2 * Copyright (C) 2016-2017 Red Hat, Inc. All rights reserved.
3 * Copyright (C) 2016-2017 Milan Broz
4 * Copyright (C) 2016-2017 Mikulas Patocka
6 * This file is released under the GPL.
9 #include "dm-bio-record.h"
11 #include <linux/compiler.h>
12 #include <linux/module.h>
13 #include <linux/device-mapper.h>
14 #include <linux/dm-io.h>
15 #include <linux/vmalloc.h>
16 #include <linux/sort.h>
17 #include <linux/rbtree.h>
18 #include <linux/delay.h>
19 #include <linux/random.h>
20 #include <linux/reboot.h>
21 #include <crypto/hash.h>
22 #include <crypto/skcipher.h>
23 #include <linux/async_tx.h>
24 #include <linux/dm-bufio.h>
28 #define DM_MSG_PREFIX "integrity"
30 #define DEFAULT_INTERLEAVE_SECTORS 32768
31 #define DEFAULT_JOURNAL_SIZE_FACTOR 7
32 #define DEFAULT_SECTORS_PER_BITMAP_BIT 32768
33 #define DEFAULT_BUFFER_SECTORS 128
34 #define DEFAULT_JOURNAL_WATERMARK 50
35 #define DEFAULT_SYNC_MSEC 10000
36 #define DEFAULT_MAX_JOURNAL_SECTORS 131072
37 #define MIN_LOG2_INTERLEAVE_SECTORS 3
38 #define MAX_LOG2_INTERLEAVE_SECTORS 31
39 #define METADATA_WORKQUEUE_MAX_ACTIVE 16
40 #define RECALC_SECTORS 32768
41 #define RECALC_WRITE_SUPER 16
42 #define BITMAP_BLOCK_SIZE 4096 /* don't change it */
43 #define BITMAP_FLUSH_INTERVAL (10 * HZ)
44 #define DISCARD_FILLER 0xf6
48 * Warning - DEBUG_PRINT prints security-sensitive data to the log,
49 * so it should not be enabled in the official kernel
52 //#define INTERNAL_VERIFY
58 #define SB_MAGIC "integrt"
59 #define SB_VERSION_1 1
60 #define SB_VERSION_2 2
61 #define SB_VERSION_3 3
62 #define SB_VERSION_4 4
63 #define SB_VERSION_5 5
65 #define MAX_SECTORS_PER_BLOCK 8
70 __u8 log2_interleave_sectors;
71 __le16 integrity_tag_size;
72 __le32 journal_sections;
73 __le64 provided_data_sectors; /* userspace uses this value */
75 __u8 log2_sectors_per_block;
76 __u8 log2_blocks_per_bitmap_bit;
83 #define SB_FLAG_HAVE_JOURNAL_MAC 0x1
84 #define SB_FLAG_RECALCULATING 0x2
85 #define SB_FLAG_DIRTY_BITMAP 0x4
86 #define SB_FLAG_FIXED_PADDING 0x8
87 #define SB_FLAG_FIXED_HMAC 0x10
89 #define JOURNAL_ENTRY_ROUNDUP 8
91 typedef __le64 commit_id_t;
92 #define JOURNAL_MAC_PER_SECTOR 8
94 struct journal_entry {
102 commit_id_t last_bytes[];
106 #define journal_entry_tag(ic, je) ((__u8 *)&(je)->last_bytes[(ic)->sectors_per_block])
108 #if BITS_PER_LONG == 64
109 #define journal_entry_set_sector(je, x) do { smp_wmb(); WRITE_ONCE((je)->u.sector, cpu_to_le64(x)); } while (0)
111 #define journal_entry_set_sector(je, x) do { (je)->u.s.sector_lo = cpu_to_le32(x); smp_wmb(); WRITE_ONCE((je)->u.s.sector_hi, cpu_to_le32((x) >> 32)); } while (0)
113 #define journal_entry_get_sector(je) le64_to_cpu((je)->u.sector)
114 #define journal_entry_is_unused(je) ((je)->u.s.sector_hi == cpu_to_le32(-1))
115 #define journal_entry_set_unused(je) do { ((je)->u.s.sector_hi = cpu_to_le32(-1)); } while (0)
116 #define journal_entry_is_inprogress(je) ((je)->u.s.sector_hi == cpu_to_le32(-2))
117 #define journal_entry_set_inprogress(je) do { ((je)->u.s.sector_hi = cpu_to_le32(-2)); } while (0)
119 #define JOURNAL_BLOCK_SECTORS 8
120 #define JOURNAL_SECTOR_DATA ((1 << SECTOR_SHIFT) - sizeof(commit_id_t))
121 #define JOURNAL_MAC_SIZE (JOURNAL_MAC_PER_SECTOR * JOURNAL_BLOCK_SECTORS)
123 struct journal_sector {
124 struct_group(sectors,
125 __u8 entries[JOURNAL_SECTOR_DATA - JOURNAL_MAC_PER_SECTOR];
126 __u8 mac[JOURNAL_MAC_PER_SECTOR];
128 commit_id_t commit_id;
131 #define MAX_TAG_SIZE (JOURNAL_SECTOR_DATA - JOURNAL_MAC_PER_SECTOR - offsetof(struct journal_entry, last_bytes[MAX_SECTORS_PER_BLOCK]))
133 #define METADATA_PADDING_SECTORS 8
135 #define N_COMMIT_IDS 4
137 static unsigned char prev_commit_seq(unsigned char seq)
139 return (seq + N_COMMIT_IDS - 1) % N_COMMIT_IDS;
142 static unsigned char next_commit_seq(unsigned char seq)
144 return (seq + 1) % N_COMMIT_IDS;
148 * In-memory structures
151 struct journal_node {
160 unsigned int key_size;
163 struct dm_integrity_c {
165 struct dm_dev *meta_dev;
166 unsigned int tag_size;
169 mempool_t journal_io_mempool;
170 struct dm_io_client *io;
171 struct dm_bufio_client *bufio;
172 struct workqueue_struct *metadata_wq;
173 struct superblock *sb;
174 unsigned int journal_pages;
175 unsigned int n_bitmap_blocks;
177 struct page_list *journal;
178 struct page_list *journal_io;
179 struct page_list *journal_xor;
180 struct page_list *recalc_bitmap;
181 struct page_list *may_write_bitmap;
182 struct bitmap_block_status *bbs;
183 unsigned int bitmap_flush_interval;
184 int synchronous_mode;
185 struct bio_list synchronous_bios;
186 struct delayed_work bitmap_flush_work;
188 struct crypto_skcipher *journal_crypt;
189 struct scatterlist **journal_scatterlist;
190 struct scatterlist **journal_io_scatterlist;
191 struct skcipher_request **sk_requests;
193 struct crypto_shash *journal_mac;
195 struct journal_node *journal_tree;
196 struct rb_root journal_tree_root;
198 sector_t provided_data_sectors;
200 unsigned short journal_entry_size;
201 unsigned char journal_entries_per_sector;
202 unsigned char journal_section_entries;
203 unsigned short journal_section_sectors;
204 unsigned int journal_sections;
205 unsigned int journal_entries;
206 sector_t data_device_sectors;
207 sector_t meta_device_sectors;
208 unsigned int initial_sectors;
209 unsigned int metadata_run;
210 __s8 log2_metadata_run;
211 __u8 log2_buffer_sectors;
212 __u8 sectors_per_block;
213 __u8 log2_blocks_per_bitmap_bit;
219 struct crypto_shash *internal_hash;
221 struct dm_target *ti;
223 /* these variables are locked with endio_wait.lock */
224 struct rb_root in_progress;
225 struct list_head wait_list;
226 wait_queue_head_t endio_wait;
227 struct workqueue_struct *wait_wq;
228 struct workqueue_struct *offload_wq;
230 unsigned char commit_seq;
231 commit_id_t commit_ids[N_COMMIT_IDS];
233 unsigned int committed_section;
234 unsigned int n_committed_sections;
236 unsigned int uncommitted_section;
237 unsigned int n_uncommitted_sections;
239 unsigned int free_section;
240 unsigned char free_section_entry;
241 unsigned int free_sectors;
243 unsigned int free_sectors_threshold;
245 struct workqueue_struct *commit_wq;
246 struct work_struct commit_work;
248 struct workqueue_struct *writer_wq;
249 struct work_struct writer_work;
251 struct workqueue_struct *recalc_wq;
252 struct work_struct recalc_work;
256 struct bio_list flush_bio_list;
258 unsigned long autocommit_jiffies;
259 struct timer_list autocommit_timer;
260 unsigned int autocommit_msec;
262 wait_queue_head_t copy_to_journal_wait;
264 struct completion crypto_backoff;
266 bool wrote_to_journal;
267 bool journal_uptodate;
269 bool recalculate_flag;
270 bool reset_recalculate_flag;
274 bool legacy_recalculate;
276 struct alg_spec internal_hash_alg;
277 struct alg_spec journal_crypt_alg;
278 struct alg_spec journal_mac_alg;
280 atomic64_t number_of_mismatches;
282 struct notifier_block reboot_notifier;
285 struct dm_integrity_range {
286 sector_t logical_sector;
292 struct task_struct *task;
293 struct list_head wait_entry;
298 struct dm_integrity_io {
299 struct work_struct work;
301 struct dm_integrity_c *ic;
305 struct dm_integrity_range range;
307 sector_t metadata_block;
308 unsigned int metadata_offset;
311 blk_status_t bi_status;
313 struct completion *completion;
315 struct dm_bio_details bio_details;
318 struct journal_completion {
319 struct dm_integrity_c *ic;
321 struct completion comp;
325 struct dm_integrity_range range;
326 struct journal_completion *comp;
329 struct bitmap_block_status {
330 struct work_struct work;
331 struct dm_integrity_c *ic;
333 unsigned long *bitmap;
334 struct bio_list bio_queue;
335 spinlock_t bio_queue_lock;
339 static struct kmem_cache *journal_io_cache;
341 #define JOURNAL_IO_MEMPOOL 32
344 #define DEBUG_print(x, ...) printk(KERN_DEBUG x, ##__VA_ARGS__)
345 static void __DEBUG_bytes(__u8 *bytes, size_t len, const char *msg, ...)
354 pr_cont(" %02x", *bytes);
360 #define DEBUG_bytes(bytes, len, msg, ...) __DEBUG_bytes(bytes, len, KERN_DEBUG msg, ##__VA_ARGS__)
362 #define DEBUG_print(x, ...) do { } while (0)
363 #define DEBUG_bytes(bytes, len, msg, ...) do { } while (0)
366 static void dm_integrity_prepare(struct request *rq)
370 static void dm_integrity_complete(struct request *rq, unsigned int nr_bytes)
375 * DM Integrity profile, protection is performed layer above (dm-crypt)
377 static const struct blk_integrity_profile dm_integrity_profile = {
378 .name = "DM-DIF-EXT-TAG",
381 .prepare_fn = dm_integrity_prepare,
382 .complete_fn = dm_integrity_complete,
385 static void dm_integrity_map_continue(struct dm_integrity_io *dio, bool from_map);
386 static void integrity_bio_wait(struct work_struct *w);
387 static void dm_integrity_dtr(struct dm_target *ti);
389 static void dm_integrity_io_error(struct dm_integrity_c *ic, const char *msg, int err)
392 atomic64_inc(&ic->number_of_mismatches);
393 if (!cmpxchg(&ic->failed, 0, err))
394 DMERR("Error on %s: %d", msg, err);
397 static int dm_integrity_failed(struct dm_integrity_c *ic)
399 return READ_ONCE(ic->failed);
402 static bool dm_integrity_disable_recalculate(struct dm_integrity_c *ic)
404 if (ic->legacy_recalculate)
406 if (!(ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) ?
407 ic->internal_hash_alg.key || ic->journal_mac_alg.key :
408 ic->internal_hash_alg.key && !ic->journal_mac_alg.key)
413 static commit_id_t dm_integrity_commit_id(struct dm_integrity_c *ic, unsigned int i,
414 unsigned int j, unsigned char seq)
417 * Xor the number with section and sector, so that if a piece of
418 * journal is written at wrong place, it is detected.
420 return ic->commit_ids[seq] ^ cpu_to_le64(((__u64)i << 32) ^ j);
423 static void get_area_and_offset(struct dm_integrity_c *ic, sector_t data_sector,
424 sector_t *area, sector_t *offset)
427 __u8 log2_interleave_sectors = ic->sb->log2_interleave_sectors;
428 *area = data_sector >> log2_interleave_sectors;
429 *offset = (unsigned int)data_sector & ((1U << log2_interleave_sectors) - 1);
432 *offset = data_sector;
436 #define sector_to_block(ic, n) \
438 BUG_ON((n) & (unsigned int)((ic)->sectors_per_block - 1)); \
439 (n) >>= (ic)->sb->log2_sectors_per_block; \
442 static __u64 get_metadata_sector_and_offset(struct dm_integrity_c *ic, sector_t area,
443 sector_t offset, unsigned int *metadata_offset)
448 ms = area << ic->sb->log2_interleave_sectors;
449 if (likely(ic->log2_metadata_run >= 0))
450 ms += area << ic->log2_metadata_run;
452 ms += area * ic->metadata_run;
453 ms >>= ic->log2_buffer_sectors;
455 sector_to_block(ic, offset);
457 if (likely(ic->log2_tag_size >= 0)) {
458 ms += offset >> (SECTOR_SHIFT + ic->log2_buffer_sectors - ic->log2_tag_size);
459 mo = (offset << ic->log2_tag_size) & ((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - 1);
461 ms += (__u64)offset * ic->tag_size >> (SECTOR_SHIFT + ic->log2_buffer_sectors);
462 mo = (offset * ic->tag_size) & ((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - 1);
464 *metadata_offset = mo;
468 static sector_t get_data_sector(struct dm_integrity_c *ic, sector_t area, sector_t offset)
475 result = area << ic->sb->log2_interleave_sectors;
476 if (likely(ic->log2_metadata_run >= 0))
477 result += (area + 1) << ic->log2_metadata_run;
479 result += (area + 1) * ic->metadata_run;
481 result += (sector_t)ic->initial_sectors + offset;
487 static void wraparound_section(struct dm_integrity_c *ic, unsigned int *sec_ptr)
489 if (unlikely(*sec_ptr >= ic->journal_sections))
490 *sec_ptr -= ic->journal_sections;
493 static void sb_set_version(struct dm_integrity_c *ic)
495 if (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC))
496 ic->sb->version = SB_VERSION_5;
497 else if (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING))
498 ic->sb->version = SB_VERSION_4;
499 else if (ic->mode == 'B' || ic->sb->flags & cpu_to_le32(SB_FLAG_DIRTY_BITMAP))
500 ic->sb->version = SB_VERSION_3;
501 else if (ic->meta_dev || ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
502 ic->sb->version = SB_VERSION_2;
504 ic->sb->version = SB_VERSION_1;
507 static int sb_mac(struct dm_integrity_c *ic, bool wr)
509 SHASH_DESC_ON_STACK(desc, ic->journal_mac);
511 unsigned int size = crypto_shash_digestsize(ic->journal_mac);
513 if (sizeof(struct superblock) + size > 1 << SECTOR_SHIFT) {
514 dm_integrity_io_error(ic, "digest is too long", -EINVAL);
518 desc->tfm = ic->journal_mac;
520 r = crypto_shash_init(desc);
521 if (unlikely(r < 0)) {
522 dm_integrity_io_error(ic, "crypto_shash_init", r);
526 r = crypto_shash_update(desc, (__u8 *)ic->sb, (1 << SECTOR_SHIFT) - size);
527 if (unlikely(r < 0)) {
528 dm_integrity_io_error(ic, "crypto_shash_update", r);
533 r = crypto_shash_final(desc, (__u8 *)ic->sb + (1 << SECTOR_SHIFT) - size);
534 if (unlikely(r < 0)) {
535 dm_integrity_io_error(ic, "crypto_shash_final", r);
539 __u8 result[HASH_MAX_DIGESTSIZE];
540 r = crypto_shash_final(desc, result);
541 if (unlikely(r < 0)) {
542 dm_integrity_io_error(ic, "crypto_shash_final", r);
545 if (memcmp((__u8 *)ic->sb + (1 << SECTOR_SHIFT) - size, result, size)) {
546 dm_integrity_io_error(ic, "superblock mac", -EILSEQ);
547 dm_audit_log_target(DM_MSG_PREFIX, "mac-superblock", ic->ti, 0);
555 static int sync_rw_sb(struct dm_integrity_c *ic, blk_opf_t opf)
557 struct dm_io_request io_req;
558 struct dm_io_region io_loc;
559 const enum req_op op = opf & REQ_OP_MASK;
563 io_req.mem.type = DM_IO_KMEM;
564 io_req.mem.ptr.addr = ic->sb;
565 io_req.notify.fn = NULL;
566 io_req.client = ic->io;
567 io_loc.bdev = ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev;
568 io_loc.sector = ic->start;
569 io_loc.count = SB_SECTORS;
571 if (op == REQ_OP_WRITE) {
573 if (ic->journal_mac && ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) {
574 r = sb_mac(ic, true);
580 r = dm_io(&io_req, 1, &io_loc, NULL);
584 if (op == REQ_OP_READ) {
585 if (ic->mode != 'R' && ic->journal_mac && ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) {
586 r = sb_mac(ic, false);
595 #define BITMAP_OP_TEST_ALL_SET 0
596 #define BITMAP_OP_TEST_ALL_CLEAR 1
597 #define BITMAP_OP_SET 2
598 #define BITMAP_OP_CLEAR 3
600 static bool block_bitmap_op(struct dm_integrity_c *ic, struct page_list *bitmap,
601 sector_t sector, sector_t n_sectors, int mode)
603 unsigned long bit, end_bit, this_end_bit, page, end_page;
606 if (unlikely(((sector | n_sectors) & ((1 << ic->sb->log2_sectors_per_block) - 1)) != 0)) {
607 DMCRIT("invalid bitmap access (%llx,%llx,%d,%d,%d)",
610 ic->sb->log2_sectors_per_block,
611 ic->log2_blocks_per_bitmap_bit,
616 if (unlikely(!n_sectors))
619 bit = sector >> (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
620 end_bit = (sector + n_sectors - 1) >>
621 (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
623 page = bit / (PAGE_SIZE * 8);
624 bit %= PAGE_SIZE * 8;
626 end_page = end_bit / (PAGE_SIZE * 8);
627 end_bit %= PAGE_SIZE * 8;
630 if (page < end_page) {
631 this_end_bit = PAGE_SIZE * 8 - 1;
633 this_end_bit = end_bit;
636 data = lowmem_page_address(bitmap[page].page);
638 if (mode == BITMAP_OP_TEST_ALL_SET) {
639 while (bit <= this_end_bit) {
640 if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
642 if (data[bit / BITS_PER_LONG] != -1)
644 bit += BITS_PER_LONG;
645 } while (this_end_bit >= bit + BITS_PER_LONG - 1);
648 if (!test_bit(bit, data))
652 } else if (mode == BITMAP_OP_TEST_ALL_CLEAR) {
653 while (bit <= this_end_bit) {
654 if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
656 if (data[bit / BITS_PER_LONG] != 0)
658 bit += BITS_PER_LONG;
659 } while (this_end_bit >= bit + BITS_PER_LONG - 1);
662 if (test_bit(bit, data))
666 } else if (mode == BITMAP_OP_SET) {
667 while (bit <= this_end_bit) {
668 if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
670 data[bit / BITS_PER_LONG] = -1;
671 bit += BITS_PER_LONG;
672 } while (this_end_bit >= bit + BITS_PER_LONG - 1);
675 __set_bit(bit, data);
678 } else if (mode == BITMAP_OP_CLEAR) {
679 if (!bit && this_end_bit == PAGE_SIZE * 8 - 1)
681 else while (bit <= this_end_bit) {
682 if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
684 data[bit / BITS_PER_LONG] = 0;
685 bit += BITS_PER_LONG;
686 } while (this_end_bit >= bit + BITS_PER_LONG - 1);
689 __clear_bit(bit, data);
696 if (unlikely(page < end_page)) {
705 static void block_bitmap_copy(struct dm_integrity_c *ic, struct page_list *dst, struct page_list *src)
707 unsigned int n_bitmap_pages = DIV_ROUND_UP(ic->n_bitmap_blocks, PAGE_SIZE / BITMAP_BLOCK_SIZE);
710 for (i = 0; i < n_bitmap_pages; i++) {
711 unsigned long *dst_data = lowmem_page_address(dst[i].page);
712 unsigned long *src_data = lowmem_page_address(src[i].page);
713 copy_page(dst_data, src_data);
717 static struct bitmap_block_status *sector_to_bitmap_block(struct dm_integrity_c *ic, sector_t sector)
719 unsigned int bit = sector >> (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
720 unsigned int bitmap_block = bit / (BITMAP_BLOCK_SIZE * 8);
722 BUG_ON(bitmap_block >= ic->n_bitmap_blocks);
723 return &ic->bbs[bitmap_block];
726 static void access_journal_check(struct dm_integrity_c *ic, unsigned int section, unsigned int offset,
727 bool e, const char *function)
729 #if defined(CONFIG_DM_DEBUG) || defined(INTERNAL_VERIFY)
730 unsigned int limit = e ? ic->journal_section_entries : ic->journal_section_sectors;
732 if (unlikely(section >= ic->journal_sections) ||
733 unlikely(offset >= limit)) {
734 DMCRIT("%s: invalid access at (%u,%u), limit (%u,%u)",
735 function, section, offset, ic->journal_sections, limit);
741 static void page_list_location(struct dm_integrity_c *ic, unsigned int section, unsigned int offset,
742 unsigned int *pl_index, unsigned int *pl_offset)
746 access_journal_check(ic, section, offset, false, "page_list_location");
748 sector = section * ic->journal_section_sectors + offset;
750 *pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
751 *pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
754 static struct journal_sector *access_page_list(struct dm_integrity_c *ic, struct page_list *pl,
755 unsigned int section, unsigned int offset, unsigned int *n_sectors)
757 unsigned int pl_index, pl_offset;
760 page_list_location(ic, section, offset, &pl_index, &pl_offset);
763 *n_sectors = (PAGE_SIZE - pl_offset) >> SECTOR_SHIFT;
765 va = lowmem_page_address(pl[pl_index].page);
767 return (struct journal_sector *)(va + pl_offset);
770 static struct journal_sector *access_journal(struct dm_integrity_c *ic, unsigned int section, unsigned int offset)
772 return access_page_list(ic, ic->journal, section, offset, NULL);
775 static struct journal_entry *access_journal_entry(struct dm_integrity_c *ic, unsigned int section, unsigned int n)
777 unsigned int rel_sector, offset;
778 struct journal_sector *js;
780 access_journal_check(ic, section, n, true, "access_journal_entry");
782 rel_sector = n % JOURNAL_BLOCK_SECTORS;
783 offset = n / JOURNAL_BLOCK_SECTORS;
785 js = access_journal(ic, section, rel_sector);
786 return (struct journal_entry *)((char *)js + offset * ic->journal_entry_size);
789 static struct journal_sector *access_journal_data(struct dm_integrity_c *ic, unsigned int section, unsigned int n)
791 n <<= ic->sb->log2_sectors_per_block;
793 n += JOURNAL_BLOCK_SECTORS;
795 access_journal_check(ic, section, n, false, "access_journal_data");
797 return access_journal(ic, section, n);
800 static void section_mac(struct dm_integrity_c *ic, unsigned int section, __u8 result[JOURNAL_MAC_SIZE])
802 SHASH_DESC_ON_STACK(desc, ic->journal_mac);
804 unsigned int j, size;
806 desc->tfm = ic->journal_mac;
808 r = crypto_shash_init(desc);
809 if (unlikely(r < 0)) {
810 dm_integrity_io_error(ic, "crypto_shash_init", r);
814 if (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) {
817 r = crypto_shash_update(desc, (__u8 *)&ic->sb->salt, SALT_SIZE);
818 if (unlikely(r < 0)) {
819 dm_integrity_io_error(ic, "crypto_shash_update", r);
823 section_le = cpu_to_le64(section);
824 r = crypto_shash_update(desc, (__u8 *)§ion_le, sizeof section_le);
825 if (unlikely(r < 0)) {
826 dm_integrity_io_error(ic, "crypto_shash_update", r);
831 for (j = 0; j < ic->journal_section_entries; j++) {
832 struct journal_entry *je = access_journal_entry(ic, section, j);
833 r = crypto_shash_update(desc, (__u8 *)&je->u.sector, sizeof je->u.sector);
834 if (unlikely(r < 0)) {
835 dm_integrity_io_error(ic, "crypto_shash_update", r);
840 size = crypto_shash_digestsize(ic->journal_mac);
842 if (likely(size <= JOURNAL_MAC_SIZE)) {
843 r = crypto_shash_final(desc, result);
844 if (unlikely(r < 0)) {
845 dm_integrity_io_error(ic, "crypto_shash_final", r);
848 memset(result + size, 0, JOURNAL_MAC_SIZE - size);
850 __u8 digest[HASH_MAX_DIGESTSIZE];
852 if (WARN_ON(size > sizeof(digest))) {
853 dm_integrity_io_error(ic, "digest_size", -EINVAL);
856 r = crypto_shash_final(desc, digest);
857 if (unlikely(r < 0)) {
858 dm_integrity_io_error(ic, "crypto_shash_final", r);
861 memcpy(result, digest, JOURNAL_MAC_SIZE);
866 memset(result, 0, JOURNAL_MAC_SIZE);
869 static void rw_section_mac(struct dm_integrity_c *ic, unsigned int section, bool wr)
871 __u8 result[JOURNAL_MAC_SIZE];
874 if (!ic->journal_mac)
877 section_mac(ic, section, result);
879 for (j = 0; j < JOURNAL_BLOCK_SECTORS; j++) {
880 struct journal_sector *js = access_journal(ic, section, j);
883 memcpy(&js->mac, result + (j * JOURNAL_MAC_PER_SECTOR), JOURNAL_MAC_PER_SECTOR);
885 if (memcmp(&js->mac, result + (j * JOURNAL_MAC_PER_SECTOR), JOURNAL_MAC_PER_SECTOR)) {
886 dm_integrity_io_error(ic, "journal mac", -EILSEQ);
887 dm_audit_log_target(DM_MSG_PREFIX, "mac-journal", ic->ti, 0);
893 static void complete_journal_op(void *context)
895 struct journal_completion *comp = context;
896 BUG_ON(!atomic_read(&comp->in_flight));
897 if (likely(atomic_dec_and_test(&comp->in_flight)))
898 complete(&comp->comp);
901 static void xor_journal(struct dm_integrity_c *ic, bool encrypt, unsigned int section,
902 unsigned int n_sections, struct journal_completion *comp)
904 struct async_submit_ctl submit;
905 size_t n_bytes = (size_t)(n_sections * ic->journal_section_sectors) << SECTOR_SHIFT;
906 unsigned int pl_index, pl_offset, section_index;
907 struct page_list *source_pl, *target_pl;
909 if (likely(encrypt)) {
910 source_pl = ic->journal;
911 target_pl = ic->journal_io;
913 source_pl = ic->journal_io;
914 target_pl = ic->journal;
917 page_list_location(ic, section, 0, &pl_index, &pl_offset);
919 atomic_add(roundup(pl_offset + n_bytes, PAGE_SIZE) >> PAGE_SHIFT, &comp->in_flight);
921 init_async_submit(&submit, ASYNC_TX_XOR_ZERO_DST, NULL, complete_journal_op, comp, NULL);
923 section_index = pl_index;
927 struct page *src_pages[2];
928 struct page *dst_page;
930 while (unlikely(pl_index == section_index)) {
933 rw_section_mac(ic, section, true);
938 page_list_location(ic, section, 0, §ion_index, &dummy);
941 this_step = min(n_bytes, (size_t)PAGE_SIZE - pl_offset);
942 dst_page = target_pl[pl_index].page;
943 src_pages[0] = source_pl[pl_index].page;
944 src_pages[1] = ic->journal_xor[pl_index].page;
946 async_xor(dst_page, src_pages, pl_offset, 2, this_step, &submit);
950 n_bytes -= this_step;
955 async_tx_issue_pending_all();
958 static void complete_journal_encrypt(struct crypto_async_request *req, int err)
960 struct journal_completion *comp = req->data;
962 if (likely(err == -EINPROGRESS)) {
963 complete(&comp->ic->crypto_backoff);
966 dm_integrity_io_error(comp->ic, "asynchronous encrypt", err);
968 complete_journal_op(comp);
971 static bool do_crypt(bool encrypt, struct skcipher_request *req, struct journal_completion *comp)
974 skcipher_request_set_callback(req, CRYPTO_TFM_REQ_MAY_BACKLOG,
975 complete_journal_encrypt, comp);
977 r = crypto_skcipher_encrypt(req);
979 r = crypto_skcipher_decrypt(req);
982 if (likely(r == -EINPROGRESS))
984 if (likely(r == -EBUSY)) {
985 wait_for_completion(&comp->ic->crypto_backoff);
986 reinit_completion(&comp->ic->crypto_backoff);
989 dm_integrity_io_error(comp->ic, "encrypt", r);
993 static void crypt_journal(struct dm_integrity_c *ic, bool encrypt, unsigned int section,
994 unsigned int n_sections, struct journal_completion *comp)
996 struct scatterlist **source_sg;
997 struct scatterlist **target_sg;
999 atomic_add(2, &comp->in_flight);
1001 if (likely(encrypt)) {
1002 source_sg = ic->journal_scatterlist;
1003 target_sg = ic->journal_io_scatterlist;
1005 source_sg = ic->journal_io_scatterlist;
1006 target_sg = ic->journal_scatterlist;
1010 struct skcipher_request *req;
1011 unsigned int ivsize;
1014 if (likely(encrypt))
1015 rw_section_mac(ic, section, true);
1017 req = ic->sk_requests[section];
1018 ivsize = crypto_skcipher_ivsize(ic->journal_crypt);
1021 memcpy(iv, iv + ivsize, ivsize);
1023 req->src = source_sg[section];
1024 req->dst = target_sg[section];
1026 if (unlikely(do_crypt(encrypt, req, comp)))
1027 atomic_inc(&comp->in_flight);
1031 } while (n_sections);
1033 atomic_dec(&comp->in_flight);
1034 complete_journal_op(comp);
1037 static void encrypt_journal(struct dm_integrity_c *ic, bool encrypt, unsigned int section,
1038 unsigned int n_sections, struct journal_completion *comp)
1040 if (ic->journal_xor)
1041 return xor_journal(ic, encrypt, section, n_sections, comp);
1043 return crypt_journal(ic, encrypt, section, n_sections, comp);
1046 static void complete_journal_io(unsigned long error, void *context)
1048 struct journal_completion *comp = context;
1049 if (unlikely(error != 0))
1050 dm_integrity_io_error(comp->ic, "writing journal", -EIO);
1051 complete_journal_op(comp);
1054 static void rw_journal_sectors(struct dm_integrity_c *ic, blk_opf_t opf,
1055 unsigned int sector, unsigned int n_sectors,
1056 struct journal_completion *comp)
1058 struct dm_io_request io_req;
1059 struct dm_io_region io_loc;
1060 unsigned int pl_index, pl_offset;
1063 if (unlikely(dm_integrity_failed(ic))) {
1065 complete_journal_io(-1UL, comp);
1069 pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
1070 pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
1072 io_req.bi_opf = opf;
1073 io_req.mem.type = DM_IO_PAGE_LIST;
1075 io_req.mem.ptr.pl = &ic->journal_io[pl_index];
1077 io_req.mem.ptr.pl = &ic->journal[pl_index];
1078 io_req.mem.offset = pl_offset;
1079 if (likely(comp != NULL)) {
1080 io_req.notify.fn = complete_journal_io;
1081 io_req.notify.context = comp;
1083 io_req.notify.fn = NULL;
1085 io_req.client = ic->io;
1086 io_loc.bdev = ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev;
1087 io_loc.sector = ic->start + SB_SECTORS + sector;
1088 io_loc.count = n_sectors;
1090 r = dm_io(&io_req, 1, &io_loc, NULL);
1092 dm_integrity_io_error(ic, (opf & REQ_OP_MASK) == REQ_OP_READ ?
1093 "reading journal" : "writing journal", r);
1095 WARN_ONCE(1, "asynchronous dm_io failed: %d", r);
1096 complete_journal_io(-1UL, comp);
1101 static void rw_journal(struct dm_integrity_c *ic, blk_opf_t opf,
1102 unsigned int section, unsigned int n_sections,
1103 struct journal_completion *comp)
1105 unsigned int sector, n_sectors;
1107 sector = section * ic->journal_section_sectors;
1108 n_sectors = n_sections * ic->journal_section_sectors;
1110 rw_journal_sectors(ic, opf, sector, n_sectors, comp);
1113 static void write_journal(struct dm_integrity_c *ic, unsigned int commit_start, unsigned int commit_sections)
1115 struct journal_completion io_comp;
1116 struct journal_completion crypt_comp_1;
1117 struct journal_completion crypt_comp_2;
1121 init_completion(&io_comp.comp);
1123 if (commit_start + commit_sections <= ic->journal_sections) {
1124 io_comp.in_flight = (atomic_t)ATOMIC_INIT(1);
1125 if (ic->journal_io) {
1126 crypt_comp_1.ic = ic;
1127 init_completion(&crypt_comp_1.comp);
1128 crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
1129 encrypt_journal(ic, true, commit_start, commit_sections, &crypt_comp_1);
1130 wait_for_completion_io(&crypt_comp_1.comp);
1132 for (i = 0; i < commit_sections; i++)
1133 rw_section_mac(ic, commit_start + i, true);
1135 rw_journal(ic, REQ_OP_WRITE | REQ_FUA | REQ_SYNC, commit_start,
1136 commit_sections, &io_comp);
1138 unsigned int to_end;
1139 io_comp.in_flight = (atomic_t)ATOMIC_INIT(2);
1140 to_end = ic->journal_sections - commit_start;
1141 if (ic->journal_io) {
1142 crypt_comp_1.ic = ic;
1143 init_completion(&crypt_comp_1.comp);
1144 crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
1145 encrypt_journal(ic, true, commit_start, to_end, &crypt_comp_1);
1146 if (try_wait_for_completion(&crypt_comp_1.comp)) {
1147 rw_journal(ic, REQ_OP_WRITE | REQ_FUA,
1148 commit_start, to_end, &io_comp);
1149 reinit_completion(&crypt_comp_1.comp);
1150 crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
1151 encrypt_journal(ic, true, 0, commit_sections - to_end, &crypt_comp_1);
1152 wait_for_completion_io(&crypt_comp_1.comp);
1154 crypt_comp_2.ic = ic;
1155 init_completion(&crypt_comp_2.comp);
1156 crypt_comp_2.in_flight = (atomic_t)ATOMIC_INIT(0);
1157 encrypt_journal(ic, true, 0, commit_sections - to_end, &crypt_comp_2);
1158 wait_for_completion_io(&crypt_comp_1.comp);
1159 rw_journal(ic, REQ_OP_WRITE | REQ_FUA, commit_start, to_end, &io_comp);
1160 wait_for_completion_io(&crypt_comp_2.comp);
1163 for (i = 0; i < to_end; i++)
1164 rw_section_mac(ic, commit_start + i, true);
1165 rw_journal(ic, REQ_OP_WRITE | REQ_FUA, commit_start, to_end, &io_comp);
1166 for (i = 0; i < commit_sections - to_end; i++)
1167 rw_section_mac(ic, i, true);
1169 rw_journal(ic, REQ_OP_WRITE | REQ_FUA, 0, commit_sections - to_end, &io_comp);
1172 wait_for_completion_io(&io_comp.comp);
1175 static void copy_from_journal(struct dm_integrity_c *ic, unsigned int section, unsigned int offset,
1176 unsigned int n_sectors, sector_t target, io_notify_fn fn, void *data)
1178 struct dm_io_request io_req;
1179 struct dm_io_region io_loc;
1181 unsigned int sector, pl_index, pl_offset;
1183 BUG_ON((target | n_sectors | offset) & (unsigned int)(ic->sectors_per_block - 1));
1185 if (unlikely(dm_integrity_failed(ic))) {
1190 sector = section * ic->journal_section_sectors + JOURNAL_BLOCK_SECTORS + offset;
1192 pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
1193 pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
1195 io_req.bi_opf = REQ_OP_WRITE;
1196 io_req.mem.type = DM_IO_PAGE_LIST;
1197 io_req.mem.ptr.pl = &ic->journal[pl_index];
1198 io_req.mem.offset = pl_offset;
1199 io_req.notify.fn = fn;
1200 io_req.notify.context = data;
1201 io_req.client = ic->io;
1202 io_loc.bdev = ic->dev->bdev;
1203 io_loc.sector = target;
1204 io_loc.count = n_sectors;
1206 r = dm_io(&io_req, 1, &io_loc, NULL);
1208 WARN_ONCE(1, "asynchronous dm_io failed: %d", r);
1213 static bool ranges_overlap(struct dm_integrity_range *range1, struct dm_integrity_range *range2)
1215 return range1->logical_sector < range2->logical_sector + range2->n_sectors &&
1216 range1->logical_sector + range1->n_sectors > range2->logical_sector;
1219 static bool add_new_range(struct dm_integrity_c *ic, struct dm_integrity_range *new_range, bool check_waiting)
1221 struct rb_node **n = &ic->in_progress.rb_node;
1222 struct rb_node *parent;
1224 BUG_ON((new_range->logical_sector | new_range->n_sectors) & (unsigned int)(ic->sectors_per_block - 1));
1226 if (likely(check_waiting)) {
1227 struct dm_integrity_range *range;
1228 list_for_each_entry(range, &ic->wait_list, wait_entry) {
1229 if (unlikely(ranges_overlap(range, new_range)))
1237 struct dm_integrity_range *range = container_of(*n, struct dm_integrity_range, node);
1240 if (new_range->logical_sector + new_range->n_sectors <= range->logical_sector) {
1241 n = &range->node.rb_left;
1242 } else if (new_range->logical_sector >= range->logical_sector + range->n_sectors) {
1243 n = &range->node.rb_right;
1249 rb_link_node(&new_range->node, parent, n);
1250 rb_insert_color(&new_range->node, &ic->in_progress);
1255 static void remove_range_unlocked(struct dm_integrity_c *ic, struct dm_integrity_range *range)
1257 rb_erase(&range->node, &ic->in_progress);
1258 while (unlikely(!list_empty(&ic->wait_list))) {
1259 struct dm_integrity_range *last_range =
1260 list_first_entry(&ic->wait_list, struct dm_integrity_range, wait_entry);
1261 struct task_struct *last_range_task;
1262 last_range_task = last_range->task;
1263 list_del(&last_range->wait_entry);
1264 if (!add_new_range(ic, last_range, false)) {
1265 last_range->task = last_range_task;
1266 list_add(&last_range->wait_entry, &ic->wait_list);
1269 last_range->waiting = false;
1270 wake_up_process(last_range_task);
1274 static void remove_range(struct dm_integrity_c *ic, struct dm_integrity_range *range)
1276 unsigned long flags;
1278 spin_lock_irqsave(&ic->endio_wait.lock, flags);
1279 remove_range_unlocked(ic, range);
1280 spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
1283 static void wait_and_add_new_range(struct dm_integrity_c *ic, struct dm_integrity_range *new_range)
1285 new_range->waiting = true;
1286 list_add_tail(&new_range->wait_entry, &ic->wait_list);
1287 new_range->task = current;
1289 __set_current_state(TASK_UNINTERRUPTIBLE);
1290 spin_unlock_irq(&ic->endio_wait.lock);
1292 spin_lock_irq(&ic->endio_wait.lock);
1293 } while (unlikely(new_range->waiting));
1296 static void add_new_range_and_wait(struct dm_integrity_c *ic, struct dm_integrity_range *new_range)
1298 if (unlikely(!add_new_range(ic, new_range, true)))
1299 wait_and_add_new_range(ic, new_range);
1302 static void init_journal_node(struct journal_node *node)
1304 RB_CLEAR_NODE(&node->node);
1305 node->sector = (sector_t)-1;
1308 static void add_journal_node(struct dm_integrity_c *ic, struct journal_node *node, sector_t sector)
1310 struct rb_node **link;
1311 struct rb_node *parent;
1313 node->sector = sector;
1314 BUG_ON(!RB_EMPTY_NODE(&node->node));
1316 link = &ic->journal_tree_root.rb_node;
1320 struct journal_node *j;
1322 j = container_of(parent, struct journal_node, node);
1323 if (sector < j->sector)
1324 link = &j->node.rb_left;
1326 link = &j->node.rb_right;
1329 rb_link_node(&node->node, parent, link);
1330 rb_insert_color(&node->node, &ic->journal_tree_root);
1333 static void remove_journal_node(struct dm_integrity_c *ic, struct journal_node *node)
1335 BUG_ON(RB_EMPTY_NODE(&node->node));
1336 rb_erase(&node->node, &ic->journal_tree_root);
1337 init_journal_node(node);
1340 #define NOT_FOUND (-1U)
1342 static unsigned int find_journal_node(struct dm_integrity_c *ic, sector_t sector, sector_t *next_sector)
1344 struct rb_node *n = ic->journal_tree_root.rb_node;
1345 unsigned int found = NOT_FOUND;
1346 *next_sector = (sector_t)-1;
1348 struct journal_node *j = container_of(n, struct journal_node, node);
1349 if (sector == j->sector) {
1350 found = j - ic->journal_tree;
1352 if (sector < j->sector) {
1353 *next_sector = j->sector;
1354 n = j->node.rb_left;
1356 n = j->node.rb_right;
1363 static bool test_journal_node(struct dm_integrity_c *ic, unsigned int pos, sector_t sector)
1365 struct journal_node *node, *next_node;
1366 struct rb_node *next;
1368 if (unlikely(pos >= ic->journal_entries))
1370 node = &ic->journal_tree[pos];
1371 if (unlikely(RB_EMPTY_NODE(&node->node)))
1373 if (unlikely(node->sector != sector))
1376 next = rb_next(&node->node);
1377 if (unlikely(!next))
1380 next_node = container_of(next, struct journal_node, node);
1381 return next_node->sector != sector;
1384 static bool find_newer_committed_node(struct dm_integrity_c *ic, struct journal_node *node)
1386 struct rb_node *next;
1387 struct journal_node *next_node;
1388 unsigned int next_section;
1390 BUG_ON(RB_EMPTY_NODE(&node->node));
1392 next = rb_next(&node->node);
1393 if (unlikely(!next))
1396 next_node = container_of(next, struct journal_node, node);
1398 if (next_node->sector != node->sector)
1401 next_section = (unsigned int)(next_node - ic->journal_tree) / ic->journal_section_entries;
1402 if (next_section >= ic->committed_section &&
1403 next_section < ic->committed_section + ic->n_committed_sections)
1405 if (next_section + ic->journal_sections < ic->committed_section + ic->n_committed_sections)
1415 static int dm_integrity_rw_tag(struct dm_integrity_c *ic, unsigned char *tag, sector_t *metadata_block,
1416 unsigned int *metadata_offset, unsigned int total_size, int op)
1418 #define MAY_BE_FILLER 1
1419 #define MAY_BE_HASH 2
1420 unsigned int hash_offset = 0;
1421 unsigned int may_be = MAY_BE_HASH | (ic->discard ? MAY_BE_FILLER : 0);
1424 unsigned char *data, *dp;
1425 struct dm_buffer *b;
1426 unsigned int to_copy;
1429 r = dm_integrity_failed(ic);
1433 data = dm_bufio_read(ic->bufio, *metadata_block, &b);
1435 return PTR_ERR(data);
1437 to_copy = min((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - *metadata_offset, total_size);
1438 dp = data + *metadata_offset;
1439 if (op == TAG_READ) {
1440 memcpy(tag, dp, to_copy);
1441 } else if (op == TAG_WRITE) {
1442 if (memcmp(dp, tag, to_copy)) {
1443 memcpy(dp, tag, to_copy);
1444 dm_bufio_mark_partial_buffer_dirty(b, *metadata_offset, *metadata_offset + to_copy);
1447 /* e.g.: op == TAG_CMP */
1449 if (likely(is_power_of_2(ic->tag_size))) {
1450 if (unlikely(memcmp(dp, tag, to_copy)))
1451 if (unlikely(!ic->discard) ||
1452 unlikely(memchr_inv(dp, DISCARD_FILLER, to_copy) != NULL)) {
1460 for (i = 0; i < to_copy; i++, ts--) {
1461 if (unlikely(dp[i] != tag[i]))
1462 may_be &= ~MAY_BE_HASH;
1463 if (likely(dp[i] != DISCARD_FILLER))
1464 may_be &= ~MAY_BE_FILLER;
1466 if (unlikely(hash_offset == ic->tag_size)) {
1467 if (unlikely(!may_be)) {
1468 dm_bufio_release(b);
1472 may_be = MAY_BE_HASH | (ic->discard ? MAY_BE_FILLER : 0);
1477 dm_bufio_release(b);
1480 *metadata_offset += to_copy;
1481 if (unlikely(*metadata_offset == 1U << SECTOR_SHIFT << ic->log2_buffer_sectors)) {
1482 (*metadata_block)++;
1483 *metadata_offset = 0;
1486 if (unlikely(!is_power_of_2(ic->tag_size))) {
1487 hash_offset = (hash_offset + to_copy) % ic->tag_size;
1490 total_size -= to_copy;
1491 } while (unlikely(total_size));
1494 #undef MAY_BE_FILLER
1498 struct flush_request {
1499 struct dm_io_request io_req;
1500 struct dm_io_region io_reg;
1501 struct dm_integrity_c *ic;
1502 struct completion comp;
1505 static void flush_notify(unsigned long error, void *fr_)
1507 struct flush_request *fr = fr_;
1508 if (unlikely(error != 0))
1509 dm_integrity_io_error(fr->ic, "flushing disk cache", -EIO);
1510 complete(&fr->comp);
1513 static void dm_integrity_flush_buffers(struct dm_integrity_c *ic, bool flush_data)
1517 struct flush_request fr;
1522 fr.io_req.bi_opf = REQ_OP_WRITE | REQ_PREFLUSH | REQ_SYNC,
1523 fr.io_req.mem.type = DM_IO_KMEM,
1524 fr.io_req.mem.ptr.addr = NULL,
1525 fr.io_req.notify.fn = flush_notify,
1526 fr.io_req.notify.context = &fr;
1527 fr.io_req.client = dm_bufio_get_dm_io_client(ic->bufio),
1528 fr.io_reg.bdev = ic->dev->bdev,
1529 fr.io_reg.sector = 0,
1530 fr.io_reg.count = 0,
1532 init_completion(&fr.comp);
1533 r = dm_io(&fr.io_req, 1, &fr.io_reg, NULL);
1537 r = dm_bufio_write_dirty_buffers(ic->bufio);
1539 dm_integrity_io_error(ic, "writing tags", r);
1542 wait_for_completion(&fr.comp);
1545 static void sleep_on_endio_wait(struct dm_integrity_c *ic)
1547 DECLARE_WAITQUEUE(wait, current);
1548 __add_wait_queue(&ic->endio_wait, &wait);
1549 __set_current_state(TASK_UNINTERRUPTIBLE);
1550 spin_unlock_irq(&ic->endio_wait.lock);
1552 spin_lock_irq(&ic->endio_wait.lock);
1553 __remove_wait_queue(&ic->endio_wait, &wait);
1556 static void autocommit_fn(struct timer_list *t)
1558 struct dm_integrity_c *ic = from_timer(ic, t, autocommit_timer);
1560 if (likely(!dm_integrity_failed(ic)))
1561 queue_work(ic->commit_wq, &ic->commit_work);
1564 static void schedule_autocommit(struct dm_integrity_c *ic)
1566 if (!timer_pending(&ic->autocommit_timer))
1567 mod_timer(&ic->autocommit_timer, jiffies + ic->autocommit_jiffies);
1570 static void submit_flush_bio(struct dm_integrity_c *ic, struct dm_integrity_io *dio)
1573 unsigned long flags;
1575 spin_lock_irqsave(&ic->endio_wait.lock, flags);
1576 bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1577 bio_list_add(&ic->flush_bio_list, bio);
1578 spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
1580 queue_work(ic->commit_wq, &ic->commit_work);
1583 static void do_endio(struct dm_integrity_c *ic, struct bio *bio)
1585 int r = dm_integrity_failed(ic);
1586 if (unlikely(r) && !bio->bi_status)
1587 bio->bi_status = errno_to_blk_status(r);
1588 if (unlikely(ic->synchronous_mode) && bio_op(bio) == REQ_OP_WRITE) {
1589 unsigned long flags;
1590 spin_lock_irqsave(&ic->endio_wait.lock, flags);
1591 bio_list_add(&ic->synchronous_bios, bio);
1592 queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, 0);
1593 spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
1599 static void do_endio_flush(struct dm_integrity_c *ic, struct dm_integrity_io *dio)
1601 struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1603 if (unlikely(dio->fua) && likely(!bio->bi_status) && likely(!dm_integrity_failed(ic)))
1604 submit_flush_bio(ic, dio);
1609 static void dec_in_flight(struct dm_integrity_io *dio)
1611 if (atomic_dec_and_test(&dio->in_flight)) {
1612 struct dm_integrity_c *ic = dio->ic;
1615 remove_range(ic, &dio->range);
1617 if (dio->op == REQ_OP_WRITE || unlikely(dio->op == REQ_OP_DISCARD))
1618 schedule_autocommit(ic);
1620 bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1622 if (unlikely(dio->bi_status) && !bio->bi_status)
1623 bio->bi_status = dio->bi_status;
1624 if (likely(!bio->bi_status) && unlikely(bio_sectors(bio) != dio->range.n_sectors)) {
1625 dio->range.logical_sector += dio->range.n_sectors;
1626 bio_advance(bio, dio->range.n_sectors << SECTOR_SHIFT);
1627 INIT_WORK(&dio->work, integrity_bio_wait);
1628 queue_work(ic->offload_wq, &dio->work);
1631 do_endio_flush(ic, dio);
1635 static void integrity_end_io(struct bio *bio)
1637 struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
1639 dm_bio_restore(&dio->bio_details, bio);
1640 if (bio->bi_integrity)
1641 bio->bi_opf |= REQ_INTEGRITY;
1643 if (dio->completion)
1644 complete(dio->completion);
1649 static void integrity_sector_checksum(struct dm_integrity_c *ic, sector_t sector,
1650 const char *data, char *result)
1652 __le64 sector_le = cpu_to_le64(sector);
1653 SHASH_DESC_ON_STACK(req, ic->internal_hash);
1655 unsigned int digest_size;
1657 req->tfm = ic->internal_hash;
1659 r = crypto_shash_init(req);
1660 if (unlikely(r < 0)) {
1661 dm_integrity_io_error(ic, "crypto_shash_init", r);
1665 if (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) {
1666 r = crypto_shash_update(req, (__u8 *)&ic->sb->salt, SALT_SIZE);
1667 if (unlikely(r < 0)) {
1668 dm_integrity_io_error(ic, "crypto_shash_update", r);
1673 r = crypto_shash_update(req, (const __u8 *)§or_le, sizeof sector_le);
1674 if (unlikely(r < 0)) {
1675 dm_integrity_io_error(ic, "crypto_shash_update", r);
1679 r = crypto_shash_update(req, data, ic->sectors_per_block << SECTOR_SHIFT);
1680 if (unlikely(r < 0)) {
1681 dm_integrity_io_error(ic, "crypto_shash_update", r);
1685 r = crypto_shash_final(req, result);
1686 if (unlikely(r < 0)) {
1687 dm_integrity_io_error(ic, "crypto_shash_final", r);
1691 digest_size = crypto_shash_digestsize(ic->internal_hash);
1692 if (unlikely(digest_size < ic->tag_size))
1693 memset(result + digest_size, 0, ic->tag_size - digest_size);
1698 /* this shouldn't happen anyway, the hash functions have no reason to fail */
1699 get_random_bytes(result, ic->tag_size);
1702 static void integrity_metadata(struct work_struct *w)
1704 struct dm_integrity_io *dio = container_of(w, struct dm_integrity_io, work);
1705 struct dm_integrity_c *ic = dio->ic;
1709 if (ic->internal_hash) {
1710 struct bvec_iter iter;
1712 unsigned int digest_size = crypto_shash_digestsize(ic->internal_hash);
1713 struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1715 unsigned int extra_space = unlikely(digest_size > ic->tag_size) ? digest_size - ic->tag_size : 0;
1716 char checksums_onstack[max((size_t)HASH_MAX_DIGESTSIZE, MAX_TAG_SIZE)];
1718 unsigned int sectors_to_process;
1720 if (unlikely(ic->mode == 'R'))
1723 if (likely(dio->op != REQ_OP_DISCARD))
1724 checksums = kmalloc((PAGE_SIZE >> SECTOR_SHIFT >> ic->sb->log2_sectors_per_block) * ic->tag_size + extra_space,
1725 GFP_NOIO | __GFP_NORETRY | __GFP_NOWARN);
1727 checksums = kmalloc(PAGE_SIZE, GFP_NOIO | __GFP_NORETRY | __GFP_NOWARN);
1729 checksums = checksums_onstack;
1730 if (WARN_ON(extra_space &&
1731 digest_size > sizeof(checksums_onstack))) {
1737 if (unlikely(dio->op == REQ_OP_DISCARD)) {
1738 unsigned int bi_size = dio->bio_details.bi_iter.bi_size;
1739 unsigned int max_size = likely(checksums != checksums_onstack) ? PAGE_SIZE : HASH_MAX_DIGESTSIZE;
1740 unsigned int max_blocks = max_size / ic->tag_size;
1741 memset(checksums, DISCARD_FILLER, max_size);
1744 unsigned int this_step_blocks = bi_size >> (SECTOR_SHIFT + ic->sb->log2_sectors_per_block);
1745 this_step_blocks = min(this_step_blocks, max_blocks);
1746 r = dm_integrity_rw_tag(ic, checksums, &dio->metadata_block, &dio->metadata_offset,
1747 this_step_blocks * ic->tag_size, TAG_WRITE);
1749 if (likely(checksums != checksums_onstack))
1754 bi_size -= this_step_blocks << (SECTOR_SHIFT + ic->sb->log2_sectors_per_block);
1757 if (likely(checksums != checksums_onstack))
1762 sector = dio->range.logical_sector;
1763 sectors_to_process = dio->range.n_sectors;
1765 __bio_for_each_segment(bv, bio, iter, dio->bio_details.bi_iter) {
1767 char *mem, *checksums_ptr;
1770 mem = bvec_kmap_local(&bv);
1772 checksums_ptr = checksums;
1774 integrity_sector_checksum(ic, sector, mem + pos, checksums_ptr);
1775 checksums_ptr += ic->tag_size;
1776 sectors_to_process -= ic->sectors_per_block;
1777 pos += ic->sectors_per_block << SECTOR_SHIFT;
1778 sector += ic->sectors_per_block;
1779 } while (pos < bv.bv_len && sectors_to_process && checksums != checksums_onstack);
1782 r = dm_integrity_rw_tag(ic, checksums, &dio->metadata_block, &dio->metadata_offset,
1783 checksums_ptr - checksums, dio->op == REQ_OP_READ ? TAG_CMP : TAG_WRITE);
1788 s = sector - ((r + ic->tag_size - 1) / ic->tag_size);
1789 DMERR_LIMIT("%pg: Checksum failed at sector 0x%llx",
1792 atomic64_inc(&ic->number_of_mismatches);
1793 dm_audit_log_bio(DM_MSG_PREFIX, "integrity-checksum",
1796 if (likely(checksums != checksums_onstack))
1801 if (!sectors_to_process)
1804 if (unlikely(pos < bv.bv_len)) {
1805 bv.bv_offset += pos;
1811 if (likely(checksums != checksums_onstack))
1814 struct bio_integrity_payload *bip = dio->bio_details.bi_integrity;
1818 struct bvec_iter iter;
1819 unsigned int data_to_process = dio->range.n_sectors;
1820 sector_to_block(ic, data_to_process);
1821 data_to_process *= ic->tag_size;
1823 bip_for_each_vec(biv, bip, iter) {
1825 unsigned int this_len;
1827 BUG_ON(PageHighMem(biv.bv_page));
1828 tag = bvec_virt(&biv);
1829 this_len = min(biv.bv_len, data_to_process);
1830 r = dm_integrity_rw_tag(ic, tag, &dio->metadata_block, &dio->metadata_offset,
1831 this_len, dio->op == REQ_OP_READ ? TAG_READ : TAG_WRITE);
1834 data_to_process -= this_len;
1835 if (!data_to_process)
1844 dio->bi_status = errno_to_blk_status(r);
1848 static int dm_integrity_map(struct dm_target *ti, struct bio *bio)
1850 struct dm_integrity_c *ic = ti->private;
1851 struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
1852 struct bio_integrity_payload *bip;
1854 sector_t area, offset;
1858 dio->op = bio_op(bio);
1860 if (unlikely(dio->op == REQ_OP_DISCARD)) {
1861 if (ti->max_io_len) {
1862 sector_t sec = dm_target_offset(ti, bio->bi_iter.bi_sector);
1863 unsigned int log2_max_io_len = __fls(ti->max_io_len);
1864 sector_t start_boundary = sec >> log2_max_io_len;
1865 sector_t end_boundary = (sec + bio_sectors(bio) - 1) >> log2_max_io_len;
1866 if (start_boundary < end_boundary) {
1867 sector_t len = ti->max_io_len - (sec & (ti->max_io_len - 1));
1868 dm_accept_partial_bio(bio, len);
1873 if (unlikely(bio->bi_opf & REQ_PREFLUSH)) {
1874 submit_flush_bio(ic, dio);
1875 return DM_MAPIO_SUBMITTED;
1878 dio->range.logical_sector = dm_target_offset(ti, bio->bi_iter.bi_sector);
1879 dio->fua = dio->op == REQ_OP_WRITE && bio->bi_opf & REQ_FUA;
1880 if (unlikely(dio->fua)) {
1882 * Don't pass down the FUA flag because we have to flush
1883 * disk cache anyway.
1885 bio->bi_opf &= ~REQ_FUA;
1887 if (unlikely(dio->range.logical_sector + bio_sectors(bio) > ic->provided_data_sectors)) {
1888 DMERR("Too big sector number: 0x%llx + 0x%x > 0x%llx",
1889 dio->range.logical_sector, bio_sectors(bio),
1890 ic->provided_data_sectors);
1891 return DM_MAPIO_KILL;
1893 if (unlikely((dio->range.logical_sector | bio_sectors(bio)) & (unsigned int)(ic->sectors_per_block - 1))) {
1894 DMERR("Bio not aligned on %u sectors: 0x%llx, 0x%x",
1895 ic->sectors_per_block,
1896 dio->range.logical_sector, bio_sectors(bio));
1897 return DM_MAPIO_KILL;
1900 if (ic->sectors_per_block > 1 && likely(dio->op != REQ_OP_DISCARD)) {
1901 struct bvec_iter iter;
1903 bio_for_each_segment(bv, bio, iter) {
1904 if (unlikely(bv.bv_len & ((ic->sectors_per_block << SECTOR_SHIFT) - 1))) {
1905 DMERR("Bio vector (%u,%u) is not aligned on %u-sector boundary",
1906 bv.bv_offset, bv.bv_len, ic->sectors_per_block);
1907 return DM_MAPIO_KILL;
1912 bip = bio_integrity(bio);
1913 if (!ic->internal_hash) {
1915 unsigned int wanted_tag_size = bio_sectors(bio) >> ic->sb->log2_sectors_per_block;
1916 if (ic->log2_tag_size >= 0)
1917 wanted_tag_size <<= ic->log2_tag_size;
1919 wanted_tag_size *= ic->tag_size;
1920 if (unlikely(wanted_tag_size != bip->bip_iter.bi_size)) {
1921 DMERR("Invalid integrity data size %u, expected %u",
1922 bip->bip_iter.bi_size, wanted_tag_size);
1923 return DM_MAPIO_KILL;
1927 if (unlikely(bip != NULL)) {
1928 DMERR("Unexpected integrity data when using internal hash");
1929 return DM_MAPIO_KILL;
1933 if (unlikely(ic->mode == 'R') && unlikely(dio->op != REQ_OP_READ))
1934 return DM_MAPIO_KILL;
1936 get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
1937 dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset, &dio->metadata_offset);
1938 bio->bi_iter.bi_sector = get_data_sector(ic, area, offset);
1940 dm_integrity_map_continue(dio, true);
1941 return DM_MAPIO_SUBMITTED;
1944 static bool __journal_read_write(struct dm_integrity_io *dio, struct bio *bio,
1945 unsigned int journal_section, unsigned int journal_entry)
1947 struct dm_integrity_c *ic = dio->ic;
1948 sector_t logical_sector;
1949 unsigned int n_sectors;
1951 logical_sector = dio->range.logical_sector;
1952 n_sectors = dio->range.n_sectors;
1954 struct bio_vec bv = bio_iovec(bio);
1957 if (unlikely(bv.bv_len >> SECTOR_SHIFT > n_sectors))
1958 bv.bv_len = n_sectors << SECTOR_SHIFT;
1959 n_sectors -= bv.bv_len >> SECTOR_SHIFT;
1960 bio_advance_iter(bio, &bio->bi_iter, bv.bv_len);
1962 mem = kmap_local_page(bv.bv_page);
1963 if (likely(dio->op == REQ_OP_WRITE))
1964 flush_dcache_page(bv.bv_page);
1967 struct journal_entry *je = access_journal_entry(ic, journal_section, journal_entry);
1969 if (unlikely(dio->op == REQ_OP_READ)) {
1970 struct journal_sector *js;
1974 if (unlikely(journal_entry_is_inprogress(je))) {
1975 flush_dcache_page(bv.bv_page);
1978 __io_wait_event(ic->copy_to_journal_wait, !journal_entry_is_inprogress(je));
1982 BUG_ON(journal_entry_get_sector(je) != logical_sector);
1983 js = access_journal_data(ic, journal_section, journal_entry);
1984 mem_ptr = mem + bv.bv_offset;
1987 memcpy(mem_ptr, js, JOURNAL_SECTOR_DATA);
1988 *(commit_id_t *)(mem_ptr + JOURNAL_SECTOR_DATA) = je->last_bytes[s];
1990 mem_ptr += 1 << SECTOR_SHIFT;
1991 } while (++s < ic->sectors_per_block);
1992 #ifdef INTERNAL_VERIFY
1993 if (ic->internal_hash) {
1994 char checksums_onstack[max((size_t)HASH_MAX_DIGESTSIZE, MAX_TAG_SIZE)];
1996 integrity_sector_checksum(ic, logical_sector, mem + bv.bv_offset, checksums_onstack);
1997 if (unlikely(memcmp(checksums_onstack, journal_entry_tag(ic, je), ic->tag_size))) {
1998 DMERR_LIMIT("Checksum failed when reading from journal, at sector 0x%llx",
2000 dm_audit_log_bio(DM_MSG_PREFIX, "journal-checksum",
2001 bio, logical_sector, 0);
2007 if (!ic->internal_hash) {
2008 struct bio_integrity_payload *bip = bio_integrity(bio);
2009 unsigned int tag_todo = ic->tag_size;
2010 char *tag_ptr = journal_entry_tag(ic, je);
2013 struct bio_vec biv = bvec_iter_bvec(bip->bip_vec, bip->bip_iter);
2014 unsigned int tag_now = min(biv.bv_len, tag_todo);
2016 BUG_ON(PageHighMem(biv.bv_page));
2017 tag_addr = bvec_virt(&biv);
2018 if (likely(dio->op == REQ_OP_WRITE))
2019 memcpy(tag_ptr, tag_addr, tag_now);
2021 memcpy(tag_addr, tag_ptr, tag_now);
2022 bvec_iter_advance(bip->bip_vec, &bip->bip_iter, tag_now);
2024 tag_todo -= tag_now;
2025 } while (unlikely(tag_todo)); else {
2026 if (likely(dio->op == REQ_OP_WRITE))
2027 memset(tag_ptr, 0, tag_todo);
2031 if (likely(dio->op == REQ_OP_WRITE)) {
2032 struct journal_sector *js;
2035 js = access_journal_data(ic, journal_section, journal_entry);
2036 memcpy(js, mem + bv.bv_offset, ic->sectors_per_block << SECTOR_SHIFT);
2040 je->last_bytes[s] = js[s].commit_id;
2041 } while (++s < ic->sectors_per_block);
2043 if (ic->internal_hash) {
2044 unsigned int digest_size = crypto_shash_digestsize(ic->internal_hash);
2045 if (unlikely(digest_size > ic->tag_size)) {
2046 char checksums_onstack[HASH_MAX_DIGESTSIZE];
2047 integrity_sector_checksum(ic, logical_sector, (char *)js, checksums_onstack);
2048 memcpy(journal_entry_tag(ic, je), checksums_onstack, ic->tag_size);
2050 integrity_sector_checksum(ic, logical_sector, (char *)js, journal_entry_tag(ic, je));
2053 journal_entry_set_sector(je, logical_sector);
2055 logical_sector += ic->sectors_per_block;
2058 if (unlikely(journal_entry == ic->journal_section_entries)) {
2061 wraparound_section(ic, &journal_section);
2064 bv.bv_offset += ic->sectors_per_block << SECTOR_SHIFT;
2065 } while (bv.bv_len -= ic->sectors_per_block << SECTOR_SHIFT);
2067 if (unlikely(dio->op == REQ_OP_READ))
2068 flush_dcache_page(bv.bv_page);
2070 } while (n_sectors);
2072 if (likely(dio->op == REQ_OP_WRITE)) {
2074 if (unlikely(waitqueue_active(&ic->copy_to_journal_wait)))
2075 wake_up(&ic->copy_to_journal_wait);
2076 if (READ_ONCE(ic->free_sectors) <= ic->free_sectors_threshold) {
2077 queue_work(ic->commit_wq, &ic->commit_work);
2079 schedule_autocommit(ic);
2082 remove_range(ic, &dio->range);
2085 if (unlikely(bio->bi_iter.bi_size)) {
2086 sector_t area, offset;
2088 dio->range.logical_sector = logical_sector;
2089 get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
2090 dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset, &dio->metadata_offset);
2097 static void dm_integrity_map_continue(struct dm_integrity_io *dio, bool from_map)
2099 struct dm_integrity_c *ic = dio->ic;
2100 struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
2101 unsigned int journal_section, journal_entry;
2102 unsigned int journal_read_pos;
2103 struct completion read_comp;
2104 bool discard_retried = false;
2105 bool need_sync_io = ic->internal_hash && dio->op == REQ_OP_READ;
2106 if (unlikely(dio->op == REQ_OP_DISCARD) && ic->mode != 'D')
2107 need_sync_io = true;
2109 if (need_sync_io && from_map) {
2110 INIT_WORK(&dio->work, integrity_bio_wait);
2111 queue_work(ic->offload_wq, &dio->work);
2116 spin_lock_irq(&ic->endio_wait.lock);
2118 if (unlikely(dm_integrity_failed(ic))) {
2119 spin_unlock_irq(&ic->endio_wait.lock);
2123 dio->range.n_sectors = bio_sectors(bio);
2124 journal_read_pos = NOT_FOUND;
2125 if (ic->mode == 'J' && likely(dio->op != REQ_OP_DISCARD)) {
2126 if (dio->op == REQ_OP_WRITE) {
2127 unsigned int next_entry, i, pos;
2128 unsigned int ws, we, range_sectors;
2130 dio->range.n_sectors = min(dio->range.n_sectors,
2131 (sector_t)ic->free_sectors << ic->sb->log2_sectors_per_block);
2132 if (unlikely(!dio->range.n_sectors)) {
2134 goto offload_to_thread;
2135 sleep_on_endio_wait(ic);
2138 range_sectors = dio->range.n_sectors >> ic->sb->log2_sectors_per_block;
2139 ic->free_sectors -= range_sectors;
2140 journal_section = ic->free_section;
2141 journal_entry = ic->free_section_entry;
2143 next_entry = ic->free_section_entry + range_sectors;
2144 ic->free_section_entry = next_entry % ic->journal_section_entries;
2145 ic->free_section += next_entry / ic->journal_section_entries;
2146 ic->n_uncommitted_sections += next_entry / ic->journal_section_entries;
2147 wraparound_section(ic, &ic->free_section);
2149 pos = journal_section * ic->journal_section_entries + journal_entry;
2150 ws = journal_section;
2154 struct journal_entry *je;
2156 add_journal_node(ic, &ic->journal_tree[pos], dio->range.logical_sector + i);
2158 if (unlikely(pos >= ic->journal_entries))
2161 je = access_journal_entry(ic, ws, we);
2162 BUG_ON(!journal_entry_is_unused(je));
2163 journal_entry_set_inprogress(je);
2165 if (unlikely(we == ic->journal_section_entries)) {
2168 wraparound_section(ic, &ws);
2170 } while ((i += ic->sectors_per_block) < dio->range.n_sectors);
2172 spin_unlock_irq(&ic->endio_wait.lock);
2173 goto journal_read_write;
2175 sector_t next_sector;
2176 journal_read_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
2177 if (likely(journal_read_pos == NOT_FOUND)) {
2178 if (unlikely(dio->range.n_sectors > next_sector - dio->range.logical_sector))
2179 dio->range.n_sectors = next_sector - dio->range.logical_sector;
2182 unsigned int jp = journal_read_pos + 1;
2183 for (i = ic->sectors_per_block; i < dio->range.n_sectors; i += ic->sectors_per_block, jp++) {
2184 if (!test_journal_node(ic, jp, dio->range.logical_sector + i))
2187 dio->range.n_sectors = i;
2191 if (unlikely(!add_new_range(ic, &dio->range, true))) {
2193 * We must not sleep in the request routine because it could
2194 * stall bios on current->bio_list.
2195 * So, we offload the bio to a workqueue if we have to sleep.
2199 spin_unlock_irq(&ic->endio_wait.lock);
2200 INIT_WORK(&dio->work, integrity_bio_wait);
2201 queue_work(ic->wait_wq, &dio->work);
2204 if (journal_read_pos != NOT_FOUND)
2205 dio->range.n_sectors = ic->sectors_per_block;
2206 wait_and_add_new_range(ic, &dio->range);
2208 * wait_and_add_new_range drops the spinlock, so the journal
2209 * may have been changed arbitrarily. We need to recheck.
2210 * To simplify the code, we restrict I/O size to just one block.
2212 if (journal_read_pos != NOT_FOUND) {
2213 sector_t next_sector;
2214 unsigned int new_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
2215 if (unlikely(new_pos != journal_read_pos)) {
2216 remove_range_unlocked(ic, &dio->range);
2221 if (ic->mode == 'J' && likely(dio->op == REQ_OP_DISCARD) && !discard_retried) {
2222 sector_t next_sector;
2223 unsigned int new_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
2224 if (unlikely(new_pos != NOT_FOUND) ||
2225 unlikely(next_sector < dio->range.logical_sector - dio->range.n_sectors)) {
2226 remove_range_unlocked(ic, &dio->range);
2227 spin_unlock_irq(&ic->endio_wait.lock);
2228 queue_work(ic->commit_wq, &ic->commit_work);
2229 flush_workqueue(ic->commit_wq);
2230 queue_work(ic->writer_wq, &ic->writer_work);
2231 flush_workqueue(ic->writer_wq);
2232 discard_retried = true;
2236 spin_unlock_irq(&ic->endio_wait.lock);
2238 if (unlikely(journal_read_pos != NOT_FOUND)) {
2239 journal_section = journal_read_pos / ic->journal_section_entries;
2240 journal_entry = journal_read_pos % ic->journal_section_entries;
2241 goto journal_read_write;
2244 if (ic->mode == 'B' && (dio->op == REQ_OP_WRITE || unlikely(dio->op == REQ_OP_DISCARD))) {
2245 if (!block_bitmap_op(ic, ic->may_write_bitmap, dio->range.logical_sector,
2246 dio->range.n_sectors, BITMAP_OP_TEST_ALL_SET)) {
2247 struct bitmap_block_status *bbs;
2249 bbs = sector_to_bitmap_block(ic, dio->range.logical_sector);
2250 spin_lock(&bbs->bio_queue_lock);
2251 bio_list_add(&bbs->bio_queue, bio);
2252 spin_unlock(&bbs->bio_queue_lock);
2253 queue_work(ic->writer_wq, &bbs->work);
2258 dio->in_flight = (atomic_t)ATOMIC_INIT(2);
2261 init_completion(&read_comp);
2262 dio->completion = &read_comp;
2264 dio->completion = NULL;
2266 dm_bio_record(&dio->bio_details, bio);
2267 bio_set_dev(bio, ic->dev->bdev);
2268 bio->bi_integrity = NULL;
2269 bio->bi_opf &= ~REQ_INTEGRITY;
2270 bio->bi_end_io = integrity_end_io;
2271 bio->bi_iter.bi_size = dio->range.n_sectors << SECTOR_SHIFT;
2273 if (unlikely(dio->op == REQ_OP_DISCARD) && likely(ic->mode != 'D')) {
2274 integrity_metadata(&dio->work);
2275 dm_integrity_flush_buffers(ic, false);
2277 dio->in_flight = (atomic_t)ATOMIC_INIT(1);
2278 dio->completion = NULL;
2280 submit_bio_noacct(bio);
2285 submit_bio_noacct(bio);
2288 wait_for_completion_io(&read_comp);
2289 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING) &&
2290 dio->range.logical_sector + dio->range.n_sectors > le64_to_cpu(ic->sb->recalc_sector))
2292 if (ic->mode == 'B') {
2293 if (!block_bitmap_op(ic, ic->recalc_bitmap, dio->range.logical_sector,
2294 dio->range.n_sectors, BITMAP_OP_TEST_ALL_CLEAR))
2298 if (likely(!bio->bi_status))
2299 integrity_metadata(&dio->work);
2305 INIT_WORK(&dio->work, integrity_metadata);
2306 queue_work(ic->metadata_wq, &dio->work);
2312 if (unlikely(__journal_read_write(dio, bio, journal_section, journal_entry)))
2315 do_endio_flush(ic, dio);
2319 static void integrity_bio_wait(struct work_struct *w)
2321 struct dm_integrity_io *dio = container_of(w, struct dm_integrity_io, work);
2323 dm_integrity_map_continue(dio, false);
2326 static void pad_uncommitted(struct dm_integrity_c *ic)
2328 if (ic->free_section_entry) {
2329 ic->free_sectors -= ic->journal_section_entries - ic->free_section_entry;
2330 ic->free_section_entry = 0;
2332 wraparound_section(ic, &ic->free_section);
2333 ic->n_uncommitted_sections++;
2335 if (WARN_ON(ic->journal_sections * ic->journal_section_entries !=
2336 (ic->n_uncommitted_sections + ic->n_committed_sections) *
2337 ic->journal_section_entries + ic->free_sectors)) {
2338 DMCRIT("journal_sections %u, journal_section_entries %u, "
2339 "n_uncommitted_sections %u, n_committed_sections %u, "
2340 "journal_section_entries %u, free_sectors %u",
2341 ic->journal_sections, ic->journal_section_entries,
2342 ic->n_uncommitted_sections, ic->n_committed_sections,
2343 ic->journal_section_entries, ic->free_sectors);
2347 static void integrity_commit(struct work_struct *w)
2349 struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, commit_work);
2350 unsigned int commit_start, commit_sections;
2351 unsigned int i, j, n;
2352 struct bio *flushes;
2354 del_timer(&ic->autocommit_timer);
2356 spin_lock_irq(&ic->endio_wait.lock);
2357 flushes = bio_list_get(&ic->flush_bio_list);
2358 if (unlikely(ic->mode != 'J')) {
2359 spin_unlock_irq(&ic->endio_wait.lock);
2360 dm_integrity_flush_buffers(ic, true);
2361 goto release_flush_bios;
2364 pad_uncommitted(ic);
2365 commit_start = ic->uncommitted_section;
2366 commit_sections = ic->n_uncommitted_sections;
2367 spin_unlock_irq(&ic->endio_wait.lock);
2369 if (!commit_sections)
2370 goto release_flush_bios;
2372 ic->wrote_to_journal = true;
2375 for (n = 0; n < commit_sections; n++) {
2376 for (j = 0; j < ic->journal_section_entries; j++) {
2377 struct journal_entry *je;
2378 je = access_journal_entry(ic, i, j);
2379 io_wait_event(ic->copy_to_journal_wait, !journal_entry_is_inprogress(je));
2381 for (j = 0; j < ic->journal_section_sectors; j++) {
2382 struct journal_sector *js;
2383 js = access_journal(ic, i, j);
2384 js->commit_id = dm_integrity_commit_id(ic, i, j, ic->commit_seq);
2387 if (unlikely(i >= ic->journal_sections))
2388 ic->commit_seq = next_commit_seq(ic->commit_seq);
2389 wraparound_section(ic, &i);
2393 write_journal(ic, commit_start, commit_sections);
2395 spin_lock_irq(&ic->endio_wait.lock);
2396 ic->uncommitted_section += commit_sections;
2397 wraparound_section(ic, &ic->uncommitted_section);
2398 ic->n_uncommitted_sections -= commit_sections;
2399 ic->n_committed_sections += commit_sections;
2400 spin_unlock_irq(&ic->endio_wait.lock);
2402 if (READ_ONCE(ic->free_sectors) <= ic->free_sectors_threshold)
2403 queue_work(ic->writer_wq, &ic->writer_work);
2407 struct bio *next = flushes->bi_next;
2408 flushes->bi_next = NULL;
2409 do_endio(ic, flushes);
2414 static void complete_copy_from_journal(unsigned long error, void *context)
2416 struct journal_io *io = context;
2417 struct journal_completion *comp = io->comp;
2418 struct dm_integrity_c *ic = comp->ic;
2419 remove_range(ic, &io->range);
2420 mempool_free(io, &ic->journal_io_mempool);
2421 if (unlikely(error != 0))
2422 dm_integrity_io_error(ic, "copying from journal", -EIO);
2423 complete_journal_op(comp);
2426 static void restore_last_bytes(struct dm_integrity_c *ic, struct journal_sector *js,
2427 struct journal_entry *je)
2431 js->commit_id = je->last_bytes[s];
2433 } while (++s < ic->sectors_per_block);
2436 static void do_journal_write(struct dm_integrity_c *ic, unsigned int write_start,
2437 unsigned int write_sections, bool from_replay)
2439 unsigned int i, j, n;
2440 struct journal_completion comp;
2441 struct blk_plug plug;
2443 blk_start_plug(&plug);
2446 comp.in_flight = (atomic_t)ATOMIC_INIT(1);
2447 init_completion(&comp.comp);
2450 for (n = 0; n < write_sections; n++, i++, wraparound_section(ic, &i)) {
2451 #ifndef INTERNAL_VERIFY
2452 if (unlikely(from_replay))
2454 rw_section_mac(ic, i, false);
2455 for (j = 0; j < ic->journal_section_entries; j++) {
2456 struct journal_entry *je = access_journal_entry(ic, i, j);
2457 sector_t sec, area, offset;
2458 unsigned int k, l, next_loop;
2459 sector_t metadata_block;
2460 unsigned int metadata_offset;
2461 struct journal_io *io;
2463 if (journal_entry_is_unused(je))
2465 BUG_ON(unlikely(journal_entry_is_inprogress(je)) && !from_replay);
2466 sec = journal_entry_get_sector(je);
2467 if (unlikely(from_replay)) {
2468 if (unlikely(sec & (unsigned int)(ic->sectors_per_block - 1))) {
2469 dm_integrity_io_error(ic, "invalid sector in journal", -EIO);
2470 sec &= ~(sector_t)(ic->sectors_per_block - 1);
2472 if (unlikely(sec >= ic->provided_data_sectors)) {
2473 journal_entry_set_unused(je);
2477 get_area_and_offset(ic, sec, &area, &offset);
2478 restore_last_bytes(ic, access_journal_data(ic, i, j), je);
2479 for (k = j + 1; k < ic->journal_section_entries; k++) {
2480 struct journal_entry *je2 = access_journal_entry(ic, i, k);
2481 sector_t sec2, area2, offset2;
2482 if (journal_entry_is_unused(je2))
2484 BUG_ON(unlikely(journal_entry_is_inprogress(je2)) && !from_replay);
2485 sec2 = journal_entry_get_sector(je2);
2486 if (unlikely(sec2 >= ic->provided_data_sectors))
2488 get_area_and_offset(ic, sec2, &area2, &offset2);
2489 if (area2 != area || offset2 != offset + ((k - j) << ic->sb->log2_sectors_per_block))
2491 restore_last_bytes(ic, access_journal_data(ic, i, k), je2);
2495 io = mempool_alloc(&ic->journal_io_mempool, GFP_NOIO);
2497 io->range.logical_sector = sec;
2498 io->range.n_sectors = (k - j) << ic->sb->log2_sectors_per_block;
2500 spin_lock_irq(&ic->endio_wait.lock);
2501 add_new_range_and_wait(ic, &io->range);
2503 if (likely(!from_replay)) {
2504 struct journal_node *section_node = &ic->journal_tree[i * ic->journal_section_entries];
2506 /* don't write if there is newer committed sector */
2507 while (j < k && find_newer_committed_node(ic, §ion_node[j])) {
2508 struct journal_entry *je2 = access_journal_entry(ic, i, j);
2510 journal_entry_set_unused(je2);
2511 remove_journal_node(ic, §ion_node[j]);
2513 sec += ic->sectors_per_block;
2514 offset += ic->sectors_per_block;
2516 while (j < k && find_newer_committed_node(ic, §ion_node[k - 1])) {
2517 struct journal_entry *je2 = access_journal_entry(ic, i, k - 1);
2519 journal_entry_set_unused(je2);
2520 remove_journal_node(ic, §ion_node[k - 1]);
2524 remove_range_unlocked(ic, &io->range);
2525 spin_unlock_irq(&ic->endio_wait.lock);
2526 mempool_free(io, &ic->journal_io_mempool);
2529 for (l = j; l < k; l++) {
2530 remove_journal_node(ic, §ion_node[l]);
2533 spin_unlock_irq(&ic->endio_wait.lock);
2535 metadata_block = get_metadata_sector_and_offset(ic, area, offset, &metadata_offset);
2536 for (l = j; l < k; l++) {
2538 struct journal_entry *je2 = access_journal_entry(ic, i, l);
2541 #ifndef INTERNAL_VERIFY
2542 unlikely(from_replay) &&
2544 ic->internal_hash) {
2545 char test_tag[max_t(size_t, HASH_MAX_DIGESTSIZE, MAX_TAG_SIZE)];
2547 integrity_sector_checksum(ic, sec + ((l - j) << ic->sb->log2_sectors_per_block),
2548 (char *)access_journal_data(ic, i, l), test_tag);
2549 if (unlikely(memcmp(test_tag, journal_entry_tag(ic, je2), ic->tag_size))) {
2550 dm_integrity_io_error(ic, "tag mismatch when replaying journal", -EILSEQ);
2551 dm_audit_log_target(DM_MSG_PREFIX, "integrity-replay-journal", ic->ti, 0);
2555 journal_entry_set_unused(je2);
2556 r = dm_integrity_rw_tag(ic, journal_entry_tag(ic, je2), &metadata_block, &metadata_offset,
2557 ic->tag_size, TAG_WRITE);
2559 dm_integrity_io_error(ic, "reading tags", r);
2563 atomic_inc(&comp.in_flight);
2564 copy_from_journal(ic, i, j << ic->sb->log2_sectors_per_block,
2565 (k - j) << ic->sb->log2_sectors_per_block,
2566 get_data_sector(ic, area, offset),
2567 complete_copy_from_journal, io);
2573 dm_bufio_write_dirty_buffers_async(ic->bufio);
2575 blk_finish_plug(&plug);
2577 complete_journal_op(&comp);
2578 wait_for_completion_io(&comp.comp);
2580 dm_integrity_flush_buffers(ic, true);
2583 static void integrity_writer(struct work_struct *w)
2585 struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, writer_work);
2586 unsigned int write_start, write_sections;
2588 unsigned int prev_free_sectors;
2590 spin_lock_irq(&ic->endio_wait.lock);
2591 write_start = ic->committed_section;
2592 write_sections = ic->n_committed_sections;
2593 spin_unlock_irq(&ic->endio_wait.lock);
2595 if (!write_sections)
2598 do_journal_write(ic, write_start, write_sections, false);
2600 spin_lock_irq(&ic->endio_wait.lock);
2602 ic->committed_section += write_sections;
2603 wraparound_section(ic, &ic->committed_section);
2604 ic->n_committed_sections -= write_sections;
2606 prev_free_sectors = ic->free_sectors;
2607 ic->free_sectors += write_sections * ic->journal_section_entries;
2608 if (unlikely(!prev_free_sectors))
2609 wake_up_locked(&ic->endio_wait);
2611 spin_unlock_irq(&ic->endio_wait.lock);
2614 static void recalc_write_super(struct dm_integrity_c *ic)
2618 dm_integrity_flush_buffers(ic, false);
2619 if (dm_integrity_failed(ic))
2622 r = sync_rw_sb(ic, REQ_OP_WRITE);
2624 dm_integrity_io_error(ic, "writing superblock", r);
2627 static void integrity_recalc(struct work_struct *w)
2629 struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, recalc_work);
2630 struct dm_integrity_range range;
2631 struct dm_io_request io_req;
2632 struct dm_io_region io_loc;
2633 sector_t area, offset;
2634 sector_t metadata_block;
2635 unsigned int metadata_offset;
2636 sector_t logical_sector, n_sectors;
2640 unsigned int super_counter = 0;
2642 DEBUG_print("start recalculation... (position %llx)\n", le64_to_cpu(ic->sb->recalc_sector));
2644 spin_lock_irq(&ic->endio_wait.lock);
2648 if (unlikely(dm_post_suspending(ic->ti)))
2651 range.logical_sector = le64_to_cpu(ic->sb->recalc_sector);
2652 if (unlikely(range.logical_sector >= ic->provided_data_sectors)) {
2653 if (ic->mode == 'B') {
2654 block_bitmap_op(ic, ic->recalc_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_CLEAR);
2655 DEBUG_print("queue_delayed_work: bitmap_flush_work\n");
2656 queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, 0);
2661 get_area_and_offset(ic, range.logical_sector, &area, &offset);
2662 range.n_sectors = min((sector_t)RECALC_SECTORS, ic->provided_data_sectors - range.logical_sector);
2664 range.n_sectors = min(range.n_sectors, ((sector_t)1U << ic->sb->log2_interleave_sectors) - (unsigned int)offset);
2666 add_new_range_and_wait(ic, &range);
2667 spin_unlock_irq(&ic->endio_wait.lock);
2668 logical_sector = range.logical_sector;
2669 n_sectors = range.n_sectors;
2671 if (ic->mode == 'B') {
2672 if (block_bitmap_op(ic, ic->recalc_bitmap, logical_sector, n_sectors, BITMAP_OP_TEST_ALL_CLEAR)) {
2673 goto advance_and_next;
2675 while (block_bitmap_op(ic, ic->recalc_bitmap, logical_sector,
2676 ic->sectors_per_block, BITMAP_OP_TEST_ALL_CLEAR)) {
2677 logical_sector += ic->sectors_per_block;
2678 n_sectors -= ic->sectors_per_block;
2681 while (block_bitmap_op(ic, ic->recalc_bitmap, logical_sector + n_sectors - ic->sectors_per_block,
2682 ic->sectors_per_block, BITMAP_OP_TEST_ALL_CLEAR)) {
2683 n_sectors -= ic->sectors_per_block;
2686 get_area_and_offset(ic, logical_sector, &area, &offset);
2689 DEBUG_print("recalculating: %llx, %llx\n", logical_sector, n_sectors);
2691 if (unlikely(++super_counter == RECALC_WRITE_SUPER)) {
2692 recalc_write_super(ic);
2693 if (ic->mode == 'B') {
2694 queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, ic->bitmap_flush_interval);
2699 if (unlikely(dm_integrity_failed(ic)))
2702 io_req.bi_opf = REQ_OP_READ;
2703 io_req.mem.type = DM_IO_VMA;
2704 io_req.mem.ptr.addr = ic->recalc_buffer;
2705 io_req.notify.fn = NULL;
2706 io_req.client = ic->io;
2707 io_loc.bdev = ic->dev->bdev;
2708 io_loc.sector = get_data_sector(ic, area, offset);
2709 io_loc.count = n_sectors;
2711 r = dm_io(&io_req, 1, &io_loc, NULL);
2713 dm_integrity_io_error(ic, "reading data", r);
2717 t = ic->recalc_tags;
2718 for (i = 0; i < n_sectors; i += ic->sectors_per_block) {
2719 integrity_sector_checksum(ic, logical_sector + i, ic->recalc_buffer + (i << SECTOR_SHIFT), t);
2723 metadata_block = get_metadata_sector_and_offset(ic, area, offset, &metadata_offset);
2725 r = dm_integrity_rw_tag(ic, ic->recalc_tags, &metadata_block, &metadata_offset, t - ic->recalc_tags, TAG_WRITE);
2727 dm_integrity_io_error(ic, "writing tags", r);
2731 if (ic->mode == 'B') {
2732 sector_t start, end;
2733 start = (range.logical_sector >>
2734 (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit)) <<
2735 (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
2736 end = ((range.logical_sector + range.n_sectors) >>
2737 (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit)) <<
2738 (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
2739 block_bitmap_op(ic, ic->recalc_bitmap, start, end - start, BITMAP_OP_CLEAR);
2745 spin_lock_irq(&ic->endio_wait.lock);
2746 remove_range_unlocked(ic, &range);
2747 ic->sb->recalc_sector = cpu_to_le64(range.logical_sector + range.n_sectors);
2751 remove_range(ic, &range);
2755 spin_unlock_irq(&ic->endio_wait.lock);
2757 recalc_write_super(ic);
2760 static void bitmap_block_work(struct work_struct *w)
2762 struct bitmap_block_status *bbs = container_of(w, struct bitmap_block_status, work);
2763 struct dm_integrity_c *ic = bbs->ic;
2765 struct bio_list bio_queue;
2766 struct bio_list waiting;
2768 bio_list_init(&waiting);
2770 spin_lock(&bbs->bio_queue_lock);
2771 bio_queue = bbs->bio_queue;
2772 bio_list_init(&bbs->bio_queue);
2773 spin_unlock(&bbs->bio_queue_lock);
2775 while ((bio = bio_list_pop(&bio_queue))) {
2776 struct dm_integrity_io *dio;
2778 dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
2780 if (block_bitmap_op(ic, ic->may_write_bitmap, dio->range.logical_sector,
2781 dio->range.n_sectors, BITMAP_OP_TEST_ALL_SET)) {
2782 remove_range(ic, &dio->range);
2783 INIT_WORK(&dio->work, integrity_bio_wait);
2784 queue_work(ic->offload_wq, &dio->work);
2786 block_bitmap_op(ic, ic->journal, dio->range.logical_sector,
2787 dio->range.n_sectors, BITMAP_OP_SET);
2788 bio_list_add(&waiting, bio);
2792 if (bio_list_empty(&waiting))
2795 rw_journal_sectors(ic, REQ_OP_WRITE | REQ_FUA | REQ_SYNC,
2796 bbs->idx * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT),
2797 BITMAP_BLOCK_SIZE >> SECTOR_SHIFT, NULL);
2799 while ((bio = bio_list_pop(&waiting))) {
2800 struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
2802 block_bitmap_op(ic, ic->may_write_bitmap, dio->range.logical_sector,
2803 dio->range.n_sectors, BITMAP_OP_SET);
2805 remove_range(ic, &dio->range);
2806 INIT_WORK(&dio->work, integrity_bio_wait);
2807 queue_work(ic->offload_wq, &dio->work);
2810 queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, ic->bitmap_flush_interval);
2813 static void bitmap_flush_work(struct work_struct *work)
2815 struct dm_integrity_c *ic = container_of(work, struct dm_integrity_c, bitmap_flush_work.work);
2816 struct dm_integrity_range range;
2817 unsigned long limit;
2820 dm_integrity_flush_buffers(ic, false);
2822 range.logical_sector = 0;
2823 range.n_sectors = ic->provided_data_sectors;
2825 spin_lock_irq(&ic->endio_wait.lock);
2826 add_new_range_and_wait(ic, &range);
2827 spin_unlock_irq(&ic->endio_wait.lock);
2829 dm_integrity_flush_buffers(ic, true);
2831 limit = ic->provided_data_sectors;
2832 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
2833 limit = le64_to_cpu(ic->sb->recalc_sector)
2834 >> (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit)
2835 << (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
2837 /*DEBUG_print("zeroing journal\n");*/
2838 block_bitmap_op(ic, ic->journal, 0, limit, BITMAP_OP_CLEAR);
2839 block_bitmap_op(ic, ic->may_write_bitmap, 0, limit, BITMAP_OP_CLEAR);
2841 rw_journal_sectors(ic, REQ_OP_WRITE | REQ_FUA | REQ_SYNC, 0,
2842 ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
2844 spin_lock_irq(&ic->endio_wait.lock);
2845 remove_range_unlocked(ic, &range);
2846 while (unlikely((bio = bio_list_pop(&ic->synchronous_bios)) != NULL)) {
2848 spin_unlock_irq(&ic->endio_wait.lock);
2849 spin_lock_irq(&ic->endio_wait.lock);
2851 spin_unlock_irq(&ic->endio_wait.lock);
2855 static void init_journal(struct dm_integrity_c *ic, unsigned int start_section,
2856 unsigned int n_sections, unsigned char commit_seq)
2858 unsigned int i, j, n;
2863 for (n = 0; n < n_sections; n++) {
2864 i = start_section + n;
2865 wraparound_section(ic, &i);
2866 for (j = 0; j < ic->journal_section_sectors; j++) {
2867 struct journal_sector *js = access_journal(ic, i, j);
2868 BUILD_BUG_ON(sizeof(js->sectors) != JOURNAL_SECTOR_DATA);
2869 memset(&js->sectors, 0, sizeof(js->sectors));
2870 js->commit_id = dm_integrity_commit_id(ic, i, j, commit_seq);
2872 for (j = 0; j < ic->journal_section_entries; j++) {
2873 struct journal_entry *je = access_journal_entry(ic, i, j);
2874 journal_entry_set_unused(je);
2878 write_journal(ic, start_section, n_sections);
2881 static int find_commit_seq(struct dm_integrity_c *ic, unsigned int i, unsigned int j, commit_id_t id)
2884 for (k = 0; k < N_COMMIT_IDS; k++) {
2885 if (dm_integrity_commit_id(ic, i, j, k) == id)
2888 dm_integrity_io_error(ic, "journal commit id", -EIO);
2892 static void replay_journal(struct dm_integrity_c *ic)
2895 bool used_commit_ids[N_COMMIT_IDS];
2896 unsigned int max_commit_id_sections[N_COMMIT_IDS];
2897 unsigned int write_start, write_sections;
2898 unsigned int continue_section;
2900 unsigned char unused, last_used, want_commit_seq;
2902 if (ic->mode == 'R')
2905 if (ic->journal_uptodate)
2911 if (!ic->just_formatted) {
2912 DEBUG_print("reading journal\n");
2913 rw_journal(ic, REQ_OP_READ, 0, ic->journal_sections, NULL);
2915 DEBUG_bytes(lowmem_page_address(ic->journal_io[0].page), 64, "read journal");
2916 if (ic->journal_io) {
2917 struct journal_completion crypt_comp;
2919 init_completion(&crypt_comp.comp);
2920 crypt_comp.in_flight = (atomic_t)ATOMIC_INIT(0);
2921 encrypt_journal(ic, false, 0, ic->journal_sections, &crypt_comp);
2922 wait_for_completion(&crypt_comp.comp);
2924 DEBUG_bytes(lowmem_page_address(ic->journal[0].page), 64, "decrypted journal");
2927 if (dm_integrity_failed(ic))
2930 journal_empty = true;
2931 memset(used_commit_ids, 0, sizeof used_commit_ids);
2932 memset(max_commit_id_sections, 0, sizeof max_commit_id_sections);
2933 for (i = 0; i < ic->journal_sections; i++) {
2934 for (j = 0; j < ic->journal_section_sectors; j++) {
2936 struct journal_sector *js = access_journal(ic, i, j);
2937 k = find_commit_seq(ic, i, j, js->commit_id);
2940 used_commit_ids[k] = true;
2941 max_commit_id_sections[k] = i;
2943 if (journal_empty) {
2944 for (j = 0; j < ic->journal_section_entries; j++) {
2945 struct journal_entry *je = access_journal_entry(ic, i, j);
2946 if (!journal_entry_is_unused(je)) {
2947 journal_empty = false;
2954 if (!used_commit_ids[N_COMMIT_IDS - 1]) {
2955 unused = N_COMMIT_IDS - 1;
2956 while (unused && !used_commit_ids[unused - 1])
2959 for (unused = 0; unused < N_COMMIT_IDS; unused++)
2960 if (!used_commit_ids[unused])
2962 if (unused == N_COMMIT_IDS) {
2963 dm_integrity_io_error(ic, "journal commit ids", -EIO);
2967 DEBUG_print("first unused commit seq %d [%d,%d,%d,%d]\n",
2968 unused, used_commit_ids[0], used_commit_ids[1],
2969 used_commit_ids[2], used_commit_ids[3]);
2971 last_used = prev_commit_seq(unused);
2972 want_commit_seq = prev_commit_seq(last_used);
2974 if (!used_commit_ids[want_commit_seq] && used_commit_ids[prev_commit_seq(want_commit_seq)])
2975 journal_empty = true;
2977 write_start = max_commit_id_sections[last_used] + 1;
2978 if (unlikely(write_start >= ic->journal_sections))
2979 want_commit_seq = next_commit_seq(want_commit_seq);
2980 wraparound_section(ic, &write_start);
2983 for (write_sections = 0; write_sections < ic->journal_sections; write_sections++) {
2984 for (j = 0; j < ic->journal_section_sectors; j++) {
2985 struct journal_sector *js = access_journal(ic, i, j);
2987 if (js->commit_id != dm_integrity_commit_id(ic, i, j, want_commit_seq)) {
2989 * This could be caused by crash during writing.
2990 * We won't replay the inconsistent part of the
2993 DEBUG_print("commit id mismatch at position (%u, %u): %d != %d\n",
2994 i, j, find_commit_seq(ic, i, j, js->commit_id), want_commit_seq);
2999 if (unlikely(i >= ic->journal_sections))
3000 want_commit_seq = next_commit_seq(want_commit_seq);
3001 wraparound_section(ic, &i);
3005 if (!journal_empty) {
3006 DEBUG_print("replaying %u sections, starting at %u, commit seq %d\n",
3007 write_sections, write_start, want_commit_seq);
3008 do_journal_write(ic, write_start, write_sections, true);
3011 if (write_sections == ic->journal_sections && (ic->mode == 'J' || journal_empty)) {
3012 continue_section = write_start;
3013 ic->commit_seq = want_commit_seq;
3014 DEBUG_print("continuing from section %u, commit seq %d\n", write_start, ic->commit_seq);
3017 unsigned char erase_seq;
3019 DEBUG_print("clearing journal\n");
3021 erase_seq = prev_commit_seq(prev_commit_seq(last_used));
3023 init_journal(ic, s, 1, erase_seq);
3025 wraparound_section(ic, &s);
3026 if (ic->journal_sections >= 2) {
3027 init_journal(ic, s, ic->journal_sections - 2, erase_seq);
3028 s += ic->journal_sections - 2;
3029 wraparound_section(ic, &s);
3030 init_journal(ic, s, 1, erase_seq);
3033 continue_section = 0;
3034 ic->commit_seq = next_commit_seq(erase_seq);
3037 ic->committed_section = continue_section;
3038 ic->n_committed_sections = 0;
3040 ic->uncommitted_section = continue_section;
3041 ic->n_uncommitted_sections = 0;
3043 ic->free_section = continue_section;
3044 ic->free_section_entry = 0;
3045 ic->free_sectors = ic->journal_entries;
3047 ic->journal_tree_root = RB_ROOT;
3048 for (i = 0; i < ic->journal_entries; i++)
3049 init_journal_node(&ic->journal_tree[i]);
3052 static void dm_integrity_enter_synchronous_mode(struct dm_integrity_c *ic)
3054 DEBUG_print("dm_integrity_enter_synchronous_mode\n");
3056 if (ic->mode == 'B') {
3057 ic->bitmap_flush_interval = msecs_to_jiffies(10) + 1;
3058 ic->synchronous_mode = 1;
3060 cancel_delayed_work_sync(&ic->bitmap_flush_work);
3061 queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, 0);
3062 flush_workqueue(ic->commit_wq);
3066 static int dm_integrity_reboot(struct notifier_block *n, unsigned long code, void *x)
3068 struct dm_integrity_c *ic = container_of(n, struct dm_integrity_c, reboot_notifier);
3070 DEBUG_print("dm_integrity_reboot\n");
3072 dm_integrity_enter_synchronous_mode(ic);
3077 static void dm_integrity_postsuspend(struct dm_target *ti)
3079 struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
3082 WARN_ON(unregister_reboot_notifier(&ic->reboot_notifier));
3084 del_timer_sync(&ic->autocommit_timer);
3087 drain_workqueue(ic->recalc_wq);
3089 if (ic->mode == 'B')
3090 cancel_delayed_work_sync(&ic->bitmap_flush_work);
3092 queue_work(ic->commit_wq, &ic->commit_work);
3093 drain_workqueue(ic->commit_wq);
3095 if (ic->mode == 'J') {
3096 queue_work(ic->writer_wq, &ic->writer_work);
3097 drain_workqueue(ic->writer_wq);
3098 dm_integrity_flush_buffers(ic, true);
3099 if (ic->wrote_to_journal) {
3100 init_journal(ic, ic->free_section,
3101 ic->journal_sections - ic->free_section, ic->commit_seq);
3102 if (ic->free_section) {
3103 init_journal(ic, 0, ic->free_section,
3104 next_commit_seq(ic->commit_seq));
3109 if (ic->mode == 'B') {
3110 dm_integrity_flush_buffers(ic, true);
3112 /* set to 0 to test bitmap replay code */
3113 init_journal(ic, 0, ic->journal_sections, 0);
3114 ic->sb->flags &= ~cpu_to_le32(SB_FLAG_DIRTY_BITMAP);
3115 r = sync_rw_sb(ic, REQ_OP_WRITE | REQ_FUA);
3117 dm_integrity_io_error(ic, "writing superblock", r);
3121 BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
3123 ic->journal_uptodate = true;
3126 static void dm_integrity_resume(struct dm_target *ti)
3128 struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
3129 __u64 old_provided_data_sectors = le64_to_cpu(ic->sb->provided_data_sectors);
3132 DEBUG_print("resume\n");
3134 ic->wrote_to_journal = false;
3136 if (ic->provided_data_sectors != old_provided_data_sectors) {
3137 if (ic->provided_data_sectors > old_provided_data_sectors &&
3139 ic->sb->log2_blocks_per_bitmap_bit == ic->log2_blocks_per_bitmap_bit) {
3140 rw_journal_sectors(ic, REQ_OP_READ, 0,
3141 ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3142 block_bitmap_op(ic, ic->journal, old_provided_data_sectors,
3143 ic->provided_data_sectors - old_provided_data_sectors, BITMAP_OP_SET);
3144 rw_journal_sectors(ic, REQ_OP_WRITE | REQ_FUA | REQ_SYNC, 0,
3145 ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3148 ic->sb->provided_data_sectors = cpu_to_le64(ic->provided_data_sectors);
3149 r = sync_rw_sb(ic, REQ_OP_WRITE | REQ_FUA);
3151 dm_integrity_io_error(ic, "writing superblock", r);
3154 if (ic->sb->flags & cpu_to_le32(SB_FLAG_DIRTY_BITMAP)) {
3155 DEBUG_print("resume dirty_bitmap\n");
3156 rw_journal_sectors(ic, REQ_OP_READ, 0,
3157 ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3158 if (ic->mode == 'B') {
3159 if (ic->sb->log2_blocks_per_bitmap_bit == ic->log2_blocks_per_bitmap_bit &&
3160 !ic->reset_recalculate_flag) {
3161 block_bitmap_copy(ic, ic->recalc_bitmap, ic->journal);
3162 block_bitmap_copy(ic, ic->may_write_bitmap, ic->journal);
3163 if (!block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors,
3164 BITMAP_OP_TEST_ALL_CLEAR)) {
3165 ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
3166 ic->sb->recalc_sector = cpu_to_le64(0);
3169 DEBUG_print("non-matching blocks_per_bitmap_bit: %u, %u\n",
3170 ic->sb->log2_blocks_per_bitmap_bit, ic->log2_blocks_per_bitmap_bit);
3171 ic->sb->log2_blocks_per_bitmap_bit = ic->log2_blocks_per_bitmap_bit;
3172 block_bitmap_op(ic, ic->recalc_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_SET);
3173 block_bitmap_op(ic, ic->may_write_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_SET);
3174 block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors, BITMAP_OP_SET);
3175 rw_journal_sectors(ic, REQ_OP_WRITE | REQ_FUA | REQ_SYNC, 0,
3176 ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3177 ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
3178 ic->sb->recalc_sector = cpu_to_le64(0);
3181 if (!(ic->sb->log2_blocks_per_bitmap_bit == ic->log2_blocks_per_bitmap_bit &&
3182 block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors, BITMAP_OP_TEST_ALL_CLEAR)) ||
3183 ic->reset_recalculate_flag) {
3184 ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
3185 ic->sb->recalc_sector = cpu_to_le64(0);
3187 init_journal(ic, 0, ic->journal_sections, 0);
3189 ic->sb->flags &= ~cpu_to_le32(SB_FLAG_DIRTY_BITMAP);
3191 r = sync_rw_sb(ic, REQ_OP_WRITE | REQ_FUA);
3193 dm_integrity_io_error(ic, "writing superblock", r);
3196 if (ic->reset_recalculate_flag) {
3197 ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
3198 ic->sb->recalc_sector = cpu_to_le64(0);
3200 if (ic->mode == 'B') {
3201 ic->sb->flags |= cpu_to_le32(SB_FLAG_DIRTY_BITMAP);
3202 ic->sb->log2_blocks_per_bitmap_bit = ic->log2_blocks_per_bitmap_bit;
3203 r = sync_rw_sb(ic, REQ_OP_WRITE | REQ_FUA);
3205 dm_integrity_io_error(ic, "writing superblock", r);
3207 block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors, BITMAP_OP_CLEAR);
3208 block_bitmap_op(ic, ic->recalc_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_CLEAR);
3209 block_bitmap_op(ic, ic->may_write_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_CLEAR);
3210 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING) &&
3211 le64_to_cpu(ic->sb->recalc_sector) < ic->provided_data_sectors) {
3212 block_bitmap_op(ic, ic->journal, le64_to_cpu(ic->sb->recalc_sector),
3213 ic->provided_data_sectors - le64_to_cpu(ic->sb->recalc_sector), BITMAP_OP_SET);
3214 block_bitmap_op(ic, ic->recalc_bitmap, le64_to_cpu(ic->sb->recalc_sector),
3215 ic->provided_data_sectors - le64_to_cpu(ic->sb->recalc_sector), BITMAP_OP_SET);
3216 block_bitmap_op(ic, ic->may_write_bitmap, le64_to_cpu(ic->sb->recalc_sector),
3217 ic->provided_data_sectors - le64_to_cpu(ic->sb->recalc_sector), BITMAP_OP_SET);
3219 rw_journal_sectors(ic, REQ_OP_WRITE | REQ_FUA | REQ_SYNC, 0,
3220 ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3224 DEBUG_print("testing recalc: %x\n", ic->sb->flags);
3225 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
3226 __u64 recalc_pos = le64_to_cpu(ic->sb->recalc_sector);
3227 DEBUG_print("recalc pos: %llx / %llx\n", recalc_pos, ic->provided_data_sectors);
3228 if (recalc_pos < ic->provided_data_sectors) {
3229 queue_work(ic->recalc_wq, &ic->recalc_work);
3230 } else if (recalc_pos > ic->provided_data_sectors) {
3231 ic->sb->recalc_sector = cpu_to_le64(ic->provided_data_sectors);
3232 recalc_write_super(ic);
3236 ic->reboot_notifier.notifier_call = dm_integrity_reboot;
3237 ic->reboot_notifier.next = NULL;
3238 ic->reboot_notifier.priority = INT_MAX - 1; /* be notified after md and before hardware drivers */
3239 WARN_ON(register_reboot_notifier(&ic->reboot_notifier));
3242 /* set to 1 to stress test synchronous mode */
3243 dm_integrity_enter_synchronous_mode(ic);
3247 static void dm_integrity_status(struct dm_target *ti, status_type_t type,
3248 unsigned int status_flags, char *result, unsigned int maxlen)
3250 struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
3251 unsigned int arg_count;
3255 case STATUSTYPE_INFO:
3257 (unsigned long long)atomic64_read(&ic->number_of_mismatches),
3258 ic->provided_data_sectors);
3259 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
3260 DMEMIT(" %llu", le64_to_cpu(ic->sb->recalc_sector));
3265 case STATUSTYPE_TABLE: {
3266 __u64 watermark_percentage = (__u64)(ic->journal_entries - ic->free_sectors_threshold) * 100;
3267 watermark_percentage += ic->journal_entries / 2;
3268 do_div(watermark_percentage, ic->journal_entries);
3270 arg_count += !!ic->meta_dev;
3271 arg_count += ic->sectors_per_block != 1;
3272 arg_count += !!(ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING));
3273 arg_count += ic->reset_recalculate_flag;
3274 arg_count += ic->discard;
3275 arg_count += ic->mode == 'J';
3276 arg_count += ic->mode == 'J';
3277 arg_count += ic->mode == 'B';
3278 arg_count += ic->mode == 'B';
3279 arg_count += !!ic->internal_hash_alg.alg_string;
3280 arg_count += !!ic->journal_crypt_alg.alg_string;
3281 arg_count += !!ic->journal_mac_alg.alg_string;
3282 arg_count += (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING)) != 0;
3283 arg_count += (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) != 0;
3284 arg_count += ic->legacy_recalculate;
3285 DMEMIT("%s %llu %u %c %u", ic->dev->name, ic->start,
3286 ic->tag_size, ic->mode, arg_count);
3288 DMEMIT(" meta_device:%s", ic->meta_dev->name);
3289 if (ic->sectors_per_block != 1)
3290 DMEMIT(" block_size:%u", ic->sectors_per_block << SECTOR_SHIFT);
3291 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
3292 DMEMIT(" recalculate");
3293 if (ic->reset_recalculate_flag)
3294 DMEMIT(" reset_recalculate");
3296 DMEMIT(" allow_discards");
3297 DMEMIT(" journal_sectors:%u", ic->initial_sectors - SB_SECTORS);
3298 DMEMIT(" interleave_sectors:%u", 1U << ic->sb->log2_interleave_sectors);
3299 DMEMIT(" buffer_sectors:%u", 1U << ic->log2_buffer_sectors);
3300 if (ic->mode == 'J') {
3301 DMEMIT(" journal_watermark:%u", (unsigned int)watermark_percentage);
3302 DMEMIT(" commit_time:%u", ic->autocommit_msec);
3304 if (ic->mode == 'B') {
3305 DMEMIT(" sectors_per_bit:%llu", (sector_t)ic->sectors_per_block << ic->log2_blocks_per_bitmap_bit);
3306 DMEMIT(" bitmap_flush_interval:%u", jiffies_to_msecs(ic->bitmap_flush_interval));
3308 if ((ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING)) != 0)
3309 DMEMIT(" fix_padding");
3310 if ((ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) != 0)
3311 DMEMIT(" fix_hmac");
3312 if (ic->legacy_recalculate)
3313 DMEMIT(" legacy_recalculate");
3315 #define EMIT_ALG(a, n) \
3317 if (ic->a.alg_string) { \
3318 DMEMIT(" %s:%s", n, ic->a.alg_string); \
3319 if (ic->a.key_string) \
3320 DMEMIT(":%s", ic->a.key_string);\
3323 EMIT_ALG(internal_hash_alg, "internal_hash");
3324 EMIT_ALG(journal_crypt_alg, "journal_crypt");
3325 EMIT_ALG(journal_mac_alg, "journal_mac");
3328 case STATUSTYPE_IMA:
3329 DMEMIT_TARGET_NAME_VERSION(ti->type);
3330 DMEMIT(",dev_name=%s,start=%llu,tag_size=%u,mode=%c",
3331 ic->dev->name, ic->start, ic->tag_size, ic->mode);
3334 DMEMIT(",meta_device=%s", ic->meta_dev->name);
3335 if (ic->sectors_per_block != 1)
3336 DMEMIT(",block_size=%u", ic->sectors_per_block << SECTOR_SHIFT);
3338 DMEMIT(",recalculate=%c", (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) ?
3340 DMEMIT(",allow_discards=%c", ic->discard ? 'y' : 'n');
3341 DMEMIT(",fix_padding=%c",
3342 ((ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING)) != 0) ? 'y' : 'n');
3343 DMEMIT(",fix_hmac=%c",
3344 ((ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_HMAC)) != 0) ? 'y' : 'n');
3345 DMEMIT(",legacy_recalculate=%c", ic->legacy_recalculate ? 'y' : 'n');
3347 DMEMIT(",journal_sectors=%u", ic->initial_sectors - SB_SECTORS);
3348 DMEMIT(",interleave_sectors=%u", 1U << ic->sb->log2_interleave_sectors);
3349 DMEMIT(",buffer_sectors=%u", 1U << ic->log2_buffer_sectors);
3355 static int dm_integrity_iterate_devices(struct dm_target *ti,
3356 iterate_devices_callout_fn fn, void *data)
3358 struct dm_integrity_c *ic = ti->private;
3361 return fn(ti, ic->dev, ic->start + ic->initial_sectors + ic->metadata_run, ti->len, data);
3363 return fn(ti, ic->dev, 0, ti->len, data);
3366 static void dm_integrity_io_hints(struct dm_target *ti, struct queue_limits *limits)
3368 struct dm_integrity_c *ic = ti->private;
3370 if (ic->sectors_per_block > 1) {
3371 limits->logical_block_size = ic->sectors_per_block << SECTOR_SHIFT;
3372 limits->physical_block_size = ic->sectors_per_block << SECTOR_SHIFT;
3373 blk_limits_io_min(limits, ic->sectors_per_block << SECTOR_SHIFT);
3374 limits->dma_alignment = limits->logical_block_size - 1;
3378 static void calculate_journal_section_size(struct dm_integrity_c *ic)
3380 unsigned int sector_space = JOURNAL_SECTOR_DATA;
3382 ic->journal_sections = le32_to_cpu(ic->sb->journal_sections);
3383 ic->journal_entry_size = roundup(offsetof(struct journal_entry, last_bytes[ic->sectors_per_block]) + ic->tag_size,
3384 JOURNAL_ENTRY_ROUNDUP);
3386 if (ic->sb->flags & cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC))
3387 sector_space -= JOURNAL_MAC_PER_SECTOR;
3388 ic->journal_entries_per_sector = sector_space / ic->journal_entry_size;
3389 ic->journal_section_entries = ic->journal_entries_per_sector * JOURNAL_BLOCK_SECTORS;
3390 ic->journal_section_sectors = (ic->journal_section_entries << ic->sb->log2_sectors_per_block) + JOURNAL_BLOCK_SECTORS;
3391 ic->journal_entries = ic->journal_section_entries * ic->journal_sections;
3394 static int calculate_device_limits(struct dm_integrity_c *ic)
3396 __u64 initial_sectors;
3398 calculate_journal_section_size(ic);
3399 initial_sectors = SB_SECTORS + (__u64)ic->journal_section_sectors * ic->journal_sections;
3400 if (initial_sectors + METADATA_PADDING_SECTORS >= ic->meta_device_sectors || initial_sectors > UINT_MAX)
3402 ic->initial_sectors = initial_sectors;
3404 if (!ic->meta_dev) {
3405 sector_t last_sector, last_area, last_offset;
3407 /* we have to maintain excessive padding for compatibility with existing volumes */
3408 __u64 metadata_run_padding =
3409 ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING) ?
3410 (__u64)(METADATA_PADDING_SECTORS << SECTOR_SHIFT) :
3411 (__u64)(1 << SECTOR_SHIFT << METADATA_PADDING_SECTORS);
3413 ic->metadata_run = round_up((__u64)ic->tag_size << (ic->sb->log2_interleave_sectors - ic->sb->log2_sectors_per_block),
3414 metadata_run_padding) >> SECTOR_SHIFT;
3415 if (!(ic->metadata_run & (ic->metadata_run - 1)))
3416 ic->log2_metadata_run = __ffs(ic->metadata_run);
3418 ic->log2_metadata_run = -1;
3420 get_area_and_offset(ic, ic->provided_data_sectors - 1, &last_area, &last_offset);
3421 last_sector = get_data_sector(ic, last_area, last_offset);
3422 if (last_sector < ic->start || last_sector >= ic->meta_device_sectors)
3425 __u64 meta_size = (ic->provided_data_sectors >> ic->sb->log2_sectors_per_block) * ic->tag_size;
3426 meta_size = (meta_size + ((1U << (ic->log2_buffer_sectors + SECTOR_SHIFT)) - 1))
3427 >> (ic->log2_buffer_sectors + SECTOR_SHIFT);
3428 meta_size <<= ic->log2_buffer_sectors;
3429 if (ic->initial_sectors + meta_size < ic->initial_sectors ||
3430 ic->initial_sectors + meta_size > ic->meta_device_sectors)
3432 ic->metadata_run = 1;
3433 ic->log2_metadata_run = 0;
3439 static void get_provided_data_sectors(struct dm_integrity_c *ic)
3441 if (!ic->meta_dev) {
3443 ic->provided_data_sectors = 0;
3444 for (test_bit = fls64(ic->meta_device_sectors) - 1; test_bit >= 3; test_bit--) {
3445 __u64 prev_data_sectors = ic->provided_data_sectors;
3447 ic->provided_data_sectors |= (sector_t)1 << test_bit;
3448 if (calculate_device_limits(ic))
3449 ic->provided_data_sectors = prev_data_sectors;
3452 ic->provided_data_sectors = ic->data_device_sectors;
3453 ic->provided_data_sectors &= ~(sector_t)(ic->sectors_per_block - 1);
3457 static int initialize_superblock(struct dm_integrity_c *ic,
3458 unsigned int journal_sectors, unsigned int interleave_sectors)
3460 unsigned int journal_sections;
3463 memset(ic->sb, 0, SB_SECTORS << SECTOR_SHIFT);
3464 memcpy(ic->sb->magic, SB_MAGIC, 8);
3465 ic->sb->integrity_tag_size = cpu_to_le16(ic->tag_size);
3466 ic->sb->log2_sectors_per_block = __ffs(ic->sectors_per_block);
3467 if (ic->journal_mac_alg.alg_string)
3468 ic->sb->flags |= cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC);
3470 calculate_journal_section_size(ic);
3471 journal_sections = journal_sectors / ic->journal_section_sectors;
3472 if (!journal_sections)
3473 journal_sections = 1;
3475 if (ic->fix_hmac && (ic->internal_hash_alg.alg_string || ic->journal_mac_alg.alg_string)) {
3476 ic->sb->flags |= cpu_to_le32(SB_FLAG_FIXED_HMAC);
3477 get_random_bytes(ic->sb->salt, SALT_SIZE);
3480 if (!ic->meta_dev) {
3481 if (ic->fix_padding)
3482 ic->sb->flags |= cpu_to_le32(SB_FLAG_FIXED_PADDING);
3483 ic->sb->journal_sections = cpu_to_le32(journal_sections);
3484 if (!interleave_sectors)
3485 interleave_sectors = DEFAULT_INTERLEAVE_SECTORS;
3486 ic->sb->log2_interleave_sectors = __fls(interleave_sectors);
3487 ic->sb->log2_interleave_sectors = max((__u8)MIN_LOG2_INTERLEAVE_SECTORS, ic->sb->log2_interleave_sectors);
3488 ic->sb->log2_interleave_sectors = min((__u8)MAX_LOG2_INTERLEAVE_SECTORS, ic->sb->log2_interleave_sectors);
3490 get_provided_data_sectors(ic);
3491 if (!ic->provided_data_sectors)
3494 ic->sb->log2_interleave_sectors = 0;
3496 get_provided_data_sectors(ic);
3497 if (!ic->provided_data_sectors)
3501 ic->sb->journal_sections = cpu_to_le32(0);
3502 for (test_bit = fls(journal_sections) - 1; test_bit >= 0; test_bit--) {
3503 __u32 prev_journal_sections = le32_to_cpu(ic->sb->journal_sections);
3504 __u32 test_journal_sections = prev_journal_sections | (1U << test_bit);
3505 if (test_journal_sections > journal_sections)
3507 ic->sb->journal_sections = cpu_to_le32(test_journal_sections);
3508 if (calculate_device_limits(ic))
3509 ic->sb->journal_sections = cpu_to_le32(prev_journal_sections);
3512 if (!le32_to_cpu(ic->sb->journal_sections)) {
3513 if (ic->log2_buffer_sectors > 3) {
3514 ic->log2_buffer_sectors--;
3515 goto try_smaller_buffer;
3521 ic->sb->provided_data_sectors = cpu_to_le64(ic->provided_data_sectors);
3528 static void dm_integrity_set(struct dm_target *ti, struct dm_integrity_c *ic)
3530 struct gendisk *disk = dm_disk(dm_table_get_md(ti->table));
3531 struct blk_integrity bi;
3533 memset(&bi, 0, sizeof(bi));
3534 bi.profile = &dm_integrity_profile;
3535 bi.tuple_size = ic->tag_size;
3536 bi.tag_size = bi.tuple_size;
3537 bi.interval_exp = ic->sb->log2_sectors_per_block + SECTOR_SHIFT;
3539 blk_integrity_register(disk, &bi);
3540 blk_queue_max_integrity_segments(disk->queue, UINT_MAX);
3543 static void dm_integrity_free_page_list(struct page_list *pl)
3549 for (i = 0; pl[i].page; i++)
3550 __free_page(pl[i].page);
3554 static struct page_list *dm_integrity_alloc_page_list(unsigned int n_pages)
3556 struct page_list *pl;
3559 pl = kvmalloc_array(n_pages + 1, sizeof(struct page_list), GFP_KERNEL | __GFP_ZERO);
3563 for (i = 0; i < n_pages; i++) {
3564 pl[i].page = alloc_page(GFP_KERNEL);
3566 dm_integrity_free_page_list(pl);
3570 pl[i - 1].next = &pl[i];
3578 static void dm_integrity_free_journal_scatterlist(struct dm_integrity_c *ic, struct scatterlist **sl)
3581 for (i = 0; i < ic->journal_sections; i++)
3586 static struct scatterlist **dm_integrity_alloc_journal_scatterlist(struct dm_integrity_c *ic,
3587 struct page_list *pl)
3589 struct scatterlist **sl;
3592 sl = kvmalloc_array(ic->journal_sections,
3593 sizeof(struct scatterlist *),
3594 GFP_KERNEL | __GFP_ZERO);
3598 for (i = 0; i < ic->journal_sections; i++) {
3599 struct scatterlist *s;
3600 unsigned int start_index, start_offset;
3601 unsigned int end_index, end_offset;
3602 unsigned int n_pages;
3605 page_list_location(ic, i, 0, &start_index, &start_offset);
3606 page_list_location(ic, i, ic->journal_section_sectors - 1,
3607 &end_index, &end_offset);
3609 n_pages = (end_index - start_index + 1);
3611 s = kvmalloc_array(n_pages, sizeof(struct scatterlist),
3614 dm_integrity_free_journal_scatterlist(ic, sl);
3618 sg_init_table(s, n_pages);
3619 for (idx = start_index; idx <= end_index; idx++) {
3620 char *va = lowmem_page_address(pl[idx].page);
3621 unsigned int start = 0, end = PAGE_SIZE;
3622 if (idx == start_index)
3623 start = start_offset;
3624 if (idx == end_index)
3625 end = end_offset + (1 << SECTOR_SHIFT);
3626 sg_set_buf(&s[idx - start_index], va + start, end - start);
3635 static void free_alg(struct alg_spec *a)
3637 kfree_sensitive(a->alg_string);
3638 kfree_sensitive(a->key);
3639 memset(a, 0, sizeof *a);
3642 static int get_alg_and_key(const char *arg, struct alg_spec *a, char **error, char *error_inval)
3648 a->alg_string = kstrdup(strchr(arg, ':') + 1, GFP_KERNEL);
3652 k = strchr(a->alg_string, ':');
3655 a->key_string = k + 1;
3656 if (strlen(a->key_string) & 1)
3659 a->key_size = strlen(a->key_string) / 2;
3660 a->key = kmalloc(a->key_size, GFP_KERNEL);
3663 if (hex2bin(a->key, a->key_string, a->key_size))
3669 *error = error_inval;
3672 *error = "Out of memory for an argument";
3676 static int get_mac(struct crypto_shash **hash, struct alg_spec *a, char **error,
3677 char *error_alg, char *error_key)
3681 if (a->alg_string) {
3682 *hash = crypto_alloc_shash(a->alg_string, 0, CRYPTO_ALG_ALLOCATES_MEMORY);
3683 if (IS_ERR(*hash)) {
3691 r = crypto_shash_setkey(*hash, a->key, a->key_size);
3696 } else if (crypto_shash_get_flags(*hash) & CRYPTO_TFM_NEED_KEY) {
3705 static int create_journal(struct dm_integrity_c *ic, char **error)
3709 __u64 journal_pages, journal_desc_size, journal_tree_size;
3710 unsigned char *crypt_data = NULL, *crypt_iv = NULL;
3711 struct skcipher_request *req = NULL;
3713 ic->commit_ids[0] = cpu_to_le64(0x1111111111111111ULL);
3714 ic->commit_ids[1] = cpu_to_le64(0x2222222222222222ULL);
3715 ic->commit_ids[2] = cpu_to_le64(0x3333333333333333ULL);
3716 ic->commit_ids[3] = cpu_to_le64(0x4444444444444444ULL);
3718 journal_pages = roundup((__u64)ic->journal_sections * ic->journal_section_sectors,
3719 PAGE_SIZE >> SECTOR_SHIFT) >> (PAGE_SHIFT - SECTOR_SHIFT);
3720 journal_desc_size = journal_pages * sizeof(struct page_list);
3721 if (journal_pages >= totalram_pages() - totalhigh_pages() || journal_desc_size > ULONG_MAX) {
3722 *error = "Journal doesn't fit into memory";
3726 ic->journal_pages = journal_pages;
3728 ic->journal = dm_integrity_alloc_page_list(ic->journal_pages);
3730 *error = "Could not allocate memory for journal";
3734 if (ic->journal_crypt_alg.alg_string) {
3735 unsigned int ivsize, blocksize;
3736 struct journal_completion comp;
3739 ic->journal_crypt = crypto_alloc_skcipher(ic->journal_crypt_alg.alg_string, 0, CRYPTO_ALG_ALLOCATES_MEMORY);
3740 if (IS_ERR(ic->journal_crypt)) {
3741 *error = "Invalid journal cipher";
3742 r = PTR_ERR(ic->journal_crypt);
3743 ic->journal_crypt = NULL;
3746 ivsize = crypto_skcipher_ivsize(ic->journal_crypt);
3747 blocksize = crypto_skcipher_blocksize(ic->journal_crypt);
3749 if (ic->journal_crypt_alg.key) {
3750 r = crypto_skcipher_setkey(ic->journal_crypt, ic->journal_crypt_alg.key,
3751 ic->journal_crypt_alg.key_size);
3753 *error = "Error setting encryption key";
3757 DEBUG_print("cipher %s, block size %u iv size %u\n",
3758 ic->journal_crypt_alg.alg_string, blocksize, ivsize);
3760 ic->journal_io = dm_integrity_alloc_page_list(ic->journal_pages);
3761 if (!ic->journal_io) {
3762 *error = "Could not allocate memory for journal io";
3767 if (blocksize == 1) {
3768 struct scatterlist *sg;
3770 req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
3772 *error = "Could not allocate crypt request";
3777 crypt_iv = kzalloc(ivsize, GFP_KERNEL);
3779 *error = "Could not allocate iv";
3784 ic->journal_xor = dm_integrity_alloc_page_list(ic->journal_pages);
3785 if (!ic->journal_xor) {
3786 *error = "Could not allocate memory for journal xor";
3791 sg = kvmalloc_array(ic->journal_pages + 1,
3792 sizeof(struct scatterlist),
3795 *error = "Unable to allocate sg list";
3799 sg_init_table(sg, ic->journal_pages + 1);
3800 for (i = 0; i < ic->journal_pages; i++) {
3801 char *va = lowmem_page_address(ic->journal_xor[i].page);
3803 sg_set_buf(&sg[i], va, PAGE_SIZE);
3805 sg_set_buf(&sg[i], &ic->commit_ids, sizeof ic->commit_ids);
3807 skcipher_request_set_crypt(req, sg, sg,
3808 PAGE_SIZE * ic->journal_pages + sizeof ic->commit_ids, crypt_iv);
3809 init_completion(&comp.comp);
3810 comp.in_flight = (atomic_t)ATOMIC_INIT(1);
3811 if (do_crypt(true, req, &comp))
3812 wait_for_completion(&comp.comp);
3814 r = dm_integrity_failed(ic);
3816 *error = "Unable to encrypt journal";
3819 DEBUG_bytes(lowmem_page_address(ic->journal_xor[0].page), 64, "xor data");
3821 crypto_free_skcipher(ic->journal_crypt);
3822 ic->journal_crypt = NULL;
3824 unsigned int crypt_len = roundup(ivsize, blocksize);
3826 req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
3828 *error = "Could not allocate crypt request";
3833 crypt_iv = kmalloc(ivsize, GFP_KERNEL);
3835 *error = "Could not allocate iv";
3840 crypt_data = kmalloc(crypt_len, GFP_KERNEL);
3842 *error = "Unable to allocate crypt data";
3847 ic->journal_scatterlist = dm_integrity_alloc_journal_scatterlist(ic, ic->journal);
3848 if (!ic->journal_scatterlist) {
3849 *error = "Unable to allocate sg list";
3853 ic->journal_io_scatterlist = dm_integrity_alloc_journal_scatterlist(ic, ic->journal_io);
3854 if (!ic->journal_io_scatterlist) {
3855 *error = "Unable to allocate sg list";
3859 ic->sk_requests = kvmalloc_array(ic->journal_sections,
3860 sizeof(struct skcipher_request *),
3861 GFP_KERNEL | __GFP_ZERO);
3862 if (!ic->sk_requests) {
3863 *error = "Unable to allocate sk requests";
3867 for (i = 0; i < ic->journal_sections; i++) {
3868 struct scatterlist sg;
3869 struct skcipher_request *section_req;
3870 __le32 section_le = cpu_to_le32(i);
3872 memset(crypt_iv, 0x00, ivsize);
3873 memset(crypt_data, 0x00, crypt_len);
3874 memcpy(crypt_data, §ion_le, min((size_t)crypt_len, sizeof(section_le)));
3876 sg_init_one(&sg, crypt_data, crypt_len);
3877 skcipher_request_set_crypt(req, &sg, &sg, crypt_len, crypt_iv);
3878 init_completion(&comp.comp);
3879 comp.in_flight = (atomic_t)ATOMIC_INIT(1);
3880 if (do_crypt(true, req, &comp))
3881 wait_for_completion(&comp.comp);
3883 r = dm_integrity_failed(ic);
3885 *error = "Unable to generate iv";
3889 section_req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
3891 *error = "Unable to allocate crypt request";
3895 section_req->iv = kmalloc_array(ivsize, 2,
3897 if (!section_req->iv) {
3898 skcipher_request_free(section_req);
3899 *error = "Unable to allocate iv";
3903 memcpy(section_req->iv + ivsize, crypt_data, ivsize);
3904 section_req->cryptlen = (size_t)ic->journal_section_sectors << SECTOR_SHIFT;
3905 ic->sk_requests[i] = section_req;
3906 DEBUG_bytes(crypt_data, ivsize, "iv(%u)", i);
3911 for (i = 0; i < N_COMMIT_IDS; i++) {
3914 for (j = 0; j < i; j++) {
3915 if (ic->commit_ids[j] == ic->commit_ids[i]) {
3916 ic->commit_ids[i] = cpu_to_le64(le64_to_cpu(ic->commit_ids[i]) + 1);
3917 goto retest_commit_id;
3920 DEBUG_print("commit id %u: %016llx\n", i, ic->commit_ids[i]);
3923 journal_tree_size = (__u64)ic->journal_entries * sizeof(struct journal_node);
3924 if (journal_tree_size > ULONG_MAX) {
3925 *error = "Journal doesn't fit into memory";
3929 ic->journal_tree = kvmalloc(journal_tree_size, GFP_KERNEL);
3930 if (!ic->journal_tree) {
3931 *error = "Could not allocate memory for journal tree";
3937 skcipher_request_free(req);
3943 * Construct a integrity mapping
3947 * offset from the start of the device
3949 * D - direct writes, J - journal writes, B - bitmap mode, R - recovery mode
3950 * number of optional arguments
3951 * optional arguments:
3953 * interleave_sectors
3960 * bitmap_flush_interval
3966 static int dm_integrity_ctr(struct dm_target *ti, unsigned int argc, char **argv)
3968 struct dm_integrity_c *ic;
3971 unsigned int extra_args;
3972 struct dm_arg_set as;
3973 static const struct dm_arg _args[] = {
3974 {0, 18, "Invalid number of feature args"},
3976 unsigned int journal_sectors, interleave_sectors, buffer_sectors, journal_watermark, sync_msec;
3977 bool should_write_sb;
3979 unsigned long long start;
3980 __s8 log2_sectors_per_bitmap_bit = -1;
3981 __s8 log2_blocks_per_bitmap_bit;
3982 __u64 bits_in_journal;
3983 __u64 n_bitmap_bits;
3985 #define DIRECT_ARGUMENTS 4
3987 if (argc <= DIRECT_ARGUMENTS) {
3988 ti->error = "Invalid argument count";
3992 ic = kzalloc(sizeof(struct dm_integrity_c), GFP_KERNEL);
3994 ti->error = "Cannot allocate integrity context";
3998 ti->per_io_data_size = sizeof(struct dm_integrity_io);
4001 ic->in_progress = RB_ROOT;
4002 INIT_LIST_HEAD(&ic->wait_list);
4003 init_waitqueue_head(&ic->endio_wait);
4004 bio_list_init(&ic->flush_bio_list);
4005 init_waitqueue_head(&ic->copy_to_journal_wait);
4006 init_completion(&ic->crypto_backoff);
4007 atomic64_set(&ic->number_of_mismatches, 0);
4008 ic->bitmap_flush_interval = BITMAP_FLUSH_INTERVAL;
4010 r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &ic->dev);
4012 ti->error = "Device lookup failed";
4016 if (sscanf(argv[1], "%llu%c", &start, &dummy) != 1 || start != (sector_t)start) {
4017 ti->error = "Invalid starting offset";
4023 if (strcmp(argv[2], "-")) {
4024 if (sscanf(argv[2], "%u%c", &ic->tag_size, &dummy) != 1 || !ic->tag_size) {
4025 ti->error = "Invalid tag size";
4031 if (!strcmp(argv[3], "J") || !strcmp(argv[3], "B") ||
4032 !strcmp(argv[3], "D") || !strcmp(argv[3], "R")) {
4033 ic->mode = argv[3][0];
4035 ti->error = "Invalid mode (expecting J, B, D, R)";
4040 journal_sectors = 0;
4041 interleave_sectors = DEFAULT_INTERLEAVE_SECTORS;
4042 buffer_sectors = DEFAULT_BUFFER_SECTORS;
4043 journal_watermark = DEFAULT_JOURNAL_WATERMARK;
4044 sync_msec = DEFAULT_SYNC_MSEC;
4045 ic->sectors_per_block = 1;
4047 as.argc = argc - DIRECT_ARGUMENTS;
4048 as.argv = argv + DIRECT_ARGUMENTS;
4049 r = dm_read_arg_group(_args, &as, &extra_args, &ti->error);
4053 while (extra_args--) {
4054 const char *opt_string;
4056 unsigned long long llval;
4057 opt_string = dm_shift_arg(&as);
4060 ti->error = "Not enough feature arguments";
4063 if (sscanf(opt_string, "journal_sectors:%u%c", &val, &dummy) == 1)
4064 journal_sectors = val ? val : 1;
4065 else if (sscanf(opt_string, "interleave_sectors:%u%c", &val, &dummy) == 1)
4066 interleave_sectors = val;
4067 else if (sscanf(opt_string, "buffer_sectors:%u%c", &val, &dummy) == 1)
4068 buffer_sectors = val;
4069 else if (sscanf(opt_string, "journal_watermark:%u%c", &val, &dummy) == 1 && val <= 100)
4070 journal_watermark = val;
4071 else if (sscanf(opt_string, "commit_time:%u%c", &val, &dummy) == 1)
4073 else if (!strncmp(opt_string, "meta_device:", strlen("meta_device:"))) {
4075 dm_put_device(ti, ic->meta_dev);
4076 ic->meta_dev = NULL;
4078 r = dm_get_device(ti, strchr(opt_string, ':') + 1,
4079 dm_table_get_mode(ti->table), &ic->meta_dev);
4081 ti->error = "Device lookup failed";
4084 } else if (sscanf(opt_string, "block_size:%u%c", &val, &dummy) == 1) {
4085 if (val < 1 << SECTOR_SHIFT ||
4086 val > MAX_SECTORS_PER_BLOCK << SECTOR_SHIFT ||
4089 ti->error = "Invalid block_size argument";
4092 ic->sectors_per_block = val >> SECTOR_SHIFT;
4093 } else if (sscanf(opt_string, "sectors_per_bit:%llu%c", &llval, &dummy) == 1) {
4094 log2_sectors_per_bitmap_bit = !llval ? 0 : __ilog2_u64(llval);
4095 } else if (sscanf(opt_string, "bitmap_flush_interval:%u%c", &val, &dummy) == 1) {
4096 if (val >= (uint64_t)UINT_MAX * 1000 / HZ) {
4098 ti->error = "Invalid bitmap_flush_interval argument";
4101 ic->bitmap_flush_interval = msecs_to_jiffies(val);
4102 } else if (!strncmp(opt_string, "internal_hash:", strlen("internal_hash:"))) {
4103 r = get_alg_and_key(opt_string, &ic->internal_hash_alg, &ti->error,
4104 "Invalid internal_hash argument");
4107 } else if (!strncmp(opt_string, "journal_crypt:", strlen("journal_crypt:"))) {
4108 r = get_alg_and_key(opt_string, &ic->journal_crypt_alg, &ti->error,
4109 "Invalid journal_crypt argument");
4112 } else if (!strncmp(opt_string, "journal_mac:", strlen("journal_mac:"))) {
4113 r = get_alg_and_key(opt_string, &ic->journal_mac_alg, &ti->error,
4114 "Invalid journal_mac argument");
4117 } else if (!strcmp(opt_string, "recalculate")) {
4118 ic->recalculate_flag = true;
4119 } else if (!strcmp(opt_string, "reset_recalculate")) {
4120 ic->recalculate_flag = true;
4121 ic->reset_recalculate_flag = true;
4122 } else if (!strcmp(opt_string, "allow_discards")) {
4124 } else if (!strcmp(opt_string, "fix_padding")) {
4125 ic->fix_padding = true;
4126 } else if (!strcmp(opt_string, "fix_hmac")) {
4127 ic->fix_hmac = true;
4128 } else if (!strcmp(opt_string, "legacy_recalculate")) {
4129 ic->legacy_recalculate = true;
4132 ti->error = "Invalid argument";
4137 ic->data_device_sectors = bdev_nr_sectors(ic->dev->bdev);
4139 ic->meta_device_sectors = ic->data_device_sectors;
4141 ic->meta_device_sectors = bdev_nr_sectors(ic->meta_dev->bdev);
4143 if (!journal_sectors) {
4144 journal_sectors = min((sector_t)DEFAULT_MAX_JOURNAL_SECTORS,
4145 ic->data_device_sectors >> DEFAULT_JOURNAL_SIZE_FACTOR);
4148 if (!buffer_sectors)
4150 ic->log2_buffer_sectors = min((int)__fls(buffer_sectors), 31 - SECTOR_SHIFT);
4152 r = get_mac(&ic->internal_hash, &ic->internal_hash_alg, &ti->error,
4153 "Invalid internal hash", "Error setting internal hash key");
4157 r = get_mac(&ic->journal_mac, &ic->journal_mac_alg, &ti->error,
4158 "Invalid journal mac", "Error setting journal mac key");
4162 if (!ic->tag_size) {
4163 if (!ic->internal_hash) {
4164 ti->error = "Unknown tag size";
4168 ic->tag_size = crypto_shash_digestsize(ic->internal_hash);
4170 if (ic->tag_size > MAX_TAG_SIZE) {
4171 ti->error = "Too big tag size";
4175 if (!(ic->tag_size & (ic->tag_size - 1)))
4176 ic->log2_tag_size = __ffs(ic->tag_size);
4178 ic->log2_tag_size = -1;
4180 if (ic->mode == 'B' && !ic->internal_hash) {
4182 ti->error = "Bitmap mode can be only used with internal hash";
4186 if (ic->discard && !ic->internal_hash) {
4188 ti->error = "Discard can be only used with internal hash";
4192 ic->autocommit_jiffies = msecs_to_jiffies(sync_msec);
4193 ic->autocommit_msec = sync_msec;
4194 timer_setup(&ic->autocommit_timer, autocommit_fn, 0);
4196 ic->io = dm_io_client_create();
4197 if (IS_ERR(ic->io)) {
4198 r = PTR_ERR(ic->io);
4200 ti->error = "Cannot allocate dm io";
4204 r = mempool_init_slab_pool(&ic->journal_io_mempool, JOURNAL_IO_MEMPOOL, journal_io_cache);
4206 ti->error = "Cannot allocate mempool";
4210 ic->metadata_wq = alloc_workqueue("dm-integrity-metadata",
4211 WQ_MEM_RECLAIM, METADATA_WORKQUEUE_MAX_ACTIVE);
4212 if (!ic->metadata_wq) {
4213 ti->error = "Cannot allocate workqueue";
4219 * If this workqueue were percpu, it would cause bio reordering
4220 * and reduced performance.
4222 ic->wait_wq = alloc_workqueue("dm-integrity-wait", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
4224 ti->error = "Cannot allocate workqueue";
4229 ic->offload_wq = alloc_workqueue("dm-integrity-offload", WQ_MEM_RECLAIM,
4230 METADATA_WORKQUEUE_MAX_ACTIVE);
4231 if (!ic->offload_wq) {
4232 ti->error = "Cannot allocate workqueue";
4237 ic->commit_wq = alloc_workqueue("dm-integrity-commit", WQ_MEM_RECLAIM, 1);
4238 if (!ic->commit_wq) {
4239 ti->error = "Cannot allocate workqueue";
4243 INIT_WORK(&ic->commit_work, integrity_commit);
4245 if (ic->mode == 'J' || ic->mode == 'B') {
4246 ic->writer_wq = alloc_workqueue("dm-integrity-writer", WQ_MEM_RECLAIM, 1);
4247 if (!ic->writer_wq) {
4248 ti->error = "Cannot allocate workqueue";
4252 INIT_WORK(&ic->writer_work, integrity_writer);
4255 ic->sb = alloc_pages_exact(SB_SECTORS << SECTOR_SHIFT, GFP_KERNEL);
4258 ti->error = "Cannot allocate superblock area";
4262 r = sync_rw_sb(ic, REQ_OP_READ);
4264 ti->error = "Error reading superblock";
4267 should_write_sb = false;
4268 if (memcmp(ic->sb->magic, SB_MAGIC, 8)) {
4269 if (ic->mode != 'R') {
4270 if (memchr_inv(ic->sb, 0, SB_SECTORS << SECTOR_SHIFT)) {
4272 ti->error = "The device is not initialized";
4277 r = initialize_superblock(ic, journal_sectors, interleave_sectors);
4279 ti->error = "Could not initialize superblock";
4282 if (ic->mode != 'R')
4283 should_write_sb = true;
4286 if (!ic->sb->version || ic->sb->version > SB_VERSION_5) {
4288 ti->error = "Unknown version";
4291 if (le16_to_cpu(ic->sb->integrity_tag_size) != ic->tag_size) {
4293 ti->error = "Tag size doesn't match the information in superblock";
4296 if (ic->sb->log2_sectors_per_block != __ffs(ic->sectors_per_block)) {
4298 ti->error = "Block size doesn't match the information in superblock";
4301 if (!le32_to_cpu(ic->sb->journal_sections)) {
4303 ti->error = "Corrupted superblock, journal_sections is 0";
4306 /* make sure that ti->max_io_len doesn't overflow */
4307 if (!ic->meta_dev) {
4308 if (ic->sb->log2_interleave_sectors < MIN_LOG2_INTERLEAVE_SECTORS ||
4309 ic->sb->log2_interleave_sectors > MAX_LOG2_INTERLEAVE_SECTORS) {
4311 ti->error = "Invalid interleave_sectors in the superblock";
4315 if (ic->sb->log2_interleave_sectors) {
4317 ti->error = "Invalid interleave_sectors in the superblock";
4321 if (!!(ic->sb->flags & cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC)) != !!ic->journal_mac_alg.alg_string) {
4323 ti->error = "Journal mac mismatch";
4327 get_provided_data_sectors(ic);
4328 if (!ic->provided_data_sectors) {
4330 ti->error = "The device is too small";
4335 r = calculate_device_limits(ic);
4338 if (ic->log2_buffer_sectors > 3) {
4339 ic->log2_buffer_sectors--;
4340 goto try_smaller_buffer;
4343 ti->error = "The device is too small";
4347 if (log2_sectors_per_bitmap_bit < 0)
4348 log2_sectors_per_bitmap_bit = __fls(DEFAULT_SECTORS_PER_BITMAP_BIT);
4349 if (log2_sectors_per_bitmap_bit < ic->sb->log2_sectors_per_block)
4350 log2_sectors_per_bitmap_bit = ic->sb->log2_sectors_per_block;
4352 bits_in_journal = ((__u64)ic->journal_section_sectors * ic->journal_sections) << (SECTOR_SHIFT + 3);
4353 if (bits_in_journal > UINT_MAX)
4354 bits_in_journal = UINT_MAX;
4355 while (bits_in_journal < (ic->provided_data_sectors + ((sector_t)1 << log2_sectors_per_bitmap_bit) - 1) >> log2_sectors_per_bitmap_bit)
4356 log2_sectors_per_bitmap_bit++;
4358 log2_blocks_per_bitmap_bit = log2_sectors_per_bitmap_bit - ic->sb->log2_sectors_per_block;
4359 ic->log2_blocks_per_bitmap_bit = log2_blocks_per_bitmap_bit;
4360 if (should_write_sb) {
4361 ic->sb->log2_blocks_per_bitmap_bit = log2_blocks_per_bitmap_bit;
4363 n_bitmap_bits = ((ic->provided_data_sectors >> ic->sb->log2_sectors_per_block)
4364 + (((sector_t)1 << log2_blocks_per_bitmap_bit) - 1)) >> log2_blocks_per_bitmap_bit;
4365 ic->n_bitmap_blocks = DIV_ROUND_UP(n_bitmap_bits, BITMAP_BLOCK_SIZE * 8);
4368 ic->log2_buffer_sectors = min(ic->log2_buffer_sectors, (__u8)__ffs(ic->metadata_run));
4370 if (ti->len > ic->provided_data_sectors) {
4372 ti->error = "Not enough provided sectors for requested mapping size";
4377 threshold = (__u64)ic->journal_entries * (100 - journal_watermark);
4379 do_div(threshold, 100);
4380 ic->free_sectors_threshold = threshold;
4382 DEBUG_print("initialized:\n");
4383 DEBUG_print(" integrity_tag_size %u\n", le16_to_cpu(ic->sb->integrity_tag_size));
4384 DEBUG_print(" journal_entry_size %u\n", ic->journal_entry_size);
4385 DEBUG_print(" journal_entries_per_sector %u\n", ic->journal_entries_per_sector);
4386 DEBUG_print(" journal_section_entries %u\n", ic->journal_section_entries);
4387 DEBUG_print(" journal_section_sectors %u\n", ic->journal_section_sectors);
4388 DEBUG_print(" journal_sections %u\n", (unsigned int)le32_to_cpu(ic->sb->journal_sections));
4389 DEBUG_print(" journal_entries %u\n", ic->journal_entries);
4390 DEBUG_print(" log2_interleave_sectors %d\n", ic->sb->log2_interleave_sectors);
4391 DEBUG_print(" data_device_sectors 0x%llx\n", bdev_nr_sectors(ic->dev->bdev));
4392 DEBUG_print(" initial_sectors 0x%x\n", ic->initial_sectors);
4393 DEBUG_print(" metadata_run 0x%x\n", ic->metadata_run);
4394 DEBUG_print(" log2_metadata_run %d\n", ic->log2_metadata_run);
4395 DEBUG_print(" provided_data_sectors 0x%llx (%llu)\n", ic->provided_data_sectors, ic->provided_data_sectors);
4396 DEBUG_print(" log2_buffer_sectors %u\n", ic->log2_buffer_sectors);
4397 DEBUG_print(" bits_in_journal %llu\n", bits_in_journal);
4399 if (ic->recalculate_flag && !(ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))) {
4400 ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
4401 ic->sb->recalc_sector = cpu_to_le64(0);
4404 if (ic->internal_hash) {
4405 size_t recalc_tags_size;
4406 ic->recalc_wq = alloc_workqueue("dm-integrity-recalc", WQ_MEM_RECLAIM, 1);
4407 if (!ic->recalc_wq ) {
4408 ti->error = "Cannot allocate workqueue";
4412 INIT_WORK(&ic->recalc_work, integrity_recalc);
4413 ic->recalc_buffer = vmalloc(RECALC_SECTORS << SECTOR_SHIFT);
4414 if (!ic->recalc_buffer) {
4415 ti->error = "Cannot allocate buffer for recalculating";
4419 recalc_tags_size = (RECALC_SECTORS >> ic->sb->log2_sectors_per_block) * ic->tag_size;
4420 if (crypto_shash_digestsize(ic->internal_hash) > ic->tag_size)
4421 recalc_tags_size += crypto_shash_digestsize(ic->internal_hash) - ic->tag_size;
4422 ic->recalc_tags = kvmalloc(recalc_tags_size, GFP_KERNEL);
4423 if (!ic->recalc_tags) {
4424 ti->error = "Cannot allocate tags for recalculating";
4429 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
4430 ti->error = "Recalculate can only be specified with internal_hash";
4436 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING) &&
4437 le64_to_cpu(ic->sb->recalc_sector) < ic->provided_data_sectors &&
4438 dm_integrity_disable_recalculate(ic)) {
4439 ti->error = "Recalculating with HMAC is disabled for security reasons - if you really need it, use the argument \"legacy_recalculate\"";
4444 ic->bufio = dm_bufio_client_create(ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev,
4445 1U << (SECTOR_SHIFT + ic->log2_buffer_sectors), 1, 0, NULL, NULL, 0);
4446 if (IS_ERR(ic->bufio)) {
4447 r = PTR_ERR(ic->bufio);
4448 ti->error = "Cannot initialize dm-bufio";
4452 dm_bufio_set_sector_offset(ic->bufio, ic->start + ic->initial_sectors);
4454 if (ic->mode != 'R') {
4455 r = create_journal(ic, &ti->error);
4461 if (ic->mode == 'B') {
4463 unsigned int n_bitmap_pages = DIV_ROUND_UP(ic->n_bitmap_blocks, PAGE_SIZE / BITMAP_BLOCK_SIZE);
4465 ic->recalc_bitmap = dm_integrity_alloc_page_list(n_bitmap_pages);
4466 if (!ic->recalc_bitmap) {
4470 ic->may_write_bitmap = dm_integrity_alloc_page_list(n_bitmap_pages);
4471 if (!ic->may_write_bitmap) {
4475 ic->bbs = kvmalloc_array(ic->n_bitmap_blocks, sizeof(struct bitmap_block_status), GFP_KERNEL);
4480 INIT_DELAYED_WORK(&ic->bitmap_flush_work, bitmap_flush_work);
4481 for (i = 0; i < ic->n_bitmap_blocks; i++) {
4482 struct bitmap_block_status *bbs = &ic->bbs[i];
4483 unsigned int sector, pl_index, pl_offset;
4485 INIT_WORK(&bbs->work, bitmap_block_work);
4488 bio_list_init(&bbs->bio_queue);
4489 spin_lock_init(&bbs->bio_queue_lock);
4491 sector = i * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT);
4492 pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
4493 pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
4495 bbs->bitmap = lowmem_page_address(ic->journal[pl_index].page) + pl_offset;
4499 if (should_write_sb) {
4500 init_journal(ic, 0, ic->journal_sections, 0);
4501 r = dm_integrity_failed(ic);
4503 ti->error = "Error initializing journal";
4506 r = sync_rw_sb(ic, REQ_OP_WRITE | REQ_FUA);
4508 ti->error = "Error initializing superblock";
4511 ic->just_formatted = true;
4514 if (!ic->meta_dev) {
4515 r = dm_set_target_max_io_len(ti, 1U << ic->sb->log2_interleave_sectors);
4519 if (ic->mode == 'B') {
4520 unsigned int max_io_len = ((sector_t)ic->sectors_per_block << ic->log2_blocks_per_bitmap_bit) * (BITMAP_BLOCK_SIZE * 8);
4522 max_io_len = 1U << 31;
4523 DEBUG_print("max_io_len: old %u, new %u\n", ti->max_io_len, max_io_len);
4524 if (!ti->max_io_len || ti->max_io_len > max_io_len) {
4525 r = dm_set_target_max_io_len(ti, max_io_len);
4531 if (!ic->internal_hash)
4532 dm_integrity_set(ti, ic);
4534 ti->num_flush_bios = 1;
4535 ti->flush_supported = true;
4537 ti->num_discard_bios = 1;
4539 dm_audit_log_ctr(DM_MSG_PREFIX, ti, 1);
4543 dm_audit_log_ctr(DM_MSG_PREFIX, ti, 0);
4544 dm_integrity_dtr(ti);
4548 static void dm_integrity_dtr(struct dm_target *ti)
4550 struct dm_integrity_c *ic = ti->private;
4552 BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
4553 BUG_ON(!list_empty(&ic->wait_list));
4555 if (ic->mode == 'B')
4556 cancel_delayed_work_sync(&ic->bitmap_flush_work);
4557 if (ic->metadata_wq)
4558 destroy_workqueue(ic->metadata_wq);
4560 destroy_workqueue(ic->wait_wq);
4562 destroy_workqueue(ic->offload_wq);
4564 destroy_workqueue(ic->commit_wq);
4566 destroy_workqueue(ic->writer_wq);
4568 destroy_workqueue(ic->recalc_wq);
4569 vfree(ic->recalc_buffer);
4570 kvfree(ic->recalc_tags);
4573 dm_bufio_client_destroy(ic->bufio);
4574 mempool_exit(&ic->journal_io_mempool);
4576 dm_io_client_destroy(ic->io);
4578 dm_put_device(ti, ic->dev);
4580 dm_put_device(ti, ic->meta_dev);
4581 dm_integrity_free_page_list(ic->journal);
4582 dm_integrity_free_page_list(ic->journal_io);
4583 dm_integrity_free_page_list(ic->journal_xor);
4584 dm_integrity_free_page_list(ic->recalc_bitmap);
4585 dm_integrity_free_page_list(ic->may_write_bitmap);
4586 if (ic->journal_scatterlist)
4587 dm_integrity_free_journal_scatterlist(ic, ic->journal_scatterlist);
4588 if (ic->journal_io_scatterlist)
4589 dm_integrity_free_journal_scatterlist(ic, ic->journal_io_scatterlist);
4590 if (ic->sk_requests) {
4593 for (i = 0; i < ic->journal_sections; i++) {
4594 struct skcipher_request *req = ic->sk_requests[i];
4596 kfree_sensitive(req->iv);
4597 skcipher_request_free(req);
4600 kvfree(ic->sk_requests);
4602 kvfree(ic->journal_tree);
4604 free_pages_exact(ic->sb, SB_SECTORS << SECTOR_SHIFT);
4606 if (ic->internal_hash)
4607 crypto_free_shash(ic->internal_hash);
4608 free_alg(&ic->internal_hash_alg);
4610 if (ic->journal_crypt)
4611 crypto_free_skcipher(ic->journal_crypt);
4612 free_alg(&ic->journal_crypt_alg);
4614 if (ic->journal_mac)
4615 crypto_free_shash(ic->journal_mac);
4616 free_alg(&ic->journal_mac_alg);
4619 dm_audit_log_dtr(DM_MSG_PREFIX, ti, 1);
4622 static struct target_type integrity_target = {
4623 .name = "integrity",
4624 .version = {1, 10, 0},
4625 .module = THIS_MODULE,
4626 .features = DM_TARGET_SINGLETON | DM_TARGET_INTEGRITY,
4627 .ctr = dm_integrity_ctr,
4628 .dtr = dm_integrity_dtr,
4629 .map = dm_integrity_map,
4630 .postsuspend = dm_integrity_postsuspend,
4631 .resume = dm_integrity_resume,
4632 .status = dm_integrity_status,
4633 .iterate_devices = dm_integrity_iterate_devices,
4634 .io_hints = dm_integrity_io_hints,
4637 static int __init dm_integrity_init(void)
4641 journal_io_cache = kmem_cache_create("integrity_journal_io",
4642 sizeof(struct journal_io), 0, 0, NULL);
4643 if (!journal_io_cache) {
4644 DMERR("can't allocate journal io cache");
4648 r = dm_register_target(&integrity_target);
4651 DMERR("register failed %d", r);
4656 static void __exit dm_integrity_exit(void)
4658 dm_unregister_target(&integrity_target);
4659 kmem_cache_destroy(journal_io_cache);
4662 module_init(dm_integrity_init);
4663 module_exit(dm_integrity_exit);
4665 MODULE_AUTHOR("Milan Broz");
4666 MODULE_AUTHOR("Mikulas Patocka");
4667 MODULE_DESCRIPTION(DM_NAME " target for integrity tags extension");
4668 MODULE_LICENSE("GPL");