 block_bitmap_op(ic, ic->journal, old_provided_data_sectors,
3128 ic->provided_data_sectors - old_provided_data_sectors, BITMAP_OP_SET);
3129 rw_journal_sectors(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, 0,
3130 ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3133 ic->sb->provided_data_sectors = cpu_to_le64(ic->provided_data_sectors);
3134 r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
3136 dm_integrity_io_error(ic, "writing superblock", r);
3139 if (ic->sb->flags & cpu_to_le32(SB_FLAG_DIRTY_BITMAP)) {
3140 DEBUG_print("resume dirty_bitmap\n");
3141 rw_journal_sectors(ic, REQ_OP_READ, 0, 0,
3142 ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3143 if (ic->mode == 'B') {
3144 if (ic->sb->log2_blocks_per_bitmap_bit == ic->log2_blocks_per_bitmap_bit &&
3145 !ic->reset_recalculate_flag) {
3146 block_bitmap_copy(ic, ic->recalc_bitmap, ic->journal);
3147 block_bitmap_copy(ic, ic->may_write_bitmap, ic->journal);
3148 if (!block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors,
3149 BITMAP_OP_TEST_ALL_CLEAR)) {
3150 ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
3151 ic->sb->recalc_sector = cpu_to_le64(0);
3154 DEBUG_print("non-matching blocks_per_bitmap_bit: %u, %u\n",
3155 ic->sb->log2_blocks_per_bitmap_bit, ic->log2_blocks_per_bitmap_bit);
3156 ic->sb->log2_blocks_per_bitmap_bit = ic->log2_blocks_per_bitmap_bit;
3157 block_bitmap_op(ic, ic->recalc_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_SET);
3158 block_bitmap_op(ic, ic->may_write_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_SET);
3159 block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors, BITMAP_OP_SET);
3160 rw_journal_sectors(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, 0,
3161 ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3162 ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
3163 ic->sb->recalc_sector = cpu_to_le64(0);
3166 if (!(ic->sb->log2_blocks_per_bitmap_bit == ic->log2_blocks_per_bitmap_bit &&
3167 block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors, BITMAP_OP_TEST_ALL_CLEAR)) ||
3168 ic->reset_recalculate_flag) {
3169 ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
3170 ic->sb->recalc_sector = cpu_to_le64(0);
3172 init_journal(ic, 0, ic->journal_sections, 0);
3174 ic->sb->flags &= ~cpu_to_le32(SB_FLAG_DIRTY_BITMAP);
3176 r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
3178 dm_integrity_io_error(ic, "writing superblock", r);
3181 if (ic->reset_recalculate_flag) {
3182 ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
3183 ic->sb->recalc_sector = cpu_to_le64(0);
3185 if (ic->mode == 'B') {
3186 ic->sb->flags |= cpu_to_le32(SB_FLAG_DIRTY_BITMAP);
3187 ic->sb->log2_blocks_per_bitmap_bit = ic->log2_blocks_per_bitmap_bit;
3188 r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
3190 dm_integrity_io_error(ic, "writing superblock", r);
3192 block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors, BITMAP_OP_CLEAR);
3193 block_bitmap_op(ic, ic->recalc_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_CLEAR);
3194 block_bitmap_op(ic, ic->may_write_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_CLEAR);
3195 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING) &&
3196 le64_to_cpu(ic->sb->recalc_sector) < ic->provided_data_sectors) {
3197 block_bitmap_op(ic, ic->journal, le64_to_cpu(ic->sb->recalc_sector),
3198 ic->provided_data_sectors - le64_to_cpu(ic->sb->recalc_sector), BITMAP_OP_SET);
3199 block_bitmap_op(ic, ic->recalc_bitmap, le64_to_cpu(ic->sb->recalc_sector),
3200 ic->provided_data_sectors - le64_to_cpu(ic->sb->recalc_sector), BITMAP_OP_SET);
3201 block_bitmap_op(ic, ic->may_write_bitmap, le64_to_cpu(ic->sb->recalc_sector),
3202 ic->provided_data_sectors - le64_to_cpu(ic->sb->recalc_sector), BITMAP_OP_SET);
3204 rw_journal_sectors(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, 0,
3205 ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3209 DEBUG_print("testing recalc: %x\n", ic->sb->flags);
3210 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
3211 __u64 recalc_pos = le64_to_cpu(ic->sb->recalc_sector);
3212 DEBUG_print("recalc pos: %llx / %llx\n", recalc_pos, ic->provided_data_sectors);
3213 if (recalc_pos < ic->provided_data_sectors) {
3214 queue_work(ic->recalc_wq, &ic->recalc_work);
3215 } else if (recalc_pos > ic->provided_data_sectors) {
3216 ic->sb->recalc_sector = cpu_to_le64(ic->provided_data_sectors);
3217 recalc_write_super(ic);
3221 ic->reboot_notifier.notifier_call = dm_integrity_reboot;
3222 ic->reboot_notifier.next = NULL;
3223 ic->reboot_notifier.priority = INT_MAX - 1; /* be notified after md and before hardware drivers */
3224 WARN_ON(register_reboot_notifier(&ic->reboot_notifier));
3227 /* set to 1 to stress test synchronous mode */
3228 dm_integrity_enter_synchronous_mode(ic);
3232 static void dm_integrity_status(struct dm_target *ti, status_type_t type,
3233 unsigned status_flags, char *result, unsigned maxlen)
3235 struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
3240 case STATUSTYPE_INFO:
3242 (unsigned long long)atomic64_read(&ic->number_of_mismatches),
3243 ic->provided_data_sectors);
3244 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
3245 DMEMIT(" %llu", le64_to_cpu(ic->sb->recalc_sector));
3250 case STATUSTYPE_TABLE: {
3251 __u64 watermark_percentage = (__u64)(ic->journal_entries - ic->free_sectors_threshold) * 100;
3252 watermark_percentage += ic->journal_entries / 2;
3253 do_div(watermark_percentage, ic->journal_entries);
3255 arg_count += !!ic->meta_dev;
3256 arg_count += ic->sectors_per_block != 1;
3257 arg_count += !!(ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING));
3258 arg_count += ic->reset_recalculate_flag;
3259 arg_count += ic->discard;
3260 arg_count += ic->mode == 'J';
3261 arg_count += ic->mode == 'J';
3262 arg_count += ic->mode == 'B';
3263 arg_count += ic->mode == 'B';
3264 arg_count += !!ic->internal_hash_alg.alg_string;
3265 arg_count += !!ic->journal_crypt_alg.alg_string;
3266 arg_count += !!ic->journal_mac_alg.alg_string;
3267 arg_count += (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING)) != 0;
3268 arg_count += (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) != 0;
3269 arg_count += ic->legacy_recalculate;
3270 DMEMIT("%s %llu %u %c %u", ic->dev->name, ic->start,
3271 ic->tag_size, ic->mode, arg_count);
3273 DMEMIT(" meta_device:%s", ic->meta_dev->name);
3274 if (ic->sectors_per_block != 1)
3275 DMEMIT(" block_size:%u", ic->sectors_per_block << SECTOR_SHIFT);
3276 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
3277 DMEMIT(" recalculate");
3278 if (ic->reset_recalculate_flag)
3279 DMEMIT(" reset_recalculate");
3281 DMEMIT(" allow_discards");
3282 DMEMIT(" journal_sectors:%u", ic->initial_sectors - SB_SECTORS);
3283 DMEMIT(" interleave_sectors:%u", 1U << ic->sb->log2_interleave_sectors);
3284 DMEMIT(" buffer_sectors:%u", 1U << ic->log2_buffer_sectors);
3285 if (ic->mode == 'J') {
3286 DMEMIT(" journal_watermark:%u", (unsigned)watermark_percentage);
3287 DMEMIT(" commit_time:%u", ic->autocommit_msec);
3289 if (ic->mode == 'B') {
3290 DMEMIT(" sectors_per_bit:%llu", (sector_t)ic->sectors_per_block << ic->log2_blocks_per_bitmap_bit);
3291 DMEMIT(" bitmap_flush_interval:%u", jiffies_to_msecs(ic->bitmap_flush_interval));
3293 if ((ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING)) != 0)
3294 DMEMIT(" fix_padding");
3295 if ((ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) != 0)
3296 DMEMIT(" fix_hmac");
3297 if (ic->legacy_recalculate)
3298 DMEMIT(" legacy_recalculate");
3300 #define EMIT_ALG(a, n) \
3302 if (ic->a.alg_string) { \
3303 DMEMIT(" %s:%s", n, ic->a.alg_string); \
3304 if (ic->a.key_string) \
3305 DMEMIT(":%s", ic->a.key_string);\
3308 EMIT_ALG(internal_hash_alg, "internal_hash");
3309 EMIT_ALG(journal_crypt_alg, "journal_crypt");
3310 EMIT_ALG(journal_mac_alg, "journal_mac");
3316 static int dm_integrity_iterate_devices(struct dm_target *ti,
3317 iterate_devices_callout_fn fn, void *data)
3319 struct dm_integrity_c *ic = ti->private;
3322 return fn(ti, ic->dev, ic->start + ic->initial_sectors + ic->metadata_run, ti->len, data);
3324 return fn(ti, ic->dev, 0, ti->len, data);
3327 static void dm_integrity_io_hints(struct dm_target *ti, struct queue_limits *limits)
3329 struct dm_integrity_c *ic = ti->private;
3331 if (ic->sectors_per_block > 1) {
3332 limits->logical_block_size = ic->sectors_per_block << SECTOR_SHIFT;
3333 limits->physical_block_size = ic->sectors_per_block << SECTOR_SHIFT;
3334 blk_limits_io_min(limits, ic->sectors_per_block << SECTOR_SHIFT);
3338 static void calculate_journal_section_size(struct dm_integrity_c *ic)
3340 unsigned sector_space = JOURNAL_SECTOR_DATA;
3342 ic->journal_sections = le32_to_cpu(ic->sb->journal_sections);
3343 ic->journal_entry_size = roundup(offsetof(struct journal_entry, last_bytes[ic->sectors_per_block]) + ic->tag_size,
3344 JOURNAL_ENTRY_ROUNDUP);
3346 if (ic->sb->flags & cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC))
3347 sector_space -= JOURNAL_MAC_PER_SECTOR;
3348 ic->journal_entries_per_sector = sector_space / ic->journal_entry_size;
3349 ic->journal_section_entries = ic->journal_entries_per_sector * JOURNAL_BLOCK_SECTORS;
3350 ic->journal_section_sectors = (ic->journal_section_entries << ic->sb->log2_sectors_per_block) + JOURNAL_BLOCK_SECTORS;
3351 ic->journal_entries = ic->journal_section_entries * ic->journal_sections;
3354 static int calculate_device_limits(struct dm_integrity_c *ic)
3356 __u64 initial_sectors;
3358 calculate_journal_section_size(ic);
3359 initial_sectors = SB_SECTORS + (__u64)ic->journal_section_sectors * ic->journal_sections;
3360 if (initial_sectors + METADATA_PADDING_SECTORS >= ic->meta_device_sectors || initial_sectors > UINT_MAX)
3362 ic->initial_sectors = initial_sectors;
3364 if (!ic->meta_dev) {
3365 sector_t last_sector, last_area, last_offset;
3367 /* we have to maintain excessive padding for compatibility with existing volumes */
3368 __u64 metadata_run_padding =
3369 ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING) ?
3370 (__u64)(METADATA_PADDING_SECTORS << SECTOR_SHIFT) :
3371 (__u64)(1 << SECTOR_SHIFT << METADATA_PADDING_SECTORS);
3373 ic->metadata_run = round_up((__u64)ic->tag_size << (ic->sb->log2_interleave_sectors - ic->sb->log2_sectors_per_block),
3374 metadata_run_padding) >> SECTOR_SHIFT;
3375 if (!(ic->metadata_run & (ic->metadata_run - 1)))
3376 ic->log2_metadata_run = __ffs(ic->metadata_run);
3378 ic->log2_metadata_run = -1;
3380 get_area_and_offset(ic, ic->provided_data_sectors - 1, &last_area, &last_offset);
3381 last_sector = get_data_sector(ic, last_area, last_offset);
3382 if (last_sector < ic->start || last_sector >= ic->meta_device_sectors)
3385 __u64 meta_size = (ic->provided_data_sectors >> ic->sb->log2_sectors_per_block) * ic->tag_size;
3386 meta_size = (meta_size + ((1U << (ic->log2_buffer_sectors + SECTOR_SHIFT)) - 1))
3387 >> (ic->log2_buffer_sectors + SECTOR_SHIFT);
3388 meta_size <<= ic->log2_buffer_sectors;
3389 if (ic->initial_sectors + meta_size < ic->initial_sectors ||
3390 ic->initial_sectors + meta_size > ic->meta_device_sectors)
3392 ic->metadata_run = 1;
3393 ic->log2_metadata_run = 0;
3399 static void get_provided_data_sectors(struct dm_integrity_c *ic)
3401 if (!ic->meta_dev) {
3403 ic->provided_data_sectors = 0;
3404 for (test_bit = fls64(ic->meta_device_sectors) - 1; test_bit >= 3; test_bit--) {
3405 __u64 prev_data_sectors = ic->provided_data_sectors;
3407 ic->provided_data_sectors |= (sector_t)1 << test_bit;
3408 if (calculate_device_limits(ic))
3409 ic->provided_data_sectors = prev_data_sectors;
3412 ic->provided_data_sectors = ic->data_device_sectors;
3413 ic->provided_data_sectors &= ~(sector_t)(ic->sectors_per_block - 1);
3417 static int initialize_superblock(struct dm_integrity_c *ic, unsigned journal_sectors, unsigned interleave_sectors)
3419 unsigned journal_sections;
3422 memset(ic->sb, 0, SB_SECTORS << SECTOR_SHIFT);
3423 memcpy(ic->sb->magic, SB_MAGIC, 8);
3424 ic->sb->integrity_tag_size = cpu_to_le16(ic->tag_size);
3425 ic->sb->log2_sectors_per_block = __ffs(ic->sectors_per_block);
3426 if (ic->journal_mac_alg.alg_string)
3427 ic->sb->flags |= cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC);
3429 calculate_journal_section_size(ic);
3430 journal_sections = journal_sectors / ic->journal_section_sectors;
3431 if (!journal_sections)
3432 journal_sections = 1;
3434 if (ic->fix_hmac && (ic->internal_hash_alg.alg_string || ic->journal_mac_alg.alg_string)) {
3435 ic->sb->flags |= cpu_to_le32(SB_FLAG_FIXED_HMAC);
3436 get_random_bytes(ic->sb->salt, SALT_SIZE);
3439 if (!ic->meta_dev) {
3440 if (ic->fix_padding)
3441 ic->sb->flags |= cpu_to_le32(SB_FLAG_FIXED_PADDING);
3442 ic->sb->journal_sections = cpu_to_le32(journal_sections);
3443 if (!interleave_sectors)
3444 interleave_sectors = DEFAULT_INTERLEAVE_SECTORS;
3445 ic->sb->log2_interleave_sectors = __fls(interleave_sectors);
3446 ic->sb->log2_interleave_sectors = max((__u8)MIN_LOG2_INTERLEAVE_SECTORS, ic->sb->log2_interleave_sectors);
3447 ic->sb->log2_interleave_sectors = min((__u8)MAX_LOG2_INTERLEAVE_SECTORS, ic->sb->log2_interleave_sectors);
3449 get_provided_data_sectors(ic);
3450 if (!ic->provided_data_sectors)
3453 ic->sb->log2_interleave_sectors = 0;
3455 get_provided_data_sectors(ic);
3456 if (!ic->provided_data_sectors)
3460 ic->sb->journal_sections = cpu_to_le32(0);
3461 for (test_bit = fls(journal_sections) - 1; test_bit >= 0; test_bit--) {
3462 __u32 prev_journal_sections = le32_to_cpu(ic->sb->journal_sections);
3463 __u32 test_journal_sections = prev_journal_sections | (1U << test_bit);
3464 if (test_journal_sections > journal_sections)
3466 ic->sb->journal_sections = cpu_to_le32(test_journal_sections);
3467 if (calculate_device_limits(ic))
3468 ic->sb->journal_sections = cpu_to_le32(prev_journal_sections);
3471 if (!le32_to_cpu(ic->sb->journal_sections)) {
3472 if (ic->log2_buffer_sectors > 3) {
3473 ic->log2_buffer_sectors--;
3474 goto try_smaller_buffer;
3480 ic->sb->provided_data_sectors = cpu_to_le64(ic->provided_data_sectors);
3487 static void dm_integrity_set(struct dm_target *ti, struct dm_integrity_c *ic)
3489 struct gendisk *disk = dm_disk(dm_table_get_md(ti->table));
3490 struct blk_integrity bi;
3492 memset(&bi, 0, sizeof(bi));
3493 bi.profile = &dm_integrity_profile;
3494 bi.tuple_size = ic->tag_size;
3495 bi.tag_size = bi.tuple_size;
3496 bi.interval_exp = ic->sb->log2_sectors_per_block + SECTOR_SHIFT;
3498 blk_integrity_register(disk, &bi);
3499 blk_queue_max_integrity_segments(disk->queue, UINT_MAX);
3502 static void dm_integrity_free_page_list(struct page_list *pl)
3508 for (i = 0; pl[i].page; i++)
3509 __free_page(pl[i].page);
3513 static struct page_list *dm_integrity_alloc_page_list(unsigned n_pages)
3515 struct page_list *pl;
3518 pl = kvmalloc_array(n_pages + 1, sizeof(struct page_list), GFP_KERNEL | __GFP_ZERO);
3522 for (i = 0; i < n_pages; i++) {
3523 pl[i].page = alloc_page(GFP_KERNEL);
3525 dm_integrity_free_page_list(pl);
3529 pl[i - 1].next = &pl[i];
3537 static void dm_integrity_free_journal_scatterlist(struct dm_integrity_c *ic, struct scatterlist **sl)
3540 for (i = 0; i < ic->journal_sections; i++)
3545 static struct scatterlist **dm_integrity_alloc_journal_scatterlist(struct dm_integrity_c *ic,
3546 struct page_list *pl)
3548 struct scatterlist **sl;
3551 sl = kvmalloc_array(ic->journal_sections,
3552 sizeof(struct scatterlist *),
3553 GFP_KERNEL | __GFP_ZERO);
3557 for (i = 0; i < ic->journal_sections; i++) {
3558 struct scatterlist *s;
3559 unsigned start_index, start_offset;
3560 unsigned end_index, end_offset;
3564 page_list_location(ic, i, 0, &start_index, &start_offset);
3565 page_list_location(ic, i, ic->journal_section_sectors - 1,
3566 &end_index, &end_offset);
3568 n_pages = (end_index - start_index + 1);
3570 s = kvmalloc_array(n_pages, sizeof(struct scatterlist),
3573 dm_integrity_free_journal_scatterlist(ic, sl);
3577 sg_init_table(s, n_pages);
3578 for (idx = start_index; idx <= end_index; idx++) {
3579 char *va = lowmem_page_address(pl[idx].page);
3580 unsigned start = 0, end = PAGE_SIZE;
3581 if (idx == start_index)
3582 start = start_offset;
3583 if (idx == end_index)
3584 end = end_offset + (1 << SECTOR_SHIFT);
3585 sg_set_buf(&s[idx - start_index], va + start, end - start);
3594 static void free_alg(struct alg_spec *a)
3596 kfree_sensitive(a->alg_string);
3597 kfree_sensitive(a->key);
3598 memset(a, 0, sizeof *a);
3601 static int get_alg_and_key(const char *arg, struct alg_spec *a, char **error, char *error_inval)
3607 a->alg_string = kstrdup(strchr(arg, ':') + 1, GFP_KERNEL);
3611 k = strchr(a->alg_string, ':');
3614 a->key_string = k + 1;
3615 if (strlen(a->key_string) & 1)
3618 a->key_size = strlen(a->key_string) / 2;
3619 a->key = kmalloc(a->key_size, GFP_KERNEL);
3622 if (hex2bin(a->key, a->key_string, a->key_size))
3628 *error = error_inval;
3631 *error = "Out of memory for an argument";
3635 static int get_mac(struct crypto_shash **hash, struct alg_spec *a, char **error,
3636 char *error_alg, char *error_key)
3640 if (a->alg_string) {
3641 *hash = crypto_alloc_shash(a->alg_string, 0, CRYPTO_ALG_ALLOCATES_MEMORY);
3642 if (IS_ERR(*hash)) {
3650 r = crypto_shash_setkey(*hash, a->key, a->key_size);
3655 } else if (crypto_shash_get_flags(*hash) & CRYPTO_TFM_NEED_KEY) {
3664 static int create_journal(struct dm_integrity_c *ic, char **error)
3668 __u64 journal_pages, journal_desc_size, journal_tree_size;
3669 unsigned char *crypt_data = NULL, *crypt_iv = NULL;
3670 struct skcipher_request *req = NULL;
3672 ic->commit_ids[0] = cpu_to_le64(0x1111111111111111ULL);
3673 ic->commit_ids[1] = cpu_to_le64(0x2222222222222222ULL);
3674 ic->commit_ids[2] = cpu_to_le64(0x3333333333333333ULL);
3675 ic->commit_ids[3] = cpu_to_le64(0x4444444444444444ULL);
3677 journal_pages = roundup((__u64)ic->journal_sections * ic->journal_section_sectors,
3678 PAGE_SIZE >> SECTOR_SHIFT) >> (PAGE_SHIFT - SECTOR_SHIFT);
3679 journal_desc_size = journal_pages * sizeof(struct page_list);
3680 if (journal_pages >= totalram_pages() - totalhigh_pages() || journal_desc_size > ULONG_MAX) {
3681 *error = "Journal doesn't fit into memory";
3685 ic->journal_pages = journal_pages;
3687 ic->journal = dm_integrity_alloc_page_list(ic->journal_pages);
3689 *error = "Could not allocate memory for journal";
3693 if (ic->journal_crypt_alg.alg_string) {
3694 unsigned ivsize, blocksize;
3695 struct journal_completion comp;
3698 ic->journal_crypt = crypto_alloc_skcipher(ic->journal_crypt_alg.alg_string, 0, CRYPTO_ALG_ALLOCATES_MEMORY);
3699 if (IS_ERR(ic->journal_crypt)) {
3700 *error = "Invalid journal cipher";
3701 r = PTR_ERR(ic->journal_crypt);
3702 ic->journal_crypt = NULL;
3705 ivsize = crypto_skcipher_ivsize(ic->journal_crypt);
3706 blocksize = crypto_skcipher_blocksize(ic->journal_crypt);
3708 if (ic->journal_crypt_alg.key) {
3709 r = crypto_skcipher_setkey(ic->journal_crypt, ic->journal_crypt_alg.key,
3710 ic->journal_crypt_alg.key_size);
3712 *error = "Error setting encryption key";
3716 DEBUG_print("cipher %s, block size %u iv size %u\n",
3717 ic->journal_crypt_alg.alg_string, blocksize, ivsize);
3719 ic->journal_io = dm_integrity_alloc_page_list(ic->journal_pages);
3720 if (!ic->journal_io) {
3721 *error = "Could not allocate memory for journal io";
3726 if (blocksize == 1) {
3727 struct scatterlist *sg;
3729 req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
3731 *error = "Could not allocate crypt request";
3736 crypt_iv = kzalloc(ivsize, GFP_KERNEL);
3738 *error = "Could not allocate iv";
3743 ic->journal_xor = dm_integrity_alloc_page_list(ic->journal_pages);
3744 if (!ic->journal_xor) {
3745 *error = "Could not allocate memory for journal xor";
3750 sg = kvmalloc_array(ic->journal_pages + 1,
3751 sizeof(struct scatterlist),
3754 *error = "Unable to allocate sg list";
3758 sg_init_table(sg, ic->journal_pages + 1);
3759 for (i = 0; i < ic->journal_pages; i++) {
3760 char *va = lowmem_page_address(ic->journal_xor[i].page);
3762 sg_set_buf(&sg[i], va, PAGE_SIZE);
3764 sg_set_buf(&sg[i], &ic->commit_ids, sizeof ic->commit_ids);
3766 skcipher_request_set_crypt(req, sg, sg,
3767 PAGE_SIZE * ic->journal_pages + sizeof ic->commit_ids, crypt_iv);
3768 init_completion(&comp.comp);
3769 comp.in_flight = (atomic_t)ATOMIC_INIT(1);
3770 if (do_crypt(true, req, &comp))
3771 wait_for_completion(&comp.comp);
3773 r = dm_integrity_failed(ic);
3775 *error = "Unable to encrypt journal";
3778 DEBUG_bytes(lowmem_page_address(ic->journal_xor[0].page), 64, "xor data");
3780 crypto_free_skcipher(ic->journal_crypt);
3781 ic->journal_crypt = NULL;
3783 unsigned crypt_len = roundup(ivsize, blocksize);
3785 req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
3787 *error = "Could not allocate crypt request";
3792 crypt_iv = kmalloc(ivsize, GFP_KERNEL);
3794 *error = "Could not allocate iv";
3799 crypt_data = kmalloc(crypt_len, GFP_KERNEL);
3801 *error = "Unable to allocate crypt data";
3806 ic->journal_scatterlist = dm_integrity_alloc_journal_scatterlist(ic, ic->journal);
3807 if (!ic->journal_scatterlist) {
3808 *error = "Unable to allocate sg list";
3812 ic->journal_io_scatterlist = dm_integrity_alloc_journal_scatterlist(ic, ic->journal_io);
3813 if (!ic->journal_io_scatterlist) {
3814 *error = "Unable to allocate sg list";
3818 ic->sk_requests = kvmalloc_array(ic->journal_sections,
3819 sizeof(struct skcipher_request *),
3820 GFP_KERNEL | __GFP_ZERO);
3821 if (!ic->sk_requests) {
3822 *error = "Unable to allocate sk requests";
3826 for (i = 0; i < ic->journal_sections; i++) {
3827 struct scatterlist sg;
3828 struct skcipher_request *section_req;
3829 __u32 section_le = cpu_to_le32(i);
3831 memset(crypt_iv, 0x00, ivsize);
3832 memset(crypt_data, 0x00, crypt_len);
3833 memcpy(crypt_data, §ion_le, min((size_t)crypt_len, sizeof(section_le)));
3835 sg_init_one(&sg, crypt_data, crypt_len);
3836 skcipher_request_set_crypt(req, &sg, &sg, crypt_len, crypt_iv);
3837 init_completion(&comp.comp);
3838 comp.in_flight = (atomic_t)ATOMIC_INIT(1);
3839 if (do_crypt(true, req, &comp))
3840 wait_for_completion(&comp.comp);
3842 r = dm_integrity_failed(ic);
3844 *error = "Unable to generate iv";
3848 section_req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
3850 *error = "Unable to allocate crypt request";
3854 section_req->iv = kmalloc_array(ivsize, 2,
3856 if (!section_req->iv) {
3857 skcipher_request_free(section_req);
3858 *error = "Unable to allocate iv";
3862 memcpy(section_req->iv + ivsize, crypt_data, ivsize);
3863 section_req->cryptlen = (size_t)ic->journal_section_sectors << SECTOR_SHIFT;
3864 ic->sk_requests[i] = section_req;
3865 DEBUG_bytes(crypt_data, ivsize, "iv(%u)", i);
3870 for (i = 0; i < N_COMMIT_IDS; i++) {
3873 for (j = 0; j < i; j++) {
3874 if (ic->commit_ids[j] == ic->commit_ids[i]) {
3875 ic->commit_ids[i] = cpu_to_le64(le64_to_cpu(ic->commit_ids[i]) + 1);
3876 goto retest_commit_id;
3879 DEBUG_print("commit id %u: %016llx\n", i, ic->commit_ids[i]);
3882 journal_tree_size = (__u64)ic->journal_entries * sizeof(struct journal_node);
3883 if (journal_tree_size > ULONG_MAX) {
3884 *error = "Journal doesn't fit into memory";
3888 ic->journal_tree = kvmalloc(journal_tree_size, GFP_KERNEL);
3889 if (!ic->journal_tree) {
3890 *error = "Could not allocate memory for journal tree";
3896 skcipher_request_free(req);
3902 * Construct a integrity mapping
3906 * offset from the start of the device
3908 * D - direct writes, J - journal writes, B - bitmap mode, R - recovery mode
3909 * number of optional arguments
3910 * optional arguments:
3912 * interleave_sectors
3919 * bitmap_flush_interval
3925 static int dm_integrity_ctr(struct dm_target *ti, unsigned argc, char **argv)
3927 struct dm_integrity_c *ic;
3930 unsigned extra_args;
3931 struct dm_arg_set as;
3932 static const struct dm_arg _args[] = {
3933 {0, 18, "Invalid number of feature args"},
3935 unsigned journal_sectors, interleave_sectors, buffer_sectors, journal_watermark, sync_msec;
3936 bool should_write_sb;
3938 unsigned long long start;
3939 __s8 log2_sectors_per_bitmap_bit = -1;
3940 __s8 log2_blocks_per_bitmap_bit;
3941 __u64 bits_in_journal;
3942 __u64 n_bitmap_bits;
3944 #define DIRECT_ARGUMENTS 4
3946 if (argc <= DIRECT_ARGUMENTS) {
3947 ti->error = "Invalid argument count";
3951 ic = kzalloc(sizeof(struct dm_integrity_c), GFP_KERNEL);
3953 ti->error = "Cannot allocate integrity context";
3957 ti->per_io_data_size = sizeof(struct dm_integrity_io);
3960 ic->in_progress = RB_ROOT;
3961 INIT_LIST_HEAD(&ic->wait_list);
3962 init_waitqueue_head(&ic->endio_wait);
3963 bio_list_init(&ic->flush_bio_list);
3964 init_waitqueue_head(&ic->copy_to_journal_wait);
3965 init_completion(&ic->crypto_backoff);
3966 atomic64_set(&ic->number_of_mismatches, 0);
3967 ic->bitmap_flush_interval = BITMAP_FLUSH_INTERVAL;
3969 r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &ic->dev);
3971 ti->error = "Device lookup failed";
3975 if (sscanf(argv[1], "%llu%c", &start, &dummy) != 1 || start != (sector_t)start) {
3976 ti->error = "Invalid starting offset";
3982 if (strcmp(argv[2], "-")) {
3983 if (sscanf(argv[2], "%u%c", &ic->tag_size, &dummy) != 1 || !ic->tag_size) {
3984 ti->error = "Invalid tag size";
3990 if (!strcmp(argv[3], "J") || !strcmp(argv[3], "B") ||
3991 !strcmp(argv[3], "D") || !strcmp(argv[3], "R")) {
3992 ic->mode = argv[3][0];
3994 ti->error = "Invalid mode (expecting J, B, D, R)";
3999 journal_sectors = 0;
4000 interleave_sectors = DEFAULT_INTERLEAVE_SECTORS;
4001 buffer_sectors = DEFAULT_BUFFER_SECTORS;
4002 journal_watermark = DEFAULT_JOURNAL_WATERMARK;
4003 sync_msec = DEFAULT_SYNC_MSEC;
4004 ic->sectors_per_block = 1;
4006 as.argc = argc - DIRECT_ARGUMENTS;
4007 as.argv = argv + DIRECT_ARGUMENTS;
4008 r = dm_read_arg_group(_args, &as, &extra_args, &ti->error);
4012 while (extra_args--) {
4013 const char *opt_string;
4015 unsigned long long llval;
4016 opt_string = dm_shift_arg(&as);
4019 ti->error = "Not enough feature arguments";
4022 if (sscanf(opt_string, "journal_sectors:%u%c", &val, &dummy) == 1)
4023 journal_sectors = val ? val : 1;
4024 else if (sscanf(opt_string, "interleave_sectors:%u%c", &val, &dummy) == 1)
4025 interleave_sectors = val;
4026 else if (sscanf(opt_string, "buffer_sectors:%u%c", &val, &dummy) == 1)
4027 buffer_sectors = val;
4028 else if (sscanf(opt_string, "journal_watermark:%u%c", &val, &dummy) == 1 && val <= 100)
4029 journal_watermark = val;
4030 else if (sscanf(opt_string, "commit_time:%u%c", &val, &dummy) == 1)
4032 else if (!strncmp(opt_string, "meta_device:", strlen("meta_device:"))) {
4034 dm_put_device(ti, ic->meta_dev);
4035 ic->meta_dev = NULL;
4037 r = dm_get_device(ti, strchr(opt_string, ':') + 1,
4038 dm_table_get_mode(ti->table), &ic->meta_dev);
4040 ti->error = "Device lookup failed";
4043 } else if (sscanf(opt_string, "block_size:%u%c", &val, &dummy) == 1) {
4044 if (val < 1 << SECTOR_SHIFT ||
4045 val > MAX_SECTORS_PER_BLOCK << SECTOR_SHIFT ||
4048 ti->error = "Invalid block_size argument";
4051 ic->sectors_per_block = val >> SECTOR_SHIFT;
4052 } else if (sscanf(opt_string, "sectors_per_bit:%llu%c", &llval, &dummy) == 1) {
4053 log2_sectors_per_bitmap_bit = !llval ? 0 : __ilog2_u64(llval);
4054 } else if (sscanf(opt_string, "bitmap_flush_interval:%u%c", &val, &dummy) == 1) {
4055 if (val >= (uint64_t)UINT_MAX * 1000 / HZ) {
4057 ti->error = "Invalid bitmap_flush_interval argument";
4060 ic->bitmap_flush_interval = msecs_to_jiffies(val);
4061 } else if (!strncmp(opt_string, "internal_hash:", strlen("internal_hash:"))) {
4062 r = get_alg_and_key(opt_string, &ic->internal_hash_alg, &ti->error,
4063 "Invalid internal_hash argument");
4066 } else if (!strncmp(opt_string, "journal_crypt:", strlen("journal_crypt:"))) {
4067 r = get_alg_and_key(opt_string, &ic->journal_crypt_alg, &ti->error,
4068 "Invalid journal_crypt argument");
4071 } else if (!strncmp(opt_string, "journal_mac:", strlen("journal_mac:"))) {
4072 r = get_alg_and_key(opt_string, &ic->journal_mac_alg, &ti->error,
4073 "Invalid journal_mac argument");
4076 } else if (!strcmp(opt_string, "recalculate")) {
4077 ic->recalculate_flag = true;
4078 } else if (!strcmp(opt_string, "reset_recalculate")) {
4079 ic->recalculate_flag = true;
4080 ic->reset_recalculate_flag = true;
4081 } else if (!strcmp(opt_string, "allow_discards")) {
4083 } else if (!strcmp(opt_string, "fix_padding")) {
4084 ic->fix_padding = true;
4085 } else if (!strcmp(opt_string, "fix_hmac")) {
4086 ic->fix_hmac = true;
4087 } else if (!strcmp(opt_string, "legacy_recalculate")) {
4088 ic->legacy_recalculate = true;
4091 ti->error = "Invalid argument";
4096 ic->data_device_sectors = i_size_read(ic->dev->bdev->bd_inode) >> SECTOR_SHIFT;
4098 ic->meta_device_sectors = ic->data_device_sectors;
4100 ic->meta_device_sectors = i_size_read(ic->meta_dev->bdev->bd_inode) >> SECTOR_SHIFT;
4102 if (!journal_sectors) {
4103 journal_sectors = min((sector_t)DEFAULT_MAX_JOURNAL_SECTORS,
4104 ic->data_device_sectors >> DEFAULT_JOURNAL_SIZE_FACTOR);
4107 if (!buffer_sectors)
4109 ic->log2_buffer_sectors = min((int)__fls(buffer_sectors), 31 - SECTOR_SHIFT);
4111 r = get_mac(&ic->internal_hash, &ic->internal_hash_alg, &ti->error,
4112 "Invalid internal hash", "Error setting internal hash key");
4116 r = get_mac(&ic->journal_mac, &ic->journal_mac_alg, &ti->error,
4117 "Invalid journal mac", "Error setting journal mac key");
4121 if (!ic->tag_size) {
4122 if (!ic->internal_hash) {
4123 ti->error = "Unknown tag size";
4127 ic->tag_size = crypto_shash_digestsize(ic->internal_hash);
4129 if (ic->tag_size > MAX_TAG_SIZE) {
4130 ti->error = "Too big tag size";
4134 if (!(ic->tag_size & (ic->tag_size - 1)))
4135 ic->log2_tag_size = __ffs(ic->tag_size);
4137 ic->log2_tag_size = -1;
4139 if (ic->mode == 'B' && !ic->internal_hash) {
4141 ti->error = "Bitmap mode can be only used with internal hash";
4145 if (ic->discard && !ic->internal_hash) {
4147 ti->error = "Discard can be only used with internal hash";
4151 ic->autocommit_jiffies = msecs_to_jiffies(sync_msec);
4152 ic->autocommit_msec = sync_msec;
4153 timer_setup(&ic->autocommit_timer, autocommit_fn, 0);
4155 ic->io = dm_io_client_create();
4156 if (IS_ERR(ic->io)) {
4157 r = PTR_ERR(ic->io);
4159 ti->error = "Cannot allocate dm io";
4163 r = mempool_init_slab_pool(&ic->journal_io_mempool, JOURNAL_IO_MEMPOOL, journal_io_cache);
4165 ti->error = "Cannot allocate mempool";
4169 ic->metadata_wq = alloc_workqueue("dm-integrity-metadata",
4170 WQ_MEM_RECLAIM, METADATA_WORKQUEUE_MAX_ACTIVE);
4171 if (!ic->metadata_wq) {
4172 ti->error = "Cannot allocate workqueue";
4178 * If this workqueue were percpu, it would cause bio reordering
4179 * and reduced performance.
4181 ic->wait_wq = alloc_workqueue("dm-integrity-wait", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
4183 ti->error = "Cannot allocate workqueue";
4188 ic->offload_wq = alloc_workqueue("dm-integrity-offload", WQ_MEM_RECLAIM,
4189 METADATA_WORKQUEUE_MAX_ACTIVE);
4190 if (!ic->offload_wq) {
4191 ti->error = "Cannot allocate workqueue";
4196 ic->commit_wq = alloc_workqueue("dm-integrity-commit", WQ_MEM_RECLAIM, 1);
4197 if (!ic->commit_wq) {
4198 ti->error = "Cannot allocate workqueue";
4202 INIT_WORK(&ic->commit_work, integrity_commit);
4204 if (ic->mode == 'J' || ic->mode == 'B') {
4205 ic->writer_wq = alloc_workqueue("dm-integrity-writer", WQ_MEM_RECLAIM, 1);
4206 if (!ic->writer_wq) {
4207 ti->error = "Cannot allocate workqueue";
4211 INIT_WORK(&ic->writer_work, integrity_writer);
4214 ic->sb = alloc_pages_exact(SB_SECTORS << SECTOR_SHIFT, GFP_KERNEL);
4217 ti->error = "Cannot allocate superblock area";
4221 r = sync_rw_sb(ic, REQ_OP_READ, 0);
4223 ti->error = "Error reading superblock";
4226 should_write_sb = false;
4227 if (memcmp(ic->sb->magic, SB_MAGIC, 8)) {
4228 if (ic->mode != 'R') {
4229 if (memchr_inv(ic->sb, 0, SB_SECTORS << SECTOR_SHIFT)) {
4231 ti->error = "The device is not initialized";
4236 r = initialize_superblock(ic, journal_sectors, interleave_sectors);
4238 ti->error = "Could not initialize superblock";
4241 if (ic->mode != 'R')
4242 should_write_sb = true;
4245 if (!ic->sb->version || ic->sb->version > SB_VERSION_5) {
4247 ti->error = "Unknown version";
4250 if (le16_to_cpu(ic->sb->integrity_tag_size) != ic->tag_size) {
4252 ti->error = "Tag size doesn't match the information in superblock";
4255 if (ic->sb->log2_sectors_per_block != __ffs(ic->sectors_per_block)) {
4257 ti->error = "Block size doesn't match the information in superblock";
4260 if (!le32_to_cpu(ic->sb->journal_sections)) {
4262 ti->error = "Corrupted superblock, journal_sections is 0";
4265 /* make sure that ti->max_io_len doesn't overflow */
4266 if (!ic->meta_dev) {
4267 if (ic->sb->log2_interleave_sectors < MIN_LOG2_INTERLEAVE_SECTORS ||
4268 ic->sb->log2_interleave_sectors > MAX_LOG2_INTERLEAVE_SECTORS) {
4270 ti->error = "Invalid interleave_sectors in the superblock";
4274 if (ic->sb->log2_interleave_sectors) {
4276 ti->error = "Invalid interleave_sectors in the superblock";
4280 if (!!(ic->sb->flags & cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC)) != !!ic->journal_mac_alg.alg_string) {
4282 ti->error = "Journal mac mismatch";
4286 get_provided_data_sectors(ic);
4287 if (!ic->provided_data_sectors) {
4289 ti->error = "The device is too small";
4294 r = calculate_device_limits(ic);
4297 if (ic->log2_buffer_sectors > 3) {
4298 ic->log2_buffer_sectors--;
4299 goto try_smaller_buffer;
4302 ti->error = "The device is too small";
4306 if (log2_sectors_per_bitmap_bit < 0)
4307 log2_sectors_per_bitmap_bit = __fls(DEFAULT_SECTORS_PER_BITMAP_BIT);
4308 if (log2_sectors_per_bitmap_bit < ic->sb->log2_sectors_per_block)
4309 log2_sectors_per_bitmap_bit = ic->sb->log2_sectors_per_block;
4311 bits_in_journal = ((__u64)ic->journal_section_sectors * ic->journal_sections) << (SECTOR_SHIFT + 3);
4312 if (bits_in_journal > UINT_MAX)
4313 bits_in_journal = UINT_MAX;
4314 while (bits_in_journal < (ic->provided_data_sectors + ((sector_t)1 << log2_sectors_per_bitmap_bit) - 1) >> log2_sectors_per_bitmap_bit)
4315 log2_sectors_per_bitmap_bit++;
4317 log2_blocks_per_bitmap_bit = log2_sectors_per_bitmap_bit - ic->sb->log2_sectors_per_block;
4318 ic->log2_blocks_per_bitmap_bit = log2_blocks_per_bitmap_bit;
4319 if (should_write_sb) {
4320 ic->sb->log2_blocks_per_bitmap_bit = log2_blocks_per_bitmap_bit;
4322 n_bitmap_bits = ((ic->provided_data_sectors >> ic->sb->log2_sectors_per_block)
4323 + (((sector_t)1 << log2_blocks_per_bitmap_bit) - 1)) >> log2_blocks_per_bitmap_bit;
4324 ic->n_bitmap_blocks = DIV_ROUND_UP(n_bitmap_bits, BITMAP_BLOCK_SIZE * 8);
4327 ic->log2_buffer_sectors = min(ic->log2_buffer_sectors, (__u8)__ffs(ic->metadata_run));
4329 if (ti->len > ic->provided_data_sectors) {
4331 ti->error = "Not enough provided sectors for requested mapping size";
4336 threshold = (__u64)ic->journal_entries * (100 - journal_watermark);
4338 do_div(threshold, 100);
4339 ic->free_sectors_threshold = threshold;
4341 DEBUG_print("initialized:\n");
4342 DEBUG_print(" integrity_tag_size %u\n", le16_to_cpu(ic->sb->integrity_tag_size));
4343 DEBUG_print(" journal_entry_size %u\n", ic->journal_entry_size);
4344 DEBUG_print(" journal_entries_per_sector %u\n", ic->journal_entries_per_sector);
4345 DEBUG_print(" journal_section_entries %u\n", ic->journal_section_entries);
4346 DEBUG_print(" journal_section_sectors %u\n", ic->journal_section_sectors);
4347 DEBUG_print(" journal_sections %u\n", (unsigned)le32_to_cpu(ic->sb->journal_sections));
4348 DEBUG_print(" journal_entries %u\n", ic->journal_entries);
4349 DEBUG_print(" log2_interleave_sectors %d\n", ic->sb->log2_interleave_sectors);
4350 DEBUG_print(" data_device_sectors 0x%llx\n", i_size_read(ic->dev->bdev->bd_inode) >> SECTOR_SHIFT);
4351 DEBUG_print(" initial_sectors 0x%x\n", ic->initial_sectors);
4352 DEBUG_print(" metadata_run 0x%x\n", ic->metadata_run);
4353 DEBUG_print(" log2_metadata_run %d\n", ic->log2_metadata_run);
4354 DEBUG_print(" provided_data_sectors 0x%llx (%llu)\n", ic->provided_data_sectors, ic->provided_data_sectors);
4355 DEBUG_print(" log2_buffer_sectors %u\n", ic->log2_buffer_sectors);
4356 DEBUG_print(" bits_in_journal %llu\n", bits_in_journal);
4358 if (ic->recalculate_flag && !(ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))) {
4359 ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
4360 ic->sb->recalc_sector = cpu_to_le64(0);
4363 if (ic->internal_hash) {
4364 ic->recalc_wq = alloc_workqueue("dm-integrity-recalc", WQ_MEM_RECLAIM, 1);
4365 if (!ic->recalc_wq ) {
4366 ti->error = "Cannot allocate workqueue";
4370 INIT_WORK(&ic->recalc_work, integrity_recalc);
4372 ic->recalc_buffer = vmalloc(RECALC_SECTORS << SECTOR_SHIFT);
4373 if (!ic->recalc_buffer) {
4374 ti->error = "Cannot allocate buffer for recalculating";
4379 ic->recalc_tags = kvmalloc_array(RECALC_SECTORS >> ic->sb->log2_sectors_per_block,
4380 ic->tag_size, GFP_KERNEL);
4381 if (!ic->recalc_tags) {
4382 ti->error = "Cannot allocate tags for recalculating";
4387 memset(ic->recalc_tags, DISCARD_FILLER,
4388 (RECALC_SECTORS >> ic->sb->log2_sectors_per_block) * ic->tag_size);
4390 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
4391 ti->error = "Recalculate can only be specified with internal_hash";
4397 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING) &&
4398 le64_to_cpu(ic->sb->recalc_sector) < ic->provided_data_sectors &&
4399 dm_integrity_disable_recalculate(ic)) {
4400 ti->error = "Recalculating with HMAC is disabled for security reasons - if you really need it, use the argument \"legacy_recalculate\"";
4405 ic->bufio = dm_bufio_client_create(ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev,
4406 1U << (SECTOR_SHIFT + ic->log2_buffer_sectors), 1, 0, NULL, NULL);
4407 if (IS_ERR(ic->bufio)) {
4408 r = PTR_ERR(ic->bufio);
4409 ti->error = "Cannot initialize dm-bufio";
4413 dm_bufio_set_sector_offset(ic->bufio, ic->start + ic->initial_sectors);
4415 if (ic->mode != 'R') {
4416 r = create_journal(ic, &ti->error);
4422 if (ic->mode == 'B') {
4424 unsigned n_bitmap_pages = DIV_ROUND_UP(ic->n_bitmap_blocks, PAGE_SIZE / BITMAP_BLOCK_SIZE);
4426 ic->recalc_bitmap = dm_integrity_alloc_page_list(n_bitmap_pages);
4427 if (!ic->recalc_bitmap) {
4431 ic->may_write_bitmap = dm_integrity_alloc_page_list(n_bitmap_pages);
4432 if (!ic->may_write_bitmap) {
4436 ic->bbs = kvmalloc_array(ic->n_bitmap_blocks, sizeof(struct bitmap_block_status), GFP_KERNEL);
4441 INIT_DELAYED_WORK(&ic->bitmap_flush_work, bitmap_flush_work);
4442 for (i = 0; i < ic->n_bitmap_blocks; i++) {
4443 struct bitmap_block_status *bbs = &ic->bbs[i];
4444 unsigned sector, pl_index, pl_offset;
4446 INIT_WORK(&bbs->work, bitmap_block_work);
4449 bio_list_init(&bbs->bio_queue);
4450 spin_lock_init(&bbs->bio_queue_lock);
4452 sector = i * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT);
4453 pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
4454 pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
4456 bbs->bitmap = lowmem_page_address(ic->journal[pl_index].page) + pl_offset;
4460 if (should_write_sb) {
4463 init_journal(ic, 0, ic->journal_sections, 0);
4464 r = dm_integrity_failed(ic);
4466 ti->error = "Error initializing journal";
4469 r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
4471 ti->error = "Error initializing superblock";
4474 ic->just_formatted = true;
4477 if (!ic->meta_dev) {
4478 r = dm_set_target_max_io_len(ti, 1U << ic->sb->log2_interleave_sectors);
4482 if (ic->mode == 'B') {
4483 unsigned max_io_len = ((sector_t)ic->sectors_per_block << ic->log2_blocks_per_bitmap_bit) * (BITMAP_BLOCK_SIZE * 8);
4485 max_io_len = 1U << 31;
4486 DEBUG_print("max_io_len: old %u, new %u\n", ti->max_io_len, max_io_len);
4487 if (!ti->max_io_len || ti->max_io_len > max_io_len) {
4488 r = dm_set_target_max_io_len(ti, max_io_len);
4494 if (!ic->internal_hash)
4495 dm_integrity_set(ti, ic);
4497 ti->num_flush_bios = 1;
4498 ti->flush_supported = true;
4500 ti->num_discard_bios = 1;
4505 dm_integrity_dtr(ti);
4509 static void dm_integrity_dtr(struct dm_target *ti)
4511 struct dm_integrity_c *ic = ti->private;
4513 BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
4514 BUG_ON(!list_empty(&ic->wait_list));
4516 if (ic->metadata_wq)
4517 destroy_workqueue(ic->metadata_wq);
4519 destroy_workqueue(ic->wait_wq);
4521 destroy_workqueue(ic->offload_wq);
4523 destroy_workqueue(ic->commit_wq);
4525 destroy_workqueue(ic->writer_wq);
4527 destroy_workqueue(ic->recalc_wq);
4528 vfree(ic->recalc_buffer);
4529 kvfree(ic->recalc_tags);
4532 dm_bufio_client_destroy(ic->bufio);
4533 mempool_exit(&ic->journal_io_mempool);
4535 dm_io_client_destroy(ic->io);
4537 dm_put_device(ti, ic->dev);
4539 dm_put_device(ti, ic->meta_dev);
4540 dm_integrity_free_page_list(ic->journal);
4541 dm_integrity_free_page_list(ic->journal_io);
4542 dm_integrity_free_page_list(ic->journal_xor);
4543 dm_integrity_free_page_list(ic->recalc_bitmap);
4544 dm_integrity_free_page_list(ic->may_write_bitmap);
4545 if (ic->journal_scatterlist)
4546 dm_integrity_free_journal_scatterlist(ic, ic->journal_scatterlist);
4547 if (ic->journal_io_scatterlist)
4548 dm_integrity_free_journal_scatterlist(ic, ic->journal_io_scatterlist);
4549 if (ic->sk_requests) {
4552 for (i = 0; i < ic->journal_sections; i++) {
4553 struct skcipher_request *req = ic->sk_requests[i];
4555 kfree_sensitive(req->iv);
4556 skcipher_request_free(req);
4559 kvfree(ic->sk_requests);
4561 kvfree(ic->journal_tree);
4563 free_pages_exact(ic->sb, SB_SECTORS << SECTOR_SHIFT);
4565 if (ic->internal_hash)
4566 crypto_free_shash(ic->internal_hash);
4567 free_alg(&ic->internal_hash_alg);
4569 if (ic->journal_crypt)
4570 crypto_free_skcipher(ic->journal_crypt);
4571 free_alg(&ic->journal_crypt_alg);
4573 if (ic->journal_mac)
4574 crypto_free_shash(ic->journal_mac);
4575 free_alg(&ic->journal_mac_alg);
4580 static struct target_type integrity_target = {
4581 .name = "integrity",
4582 .version = {1, 9, 0},
4583 .module = THIS_MODULE,
4584 .features = DM_TARGET_SINGLETON | DM_TARGET_INTEGRITY,
4585 .ctr = dm_integrity_ctr,
4586 .dtr = dm_integrity_dtr,
4587 .map = dm_integrity_map,
4588 .postsuspend = dm_integrity_postsuspend,
4589 .resume = dm_integrity_resume,
4590 .status = dm_integrity_status,
4591 .iterate_devices = dm_integrity_iterate_devices,
4592 .io_hints = dm_integrity_io_hints,
4595 static int __init dm_integrity_init(void)
4599 journal_io_cache = kmem_cache_create("integrity_journal_io",
4600 sizeof(struct journal_io), 0, 0, NULL);
4601 if (!journal_io_cache) {
4602 DMERR("can't allocate journal io cache");
4606 r = dm_register_target(&integrity_target);
4609 DMERR("register failed %d", r);
4614 static void __exit dm_integrity_exit(void)
4616 dm_unregister_target(&integrity_target);
4617 kmem_cache_destroy(journal_io_cache);
4620 module_init(dm_integrity_init);
4621 module_exit(dm_integrity_exit);
4623 MODULE_AUTHOR("Milan Broz");
4624 MODULE_AUTHOR("Mikulas Patocka");
4625 MODULE_DESCRIPTION(DM_NAME " target for integrity tags extension");
4626 MODULE_LICENSE("GPL");