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 <linux/compiler.h>
10 #include <linux/module.h>
11 #include <linux/device-mapper.h>
12 #include <linux/dm-io.h>
13 #include <linux/vmalloc.h>
14 #include <linux/sort.h>
15 #include <linux/rbtree.h>
16 #include <linux/delay.h>
17 #include <linux/random.h>
18 #include <crypto/hash.h>
19 #include <crypto/skcipher.h>
20 #include <linux/async_tx.h>
21 #include <linux/dm-bufio.h>
23 #define DM_MSG_PREFIX "integrity"
25 #define DEFAULT_INTERLEAVE_SECTORS 32768
26 #define DEFAULT_JOURNAL_SIZE_FACTOR 7
27 #define DEFAULT_BUFFER_SECTORS 128
28 #define DEFAULT_JOURNAL_WATERMARK 50
29 #define DEFAULT_SYNC_MSEC 10000
30 #define DEFAULT_MAX_JOURNAL_SECTORS 131072
31 #define MIN_LOG2_INTERLEAVE_SECTORS 3
32 #define MAX_LOG2_INTERLEAVE_SECTORS 31
33 #define METADATA_WORKQUEUE_MAX_ACTIVE 16
36 * Warning - DEBUG_PRINT prints security-sensitive data to the log,
37 * so it should not be enabled in the official kernel
40 //#define INTERNAL_VERIFY
46 #define SB_MAGIC "integrt"
49 #define MAX_SECTORS_PER_BLOCK 8
54 __u8 log2_interleave_sectors;
55 __u16 integrity_tag_size;
56 __u32 journal_sections;
57 __u64 provided_data_sectors; /* userspace uses this value */
59 __u8 log2_sectors_per_block;
62 #define SB_FLAG_HAVE_JOURNAL_MAC 0x1
64 #define JOURNAL_ENTRY_ROUNDUP 8
66 typedef __u64 commit_id_t;
67 #define JOURNAL_MAC_PER_SECTOR 8
69 struct journal_entry {
77 commit_id_t last_bytes[0];
81 #define journal_entry_tag(ic, je) ((__u8 *)&(je)->last_bytes[(ic)->sectors_per_block])
83 #if BITS_PER_LONG == 64
84 #define journal_entry_set_sector(je, x) do { smp_wmb(); WRITE_ONCE((je)->u.sector, cpu_to_le64(x)); } while (0)
85 #define journal_entry_get_sector(je) le64_to_cpu((je)->u.sector)
86 #elif defined(CONFIG_LBDAF)
87 #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)
88 #define journal_entry_get_sector(je) le64_to_cpu((je)->u.sector)
90 #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(0)); } while (0)
91 #define journal_entry_get_sector(je) le32_to_cpu((je)->u.s.sector_lo)
93 #define journal_entry_is_unused(je) ((je)->u.s.sector_hi == cpu_to_le32(-1))
94 #define journal_entry_set_unused(je) do { ((je)->u.s.sector_hi = cpu_to_le32(-1)); } while (0)
95 #define journal_entry_is_inprogress(je) ((je)->u.s.sector_hi == cpu_to_le32(-2))
96 #define journal_entry_set_inprogress(je) do { ((je)->u.s.sector_hi = cpu_to_le32(-2)); } while (0)
98 #define JOURNAL_BLOCK_SECTORS 8
99 #define JOURNAL_SECTOR_DATA ((1 << SECTOR_SHIFT) - sizeof(commit_id_t))
100 #define JOURNAL_MAC_SIZE (JOURNAL_MAC_PER_SECTOR * JOURNAL_BLOCK_SECTORS)
102 struct journal_sector {
103 __u8 entries[JOURNAL_SECTOR_DATA - JOURNAL_MAC_PER_SECTOR];
104 __u8 mac[JOURNAL_MAC_PER_SECTOR];
105 commit_id_t commit_id;
108 #define MAX_TAG_SIZE (JOURNAL_SECTOR_DATA - JOURNAL_MAC_PER_SECTOR - offsetof(struct journal_entry, last_bytes[MAX_SECTORS_PER_BLOCK]))
110 #define METADATA_PADDING_SECTORS 8
112 #define N_COMMIT_IDS 4
114 static unsigned char prev_commit_seq(unsigned char seq)
116 return (seq + N_COMMIT_IDS - 1) % N_COMMIT_IDS;
119 static unsigned char next_commit_seq(unsigned char seq)
121 return (seq + 1) % N_COMMIT_IDS;
125 * In-memory structures
128 struct journal_node {
140 struct dm_integrity_c {
145 mempool_t journal_io_mempool;
146 struct dm_io_client *io;
147 struct dm_bufio_client *bufio;
148 struct workqueue_struct *metadata_wq;
149 struct superblock *sb;
150 unsigned journal_pages;
151 struct page_list *journal;
152 struct page_list *journal_io;
153 struct page_list *journal_xor;
155 struct crypto_skcipher *journal_crypt;
156 struct scatterlist **journal_scatterlist;
157 struct scatterlist **journal_io_scatterlist;
158 struct skcipher_request **sk_requests;
160 struct crypto_shash *journal_mac;
162 struct journal_node *journal_tree;
163 struct rb_root journal_tree_root;
165 sector_t provided_data_sectors;
167 unsigned short journal_entry_size;
168 unsigned char journal_entries_per_sector;
169 unsigned char journal_section_entries;
170 unsigned short journal_section_sectors;
171 unsigned journal_sections;
172 unsigned journal_entries;
173 sector_t device_sectors;
174 unsigned initial_sectors;
175 unsigned metadata_run;
176 __s8 log2_metadata_run;
177 __u8 log2_buffer_sectors;
178 __u8 sectors_per_block;
185 struct crypto_shash *internal_hash;
187 /* these variables are locked with endio_wait.lock */
188 struct rb_root in_progress;
189 struct list_head wait_list;
190 wait_queue_head_t endio_wait;
191 struct workqueue_struct *wait_wq;
193 unsigned char commit_seq;
194 commit_id_t commit_ids[N_COMMIT_IDS];
196 unsigned committed_section;
197 unsigned n_committed_sections;
199 unsigned uncommitted_section;
200 unsigned n_uncommitted_sections;
202 unsigned free_section;
203 unsigned char free_section_entry;
204 unsigned free_sectors;
206 unsigned free_sectors_threshold;
208 struct workqueue_struct *commit_wq;
209 struct work_struct commit_work;
211 struct workqueue_struct *writer_wq;
212 struct work_struct writer_work;
214 struct bio_list flush_bio_list;
216 unsigned long autocommit_jiffies;
217 struct timer_list autocommit_timer;
218 unsigned autocommit_msec;
220 wait_queue_head_t copy_to_journal_wait;
222 struct completion crypto_backoff;
224 bool journal_uptodate;
227 struct alg_spec internal_hash_alg;
228 struct alg_spec journal_crypt_alg;
229 struct alg_spec journal_mac_alg;
231 atomic64_t number_of_mismatches;
234 struct dm_integrity_range {
235 sector_t logical_sector;
241 struct task_struct *task;
242 struct list_head wait_entry;
247 struct dm_integrity_io {
248 struct work_struct work;
250 struct dm_integrity_c *ic;
254 struct dm_integrity_range range;
256 sector_t metadata_block;
257 unsigned metadata_offset;
260 blk_status_t bi_status;
262 struct completion *completion;
264 struct gendisk *orig_bi_disk;
266 bio_end_io_t *orig_bi_end_io;
267 struct bio_integrity_payload *orig_bi_integrity;
268 struct bvec_iter orig_bi_iter;
271 struct journal_completion {
272 struct dm_integrity_c *ic;
274 struct completion comp;
278 struct dm_integrity_range range;
279 struct journal_completion *comp;
282 static struct kmem_cache *journal_io_cache;
284 #define JOURNAL_IO_MEMPOOL 32
287 #define DEBUG_print(x, ...) printk(KERN_DEBUG x, ##__VA_ARGS__)
288 static void __DEBUG_bytes(__u8 *bytes, size_t len, const char *msg, ...)
297 pr_cont(" %02x", *bytes);
303 #define DEBUG_bytes(bytes, len, msg, ...) __DEBUG_bytes(bytes, len, KERN_DEBUG msg, ##__VA_ARGS__)
305 #define DEBUG_print(x, ...) do { } while (0)
306 #define DEBUG_bytes(bytes, len, msg, ...) do { } while (0)
310 * DM Integrity profile, protection is performed layer above (dm-crypt)
312 static const struct blk_integrity_profile dm_integrity_profile = {
313 .name = "DM-DIF-EXT-TAG",
318 static void dm_integrity_map_continue(struct dm_integrity_io *dio, bool from_map);
319 static void integrity_bio_wait(struct work_struct *w);
320 static void dm_integrity_dtr(struct dm_target *ti);
322 static void dm_integrity_io_error(struct dm_integrity_c *ic, const char *msg, int err)
325 atomic64_inc(&ic->number_of_mismatches);
326 if (!cmpxchg(&ic->failed, 0, err))
327 DMERR("Error on %s: %d", msg, err);
330 static int dm_integrity_failed(struct dm_integrity_c *ic)
332 return READ_ONCE(ic->failed);
335 static commit_id_t dm_integrity_commit_id(struct dm_integrity_c *ic, unsigned i,
336 unsigned j, unsigned char seq)
339 * Xor the number with section and sector, so that if a piece of
340 * journal is written at wrong place, it is detected.
342 return ic->commit_ids[seq] ^ cpu_to_le64(((__u64)i << 32) ^ j);
345 static void get_area_and_offset(struct dm_integrity_c *ic, sector_t data_sector,
346 sector_t *area, sector_t *offset)
348 __u8 log2_interleave_sectors = ic->sb->log2_interleave_sectors;
350 *area = data_sector >> log2_interleave_sectors;
351 *offset = (unsigned)data_sector & ((1U << log2_interleave_sectors) - 1);
354 #define sector_to_block(ic, n) \
356 BUG_ON((n) & (unsigned)((ic)->sectors_per_block - 1)); \
357 (n) >>= (ic)->sb->log2_sectors_per_block; \
360 static __u64 get_metadata_sector_and_offset(struct dm_integrity_c *ic, sector_t area,
361 sector_t offset, unsigned *metadata_offset)
366 ms = area << ic->sb->log2_interleave_sectors;
367 if (likely(ic->log2_metadata_run >= 0))
368 ms += area << ic->log2_metadata_run;
370 ms += area * ic->metadata_run;
371 ms >>= ic->log2_buffer_sectors;
373 sector_to_block(ic, offset);
375 if (likely(ic->log2_tag_size >= 0)) {
376 ms += offset >> (SECTOR_SHIFT + ic->log2_buffer_sectors - ic->log2_tag_size);
377 mo = (offset << ic->log2_tag_size) & ((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - 1);
379 ms += (__u64)offset * ic->tag_size >> (SECTOR_SHIFT + ic->log2_buffer_sectors);
380 mo = (offset * ic->tag_size) & ((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - 1);
382 *metadata_offset = mo;
386 static sector_t get_data_sector(struct dm_integrity_c *ic, sector_t area, sector_t offset)
390 result = area << ic->sb->log2_interleave_sectors;
391 if (likely(ic->log2_metadata_run >= 0))
392 result += (area + 1) << ic->log2_metadata_run;
394 result += (area + 1) * ic->metadata_run;
396 result += (sector_t)ic->initial_sectors + offset;
402 static void wraparound_section(struct dm_integrity_c *ic, unsigned *sec_ptr)
404 if (unlikely(*sec_ptr >= ic->journal_sections))
405 *sec_ptr -= ic->journal_sections;
408 static int sync_rw_sb(struct dm_integrity_c *ic, int op, int op_flags)
410 struct dm_io_request io_req;
411 struct dm_io_region io_loc;
414 io_req.bi_op_flags = op_flags;
415 io_req.mem.type = DM_IO_KMEM;
416 io_req.mem.ptr.addr = ic->sb;
417 io_req.notify.fn = NULL;
418 io_req.client = ic->io;
419 io_loc.bdev = ic->dev->bdev;
420 io_loc.sector = ic->start;
421 io_loc.count = SB_SECTORS;
423 return dm_io(&io_req, 1, &io_loc, NULL);
426 static void access_journal_check(struct dm_integrity_c *ic, unsigned section, unsigned offset,
427 bool e, const char *function)
429 #if defined(CONFIG_DM_DEBUG) || defined(INTERNAL_VERIFY)
430 unsigned limit = e ? ic->journal_section_entries : ic->journal_section_sectors;
432 if (unlikely(section >= ic->journal_sections) ||
433 unlikely(offset >= limit)) {
434 printk(KERN_CRIT "%s: invalid access at (%u,%u), limit (%u,%u)\n",
435 function, section, offset, ic->journal_sections, limit);
441 static void page_list_location(struct dm_integrity_c *ic, unsigned section, unsigned offset,
442 unsigned *pl_index, unsigned *pl_offset)
446 access_journal_check(ic, section, offset, false, "page_list_location");
448 sector = section * ic->journal_section_sectors + offset;
450 *pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
451 *pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
454 static struct journal_sector *access_page_list(struct dm_integrity_c *ic, struct page_list *pl,
455 unsigned section, unsigned offset, unsigned *n_sectors)
457 unsigned pl_index, pl_offset;
460 page_list_location(ic, section, offset, &pl_index, &pl_offset);
463 *n_sectors = (PAGE_SIZE - pl_offset) >> SECTOR_SHIFT;
465 va = lowmem_page_address(pl[pl_index].page);
467 return (struct journal_sector *)(va + pl_offset);
470 static struct journal_sector *access_journal(struct dm_integrity_c *ic, unsigned section, unsigned offset)
472 return access_page_list(ic, ic->journal, section, offset, NULL);
475 static struct journal_entry *access_journal_entry(struct dm_integrity_c *ic, unsigned section, unsigned n)
477 unsigned rel_sector, offset;
478 struct journal_sector *js;
480 access_journal_check(ic, section, n, true, "access_journal_entry");
482 rel_sector = n % JOURNAL_BLOCK_SECTORS;
483 offset = n / JOURNAL_BLOCK_SECTORS;
485 js = access_journal(ic, section, rel_sector);
486 return (struct journal_entry *)((char *)js + offset * ic->journal_entry_size);
489 static struct journal_sector *access_journal_data(struct dm_integrity_c *ic, unsigned section, unsigned n)
491 n <<= ic->sb->log2_sectors_per_block;
493 n += JOURNAL_BLOCK_SECTORS;
495 access_journal_check(ic, section, n, false, "access_journal_data");
497 return access_journal(ic, section, n);
500 static void section_mac(struct dm_integrity_c *ic, unsigned section, __u8 result[JOURNAL_MAC_SIZE])
502 SHASH_DESC_ON_STACK(desc, ic->journal_mac);
506 desc->tfm = ic->journal_mac;
507 desc->flags = CRYPTO_TFM_REQ_MAY_SLEEP;
509 r = crypto_shash_init(desc);
511 dm_integrity_io_error(ic, "crypto_shash_init", r);
515 for (j = 0; j < ic->journal_section_entries; j++) {
516 struct journal_entry *je = access_journal_entry(ic, section, j);
517 r = crypto_shash_update(desc, (__u8 *)&je->u.sector, sizeof je->u.sector);
519 dm_integrity_io_error(ic, "crypto_shash_update", r);
524 size = crypto_shash_digestsize(ic->journal_mac);
526 if (likely(size <= JOURNAL_MAC_SIZE)) {
527 r = crypto_shash_final(desc, result);
529 dm_integrity_io_error(ic, "crypto_shash_final", r);
532 memset(result + size, 0, JOURNAL_MAC_SIZE - size);
535 r = crypto_shash_final(desc, digest);
537 dm_integrity_io_error(ic, "crypto_shash_final", r);
540 memcpy(result, digest, JOURNAL_MAC_SIZE);
545 memset(result, 0, JOURNAL_MAC_SIZE);
548 static void rw_section_mac(struct dm_integrity_c *ic, unsigned section, bool wr)
550 __u8 result[JOURNAL_MAC_SIZE];
553 if (!ic->journal_mac)
556 section_mac(ic, section, result);
558 for (j = 0; j < JOURNAL_BLOCK_SECTORS; j++) {
559 struct journal_sector *js = access_journal(ic, section, j);
562 memcpy(&js->mac, result + (j * JOURNAL_MAC_PER_SECTOR), JOURNAL_MAC_PER_SECTOR);
564 if (memcmp(&js->mac, result + (j * JOURNAL_MAC_PER_SECTOR), JOURNAL_MAC_PER_SECTOR))
565 dm_integrity_io_error(ic, "journal mac", -EILSEQ);
570 static void complete_journal_op(void *context)
572 struct journal_completion *comp = context;
573 BUG_ON(!atomic_read(&comp->in_flight));
574 if (likely(atomic_dec_and_test(&comp->in_flight)))
575 complete(&comp->comp);
578 static void xor_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
579 unsigned n_sections, struct journal_completion *comp)
581 struct async_submit_ctl submit;
582 size_t n_bytes = (size_t)(n_sections * ic->journal_section_sectors) << SECTOR_SHIFT;
583 unsigned pl_index, pl_offset, section_index;
584 struct page_list *source_pl, *target_pl;
586 if (likely(encrypt)) {
587 source_pl = ic->journal;
588 target_pl = ic->journal_io;
590 source_pl = ic->journal_io;
591 target_pl = ic->journal;
594 page_list_location(ic, section, 0, &pl_index, &pl_offset);
596 atomic_add(roundup(pl_offset + n_bytes, PAGE_SIZE) >> PAGE_SHIFT, &comp->in_flight);
598 init_async_submit(&submit, ASYNC_TX_XOR_ZERO_DST, NULL, complete_journal_op, comp, NULL);
600 section_index = pl_index;
604 struct page *src_pages[2];
605 struct page *dst_page;
607 while (unlikely(pl_index == section_index)) {
610 rw_section_mac(ic, section, true);
615 page_list_location(ic, section, 0, §ion_index, &dummy);
618 this_step = min(n_bytes, (size_t)PAGE_SIZE - pl_offset);
619 dst_page = target_pl[pl_index].page;
620 src_pages[0] = source_pl[pl_index].page;
621 src_pages[1] = ic->journal_xor[pl_index].page;
623 async_xor(dst_page, src_pages, pl_offset, 2, this_step, &submit);
627 n_bytes -= this_step;
632 async_tx_issue_pending_all();
635 static void complete_journal_encrypt(struct crypto_async_request *req, int err)
637 struct journal_completion *comp = req->data;
639 if (likely(err == -EINPROGRESS)) {
640 complete(&comp->ic->crypto_backoff);
643 dm_integrity_io_error(comp->ic, "asynchronous encrypt", err);
645 complete_journal_op(comp);
648 static bool do_crypt(bool encrypt, struct skcipher_request *req, struct journal_completion *comp)
651 skcipher_request_set_callback(req, CRYPTO_TFM_REQ_MAY_BACKLOG | CRYPTO_TFM_REQ_MAY_SLEEP,
652 complete_journal_encrypt, comp);
654 r = crypto_skcipher_encrypt(req);
656 r = crypto_skcipher_decrypt(req);
659 if (likely(r == -EINPROGRESS))
661 if (likely(r == -EBUSY)) {
662 wait_for_completion(&comp->ic->crypto_backoff);
663 reinit_completion(&comp->ic->crypto_backoff);
666 dm_integrity_io_error(comp->ic, "encrypt", r);
670 static void crypt_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
671 unsigned n_sections, struct journal_completion *comp)
673 struct scatterlist **source_sg;
674 struct scatterlist **target_sg;
676 atomic_add(2, &comp->in_flight);
678 if (likely(encrypt)) {
679 source_sg = ic->journal_scatterlist;
680 target_sg = ic->journal_io_scatterlist;
682 source_sg = ic->journal_io_scatterlist;
683 target_sg = ic->journal_scatterlist;
687 struct skcipher_request *req;
692 rw_section_mac(ic, section, true);
694 req = ic->sk_requests[section];
695 ivsize = crypto_skcipher_ivsize(ic->journal_crypt);
698 memcpy(iv, iv + ivsize, ivsize);
700 req->src = source_sg[section];
701 req->dst = target_sg[section];
703 if (unlikely(do_crypt(encrypt, req, comp)))
704 atomic_inc(&comp->in_flight);
708 } while (n_sections);
710 atomic_dec(&comp->in_flight);
711 complete_journal_op(comp);
714 static void encrypt_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
715 unsigned n_sections, struct journal_completion *comp)
718 return xor_journal(ic, encrypt, section, n_sections, comp);
720 return crypt_journal(ic, encrypt, section, n_sections, comp);
723 static void complete_journal_io(unsigned long error, void *context)
725 struct journal_completion *comp = context;
726 if (unlikely(error != 0))
727 dm_integrity_io_error(comp->ic, "writing journal", -EIO);
728 complete_journal_op(comp);
731 static void rw_journal(struct dm_integrity_c *ic, int op, int op_flags, unsigned section,
732 unsigned n_sections, struct journal_completion *comp)
734 struct dm_io_request io_req;
735 struct dm_io_region io_loc;
736 unsigned sector, n_sectors, pl_index, pl_offset;
739 if (unlikely(dm_integrity_failed(ic))) {
741 complete_journal_io(-1UL, comp);
745 sector = section * ic->journal_section_sectors;
746 n_sectors = n_sections * ic->journal_section_sectors;
748 pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
749 pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
752 io_req.bi_op_flags = op_flags;
753 io_req.mem.type = DM_IO_PAGE_LIST;
755 io_req.mem.ptr.pl = &ic->journal_io[pl_index];
757 io_req.mem.ptr.pl = &ic->journal[pl_index];
758 io_req.mem.offset = pl_offset;
759 if (likely(comp != NULL)) {
760 io_req.notify.fn = complete_journal_io;
761 io_req.notify.context = comp;
763 io_req.notify.fn = NULL;
765 io_req.client = ic->io;
766 io_loc.bdev = ic->dev->bdev;
767 io_loc.sector = ic->start + SB_SECTORS + sector;
768 io_loc.count = n_sectors;
770 r = dm_io(&io_req, 1, &io_loc, NULL);
772 dm_integrity_io_error(ic, op == REQ_OP_READ ? "reading journal" : "writing journal", r);
774 WARN_ONCE(1, "asynchronous dm_io failed: %d", r);
775 complete_journal_io(-1UL, comp);
780 static void write_journal(struct dm_integrity_c *ic, unsigned commit_start, unsigned commit_sections)
782 struct journal_completion io_comp;
783 struct journal_completion crypt_comp_1;
784 struct journal_completion crypt_comp_2;
788 init_completion(&io_comp.comp);
790 if (commit_start + commit_sections <= ic->journal_sections) {
791 io_comp.in_flight = (atomic_t)ATOMIC_INIT(1);
792 if (ic->journal_io) {
793 crypt_comp_1.ic = ic;
794 init_completion(&crypt_comp_1.comp);
795 crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
796 encrypt_journal(ic, true, commit_start, commit_sections, &crypt_comp_1);
797 wait_for_completion_io(&crypt_comp_1.comp);
799 for (i = 0; i < commit_sections; i++)
800 rw_section_mac(ic, commit_start + i, true);
802 rw_journal(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, commit_start,
803 commit_sections, &io_comp);
806 io_comp.in_flight = (atomic_t)ATOMIC_INIT(2);
807 to_end = ic->journal_sections - commit_start;
808 if (ic->journal_io) {
809 crypt_comp_1.ic = ic;
810 init_completion(&crypt_comp_1.comp);
811 crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
812 encrypt_journal(ic, true, commit_start, to_end, &crypt_comp_1);
813 if (try_wait_for_completion(&crypt_comp_1.comp)) {
814 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
815 reinit_completion(&crypt_comp_1.comp);
816 crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
817 encrypt_journal(ic, true, 0, commit_sections - to_end, &crypt_comp_1);
818 wait_for_completion_io(&crypt_comp_1.comp);
820 crypt_comp_2.ic = ic;
821 init_completion(&crypt_comp_2.comp);
822 crypt_comp_2.in_flight = (atomic_t)ATOMIC_INIT(0);
823 encrypt_journal(ic, true, 0, commit_sections - to_end, &crypt_comp_2);
824 wait_for_completion_io(&crypt_comp_1.comp);
825 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
826 wait_for_completion_io(&crypt_comp_2.comp);
829 for (i = 0; i < to_end; i++)
830 rw_section_mac(ic, commit_start + i, true);
831 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
832 for (i = 0; i < commit_sections - to_end; i++)
833 rw_section_mac(ic, i, true);
835 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, 0, commit_sections - to_end, &io_comp);
838 wait_for_completion_io(&io_comp.comp);
841 static void copy_from_journal(struct dm_integrity_c *ic, unsigned section, unsigned offset,
842 unsigned n_sectors, sector_t target, io_notify_fn fn, void *data)
844 struct dm_io_request io_req;
845 struct dm_io_region io_loc;
847 unsigned sector, pl_index, pl_offset;
849 BUG_ON((target | n_sectors | offset) & (unsigned)(ic->sectors_per_block - 1));
851 if (unlikely(dm_integrity_failed(ic))) {
856 sector = section * ic->journal_section_sectors + JOURNAL_BLOCK_SECTORS + offset;
858 pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
859 pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
861 io_req.bi_op = REQ_OP_WRITE;
862 io_req.bi_op_flags = 0;
863 io_req.mem.type = DM_IO_PAGE_LIST;
864 io_req.mem.ptr.pl = &ic->journal[pl_index];
865 io_req.mem.offset = pl_offset;
866 io_req.notify.fn = fn;
867 io_req.notify.context = data;
868 io_req.client = ic->io;
869 io_loc.bdev = ic->dev->bdev;
870 io_loc.sector = target;
871 io_loc.count = n_sectors;
873 r = dm_io(&io_req, 1, &io_loc, NULL);
875 WARN_ONCE(1, "asynchronous dm_io failed: %d", r);
880 static bool ranges_overlap(struct dm_integrity_range *range1, struct dm_integrity_range *range2)
882 return range1->logical_sector < range2->logical_sector + range2->n_sectors &&
883 range2->logical_sector + range2->n_sectors > range2->logical_sector;
886 static bool add_new_range(struct dm_integrity_c *ic, struct dm_integrity_range *new_range, bool check_waiting)
888 struct rb_node **n = &ic->in_progress.rb_node;
889 struct rb_node *parent;
891 BUG_ON((new_range->logical_sector | new_range->n_sectors) & (unsigned)(ic->sectors_per_block - 1));
893 if (likely(check_waiting)) {
894 struct dm_integrity_range *range;
895 list_for_each_entry(range, &ic->wait_list, wait_entry) {
896 if (unlikely(ranges_overlap(range, new_range)))
904 struct dm_integrity_range *range = container_of(*n, struct dm_integrity_range, node);
907 if (new_range->logical_sector + new_range->n_sectors <= range->logical_sector) {
908 n = &range->node.rb_left;
909 } else if (new_range->logical_sector >= range->logical_sector + range->n_sectors) {
910 n = &range->node.rb_right;
916 rb_link_node(&new_range->node, parent, n);
917 rb_insert_color(&new_range->node, &ic->in_progress);
922 static void remove_range_unlocked(struct dm_integrity_c *ic, struct dm_integrity_range *range)
924 rb_erase(&range->node, &ic->in_progress);
925 while (unlikely(!list_empty(&ic->wait_list))) {
926 struct dm_integrity_range *last_range =
927 list_first_entry(&ic->wait_list, struct dm_integrity_range, wait_entry);
928 struct task_struct *last_range_task;
929 if (!ranges_overlap(range, last_range))
931 last_range_task = last_range->task;
932 list_del(&last_range->wait_entry);
933 if (!add_new_range(ic, last_range, false)) {
934 last_range->task = last_range_task;
935 list_add(&last_range->wait_entry, &ic->wait_list);
938 last_range->waiting = false;
939 wake_up_process(last_range_task);
943 static void remove_range(struct dm_integrity_c *ic, struct dm_integrity_range *range)
947 spin_lock_irqsave(&ic->endio_wait.lock, flags);
948 remove_range_unlocked(ic, range);
949 spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
952 static void wait_and_add_new_range(struct dm_integrity_c *ic, struct dm_integrity_range *new_range)
954 new_range->waiting = true;
955 list_add_tail(&new_range->wait_entry, &ic->wait_list);
956 new_range->task = current;
958 __set_current_state(TASK_UNINTERRUPTIBLE);
959 spin_unlock_irq(&ic->endio_wait.lock);
961 spin_lock_irq(&ic->endio_wait.lock);
962 } while (unlikely(new_range->waiting));
965 static void init_journal_node(struct journal_node *node)
967 RB_CLEAR_NODE(&node->node);
968 node->sector = (sector_t)-1;
971 static void add_journal_node(struct dm_integrity_c *ic, struct journal_node *node, sector_t sector)
973 struct rb_node **link;
974 struct rb_node *parent;
976 node->sector = sector;
977 BUG_ON(!RB_EMPTY_NODE(&node->node));
979 link = &ic->journal_tree_root.rb_node;
983 struct journal_node *j;
985 j = container_of(parent, struct journal_node, node);
986 if (sector < j->sector)
987 link = &j->node.rb_left;
989 link = &j->node.rb_right;
992 rb_link_node(&node->node, parent, link);
993 rb_insert_color(&node->node, &ic->journal_tree_root);
996 static void remove_journal_node(struct dm_integrity_c *ic, struct journal_node *node)
998 BUG_ON(RB_EMPTY_NODE(&node->node));
999 rb_erase(&node->node, &ic->journal_tree_root);
1000 init_journal_node(node);
1003 #define NOT_FOUND (-1U)
1005 static unsigned find_journal_node(struct dm_integrity_c *ic, sector_t sector, sector_t *next_sector)
1007 struct rb_node *n = ic->journal_tree_root.rb_node;
1008 unsigned found = NOT_FOUND;
1009 *next_sector = (sector_t)-1;
1011 struct journal_node *j = container_of(n, struct journal_node, node);
1012 if (sector == j->sector) {
1013 found = j - ic->journal_tree;
1015 if (sector < j->sector) {
1016 *next_sector = j->sector;
1017 n = j->node.rb_left;
1019 n = j->node.rb_right;
1026 static bool test_journal_node(struct dm_integrity_c *ic, unsigned pos, sector_t sector)
1028 struct journal_node *node, *next_node;
1029 struct rb_node *next;
1031 if (unlikely(pos >= ic->journal_entries))
1033 node = &ic->journal_tree[pos];
1034 if (unlikely(RB_EMPTY_NODE(&node->node)))
1036 if (unlikely(node->sector != sector))
1039 next = rb_next(&node->node);
1040 if (unlikely(!next))
1043 next_node = container_of(next, struct journal_node, node);
1044 return next_node->sector != sector;
1047 static bool find_newer_committed_node(struct dm_integrity_c *ic, struct journal_node *node)
1049 struct rb_node *next;
1050 struct journal_node *next_node;
1051 unsigned next_section;
1053 BUG_ON(RB_EMPTY_NODE(&node->node));
1055 next = rb_next(&node->node);
1056 if (unlikely(!next))
1059 next_node = container_of(next, struct journal_node, node);
1061 if (next_node->sector != node->sector)
1064 next_section = (unsigned)(next_node - ic->journal_tree) / ic->journal_section_entries;
1065 if (next_section >= ic->committed_section &&
1066 next_section < ic->committed_section + ic->n_committed_sections)
1068 if (next_section + ic->journal_sections < ic->committed_section + ic->n_committed_sections)
1078 static int dm_integrity_rw_tag(struct dm_integrity_c *ic, unsigned char *tag, sector_t *metadata_block,
1079 unsigned *metadata_offset, unsigned total_size, int op)
1082 unsigned char *data, *dp;
1083 struct dm_buffer *b;
1087 r = dm_integrity_failed(ic);
1091 data = dm_bufio_read(ic->bufio, *metadata_block, &b);
1092 if (unlikely(IS_ERR(data)))
1093 return PTR_ERR(data);
1095 to_copy = min((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - *metadata_offset, total_size);
1096 dp = data + *metadata_offset;
1097 if (op == TAG_READ) {
1098 memcpy(tag, dp, to_copy);
1099 } else if (op == TAG_WRITE) {
1100 memcpy(dp, tag, to_copy);
1101 dm_bufio_mark_partial_buffer_dirty(b, *metadata_offset, *metadata_offset + to_copy);
1103 /* e.g.: op == TAG_CMP */
1104 if (unlikely(memcmp(dp, tag, to_copy))) {
1107 for (i = 0; i < to_copy; i++) {
1108 if (dp[i] != tag[i])
1112 dm_bufio_release(b);
1116 dm_bufio_release(b);
1119 *metadata_offset += to_copy;
1120 if (unlikely(*metadata_offset == 1U << SECTOR_SHIFT << ic->log2_buffer_sectors)) {
1121 (*metadata_block)++;
1122 *metadata_offset = 0;
1124 total_size -= to_copy;
1125 } while (unlikely(total_size));
1130 static void dm_integrity_flush_buffers(struct dm_integrity_c *ic)
1133 r = dm_bufio_write_dirty_buffers(ic->bufio);
1135 dm_integrity_io_error(ic, "writing tags", r);
1138 static void sleep_on_endio_wait(struct dm_integrity_c *ic)
1140 DECLARE_WAITQUEUE(wait, current);
1141 __add_wait_queue(&ic->endio_wait, &wait);
1142 __set_current_state(TASK_UNINTERRUPTIBLE);
1143 spin_unlock_irq(&ic->endio_wait.lock);
1145 spin_lock_irq(&ic->endio_wait.lock);
1146 __remove_wait_queue(&ic->endio_wait, &wait);
1149 static void autocommit_fn(struct timer_list *t)
1151 struct dm_integrity_c *ic = from_timer(ic, t, autocommit_timer);
1153 if (likely(!dm_integrity_failed(ic)))
1154 queue_work(ic->commit_wq, &ic->commit_work);
1157 static void schedule_autocommit(struct dm_integrity_c *ic)
1159 if (!timer_pending(&ic->autocommit_timer))
1160 mod_timer(&ic->autocommit_timer, jiffies + ic->autocommit_jiffies);
1163 static void submit_flush_bio(struct dm_integrity_c *ic, struct dm_integrity_io *dio)
1166 unsigned long flags;
1168 spin_lock_irqsave(&ic->endio_wait.lock, flags);
1169 bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1170 bio_list_add(&ic->flush_bio_list, bio);
1171 spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
1173 queue_work(ic->commit_wq, &ic->commit_work);
1176 static void do_endio(struct dm_integrity_c *ic, struct bio *bio)
1178 int r = dm_integrity_failed(ic);
1179 if (unlikely(r) && !bio->bi_status)
1180 bio->bi_status = errno_to_blk_status(r);
1184 static void do_endio_flush(struct dm_integrity_c *ic, struct dm_integrity_io *dio)
1186 struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1188 if (unlikely(dio->fua) && likely(!bio->bi_status) && likely(!dm_integrity_failed(ic)))
1189 submit_flush_bio(ic, dio);
1194 static void dec_in_flight(struct dm_integrity_io *dio)
1196 if (atomic_dec_and_test(&dio->in_flight)) {
1197 struct dm_integrity_c *ic = dio->ic;
1200 remove_range(ic, &dio->range);
1202 if (unlikely(dio->write))
1203 schedule_autocommit(ic);
1205 bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1207 if (unlikely(dio->bi_status) && !bio->bi_status)
1208 bio->bi_status = dio->bi_status;
1209 if (likely(!bio->bi_status) && unlikely(bio_sectors(bio) != dio->range.n_sectors)) {
1210 dio->range.logical_sector += dio->range.n_sectors;
1211 bio_advance(bio, dio->range.n_sectors << SECTOR_SHIFT);
1212 INIT_WORK(&dio->work, integrity_bio_wait);
1213 queue_work(ic->wait_wq, &dio->work);
1216 do_endio_flush(ic, dio);
1220 static void integrity_end_io(struct bio *bio)
1222 struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
1224 bio->bi_iter = dio->orig_bi_iter;
1225 bio->bi_disk = dio->orig_bi_disk;
1226 bio->bi_partno = dio->orig_bi_partno;
1227 if (dio->orig_bi_integrity) {
1228 bio->bi_integrity = dio->orig_bi_integrity;
1229 bio->bi_opf |= REQ_INTEGRITY;
1231 bio->bi_end_io = dio->orig_bi_end_io;
1233 if (dio->completion)
1234 complete(dio->completion);
1239 static void integrity_sector_checksum(struct dm_integrity_c *ic, sector_t sector,
1240 const char *data, char *result)
1242 __u64 sector_le = cpu_to_le64(sector);
1243 SHASH_DESC_ON_STACK(req, ic->internal_hash);
1245 unsigned digest_size;
1247 req->tfm = ic->internal_hash;
1250 r = crypto_shash_init(req);
1251 if (unlikely(r < 0)) {
1252 dm_integrity_io_error(ic, "crypto_shash_init", r);
1256 r = crypto_shash_update(req, (const __u8 *)§or_le, sizeof sector_le);
1257 if (unlikely(r < 0)) {
1258 dm_integrity_io_error(ic, "crypto_shash_update", r);
1262 r = crypto_shash_update(req, data, ic->sectors_per_block << SECTOR_SHIFT);
1263 if (unlikely(r < 0)) {
1264 dm_integrity_io_error(ic, "crypto_shash_update", r);
1268 r = crypto_shash_final(req, result);
1269 if (unlikely(r < 0)) {
1270 dm_integrity_io_error(ic, "crypto_shash_final", r);
1274 digest_size = crypto_shash_digestsize(ic->internal_hash);
1275 if (unlikely(digest_size < ic->tag_size))
1276 memset(result + digest_size, 0, ic->tag_size - digest_size);
1281 /* this shouldn't happen anyway, the hash functions have no reason to fail */
1282 get_random_bytes(result, ic->tag_size);
1285 static void integrity_metadata(struct work_struct *w)
1287 struct dm_integrity_io *dio = container_of(w, struct dm_integrity_io, work);
1288 struct dm_integrity_c *ic = dio->ic;
1292 if (ic->internal_hash) {
1293 struct bvec_iter iter;
1295 unsigned digest_size = crypto_shash_digestsize(ic->internal_hash);
1296 struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1298 unsigned extra_space = unlikely(digest_size > ic->tag_size) ? digest_size - ic->tag_size : 0;
1299 char checksums_onstack[ic->tag_size + extra_space];
1300 unsigned sectors_to_process = dio->range.n_sectors;
1301 sector_t sector = dio->range.logical_sector;
1303 if (unlikely(ic->mode == 'R'))
1306 checksums = kmalloc((PAGE_SIZE >> SECTOR_SHIFT >> ic->sb->log2_sectors_per_block) * ic->tag_size + extra_space,
1307 GFP_NOIO | __GFP_NORETRY | __GFP_NOWARN);
1309 checksums = checksums_onstack;
1311 __bio_for_each_segment(bv, bio, iter, dio->orig_bi_iter) {
1313 char *mem, *checksums_ptr;
1316 mem = (char *)kmap_atomic(bv.bv_page) + bv.bv_offset;
1318 checksums_ptr = checksums;
1320 integrity_sector_checksum(ic, sector, mem + pos, checksums_ptr);
1321 checksums_ptr += ic->tag_size;
1322 sectors_to_process -= ic->sectors_per_block;
1323 pos += ic->sectors_per_block << SECTOR_SHIFT;
1324 sector += ic->sectors_per_block;
1325 } while (pos < bv.bv_len && sectors_to_process && checksums != checksums_onstack);
1328 r = dm_integrity_rw_tag(ic, checksums, &dio->metadata_block, &dio->metadata_offset,
1329 checksums_ptr - checksums, !dio->write ? TAG_CMP : TAG_WRITE);
1332 DMERR("Checksum failed at sector 0x%llx",
1333 (unsigned long long)(sector - ((r + ic->tag_size - 1) / ic->tag_size)));
1335 atomic64_inc(&ic->number_of_mismatches);
1337 if (likely(checksums != checksums_onstack))
1342 if (!sectors_to_process)
1345 if (unlikely(pos < bv.bv_len)) {
1346 bv.bv_offset += pos;
1352 if (likely(checksums != checksums_onstack))
1355 struct bio_integrity_payload *bip = dio->orig_bi_integrity;
1359 struct bvec_iter iter;
1360 unsigned data_to_process = dio->range.n_sectors;
1361 sector_to_block(ic, data_to_process);
1362 data_to_process *= ic->tag_size;
1364 bip_for_each_vec(biv, bip, iter) {
1368 BUG_ON(PageHighMem(biv.bv_page));
1369 tag = lowmem_page_address(biv.bv_page) + biv.bv_offset;
1370 this_len = min(biv.bv_len, data_to_process);
1371 r = dm_integrity_rw_tag(ic, tag, &dio->metadata_block, &dio->metadata_offset,
1372 this_len, !dio->write ? TAG_READ : TAG_WRITE);
1375 data_to_process -= this_len;
1376 if (!data_to_process)
1385 dio->bi_status = errno_to_blk_status(r);
1389 static int dm_integrity_map(struct dm_target *ti, struct bio *bio)
1391 struct dm_integrity_c *ic = ti->private;
1392 struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
1393 struct bio_integrity_payload *bip;
1395 sector_t area, offset;
1400 if (unlikely(bio->bi_opf & REQ_PREFLUSH)) {
1401 submit_flush_bio(ic, dio);
1402 return DM_MAPIO_SUBMITTED;
1405 dio->range.logical_sector = dm_target_offset(ti, bio->bi_iter.bi_sector);
1406 dio->write = bio_op(bio) == REQ_OP_WRITE;
1407 dio->fua = dio->write && bio->bi_opf & REQ_FUA;
1408 if (unlikely(dio->fua)) {
1410 * Don't pass down the FUA flag because we have to flush
1411 * disk cache anyway.
1413 bio->bi_opf &= ~REQ_FUA;
1415 if (unlikely(dio->range.logical_sector + bio_sectors(bio) > ic->provided_data_sectors)) {
1416 DMERR("Too big sector number: 0x%llx + 0x%x > 0x%llx",
1417 (unsigned long long)dio->range.logical_sector, bio_sectors(bio),
1418 (unsigned long long)ic->provided_data_sectors);
1419 return DM_MAPIO_KILL;
1421 if (unlikely((dio->range.logical_sector | bio_sectors(bio)) & (unsigned)(ic->sectors_per_block - 1))) {
1422 DMERR("Bio not aligned on %u sectors: 0x%llx, 0x%x",
1423 ic->sectors_per_block,
1424 (unsigned long long)dio->range.logical_sector, bio_sectors(bio));
1425 return DM_MAPIO_KILL;
1428 if (ic->sectors_per_block > 1) {
1429 struct bvec_iter iter;
1431 bio_for_each_segment(bv, bio, iter) {
1432 if (unlikely(bv.bv_len & ((ic->sectors_per_block << SECTOR_SHIFT) - 1))) {
1433 DMERR("Bio vector (%u,%u) is not aligned on %u-sector boundary",
1434 bv.bv_offset, bv.bv_len, ic->sectors_per_block);
1435 return DM_MAPIO_KILL;
1440 bip = bio_integrity(bio);
1441 if (!ic->internal_hash) {
1443 unsigned wanted_tag_size = bio_sectors(bio) >> ic->sb->log2_sectors_per_block;
1444 if (ic->log2_tag_size >= 0)
1445 wanted_tag_size <<= ic->log2_tag_size;
1447 wanted_tag_size *= ic->tag_size;
1448 if (unlikely(wanted_tag_size != bip->bip_iter.bi_size)) {
1449 DMERR("Invalid integrity data size %u, expected %u", bip->bip_iter.bi_size, wanted_tag_size);
1450 return DM_MAPIO_KILL;
1454 if (unlikely(bip != NULL)) {
1455 DMERR("Unexpected integrity data when using internal hash");
1456 return DM_MAPIO_KILL;
1460 if (unlikely(ic->mode == 'R') && unlikely(dio->write))
1461 return DM_MAPIO_KILL;
1463 get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
1464 dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset, &dio->metadata_offset);
1465 bio->bi_iter.bi_sector = get_data_sector(ic, area, offset);
1467 dm_integrity_map_continue(dio, true);
1468 return DM_MAPIO_SUBMITTED;
1471 static bool __journal_read_write(struct dm_integrity_io *dio, struct bio *bio,
1472 unsigned journal_section, unsigned journal_entry)
1474 struct dm_integrity_c *ic = dio->ic;
1475 sector_t logical_sector;
1478 logical_sector = dio->range.logical_sector;
1479 n_sectors = dio->range.n_sectors;
1481 struct bio_vec bv = bio_iovec(bio);
1484 if (unlikely(bv.bv_len >> SECTOR_SHIFT > n_sectors))
1485 bv.bv_len = n_sectors << SECTOR_SHIFT;
1486 n_sectors -= bv.bv_len >> SECTOR_SHIFT;
1487 bio_advance_iter(bio, &bio->bi_iter, bv.bv_len);
1489 mem = kmap_atomic(bv.bv_page);
1490 if (likely(dio->write))
1491 flush_dcache_page(bv.bv_page);
1494 struct journal_entry *je = access_journal_entry(ic, journal_section, journal_entry);
1496 if (unlikely(!dio->write)) {
1497 struct journal_sector *js;
1501 if (unlikely(journal_entry_is_inprogress(je))) {
1502 flush_dcache_page(bv.bv_page);
1505 __io_wait_event(ic->copy_to_journal_wait, !journal_entry_is_inprogress(je));
1509 BUG_ON(journal_entry_get_sector(je) != logical_sector);
1510 js = access_journal_data(ic, journal_section, journal_entry);
1511 mem_ptr = mem + bv.bv_offset;
1514 memcpy(mem_ptr, js, JOURNAL_SECTOR_DATA);
1515 *(commit_id_t *)(mem_ptr + JOURNAL_SECTOR_DATA) = je->last_bytes[s];
1517 mem_ptr += 1 << SECTOR_SHIFT;
1518 } while (++s < ic->sectors_per_block);
1519 #ifdef INTERNAL_VERIFY
1520 if (ic->internal_hash) {
1521 char checksums_onstack[max(crypto_shash_digestsize(ic->internal_hash), ic->tag_size)];
1523 integrity_sector_checksum(ic, logical_sector, mem + bv.bv_offset, checksums_onstack);
1524 if (unlikely(memcmp(checksums_onstack, journal_entry_tag(ic, je), ic->tag_size))) {
1525 DMERR("Checksum failed when reading from journal, at sector 0x%llx",
1526 (unsigned long long)logical_sector);
1532 if (!ic->internal_hash) {
1533 struct bio_integrity_payload *bip = bio_integrity(bio);
1534 unsigned tag_todo = ic->tag_size;
1535 char *tag_ptr = journal_entry_tag(ic, je);
1538 struct bio_vec biv = bvec_iter_bvec(bip->bip_vec, bip->bip_iter);
1539 unsigned tag_now = min(biv.bv_len, tag_todo);
1541 BUG_ON(PageHighMem(biv.bv_page));
1542 tag_addr = lowmem_page_address(biv.bv_page) + biv.bv_offset;
1543 if (likely(dio->write))
1544 memcpy(tag_ptr, tag_addr, tag_now);
1546 memcpy(tag_addr, tag_ptr, tag_now);
1547 bvec_iter_advance(bip->bip_vec, &bip->bip_iter, tag_now);
1549 tag_todo -= tag_now;
1550 } while (unlikely(tag_todo)); else {
1551 if (likely(dio->write))
1552 memset(tag_ptr, 0, tag_todo);
1556 if (likely(dio->write)) {
1557 struct journal_sector *js;
1560 js = access_journal_data(ic, journal_section, journal_entry);
1561 memcpy(js, mem + bv.bv_offset, ic->sectors_per_block << SECTOR_SHIFT);
1565 je->last_bytes[s] = js[s].commit_id;
1566 } while (++s < ic->sectors_per_block);
1568 if (ic->internal_hash) {
1569 unsigned digest_size = crypto_shash_digestsize(ic->internal_hash);
1570 if (unlikely(digest_size > ic->tag_size)) {
1571 char checksums_onstack[digest_size];
1572 integrity_sector_checksum(ic, logical_sector, (char *)js, checksums_onstack);
1573 memcpy(journal_entry_tag(ic, je), checksums_onstack, ic->tag_size);
1575 integrity_sector_checksum(ic, logical_sector, (char *)js, journal_entry_tag(ic, je));
1578 journal_entry_set_sector(je, logical_sector);
1580 logical_sector += ic->sectors_per_block;
1583 if (unlikely(journal_entry == ic->journal_section_entries)) {
1586 wraparound_section(ic, &journal_section);
1589 bv.bv_offset += ic->sectors_per_block << SECTOR_SHIFT;
1590 } while (bv.bv_len -= ic->sectors_per_block << SECTOR_SHIFT);
1592 if (unlikely(!dio->write))
1593 flush_dcache_page(bv.bv_page);
1595 } while (n_sectors);
1597 if (likely(dio->write)) {
1599 if (unlikely(waitqueue_active(&ic->copy_to_journal_wait)))
1600 wake_up(&ic->copy_to_journal_wait);
1601 if (READ_ONCE(ic->free_sectors) <= ic->free_sectors_threshold) {
1602 queue_work(ic->commit_wq, &ic->commit_work);
1604 schedule_autocommit(ic);
1607 remove_range(ic, &dio->range);
1610 if (unlikely(bio->bi_iter.bi_size)) {
1611 sector_t area, offset;
1613 dio->range.logical_sector = logical_sector;
1614 get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
1615 dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset, &dio->metadata_offset);
1622 static void dm_integrity_map_continue(struct dm_integrity_io *dio, bool from_map)
1624 struct dm_integrity_c *ic = dio->ic;
1625 struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1626 unsigned journal_section, journal_entry;
1627 unsigned journal_read_pos;
1628 struct completion read_comp;
1629 bool need_sync_io = ic->internal_hash && !dio->write;
1631 if (need_sync_io && from_map) {
1632 INIT_WORK(&dio->work, integrity_bio_wait);
1633 queue_work(ic->metadata_wq, &dio->work);
1638 spin_lock_irq(&ic->endio_wait.lock);
1640 if (unlikely(dm_integrity_failed(ic))) {
1641 spin_unlock_irq(&ic->endio_wait.lock);
1645 dio->range.n_sectors = bio_sectors(bio);
1646 journal_read_pos = NOT_FOUND;
1647 if (likely(ic->mode == 'J')) {
1649 unsigned next_entry, i, pos;
1650 unsigned ws, we, range_sectors;
1652 dio->range.n_sectors = min(dio->range.n_sectors,
1653 ic->free_sectors << ic->sb->log2_sectors_per_block);
1654 if (unlikely(!dio->range.n_sectors)) {
1656 goto offload_to_thread;
1657 sleep_on_endio_wait(ic);
1660 range_sectors = dio->range.n_sectors >> ic->sb->log2_sectors_per_block;
1661 ic->free_sectors -= range_sectors;
1662 journal_section = ic->free_section;
1663 journal_entry = ic->free_section_entry;
1665 next_entry = ic->free_section_entry + range_sectors;
1666 ic->free_section_entry = next_entry % ic->journal_section_entries;
1667 ic->free_section += next_entry / ic->journal_section_entries;
1668 ic->n_uncommitted_sections += next_entry / ic->journal_section_entries;
1669 wraparound_section(ic, &ic->free_section);
1671 pos = journal_section * ic->journal_section_entries + journal_entry;
1672 ws = journal_section;
1676 struct journal_entry *je;
1678 add_journal_node(ic, &ic->journal_tree[pos], dio->range.logical_sector + i);
1680 if (unlikely(pos >= ic->journal_entries))
1683 je = access_journal_entry(ic, ws, we);
1684 BUG_ON(!journal_entry_is_unused(je));
1685 journal_entry_set_inprogress(je);
1687 if (unlikely(we == ic->journal_section_entries)) {
1690 wraparound_section(ic, &ws);
1692 } while ((i += ic->sectors_per_block) < dio->range.n_sectors);
1694 spin_unlock_irq(&ic->endio_wait.lock);
1695 goto journal_read_write;
1697 sector_t next_sector;
1698 journal_read_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
1699 if (likely(journal_read_pos == NOT_FOUND)) {
1700 if (unlikely(dio->range.n_sectors > next_sector - dio->range.logical_sector))
1701 dio->range.n_sectors = next_sector - dio->range.logical_sector;
1704 unsigned jp = journal_read_pos + 1;
1705 for (i = ic->sectors_per_block; i < dio->range.n_sectors; i += ic->sectors_per_block, jp++) {
1706 if (!test_journal_node(ic, jp, dio->range.logical_sector + i))
1709 dio->range.n_sectors = i;
1713 if (unlikely(!add_new_range(ic, &dio->range, true))) {
1715 * We must not sleep in the request routine because it could
1716 * stall bios on current->bio_list.
1717 * So, we offload the bio to a workqueue if we have to sleep.
1721 spin_unlock_irq(&ic->endio_wait.lock);
1722 INIT_WORK(&dio->work, integrity_bio_wait);
1723 queue_work(ic->wait_wq, &dio->work);
1726 wait_and_add_new_range(ic, &dio->range);
1728 spin_unlock_irq(&ic->endio_wait.lock);
1730 if (unlikely(journal_read_pos != NOT_FOUND)) {
1731 journal_section = journal_read_pos / ic->journal_section_entries;
1732 journal_entry = journal_read_pos % ic->journal_section_entries;
1733 goto journal_read_write;
1736 dio->in_flight = (atomic_t)ATOMIC_INIT(2);
1739 init_completion(&read_comp);
1740 dio->completion = &read_comp;
1742 dio->completion = NULL;
1744 dio->orig_bi_iter = bio->bi_iter;
1746 dio->orig_bi_disk = bio->bi_disk;
1747 dio->orig_bi_partno = bio->bi_partno;
1748 bio_set_dev(bio, ic->dev->bdev);
1750 dio->orig_bi_integrity = bio_integrity(bio);
1751 bio->bi_integrity = NULL;
1752 bio->bi_opf &= ~REQ_INTEGRITY;
1754 dio->orig_bi_end_io = bio->bi_end_io;
1755 bio->bi_end_io = integrity_end_io;
1757 bio->bi_iter.bi_size = dio->range.n_sectors << SECTOR_SHIFT;
1758 generic_make_request(bio);
1761 wait_for_completion_io(&read_comp);
1762 if (likely(!bio->bi_status))
1763 integrity_metadata(&dio->work);
1768 INIT_WORK(&dio->work, integrity_metadata);
1769 queue_work(ic->metadata_wq, &dio->work);
1775 if (unlikely(__journal_read_write(dio, bio, journal_section, journal_entry)))
1778 do_endio_flush(ic, dio);
1782 static void integrity_bio_wait(struct work_struct *w)
1784 struct dm_integrity_io *dio = container_of(w, struct dm_integrity_io, work);
1786 dm_integrity_map_continue(dio, false);
1789 static void pad_uncommitted(struct dm_integrity_c *ic)
1791 if (ic->free_section_entry) {
1792 ic->free_sectors -= ic->journal_section_entries - ic->free_section_entry;
1793 ic->free_section_entry = 0;
1795 wraparound_section(ic, &ic->free_section);
1796 ic->n_uncommitted_sections++;
1798 WARN_ON(ic->journal_sections * ic->journal_section_entries !=
1799 (ic->n_uncommitted_sections + ic->n_committed_sections) * ic->journal_section_entries + ic->free_sectors);
1802 static void integrity_commit(struct work_struct *w)
1804 struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, commit_work);
1805 unsigned commit_start, commit_sections;
1807 struct bio *flushes;
1809 del_timer(&ic->autocommit_timer);
1811 spin_lock_irq(&ic->endio_wait.lock);
1812 flushes = bio_list_get(&ic->flush_bio_list);
1813 if (unlikely(ic->mode != 'J')) {
1814 spin_unlock_irq(&ic->endio_wait.lock);
1815 dm_integrity_flush_buffers(ic);
1816 goto release_flush_bios;
1819 pad_uncommitted(ic);
1820 commit_start = ic->uncommitted_section;
1821 commit_sections = ic->n_uncommitted_sections;
1822 spin_unlock_irq(&ic->endio_wait.lock);
1824 if (!commit_sections)
1825 goto release_flush_bios;
1828 for (n = 0; n < commit_sections; n++) {
1829 for (j = 0; j < ic->journal_section_entries; j++) {
1830 struct journal_entry *je;
1831 je = access_journal_entry(ic, i, j);
1832 io_wait_event(ic->copy_to_journal_wait, !journal_entry_is_inprogress(je));
1834 for (j = 0; j < ic->journal_section_sectors; j++) {
1835 struct journal_sector *js;
1836 js = access_journal(ic, i, j);
1837 js->commit_id = dm_integrity_commit_id(ic, i, j, ic->commit_seq);
1840 if (unlikely(i >= ic->journal_sections))
1841 ic->commit_seq = next_commit_seq(ic->commit_seq);
1842 wraparound_section(ic, &i);
1846 write_journal(ic, commit_start, commit_sections);
1848 spin_lock_irq(&ic->endio_wait.lock);
1849 ic->uncommitted_section += commit_sections;
1850 wraparound_section(ic, &ic->uncommitted_section);
1851 ic->n_uncommitted_sections -= commit_sections;
1852 ic->n_committed_sections += commit_sections;
1853 spin_unlock_irq(&ic->endio_wait.lock);
1855 if (READ_ONCE(ic->free_sectors) <= ic->free_sectors_threshold)
1856 queue_work(ic->writer_wq, &ic->writer_work);
1860 struct bio *next = flushes->bi_next;
1861 flushes->bi_next = NULL;
1862 do_endio(ic, flushes);
1867 static void complete_copy_from_journal(unsigned long error, void *context)
1869 struct journal_io *io = context;
1870 struct journal_completion *comp = io->comp;
1871 struct dm_integrity_c *ic = comp->ic;
1872 remove_range(ic, &io->range);
1873 mempool_free(io, &ic->journal_io_mempool);
1874 if (unlikely(error != 0))
1875 dm_integrity_io_error(ic, "copying from journal", -EIO);
1876 complete_journal_op(comp);
1879 static void restore_last_bytes(struct dm_integrity_c *ic, struct journal_sector *js,
1880 struct journal_entry *je)
1884 js->commit_id = je->last_bytes[s];
1886 } while (++s < ic->sectors_per_block);
1889 static void do_journal_write(struct dm_integrity_c *ic, unsigned write_start,
1890 unsigned write_sections, bool from_replay)
1893 struct journal_completion comp;
1894 struct blk_plug plug;
1896 blk_start_plug(&plug);
1899 comp.in_flight = (atomic_t)ATOMIC_INIT(1);
1900 init_completion(&comp.comp);
1903 for (n = 0; n < write_sections; n++, i++, wraparound_section(ic, &i)) {
1904 #ifndef INTERNAL_VERIFY
1905 if (unlikely(from_replay))
1907 rw_section_mac(ic, i, false);
1908 for (j = 0; j < ic->journal_section_entries; j++) {
1909 struct journal_entry *je = access_journal_entry(ic, i, j);
1910 sector_t sec, area, offset;
1911 unsigned k, l, next_loop;
1912 sector_t metadata_block;
1913 unsigned metadata_offset;
1914 struct journal_io *io;
1916 if (journal_entry_is_unused(je))
1918 BUG_ON(unlikely(journal_entry_is_inprogress(je)) && !from_replay);
1919 sec = journal_entry_get_sector(je);
1920 if (unlikely(from_replay)) {
1921 if (unlikely(sec & (unsigned)(ic->sectors_per_block - 1))) {
1922 dm_integrity_io_error(ic, "invalid sector in journal", -EIO);
1923 sec &= ~(sector_t)(ic->sectors_per_block - 1);
1926 get_area_and_offset(ic, sec, &area, &offset);
1927 restore_last_bytes(ic, access_journal_data(ic, i, j), je);
1928 for (k = j + 1; k < ic->journal_section_entries; k++) {
1929 struct journal_entry *je2 = access_journal_entry(ic, i, k);
1930 sector_t sec2, area2, offset2;
1931 if (journal_entry_is_unused(je2))
1933 BUG_ON(unlikely(journal_entry_is_inprogress(je2)) && !from_replay);
1934 sec2 = journal_entry_get_sector(je2);
1935 get_area_and_offset(ic, sec2, &area2, &offset2);
1936 if (area2 != area || offset2 != offset + ((k - j) << ic->sb->log2_sectors_per_block))
1938 restore_last_bytes(ic, access_journal_data(ic, i, k), je2);
1942 io = mempool_alloc(&ic->journal_io_mempool, GFP_NOIO);
1944 io->range.logical_sector = sec;
1945 io->range.n_sectors = (k - j) << ic->sb->log2_sectors_per_block;
1947 spin_lock_irq(&ic->endio_wait.lock);
1948 if (unlikely(!add_new_range(ic, &io->range, true)))
1949 wait_and_add_new_range(ic, &io->range);
1951 if (likely(!from_replay)) {
1952 struct journal_node *section_node = &ic->journal_tree[i * ic->journal_section_entries];
1954 /* don't write if there is newer committed sector */
1955 while (j < k && find_newer_committed_node(ic, §ion_node[j])) {
1956 struct journal_entry *je2 = access_journal_entry(ic, i, j);
1958 journal_entry_set_unused(je2);
1959 remove_journal_node(ic, §ion_node[j]);
1961 sec += ic->sectors_per_block;
1962 offset += ic->sectors_per_block;
1964 while (j < k && find_newer_committed_node(ic, §ion_node[k - 1])) {
1965 struct journal_entry *je2 = access_journal_entry(ic, i, k - 1);
1967 journal_entry_set_unused(je2);
1968 remove_journal_node(ic, §ion_node[k - 1]);
1972 remove_range_unlocked(ic, &io->range);
1973 spin_unlock_irq(&ic->endio_wait.lock);
1974 mempool_free(io, &ic->journal_io_mempool);
1977 for (l = j; l < k; l++) {
1978 remove_journal_node(ic, §ion_node[l]);
1981 spin_unlock_irq(&ic->endio_wait.lock);
1983 metadata_block = get_metadata_sector_and_offset(ic, area, offset, &metadata_offset);
1984 for (l = j; l < k; l++) {
1986 struct journal_entry *je2 = access_journal_entry(ic, i, l);
1989 #ifndef INTERNAL_VERIFY
1990 unlikely(from_replay) &&
1992 ic->internal_hash) {
1993 char test_tag[max(crypto_shash_digestsize(ic->internal_hash), ic->tag_size)];
1995 integrity_sector_checksum(ic, sec + ((l - j) << ic->sb->log2_sectors_per_block),
1996 (char *)access_journal_data(ic, i, l), test_tag);
1997 if (unlikely(memcmp(test_tag, journal_entry_tag(ic, je2), ic->tag_size)))
1998 dm_integrity_io_error(ic, "tag mismatch when replaying journal", -EILSEQ);
2001 journal_entry_set_unused(je2);
2002 r = dm_integrity_rw_tag(ic, journal_entry_tag(ic, je2), &metadata_block, &metadata_offset,
2003 ic->tag_size, TAG_WRITE);
2005 dm_integrity_io_error(ic, "reading tags", r);
2009 atomic_inc(&comp.in_flight);
2010 copy_from_journal(ic, i, j << ic->sb->log2_sectors_per_block,
2011 (k - j) << ic->sb->log2_sectors_per_block,
2012 get_data_sector(ic, area, offset),
2013 complete_copy_from_journal, io);
2019 dm_bufio_write_dirty_buffers_async(ic->bufio);
2021 blk_finish_plug(&plug);
2023 complete_journal_op(&comp);
2024 wait_for_completion_io(&comp.comp);
2026 dm_integrity_flush_buffers(ic);
2029 static void integrity_writer(struct work_struct *w)
2031 struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, writer_work);
2032 unsigned write_start, write_sections;
2034 unsigned prev_free_sectors;
2036 /* the following test is not needed, but it tests the replay code */
2037 if (READ_ONCE(ic->suspending))
2040 spin_lock_irq(&ic->endio_wait.lock);
2041 write_start = ic->committed_section;
2042 write_sections = ic->n_committed_sections;
2043 spin_unlock_irq(&ic->endio_wait.lock);
2045 if (!write_sections)
2048 do_journal_write(ic, write_start, write_sections, false);
2050 spin_lock_irq(&ic->endio_wait.lock);
2052 ic->committed_section += write_sections;
2053 wraparound_section(ic, &ic->committed_section);
2054 ic->n_committed_sections -= write_sections;
2056 prev_free_sectors = ic->free_sectors;
2057 ic->free_sectors += write_sections * ic->journal_section_entries;
2058 if (unlikely(!prev_free_sectors))
2059 wake_up_locked(&ic->endio_wait);
2061 spin_unlock_irq(&ic->endio_wait.lock);
2064 static void init_journal(struct dm_integrity_c *ic, unsigned start_section,
2065 unsigned n_sections, unsigned char commit_seq)
2072 for (n = 0; n < n_sections; n++) {
2073 i = start_section + n;
2074 wraparound_section(ic, &i);
2075 for (j = 0; j < ic->journal_section_sectors; j++) {
2076 struct journal_sector *js = access_journal(ic, i, j);
2077 memset(&js->entries, 0, JOURNAL_SECTOR_DATA);
2078 js->commit_id = dm_integrity_commit_id(ic, i, j, commit_seq);
2080 for (j = 0; j < ic->journal_section_entries; j++) {
2081 struct journal_entry *je = access_journal_entry(ic, i, j);
2082 journal_entry_set_unused(je);
2086 write_journal(ic, start_section, n_sections);
2089 static int find_commit_seq(struct dm_integrity_c *ic, unsigned i, unsigned j, commit_id_t id)
2092 for (k = 0; k < N_COMMIT_IDS; k++) {
2093 if (dm_integrity_commit_id(ic, i, j, k) == id)
2096 dm_integrity_io_error(ic, "journal commit id", -EIO);
2100 static void replay_journal(struct dm_integrity_c *ic)
2103 bool used_commit_ids[N_COMMIT_IDS];
2104 unsigned max_commit_id_sections[N_COMMIT_IDS];
2105 unsigned write_start, write_sections;
2106 unsigned continue_section;
2108 unsigned char unused, last_used, want_commit_seq;
2110 if (ic->mode == 'R')
2113 if (ic->journal_uptodate)
2119 if (!ic->just_formatted) {
2120 DEBUG_print("reading journal\n");
2121 rw_journal(ic, REQ_OP_READ, 0, 0, ic->journal_sections, NULL);
2123 DEBUG_bytes(lowmem_page_address(ic->journal_io[0].page), 64, "read journal");
2124 if (ic->journal_io) {
2125 struct journal_completion crypt_comp;
2127 init_completion(&crypt_comp.comp);
2128 crypt_comp.in_flight = (atomic_t)ATOMIC_INIT(0);
2129 encrypt_journal(ic, false, 0, ic->journal_sections, &crypt_comp);
2130 wait_for_completion(&crypt_comp.comp);
2132 DEBUG_bytes(lowmem_page_address(ic->journal[0].page), 64, "decrypted journal");
2135 if (dm_integrity_failed(ic))
2138 journal_empty = true;
2139 memset(used_commit_ids, 0, sizeof used_commit_ids);
2140 memset(max_commit_id_sections, 0, sizeof max_commit_id_sections);
2141 for (i = 0; i < ic->journal_sections; i++) {
2142 for (j = 0; j < ic->journal_section_sectors; j++) {
2144 struct journal_sector *js = access_journal(ic, i, j);
2145 k = find_commit_seq(ic, i, j, js->commit_id);
2148 used_commit_ids[k] = true;
2149 max_commit_id_sections[k] = i;
2151 if (journal_empty) {
2152 for (j = 0; j < ic->journal_section_entries; j++) {
2153 struct journal_entry *je = access_journal_entry(ic, i, j);
2154 if (!journal_entry_is_unused(je)) {
2155 journal_empty = false;
2162 if (!used_commit_ids[N_COMMIT_IDS - 1]) {
2163 unused = N_COMMIT_IDS - 1;
2164 while (unused && !used_commit_ids[unused - 1])
2167 for (unused = 0; unused < N_COMMIT_IDS; unused++)
2168 if (!used_commit_ids[unused])
2170 if (unused == N_COMMIT_IDS) {
2171 dm_integrity_io_error(ic, "journal commit ids", -EIO);
2175 DEBUG_print("first unused commit seq %d [%d,%d,%d,%d]\n",
2176 unused, used_commit_ids[0], used_commit_ids[1],
2177 used_commit_ids[2], used_commit_ids[3]);
2179 last_used = prev_commit_seq(unused);
2180 want_commit_seq = prev_commit_seq(last_used);
2182 if (!used_commit_ids[want_commit_seq] && used_commit_ids[prev_commit_seq(want_commit_seq)])
2183 journal_empty = true;
2185 write_start = max_commit_id_sections[last_used] + 1;
2186 if (unlikely(write_start >= ic->journal_sections))
2187 want_commit_seq = next_commit_seq(want_commit_seq);
2188 wraparound_section(ic, &write_start);
2191 for (write_sections = 0; write_sections < ic->journal_sections; write_sections++) {
2192 for (j = 0; j < ic->journal_section_sectors; j++) {
2193 struct journal_sector *js = access_journal(ic, i, j);
2195 if (js->commit_id != dm_integrity_commit_id(ic, i, j, want_commit_seq)) {
2197 * This could be caused by crash during writing.
2198 * We won't replay the inconsistent part of the
2201 DEBUG_print("commit id mismatch at position (%u, %u): %d != %d\n",
2202 i, j, find_commit_seq(ic, i, j, js->commit_id), want_commit_seq);
2207 if (unlikely(i >= ic->journal_sections))
2208 want_commit_seq = next_commit_seq(want_commit_seq);
2209 wraparound_section(ic, &i);
2213 if (!journal_empty) {
2214 DEBUG_print("replaying %u sections, starting at %u, commit seq %d\n",
2215 write_sections, write_start, want_commit_seq);
2216 do_journal_write(ic, write_start, write_sections, true);
2219 if (write_sections == ic->journal_sections && (ic->mode == 'J' || journal_empty)) {
2220 continue_section = write_start;
2221 ic->commit_seq = want_commit_seq;
2222 DEBUG_print("continuing from section %u, commit seq %d\n", write_start, ic->commit_seq);
2225 unsigned char erase_seq;
2227 DEBUG_print("clearing journal\n");
2229 erase_seq = prev_commit_seq(prev_commit_seq(last_used));
2231 init_journal(ic, s, 1, erase_seq);
2233 wraparound_section(ic, &s);
2234 if (ic->journal_sections >= 2) {
2235 init_journal(ic, s, ic->journal_sections - 2, erase_seq);
2236 s += ic->journal_sections - 2;
2237 wraparound_section(ic, &s);
2238 init_journal(ic, s, 1, erase_seq);
2241 continue_section = 0;
2242 ic->commit_seq = next_commit_seq(erase_seq);
2245 ic->committed_section = continue_section;
2246 ic->n_committed_sections = 0;
2248 ic->uncommitted_section = continue_section;
2249 ic->n_uncommitted_sections = 0;
2251 ic->free_section = continue_section;
2252 ic->free_section_entry = 0;
2253 ic->free_sectors = ic->journal_entries;
2255 ic->journal_tree_root = RB_ROOT;
2256 for (i = 0; i < ic->journal_entries; i++)
2257 init_journal_node(&ic->journal_tree[i]);
2260 static void dm_integrity_postsuspend(struct dm_target *ti)
2262 struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
2264 del_timer_sync(&ic->autocommit_timer);
2266 WRITE_ONCE(ic->suspending, 1);
2268 queue_work(ic->commit_wq, &ic->commit_work);
2269 drain_workqueue(ic->commit_wq);
2271 if (ic->mode == 'J') {
2272 drain_workqueue(ic->writer_wq);
2273 dm_integrity_flush_buffers(ic);
2276 WRITE_ONCE(ic->suspending, 0);
2278 BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
2280 ic->journal_uptodate = true;
2283 static void dm_integrity_resume(struct dm_target *ti)
2285 struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
2290 static void dm_integrity_status(struct dm_target *ti, status_type_t type,
2291 unsigned status_flags, char *result, unsigned maxlen)
2293 struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
2298 case STATUSTYPE_INFO:
2300 (unsigned long long)atomic64_read(&ic->number_of_mismatches),
2301 (unsigned long long)ic->provided_data_sectors);
2304 case STATUSTYPE_TABLE: {
2305 __u64 watermark_percentage = (__u64)(ic->journal_entries - ic->free_sectors_threshold) * 100;
2306 watermark_percentage += ic->journal_entries / 2;
2307 do_div(watermark_percentage, ic->journal_entries);
2309 arg_count += ic->sectors_per_block != 1;
2310 arg_count += !!ic->internal_hash_alg.alg_string;
2311 arg_count += !!ic->journal_crypt_alg.alg_string;
2312 arg_count += !!ic->journal_mac_alg.alg_string;
2313 DMEMIT("%s %llu %u %c %u", ic->dev->name, (unsigned long long)ic->start,
2314 ic->tag_size, ic->mode, arg_count);
2315 DMEMIT(" journal_sectors:%u", ic->initial_sectors - SB_SECTORS);
2316 DMEMIT(" interleave_sectors:%u", 1U << ic->sb->log2_interleave_sectors);
2317 DMEMIT(" buffer_sectors:%u", 1U << ic->log2_buffer_sectors);
2318 DMEMIT(" journal_watermark:%u", (unsigned)watermark_percentage);
2319 DMEMIT(" commit_time:%u", ic->autocommit_msec);
2320 if (ic->sectors_per_block != 1)
2321 DMEMIT(" block_size:%u", ic->sectors_per_block << SECTOR_SHIFT);
2323 #define EMIT_ALG(a, n) \
2325 if (ic->a.alg_string) { \
2326 DMEMIT(" %s:%s", n, ic->a.alg_string); \
2327 if (ic->a.key_string) \
2328 DMEMIT(":%s", ic->a.key_string);\
2331 EMIT_ALG(internal_hash_alg, "internal_hash");
2332 EMIT_ALG(journal_crypt_alg, "journal_crypt");
2333 EMIT_ALG(journal_mac_alg, "journal_mac");
2339 static int dm_integrity_iterate_devices(struct dm_target *ti,
2340 iterate_devices_callout_fn fn, void *data)
2342 struct dm_integrity_c *ic = ti->private;
2344 return fn(ti, ic->dev, ic->start + ic->initial_sectors + ic->metadata_run, ti->len, data);
2347 static void dm_integrity_io_hints(struct dm_target *ti, struct queue_limits *limits)
2349 struct dm_integrity_c *ic = ti->private;
2351 if (ic->sectors_per_block > 1) {
2352 limits->logical_block_size = ic->sectors_per_block << SECTOR_SHIFT;
2353 limits->physical_block_size = ic->sectors_per_block << SECTOR_SHIFT;
2354 blk_limits_io_min(limits, ic->sectors_per_block << SECTOR_SHIFT);
2358 static void calculate_journal_section_size(struct dm_integrity_c *ic)
2360 unsigned sector_space = JOURNAL_SECTOR_DATA;
2362 ic->journal_sections = le32_to_cpu(ic->sb->journal_sections);
2363 ic->journal_entry_size = roundup(offsetof(struct journal_entry, last_bytes[ic->sectors_per_block]) + ic->tag_size,
2364 JOURNAL_ENTRY_ROUNDUP);
2366 if (ic->sb->flags & cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC))
2367 sector_space -= JOURNAL_MAC_PER_SECTOR;
2368 ic->journal_entries_per_sector = sector_space / ic->journal_entry_size;
2369 ic->journal_section_entries = ic->journal_entries_per_sector * JOURNAL_BLOCK_SECTORS;
2370 ic->journal_section_sectors = (ic->journal_section_entries << ic->sb->log2_sectors_per_block) + JOURNAL_BLOCK_SECTORS;
2371 ic->journal_entries = ic->journal_section_entries * ic->journal_sections;
2374 static int calculate_device_limits(struct dm_integrity_c *ic)
2376 __u64 initial_sectors;
2377 sector_t last_sector, last_area, last_offset;
2379 calculate_journal_section_size(ic);
2380 initial_sectors = SB_SECTORS + (__u64)ic->journal_section_sectors * ic->journal_sections;
2381 if (initial_sectors + METADATA_PADDING_SECTORS >= ic->device_sectors || initial_sectors > UINT_MAX)
2383 ic->initial_sectors = initial_sectors;
2385 ic->metadata_run = roundup((__u64)ic->tag_size << (ic->sb->log2_interleave_sectors - ic->sb->log2_sectors_per_block),
2386 (__u64)(1 << SECTOR_SHIFT << METADATA_PADDING_SECTORS)) >> SECTOR_SHIFT;
2387 if (!(ic->metadata_run & (ic->metadata_run - 1)))
2388 ic->log2_metadata_run = __ffs(ic->metadata_run);
2390 ic->log2_metadata_run = -1;
2392 get_area_and_offset(ic, ic->provided_data_sectors - 1, &last_area, &last_offset);
2393 last_sector = get_data_sector(ic, last_area, last_offset);
2395 if (last_sector < ic->start || last_sector >= ic->device_sectors)
2401 static int initialize_superblock(struct dm_integrity_c *ic, unsigned journal_sectors, unsigned interleave_sectors)
2403 unsigned journal_sections;
2406 memset(ic->sb, 0, SB_SECTORS << SECTOR_SHIFT);
2407 memcpy(ic->sb->magic, SB_MAGIC, 8);
2408 ic->sb->version = SB_VERSION;
2409 ic->sb->integrity_tag_size = cpu_to_le16(ic->tag_size);
2410 ic->sb->log2_sectors_per_block = __ffs(ic->sectors_per_block);
2411 if (ic->journal_mac_alg.alg_string)
2412 ic->sb->flags |= cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC);
2414 calculate_journal_section_size(ic);
2415 journal_sections = journal_sectors / ic->journal_section_sectors;
2416 if (!journal_sections)
2417 journal_sections = 1;
2418 ic->sb->journal_sections = cpu_to_le32(journal_sections);
2420 if (!interleave_sectors)
2421 interleave_sectors = DEFAULT_INTERLEAVE_SECTORS;
2422 ic->sb->log2_interleave_sectors = __fls(interleave_sectors);
2423 ic->sb->log2_interleave_sectors = max((__u8)MIN_LOG2_INTERLEAVE_SECTORS, ic->sb->log2_interleave_sectors);
2424 ic->sb->log2_interleave_sectors = min((__u8)MAX_LOG2_INTERLEAVE_SECTORS, ic->sb->log2_interleave_sectors);
2426 ic->provided_data_sectors = 0;
2427 for (test_bit = fls64(ic->device_sectors) - 1; test_bit >= 3; test_bit--) {
2428 __u64 prev_data_sectors = ic->provided_data_sectors;
2430 ic->provided_data_sectors |= (sector_t)1 << test_bit;
2431 if (calculate_device_limits(ic))
2432 ic->provided_data_sectors = prev_data_sectors;
2435 if (!ic->provided_data_sectors)
2438 ic->sb->provided_data_sectors = cpu_to_le64(ic->provided_data_sectors);
2443 static void dm_integrity_set(struct dm_target *ti, struct dm_integrity_c *ic)
2445 struct gendisk *disk = dm_disk(dm_table_get_md(ti->table));
2446 struct blk_integrity bi;
2448 memset(&bi, 0, sizeof(bi));
2449 bi.profile = &dm_integrity_profile;
2450 bi.tuple_size = ic->tag_size;
2451 bi.tag_size = bi.tuple_size;
2452 bi.interval_exp = ic->sb->log2_sectors_per_block + SECTOR_SHIFT;
2454 blk_integrity_register(disk, &bi);
2455 blk_queue_max_integrity_segments(disk->queue, UINT_MAX);
2458 static void dm_integrity_free_page_list(struct dm_integrity_c *ic, struct page_list *pl)
2464 for (i = 0; i < ic->journal_pages; i++)
2466 __free_page(pl[i].page);
2470 static struct page_list *dm_integrity_alloc_page_list(struct dm_integrity_c *ic)
2472 size_t page_list_desc_size = ic->journal_pages * sizeof(struct page_list);
2473 struct page_list *pl;
2476 pl = kvmalloc(page_list_desc_size, GFP_KERNEL | __GFP_ZERO);
2480 for (i = 0; i < ic->journal_pages; i++) {
2481 pl[i].page = alloc_page(GFP_KERNEL);
2483 dm_integrity_free_page_list(ic, pl);
2487 pl[i - 1].next = &pl[i];
2493 static void dm_integrity_free_journal_scatterlist(struct dm_integrity_c *ic, struct scatterlist **sl)
2496 for (i = 0; i < ic->journal_sections; i++)
2501 static struct scatterlist **dm_integrity_alloc_journal_scatterlist(struct dm_integrity_c *ic, struct page_list *pl)
2503 struct scatterlist **sl;
2506 sl = kvmalloc_array(ic->journal_sections,
2507 sizeof(struct scatterlist *),
2508 GFP_KERNEL | __GFP_ZERO);
2512 for (i = 0; i < ic->journal_sections; i++) {
2513 struct scatterlist *s;
2514 unsigned start_index, start_offset;
2515 unsigned end_index, end_offset;
2519 page_list_location(ic, i, 0, &start_index, &start_offset);
2520 page_list_location(ic, i, ic->journal_section_sectors - 1, &end_index, &end_offset);
2522 n_pages = (end_index - start_index + 1);
2524 s = kvmalloc_array(n_pages, sizeof(struct scatterlist),
2527 dm_integrity_free_journal_scatterlist(ic, sl);
2531 sg_init_table(s, n_pages);
2532 for (idx = start_index; idx <= end_index; idx++) {
2533 char *va = lowmem_page_address(pl[idx].page);
2534 unsigned start = 0, end = PAGE_SIZE;
2535 if (idx == start_index)
2536 start = start_offset;
2537 if (idx == end_index)
2538 end = end_offset + (1 << SECTOR_SHIFT);
2539 sg_set_buf(&s[idx - start_index], va + start, end - start);
2548 static void free_alg(struct alg_spec *a)
2550 kzfree(a->alg_string);
2552 memset(a, 0, sizeof *a);
2555 static int get_alg_and_key(const char *arg, struct alg_spec *a, char **error, char *error_inval)
2561 a->alg_string = kstrdup(strchr(arg, ':') + 1, GFP_KERNEL);
2565 k = strchr(a->alg_string, ':');
2568 a->key_string = k + 1;
2569 if (strlen(a->key_string) & 1)
2572 a->key_size = strlen(a->key_string) / 2;
2573 a->key = kmalloc(a->key_size, GFP_KERNEL);
2576 if (hex2bin(a->key, a->key_string, a->key_size))
2582 *error = error_inval;
2585 *error = "Out of memory for an argument";
2589 static int get_mac(struct crypto_shash **hash, struct alg_spec *a, char **error,
2590 char *error_alg, char *error_key)
2594 if (a->alg_string) {
2595 *hash = crypto_alloc_shash(a->alg_string, 0, CRYPTO_ALG_ASYNC);
2596 if (IS_ERR(*hash)) {
2604 r = crypto_shash_setkey(*hash, a->key, a->key_size);
2609 } else if (crypto_shash_get_flags(*hash) & CRYPTO_TFM_NEED_KEY) {
2618 static int create_journal(struct dm_integrity_c *ic, char **error)
2622 __u64 journal_pages, journal_desc_size, journal_tree_size;
2623 unsigned char *crypt_data = NULL, *crypt_iv = NULL;
2624 struct skcipher_request *req = NULL;
2626 ic->commit_ids[0] = cpu_to_le64(0x1111111111111111ULL);
2627 ic->commit_ids[1] = cpu_to_le64(0x2222222222222222ULL);
2628 ic->commit_ids[2] = cpu_to_le64(0x3333333333333333ULL);
2629 ic->commit_ids[3] = cpu_to_le64(0x4444444444444444ULL);
2631 journal_pages = roundup((__u64)ic->journal_sections * ic->journal_section_sectors,
2632 PAGE_SIZE >> SECTOR_SHIFT) >> (PAGE_SHIFT - SECTOR_SHIFT);
2633 journal_desc_size = journal_pages * sizeof(struct page_list);
2634 if (journal_pages >= totalram_pages - totalhigh_pages || journal_desc_size > ULONG_MAX) {
2635 *error = "Journal doesn't fit into memory";
2639 ic->journal_pages = journal_pages;
2641 ic->journal = dm_integrity_alloc_page_list(ic);
2643 *error = "Could not allocate memory for journal";
2647 if (ic->journal_crypt_alg.alg_string) {
2648 unsigned ivsize, blocksize;
2649 struct journal_completion comp;
2652 ic->journal_crypt = crypto_alloc_skcipher(ic->journal_crypt_alg.alg_string, 0, 0);
2653 if (IS_ERR(ic->journal_crypt)) {
2654 *error = "Invalid journal cipher";
2655 r = PTR_ERR(ic->journal_crypt);
2656 ic->journal_crypt = NULL;
2659 ivsize = crypto_skcipher_ivsize(ic->journal_crypt);
2660 blocksize = crypto_skcipher_blocksize(ic->journal_crypt);
2662 if (ic->journal_crypt_alg.key) {
2663 r = crypto_skcipher_setkey(ic->journal_crypt, ic->journal_crypt_alg.key,
2664 ic->journal_crypt_alg.key_size);
2666 *error = "Error setting encryption key";
2670 DEBUG_print("cipher %s, block size %u iv size %u\n",
2671 ic->journal_crypt_alg.alg_string, blocksize, ivsize);
2673 ic->journal_io = dm_integrity_alloc_page_list(ic);
2674 if (!ic->journal_io) {
2675 *error = "Could not allocate memory for journal io";
2680 if (blocksize == 1) {
2681 struct scatterlist *sg;
2683 req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
2685 *error = "Could not allocate crypt request";
2690 crypt_iv = kmalloc(ivsize, GFP_KERNEL);
2692 *error = "Could not allocate iv";
2697 ic->journal_xor = dm_integrity_alloc_page_list(ic);
2698 if (!ic->journal_xor) {
2699 *error = "Could not allocate memory for journal xor";
2704 sg = kvmalloc_array(ic->journal_pages + 1,
2705 sizeof(struct scatterlist),
2708 *error = "Unable to allocate sg list";
2712 sg_init_table(sg, ic->journal_pages + 1);
2713 for (i = 0; i < ic->journal_pages; i++) {
2714 char *va = lowmem_page_address(ic->journal_xor[i].page);
2716 sg_set_buf(&sg[i], va, PAGE_SIZE);
2718 sg_set_buf(&sg[i], &ic->commit_ids, sizeof ic->commit_ids);
2719 memset(crypt_iv, 0x00, ivsize);
2721 skcipher_request_set_crypt(req, sg, sg, PAGE_SIZE * ic->journal_pages + sizeof ic->commit_ids, crypt_iv);
2722 init_completion(&comp.comp);
2723 comp.in_flight = (atomic_t)ATOMIC_INIT(1);
2724 if (do_crypt(true, req, &comp))
2725 wait_for_completion(&comp.comp);
2727 r = dm_integrity_failed(ic);
2729 *error = "Unable to encrypt journal";
2732 DEBUG_bytes(lowmem_page_address(ic->journal_xor[0].page), 64, "xor data");
2734 crypto_free_skcipher(ic->journal_crypt);
2735 ic->journal_crypt = NULL;
2737 unsigned crypt_len = roundup(ivsize, blocksize);
2739 req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
2741 *error = "Could not allocate crypt request";
2746 crypt_iv = kmalloc(ivsize, GFP_KERNEL);
2748 *error = "Could not allocate iv";
2753 crypt_data = kmalloc(crypt_len, GFP_KERNEL);
2755 *error = "Unable to allocate crypt data";
2760 ic->journal_scatterlist = dm_integrity_alloc_journal_scatterlist(ic, ic->journal);
2761 if (!ic->journal_scatterlist) {
2762 *error = "Unable to allocate sg list";
2766 ic->journal_io_scatterlist = dm_integrity_alloc_journal_scatterlist(ic, ic->journal_io);
2767 if (!ic->journal_io_scatterlist) {
2768 *error = "Unable to allocate sg list";
2772 ic->sk_requests = kvmalloc_array(ic->journal_sections,
2773 sizeof(struct skcipher_request *),
2774 GFP_KERNEL | __GFP_ZERO);
2775 if (!ic->sk_requests) {
2776 *error = "Unable to allocate sk requests";
2780 for (i = 0; i < ic->journal_sections; i++) {
2781 struct scatterlist sg;
2782 struct skcipher_request *section_req;
2783 __u32 section_le = cpu_to_le32(i);
2785 memset(crypt_iv, 0x00, ivsize);
2786 memset(crypt_data, 0x00, crypt_len);
2787 memcpy(crypt_data, §ion_le, min((size_t)crypt_len, sizeof(section_le)));
2789 sg_init_one(&sg, crypt_data, crypt_len);
2790 skcipher_request_set_crypt(req, &sg, &sg, crypt_len, crypt_iv);
2791 init_completion(&comp.comp);
2792 comp.in_flight = (atomic_t)ATOMIC_INIT(1);
2793 if (do_crypt(true, req, &comp))
2794 wait_for_completion(&comp.comp);
2796 r = dm_integrity_failed(ic);
2798 *error = "Unable to generate iv";
2802 section_req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
2804 *error = "Unable to allocate crypt request";
2808 section_req->iv = kmalloc_array(ivsize, 2,
2810 if (!section_req->iv) {
2811 skcipher_request_free(section_req);
2812 *error = "Unable to allocate iv";
2816 memcpy(section_req->iv + ivsize, crypt_data, ivsize);
2817 section_req->cryptlen = (size_t)ic->journal_section_sectors << SECTOR_SHIFT;
2818 ic->sk_requests[i] = section_req;
2819 DEBUG_bytes(crypt_data, ivsize, "iv(%u)", i);
2824 for (i = 0; i < N_COMMIT_IDS; i++) {
2827 for (j = 0; j < i; j++) {
2828 if (ic->commit_ids[j] == ic->commit_ids[i]) {
2829 ic->commit_ids[i] = cpu_to_le64(le64_to_cpu(ic->commit_ids[i]) + 1);
2830 goto retest_commit_id;
2833 DEBUG_print("commit id %u: %016llx\n", i, ic->commit_ids[i]);
2836 journal_tree_size = (__u64)ic->journal_entries * sizeof(struct journal_node);
2837 if (journal_tree_size > ULONG_MAX) {
2838 *error = "Journal doesn't fit into memory";
2842 ic->journal_tree = kvmalloc(journal_tree_size, GFP_KERNEL);
2843 if (!ic->journal_tree) {
2844 *error = "Could not allocate memory for journal tree";
2850 skcipher_request_free(req);
2856 * Construct a integrity mapping
2860 * offset from the start of the device
2862 * D - direct writes, J - journal writes, R - recovery mode
2863 * number of optional arguments
2864 * optional arguments:
2866 * interleave_sectors
2875 static int dm_integrity_ctr(struct dm_target *ti, unsigned argc, char **argv)
2877 struct dm_integrity_c *ic;
2880 unsigned extra_args;
2881 struct dm_arg_set as;
2882 static const struct dm_arg _args[] = {
2883 {0, 9, "Invalid number of feature args"},
2885 unsigned journal_sectors, interleave_sectors, buffer_sectors, journal_watermark, sync_msec;
2886 bool should_write_sb;
2888 unsigned long long start;
2890 #define DIRECT_ARGUMENTS 4
2892 if (argc <= DIRECT_ARGUMENTS) {
2893 ti->error = "Invalid argument count";
2897 ic = kzalloc(sizeof(struct dm_integrity_c), GFP_KERNEL);
2899 ti->error = "Cannot allocate integrity context";
2903 ti->per_io_data_size = sizeof(struct dm_integrity_io);
2905 ic->in_progress = RB_ROOT;
2906 INIT_LIST_HEAD(&ic->wait_list);
2907 init_waitqueue_head(&ic->endio_wait);
2908 bio_list_init(&ic->flush_bio_list);
2909 init_waitqueue_head(&ic->copy_to_journal_wait);
2910 init_completion(&ic->crypto_backoff);
2911 atomic64_set(&ic->number_of_mismatches, 0);
2913 r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &ic->dev);
2915 ti->error = "Device lookup failed";
2919 if (sscanf(argv[1], "%llu%c", &start, &dummy) != 1 || start != (sector_t)start) {
2920 ti->error = "Invalid starting offset";
2926 if (strcmp(argv[2], "-")) {
2927 if (sscanf(argv[2], "%u%c", &ic->tag_size, &dummy) != 1 || !ic->tag_size) {
2928 ti->error = "Invalid tag size";
2934 if (!strcmp(argv[3], "J") || !strcmp(argv[3], "D") || !strcmp(argv[3], "R"))
2935 ic->mode = argv[3][0];
2937 ti->error = "Invalid mode (expecting J, D, R)";
2942 ic->device_sectors = i_size_read(ic->dev->bdev->bd_inode) >> SECTOR_SHIFT;
2943 journal_sectors = min((sector_t)DEFAULT_MAX_JOURNAL_SECTORS,
2944 ic->device_sectors >> DEFAULT_JOURNAL_SIZE_FACTOR);
2945 interleave_sectors = DEFAULT_INTERLEAVE_SECTORS;
2946 buffer_sectors = DEFAULT_BUFFER_SECTORS;
2947 journal_watermark = DEFAULT_JOURNAL_WATERMARK;
2948 sync_msec = DEFAULT_SYNC_MSEC;
2949 ic->sectors_per_block = 1;
2951 as.argc = argc - DIRECT_ARGUMENTS;
2952 as.argv = argv + DIRECT_ARGUMENTS;
2953 r = dm_read_arg_group(_args, &as, &extra_args, &ti->error);
2957 while (extra_args--) {
2958 const char *opt_string;
2960 opt_string = dm_shift_arg(&as);
2963 ti->error = "Not enough feature arguments";
2966 if (sscanf(opt_string, "journal_sectors:%u%c", &val, &dummy) == 1)
2967 journal_sectors = val;
2968 else if (sscanf(opt_string, "interleave_sectors:%u%c", &val, &dummy) == 1)
2969 interleave_sectors = val;
2970 else if (sscanf(opt_string, "buffer_sectors:%u%c", &val, &dummy) == 1)
2971 buffer_sectors = val;
2972 else if (sscanf(opt_string, "journal_watermark:%u%c", &val, &dummy) == 1 && val <= 100)
2973 journal_watermark = val;
2974 else if (sscanf(opt_string, "commit_time:%u%c", &val, &dummy) == 1)
2976 else if (sscanf(opt_string, "block_size:%u%c", &val, &dummy) == 1) {
2977 if (val < 1 << SECTOR_SHIFT ||
2978 val > MAX_SECTORS_PER_BLOCK << SECTOR_SHIFT ||
2981 ti->error = "Invalid block_size argument";
2984 ic->sectors_per_block = val >> SECTOR_SHIFT;
2985 } else if (!memcmp(opt_string, "internal_hash:", strlen("internal_hash:"))) {
2986 r = get_alg_and_key(opt_string, &ic->internal_hash_alg, &ti->error,
2987 "Invalid internal_hash argument");
2990 } else if (!memcmp(opt_string, "journal_crypt:", strlen("journal_crypt:"))) {
2991 r = get_alg_and_key(opt_string, &ic->journal_crypt_alg, &ti->error,
2992 "Invalid journal_crypt argument");
2995 } else if (!memcmp(opt_string, "journal_mac:", strlen("journal_mac:"))) {
2996 r = get_alg_and_key(opt_string, &ic->journal_mac_alg, &ti->error,
2997 "Invalid journal_mac argument");
3002 ti->error = "Invalid argument";
3007 r = get_mac(&ic->internal_hash, &ic->internal_hash_alg, &ti->error,
3008 "Invalid internal hash", "Error setting internal hash key");
3012 r = get_mac(&ic->journal_mac, &ic->journal_mac_alg, &ti->error,
3013 "Invalid journal mac", "Error setting journal mac key");
3017 if (!ic->tag_size) {
3018 if (!ic->internal_hash) {
3019 ti->error = "Unknown tag size";
3023 ic->tag_size = crypto_shash_digestsize(ic->internal_hash);
3025 if (ic->tag_size > MAX_TAG_SIZE) {
3026 ti->error = "Too big tag size";
3030 if (!(ic->tag_size & (ic->tag_size - 1)))
3031 ic->log2_tag_size = __ffs(ic->tag_size);
3033 ic->log2_tag_size = -1;
3035 ic->autocommit_jiffies = msecs_to_jiffies(sync_msec);
3036 ic->autocommit_msec = sync_msec;
3037 timer_setup(&ic->autocommit_timer, autocommit_fn, 0);
3039 ic->io = dm_io_client_create();
3040 if (IS_ERR(ic->io)) {
3041 r = PTR_ERR(ic->io);
3043 ti->error = "Cannot allocate dm io";
3047 r = mempool_init_slab_pool(&ic->journal_io_mempool, JOURNAL_IO_MEMPOOL, journal_io_cache);
3049 ti->error = "Cannot allocate mempool";
3053 ic->metadata_wq = alloc_workqueue("dm-integrity-metadata",
3054 WQ_MEM_RECLAIM, METADATA_WORKQUEUE_MAX_ACTIVE);
3055 if (!ic->metadata_wq) {
3056 ti->error = "Cannot allocate workqueue";
3062 * If this workqueue were percpu, it would cause bio reordering
3063 * and reduced performance.
3065 ic->wait_wq = alloc_workqueue("dm-integrity-wait", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3067 ti->error = "Cannot allocate workqueue";
3072 ic->commit_wq = alloc_workqueue("dm-integrity-commit", WQ_MEM_RECLAIM, 1);
3073 if (!ic->commit_wq) {
3074 ti->error = "Cannot allocate workqueue";
3078 INIT_WORK(&ic->commit_work, integrity_commit);
3080 if (ic->mode == 'J') {
3081 ic->writer_wq = alloc_workqueue("dm-integrity-writer", WQ_MEM_RECLAIM, 1);
3082 if (!ic->writer_wq) {
3083 ti->error = "Cannot allocate workqueue";
3087 INIT_WORK(&ic->writer_work, integrity_writer);
3090 ic->sb = alloc_pages_exact(SB_SECTORS << SECTOR_SHIFT, GFP_KERNEL);
3093 ti->error = "Cannot allocate superblock area";
3097 r = sync_rw_sb(ic, REQ_OP_READ, 0);
3099 ti->error = "Error reading superblock";
3102 should_write_sb = false;
3103 if (memcmp(ic->sb->magic, SB_MAGIC, 8)) {
3104 if (ic->mode != 'R') {
3105 if (memchr_inv(ic->sb, 0, SB_SECTORS << SECTOR_SHIFT)) {
3107 ti->error = "The device is not initialized";
3112 r = initialize_superblock(ic, journal_sectors, interleave_sectors);
3114 ti->error = "Could not initialize superblock";
3117 if (ic->mode != 'R')
3118 should_write_sb = true;
3121 if (ic->sb->version != SB_VERSION) {
3123 ti->error = "Unknown version";
3126 if (le16_to_cpu(ic->sb->integrity_tag_size) != ic->tag_size) {
3128 ti->error = "Tag size doesn't match the information in superblock";
3131 if (ic->sb->log2_sectors_per_block != __ffs(ic->sectors_per_block)) {
3133 ti->error = "Block size doesn't match the information in superblock";
3136 if (!le32_to_cpu(ic->sb->journal_sections)) {
3138 ti->error = "Corrupted superblock, journal_sections is 0";
3141 /* make sure that ti->max_io_len doesn't overflow */
3142 if (ic->sb->log2_interleave_sectors < MIN_LOG2_INTERLEAVE_SECTORS ||
3143 ic->sb->log2_interleave_sectors > MAX_LOG2_INTERLEAVE_SECTORS) {
3145 ti->error = "Invalid interleave_sectors in the superblock";
3148 ic->provided_data_sectors = le64_to_cpu(ic->sb->provided_data_sectors);
3149 if (ic->provided_data_sectors != le64_to_cpu(ic->sb->provided_data_sectors)) {
3150 /* test for overflow */
3152 ti->error = "The superblock has 64-bit device size, but the kernel was compiled with 32-bit sectors";
3155 if (!!(ic->sb->flags & cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC)) != !!ic->journal_mac_alg.alg_string) {
3157 ti->error = "Journal mac mismatch";
3160 r = calculate_device_limits(ic);
3162 ti->error = "The device is too small";
3165 if (ti->len > ic->provided_data_sectors) {
3167 ti->error = "Not enough provided sectors for requested mapping size";
3171 if (!buffer_sectors)
3173 ic->log2_buffer_sectors = min3((int)__fls(buffer_sectors), (int)__ffs(ic->metadata_run), 31 - SECTOR_SHIFT);
3175 threshold = (__u64)ic->journal_entries * (100 - journal_watermark);
3177 do_div(threshold, 100);
3178 ic->free_sectors_threshold = threshold;
3180 DEBUG_print("initialized:\n");
3181 DEBUG_print(" integrity_tag_size %u\n", le16_to_cpu(ic->sb->integrity_tag_size));
3182 DEBUG_print(" journal_entry_size %u\n", ic->journal_entry_size);
3183 DEBUG_print(" journal_entries_per_sector %u\n", ic->journal_entries_per_sector);
3184 DEBUG_print(" journal_section_entries %u\n", ic->journal_section_entries);
3185 DEBUG_print(" journal_section_sectors %u\n", ic->journal_section_sectors);
3186 DEBUG_print(" journal_sections %u\n", (unsigned)le32_to_cpu(ic->sb->journal_sections));
3187 DEBUG_print(" journal_entries %u\n", ic->journal_entries);
3188 DEBUG_print(" log2_interleave_sectors %d\n", ic->sb->log2_interleave_sectors);
3189 DEBUG_print(" device_sectors 0x%llx\n", (unsigned long long)ic->device_sectors);
3190 DEBUG_print(" initial_sectors 0x%x\n", ic->initial_sectors);
3191 DEBUG_print(" metadata_run 0x%x\n", ic->metadata_run);
3192 DEBUG_print(" log2_metadata_run %d\n", ic->log2_metadata_run);
3193 DEBUG_print(" provided_data_sectors 0x%llx (%llu)\n", (unsigned long long)ic->provided_data_sectors,
3194 (unsigned long long)ic->provided_data_sectors);
3195 DEBUG_print(" log2_buffer_sectors %u\n", ic->log2_buffer_sectors);
3197 ic->bufio = dm_bufio_client_create(ic->dev->bdev, 1U << (SECTOR_SHIFT + ic->log2_buffer_sectors),
3199 if (IS_ERR(ic->bufio)) {
3200 r = PTR_ERR(ic->bufio);
3201 ti->error = "Cannot initialize dm-bufio";
3205 dm_bufio_set_sector_offset(ic->bufio, ic->start + ic->initial_sectors);
3207 if (ic->mode != 'R') {
3208 r = create_journal(ic, &ti->error);
3213 if (should_write_sb) {
3216 init_journal(ic, 0, ic->journal_sections, 0);
3217 r = dm_integrity_failed(ic);
3219 ti->error = "Error initializing journal";
3222 r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
3224 ti->error = "Error initializing superblock";
3227 ic->just_formatted = true;
3230 r = dm_set_target_max_io_len(ti, 1U << ic->sb->log2_interleave_sectors);
3234 if (!ic->internal_hash)
3235 dm_integrity_set(ti, ic);
3237 ti->num_flush_bios = 1;
3238 ti->flush_supported = true;
3242 dm_integrity_dtr(ti);
3246 static void dm_integrity_dtr(struct dm_target *ti)
3248 struct dm_integrity_c *ic = ti->private;
3250 BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
3251 BUG_ON(!list_empty(&ic->wait_list));
3253 if (ic->metadata_wq)
3254 destroy_workqueue(ic->metadata_wq);
3256 destroy_workqueue(ic->wait_wq);
3258 destroy_workqueue(ic->commit_wq);
3260 destroy_workqueue(ic->writer_wq);
3262 dm_bufio_client_destroy(ic->bufio);
3263 mempool_exit(&ic->journal_io_mempool);
3265 dm_io_client_destroy(ic->io);
3267 dm_put_device(ti, ic->dev);
3268 dm_integrity_free_page_list(ic, ic->journal);
3269 dm_integrity_free_page_list(ic, ic->journal_io);
3270 dm_integrity_free_page_list(ic, ic->journal_xor);
3271 if (ic->journal_scatterlist)
3272 dm_integrity_free_journal_scatterlist(ic, ic->journal_scatterlist);
3273 if (ic->journal_io_scatterlist)
3274 dm_integrity_free_journal_scatterlist(ic, ic->journal_io_scatterlist);
3275 if (ic->sk_requests) {
3278 for (i = 0; i < ic->journal_sections; i++) {
3279 struct skcipher_request *req = ic->sk_requests[i];
3282 skcipher_request_free(req);
3285 kvfree(ic->sk_requests);
3287 kvfree(ic->journal_tree);
3289 free_pages_exact(ic->sb, SB_SECTORS << SECTOR_SHIFT);
3291 if (ic->internal_hash)
3292 crypto_free_shash(ic->internal_hash);
3293 free_alg(&ic->internal_hash_alg);
3295 if (ic->journal_crypt)
3296 crypto_free_skcipher(ic->journal_crypt);
3297 free_alg(&ic->journal_crypt_alg);
3299 if (ic->journal_mac)
3300 crypto_free_shash(ic->journal_mac);
3301 free_alg(&ic->journal_mac_alg);
3306 static struct target_type integrity_target = {
3307 .name = "integrity",
3308 .version = {1, 1, 0},
3309 .module = THIS_MODULE,
3310 .features = DM_TARGET_SINGLETON | DM_TARGET_INTEGRITY,
3311 .ctr = dm_integrity_ctr,
3312 .dtr = dm_integrity_dtr,
3313 .map = dm_integrity_map,
3314 .postsuspend = dm_integrity_postsuspend,
3315 .resume = dm_integrity_resume,
3316 .status = dm_integrity_status,
3317 .iterate_devices = dm_integrity_iterate_devices,
3318 .io_hints = dm_integrity_io_hints,
3321 int __init dm_integrity_init(void)
3325 journal_io_cache = kmem_cache_create("integrity_journal_io",
3326 sizeof(struct journal_io), 0, 0, NULL);
3327 if (!journal_io_cache) {
3328 DMERR("can't allocate journal io cache");
3332 r = dm_register_target(&integrity_target);
3335 DMERR("register failed %d", r);
3340 void dm_integrity_exit(void)
3342 dm_unregister_target(&integrity_target);
3343 kmem_cache_destroy(journal_io_cache);
3346 module_init(dm_integrity_init);
3347 module_exit(dm_integrity_exit);
3349 MODULE_AUTHOR("Milan Broz");
3350 MODULE_AUTHOR("Mikulas Patocka");
3351 MODULE_DESCRIPTION(DM_NAME " target for integrity tags extension");
3352 MODULE_LICENSE("GPL");