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
34 #define RECALC_SECTORS 8192
35 #define RECALC_WRITE_SUPER 16
38 * Warning - DEBUG_PRINT prints security-sensitive data to the log,
39 * so it should not be enabled in the official kernel
42 //#define INTERNAL_VERIFY
48 #define SB_MAGIC "integrt"
49 #define SB_VERSION_1 1
50 #define SB_VERSION_2 2
52 #define MAX_SECTORS_PER_BLOCK 8
57 __u8 log2_interleave_sectors;
58 __u16 integrity_tag_size;
59 __u32 journal_sections;
60 __u64 provided_data_sectors; /* userspace uses this value */
62 __u8 log2_sectors_per_block;
67 #define SB_FLAG_HAVE_JOURNAL_MAC 0x1
68 #define SB_FLAG_RECALCULATING 0x2
70 #define JOURNAL_ENTRY_ROUNDUP 8
72 typedef __u64 commit_id_t;
73 #define JOURNAL_MAC_PER_SECTOR 8
75 struct journal_entry {
83 commit_id_t last_bytes[0];
87 #define journal_entry_tag(ic, je) ((__u8 *)&(je)->last_bytes[(ic)->sectors_per_block])
89 #if BITS_PER_LONG == 64
90 #define journal_entry_set_sector(je, x) do { smp_wmb(); WRITE_ONCE((je)->u.sector, cpu_to_le64(x)); } while (0)
91 #define journal_entry_get_sector(je) le64_to_cpu((je)->u.sector)
92 #elif defined(CONFIG_LBDAF)
93 #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)
94 #define journal_entry_get_sector(je) le64_to_cpu((je)->u.sector)
96 #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)
97 #define journal_entry_get_sector(je) le32_to_cpu((je)->u.s.sector_lo)
99 #define journal_entry_is_unused(je) ((je)->u.s.sector_hi == cpu_to_le32(-1))
100 #define journal_entry_set_unused(je) do { ((je)->u.s.sector_hi = cpu_to_le32(-1)); } while (0)
101 #define journal_entry_is_inprogress(je) ((je)->u.s.sector_hi == cpu_to_le32(-2))
102 #define journal_entry_set_inprogress(je) do { ((je)->u.s.sector_hi = cpu_to_le32(-2)); } while (0)
104 #define JOURNAL_BLOCK_SECTORS 8
105 #define JOURNAL_SECTOR_DATA ((1 << SECTOR_SHIFT) - sizeof(commit_id_t))
106 #define JOURNAL_MAC_SIZE (JOURNAL_MAC_PER_SECTOR * JOURNAL_BLOCK_SECTORS)
108 struct journal_sector {
109 __u8 entries[JOURNAL_SECTOR_DATA - JOURNAL_MAC_PER_SECTOR];
110 __u8 mac[JOURNAL_MAC_PER_SECTOR];
111 commit_id_t commit_id;
114 #define MAX_TAG_SIZE (JOURNAL_SECTOR_DATA - JOURNAL_MAC_PER_SECTOR - offsetof(struct journal_entry, last_bytes[MAX_SECTORS_PER_BLOCK]))
116 #define METADATA_PADDING_SECTORS 8
118 #define N_COMMIT_IDS 4
120 static unsigned char prev_commit_seq(unsigned char seq)
122 return (seq + N_COMMIT_IDS - 1) % N_COMMIT_IDS;
125 static unsigned char next_commit_seq(unsigned char seq)
127 return (seq + 1) % N_COMMIT_IDS;
131 * In-memory structures
134 struct journal_node {
146 struct dm_integrity_c {
148 struct dm_dev *meta_dev;
152 mempool_t journal_io_mempool;
153 struct dm_io_client *io;
154 struct dm_bufio_client *bufio;
155 struct workqueue_struct *metadata_wq;
156 struct superblock *sb;
157 unsigned journal_pages;
158 struct page_list *journal;
159 struct page_list *journal_io;
160 struct page_list *journal_xor;
162 struct crypto_skcipher *journal_crypt;
163 struct scatterlist **journal_scatterlist;
164 struct scatterlist **journal_io_scatterlist;
165 struct skcipher_request **sk_requests;
167 struct crypto_shash *journal_mac;
169 struct journal_node *journal_tree;
170 struct rb_root journal_tree_root;
172 sector_t provided_data_sectors;
174 unsigned short journal_entry_size;
175 unsigned char journal_entries_per_sector;
176 unsigned char journal_section_entries;
177 unsigned short journal_section_sectors;
178 unsigned journal_sections;
179 unsigned journal_entries;
180 sector_t data_device_sectors;
181 sector_t meta_device_sectors;
182 unsigned initial_sectors;
183 unsigned metadata_run;
184 __s8 log2_metadata_run;
185 __u8 log2_buffer_sectors;
186 __u8 sectors_per_block;
193 struct crypto_shash *internal_hash;
195 /* these variables are locked with endio_wait.lock */
196 struct rb_root in_progress;
197 struct list_head wait_list;
198 wait_queue_head_t endio_wait;
199 struct workqueue_struct *wait_wq;
201 unsigned char commit_seq;
202 commit_id_t commit_ids[N_COMMIT_IDS];
204 unsigned committed_section;
205 unsigned n_committed_sections;
207 unsigned uncommitted_section;
208 unsigned n_uncommitted_sections;
210 unsigned free_section;
211 unsigned char free_section_entry;
212 unsigned free_sectors;
214 unsigned free_sectors_threshold;
216 struct workqueue_struct *commit_wq;
217 struct work_struct commit_work;
219 struct workqueue_struct *writer_wq;
220 struct work_struct writer_work;
222 struct workqueue_struct *recalc_wq;
223 struct work_struct recalc_work;
227 struct bio_list flush_bio_list;
229 unsigned long autocommit_jiffies;
230 struct timer_list autocommit_timer;
231 unsigned autocommit_msec;
233 wait_queue_head_t copy_to_journal_wait;
235 struct completion crypto_backoff;
237 bool journal_uptodate;
240 struct alg_spec internal_hash_alg;
241 struct alg_spec journal_crypt_alg;
242 struct alg_spec journal_mac_alg;
244 atomic64_t number_of_mismatches;
247 struct dm_integrity_range {
248 sector_t logical_sector;
254 struct task_struct *task;
255 struct list_head wait_entry;
260 struct dm_integrity_io {
261 struct work_struct work;
263 struct dm_integrity_c *ic;
267 struct dm_integrity_range range;
269 sector_t metadata_block;
270 unsigned metadata_offset;
273 blk_status_t bi_status;
275 struct completion *completion;
277 struct gendisk *orig_bi_disk;
279 bio_end_io_t *orig_bi_end_io;
280 struct bio_integrity_payload *orig_bi_integrity;
281 struct bvec_iter orig_bi_iter;
284 struct journal_completion {
285 struct dm_integrity_c *ic;
287 struct completion comp;
291 struct dm_integrity_range range;
292 struct journal_completion *comp;
295 static struct kmem_cache *journal_io_cache;
297 #define JOURNAL_IO_MEMPOOL 32
300 #define DEBUG_print(x, ...) printk(KERN_DEBUG x, ##__VA_ARGS__)
301 static void __DEBUG_bytes(__u8 *bytes, size_t len, const char *msg, ...)
310 pr_cont(" %02x", *bytes);
316 #define DEBUG_bytes(bytes, len, msg, ...) __DEBUG_bytes(bytes, len, KERN_DEBUG msg, ##__VA_ARGS__)
318 #define DEBUG_print(x, ...) do { } while (0)
319 #define DEBUG_bytes(bytes, len, msg, ...) do { } while (0)
323 * DM Integrity profile, protection is performed layer above (dm-crypt)
325 static const struct blk_integrity_profile dm_integrity_profile = {
326 .name = "DM-DIF-EXT-TAG",
331 static void dm_integrity_map_continue(struct dm_integrity_io *dio, bool from_map);
332 static void integrity_bio_wait(struct work_struct *w);
333 static void dm_integrity_dtr(struct dm_target *ti);
335 static void dm_integrity_io_error(struct dm_integrity_c *ic, const char *msg, int err)
338 atomic64_inc(&ic->number_of_mismatches);
339 if (!cmpxchg(&ic->failed, 0, err))
340 DMERR("Error on %s: %d", msg, err);
343 static int dm_integrity_failed(struct dm_integrity_c *ic)
345 return READ_ONCE(ic->failed);
348 static commit_id_t dm_integrity_commit_id(struct dm_integrity_c *ic, unsigned i,
349 unsigned j, unsigned char seq)
352 * Xor the number with section and sector, so that if a piece of
353 * journal is written at wrong place, it is detected.
355 return ic->commit_ids[seq] ^ cpu_to_le64(((__u64)i << 32) ^ j);
358 static void get_area_and_offset(struct dm_integrity_c *ic, sector_t data_sector,
359 sector_t *area, sector_t *offset)
362 __u8 log2_interleave_sectors = ic->sb->log2_interleave_sectors;
363 *area = data_sector >> log2_interleave_sectors;
364 *offset = (unsigned)data_sector & ((1U << log2_interleave_sectors) - 1);
367 *offset = data_sector;
371 #define sector_to_block(ic, n) \
373 BUG_ON((n) & (unsigned)((ic)->sectors_per_block - 1)); \
374 (n) >>= (ic)->sb->log2_sectors_per_block; \
377 static __u64 get_metadata_sector_and_offset(struct dm_integrity_c *ic, sector_t area,
378 sector_t offset, unsigned *metadata_offset)
383 ms = area << ic->sb->log2_interleave_sectors;
384 if (likely(ic->log2_metadata_run >= 0))
385 ms += area << ic->log2_metadata_run;
387 ms += area * ic->metadata_run;
388 ms >>= ic->log2_buffer_sectors;
390 sector_to_block(ic, offset);
392 if (likely(ic->log2_tag_size >= 0)) {
393 ms += offset >> (SECTOR_SHIFT + ic->log2_buffer_sectors - ic->log2_tag_size);
394 mo = (offset << ic->log2_tag_size) & ((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - 1);
396 ms += (__u64)offset * ic->tag_size >> (SECTOR_SHIFT + ic->log2_buffer_sectors);
397 mo = (offset * ic->tag_size) & ((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - 1);
399 *metadata_offset = mo;
403 static sector_t get_data_sector(struct dm_integrity_c *ic, sector_t area, sector_t offset)
410 result = area << ic->sb->log2_interleave_sectors;
411 if (likely(ic->log2_metadata_run >= 0))
412 result += (area + 1) << ic->log2_metadata_run;
414 result += (area + 1) * ic->metadata_run;
416 result += (sector_t)ic->initial_sectors + offset;
422 static void wraparound_section(struct dm_integrity_c *ic, unsigned *sec_ptr)
424 if (unlikely(*sec_ptr >= ic->journal_sections))
425 *sec_ptr -= ic->journal_sections;
428 static void sb_set_version(struct dm_integrity_c *ic)
430 if (ic->meta_dev || ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
431 ic->sb->version = SB_VERSION_2;
433 ic->sb->version = SB_VERSION_1;
436 static int sync_rw_sb(struct dm_integrity_c *ic, int op, int op_flags)
438 struct dm_io_request io_req;
439 struct dm_io_region io_loc;
442 io_req.bi_op_flags = op_flags;
443 io_req.mem.type = DM_IO_KMEM;
444 io_req.mem.ptr.addr = ic->sb;
445 io_req.notify.fn = NULL;
446 io_req.client = ic->io;
447 io_loc.bdev = ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev;
448 io_loc.sector = ic->start;
449 io_loc.count = SB_SECTORS;
451 return dm_io(&io_req, 1, &io_loc, NULL);
454 static void access_journal_check(struct dm_integrity_c *ic, unsigned section, unsigned offset,
455 bool e, const char *function)
457 #if defined(CONFIG_DM_DEBUG) || defined(INTERNAL_VERIFY)
458 unsigned limit = e ? ic->journal_section_entries : ic->journal_section_sectors;
460 if (unlikely(section >= ic->journal_sections) ||
461 unlikely(offset >= limit)) {
462 printk(KERN_CRIT "%s: invalid access at (%u,%u), limit (%u,%u)\n",
463 function, section, offset, ic->journal_sections, limit);
469 static void page_list_location(struct dm_integrity_c *ic, unsigned section, unsigned offset,
470 unsigned *pl_index, unsigned *pl_offset)
474 access_journal_check(ic, section, offset, false, "page_list_location");
476 sector = section * ic->journal_section_sectors + offset;
478 *pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
479 *pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
482 static struct journal_sector *access_page_list(struct dm_integrity_c *ic, struct page_list *pl,
483 unsigned section, unsigned offset, unsigned *n_sectors)
485 unsigned pl_index, pl_offset;
488 page_list_location(ic, section, offset, &pl_index, &pl_offset);
491 *n_sectors = (PAGE_SIZE - pl_offset) >> SECTOR_SHIFT;
493 va = lowmem_page_address(pl[pl_index].page);
495 return (struct journal_sector *)(va + pl_offset);
498 static struct journal_sector *access_journal(struct dm_integrity_c *ic, unsigned section, unsigned offset)
500 return access_page_list(ic, ic->journal, section, offset, NULL);
503 static struct journal_entry *access_journal_entry(struct dm_integrity_c *ic, unsigned section, unsigned n)
505 unsigned rel_sector, offset;
506 struct journal_sector *js;
508 access_journal_check(ic, section, n, true, "access_journal_entry");
510 rel_sector = n % JOURNAL_BLOCK_SECTORS;
511 offset = n / JOURNAL_BLOCK_SECTORS;
513 js = access_journal(ic, section, rel_sector);
514 return (struct journal_entry *)((char *)js + offset * ic->journal_entry_size);
517 static struct journal_sector *access_journal_data(struct dm_integrity_c *ic, unsigned section, unsigned n)
519 n <<= ic->sb->log2_sectors_per_block;
521 n += JOURNAL_BLOCK_SECTORS;
523 access_journal_check(ic, section, n, false, "access_journal_data");
525 return access_journal(ic, section, n);
528 static void section_mac(struct dm_integrity_c *ic, unsigned section, __u8 result[JOURNAL_MAC_SIZE])
530 SHASH_DESC_ON_STACK(desc, ic->journal_mac);
534 desc->tfm = ic->journal_mac;
537 r = crypto_shash_init(desc);
539 dm_integrity_io_error(ic, "crypto_shash_init", r);
543 for (j = 0; j < ic->journal_section_entries; j++) {
544 struct journal_entry *je = access_journal_entry(ic, section, j);
545 r = crypto_shash_update(desc, (__u8 *)&je->u.sector, sizeof je->u.sector);
547 dm_integrity_io_error(ic, "crypto_shash_update", r);
552 size = crypto_shash_digestsize(ic->journal_mac);
554 if (likely(size <= JOURNAL_MAC_SIZE)) {
555 r = crypto_shash_final(desc, result);
557 dm_integrity_io_error(ic, "crypto_shash_final", r);
560 memset(result + size, 0, JOURNAL_MAC_SIZE - size);
562 __u8 digest[HASH_MAX_DIGESTSIZE];
564 if (WARN_ON(size > sizeof(digest))) {
565 dm_integrity_io_error(ic, "digest_size", -EINVAL);
568 r = crypto_shash_final(desc, digest);
570 dm_integrity_io_error(ic, "crypto_shash_final", r);
573 memcpy(result, digest, JOURNAL_MAC_SIZE);
578 memset(result, 0, JOURNAL_MAC_SIZE);
581 static void rw_section_mac(struct dm_integrity_c *ic, unsigned section, bool wr)
583 __u8 result[JOURNAL_MAC_SIZE];
586 if (!ic->journal_mac)
589 section_mac(ic, section, result);
591 for (j = 0; j < JOURNAL_BLOCK_SECTORS; j++) {
592 struct journal_sector *js = access_journal(ic, section, j);
595 memcpy(&js->mac, result + (j * JOURNAL_MAC_PER_SECTOR), JOURNAL_MAC_PER_SECTOR);
597 if (memcmp(&js->mac, result + (j * JOURNAL_MAC_PER_SECTOR), JOURNAL_MAC_PER_SECTOR))
598 dm_integrity_io_error(ic, "journal mac", -EILSEQ);
603 static void complete_journal_op(void *context)
605 struct journal_completion *comp = context;
606 BUG_ON(!atomic_read(&comp->in_flight));
607 if (likely(atomic_dec_and_test(&comp->in_flight)))
608 complete(&comp->comp);
611 static void xor_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
612 unsigned n_sections, struct journal_completion *comp)
614 struct async_submit_ctl submit;
615 size_t n_bytes = (size_t)(n_sections * ic->journal_section_sectors) << SECTOR_SHIFT;
616 unsigned pl_index, pl_offset, section_index;
617 struct page_list *source_pl, *target_pl;
619 if (likely(encrypt)) {
620 source_pl = ic->journal;
621 target_pl = ic->journal_io;
623 source_pl = ic->journal_io;
624 target_pl = ic->journal;
627 page_list_location(ic, section, 0, &pl_index, &pl_offset);
629 atomic_add(roundup(pl_offset + n_bytes, PAGE_SIZE) >> PAGE_SHIFT, &comp->in_flight);
631 init_async_submit(&submit, ASYNC_TX_XOR_ZERO_DST, NULL, complete_journal_op, comp, NULL);
633 section_index = pl_index;
637 struct page *src_pages[2];
638 struct page *dst_page;
640 while (unlikely(pl_index == section_index)) {
643 rw_section_mac(ic, section, true);
648 page_list_location(ic, section, 0, §ion_index, &dummy);
651 this_step = min(n_bytes, (size_t)PAGE_SIZE - pl_offset);
652 dst_page = target_pl[pl_index].page;
653 src_pages[0] = source_pl[pl_index].page;
654 src_pages[1] = ic->journal_xor[pl_index].page;
656 async_xor(dst_page, src_pages, pl_offset, 2, this_step, &submit);
660 n_bytes -= this_step;
665 async_tx_issue_pending_all();
668 static void complete_journal_encrypt(struct crypto_async_request *req, int err)
670 struct journal_completion *comp = req->data;
672 if (likely(err == -EINPROGRESS)) {
673 complete(&comp->ic->crypto_backoff);
676 dm_integrity_io_error(comp->ic, "asynchronous encrypt", err);
678 complete_journal_op(comp);
681 static bool do_crypt(bool encrypt, struct skcipher_request *req, struct journal_completion *comp)
684 skcipher_request_set_callback(req, CRYPTO_TFM_REQ_MAY_BACKLOG,
685 complete_journal_encrypt, comp);
687 r = crypto_skcipher_encrypt(req);
689 r = crypto_skcipher_decrypt(req);
692 if (likely(r == -EINPROGRESS))
694 if (likely(r == -EBUSY)) {
695 wait_for_completion(&comp->ic->crypto_backoff);
696 reinit_completion(&comp->ic->crypto_backoff);
699 dm_integrity_io_error(comp->ic, "encrypt", r);
703 static void crypt_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
704 unsigned n_sections, struct journal_completion *comp)
706 struct scatterlist **source_sg;
707 struct scatterlist **target_sg;
709 atomic_add(2, &comp->in_flight);
711 if (likely(encrypt)) {
712 source_sg = ic->journal_scatterlist;
713 target_sg = ic->journal_io_scatterlist;
715 source_sg = ic->journal_io_scatterlist;
716 target_sg = ic->journal_scatterlist;
720 struct skcipher_request *req;
725 rw_section_mac(ic, section, true);
727 req = ic->sk_requests[section];
728 ivsize = crypto_skcipher_ivsize(ic->journal_crypt);
731 memcpy(iv, iv + ivsize, ivsize);
733 req->src = source_sg[section];
734 req->dst = target_sg[section];
736 if (unlikely(do_crypt(encrypt, req, comp)))
737 atomic_inc(&comp->in_flight);
741 } while (n_sections);
743 atomic_dec(&comp->in_flight);
744 complete_journal_op(comp);
747 static void encrypt_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
748 unsigned n_sections, struct journal_completion *comp)
751 return xor_journal(ic, encrypt, section, n_sections, comp);
753 return crypt_journal(ic, encrypt, section, n_sections, comp);
756 static void complete_journal_io(unsigned long error, void *context)
758 struct journal_completion *comp = context;
759 if (unlikely(error != 0))
760 dm_integrity_io_error(comp->ic, "writing journal", -EIO);
761 complete_journal_op(comp);
764 static void rw_journal(struct dm_integrity_c *ic, int op, int op_flags, unsigned section,
765 unsigned n_sections, struct journal_completion *comp)
767 struct dm_io_request io_req;
768 struct dm_io_region io_loc;
769 unsigned sector, n_sectors, pl_index, pl_offset;
772 if (unlikely(dm_integrity_failed(ic))) {
774 complete_journal_io(-1UL, comp);
778 sector = section * ic->journal_section_sectors;
779 n_sectors = n_sections * ic->journal_section_sectors;
781 pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
782 pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
785 io_req.bi_op_flags = op_flags;
786 io_req.mem.type = DM_IO_PAGE_LIST;
788 io_req.mem.ptr.pl = &ic->journal_io[pl_index];
790 io_req.mem.ptr.pl = &ic->journal[pl_index];
791 io_req.mem.offset = pl_offset;
792 if (likely(comp != NULL)) {
793 io_req.notify.fn = complete_journal_io;
794 io_req.notify.context = comp;
796 io_req.notify.fn = NULL;
798 io_req.client = ic->io;
799 io_loc.bdev = ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev;
800 io_loc.sector = ic->start + SB_SECTORS + sector;
801 io_loc.count = n_sectors;
803 r = dm_io(&io_req, 1, &io_loc, NULL);
805 dm_integrity_io_error(ic, op == REQ_OP_READ ? "reading journal" : "writing journal", r);
807 WARN_ONCE(1, "asynchronous dm_io failed: %d", r);
808 complete_journal_io(-1UL, comp);
813 static void write_journal(struct dm_integrity_c *ic, unsigned commit_start, unsigned commit_sections)
815 struct journal_completion io_comp;
816 struct journal_completion crypt_comp_1;
817 struct journal_completion crypt_comp_2;
821 init_completion(&io_comp.comp);
823 if (commit_start + commit_sections <= ic->journal_sections) {
824 io_comp.in_flight = (atomic_t)ATOMIC_INIT(1);
825 if (ic->journal_io) {
826 crypt_comp_1.ic = ic;
827 init_completion(&crypt_comp_1.comp);
828 crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
829 encrypt_journal(ic, true, commit_start, commit_sections, &crypt_comp_1);
830 wait_for_completion_io(&crypt_comp_1.comp);
832 for (i = 0; i < commit_sections; i++)
833 rw_section_mac(ic, commit_start + i, true);
835 rw_journal(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, commit_start,
836 commit_sections, &io_comp);
839 io_comp.in_flight = (atomic_t)ATOMIC_INIT(2);
840 to_end = ic->journal_sections - commit_start;
841 if (ic->journal_io) {
842 crypt_comp_1.ic = ic;
843 init_completion(&crypt_comp_1.comp);
844 crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
845 encrypt_journal(ic, true, commit_start, to_end, &crypt_comp_1);
846 if (try_wait_for_completion(&crypt_comp_1.comp)) {
847 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
848 reinit_completion(&crypt_comp_1.comp);
849 crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
850 encrypt_journal(ic, true, 0, commit_sections - to_end, &crypt_comp_1);
851 wait_for_completion_io(&crypt_comp_1.comp);
853 crypt_comp_2.ic = ic;
854 init_completion(&crypt_comp_2.comp);
855 crypt_comp_2.in_flight = (atomic_t)ATOMIC_INIT(0);
856 encrypt_journal(ic, true, 0, commit_sections - to_end, &crypt_comp_2);
857 wait_for_completion_io(&crypt_comp_1.comp);
858 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
859 wait_for_completion_io(&crypt_comp_2.comp);
862 for (i = 0; i < to_end; i++)
863 rw_section_mac(ic, commit_start + i, true);
864 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
865 for (i = 0; i < commit_sections - to_end; i++)
866 rw_section_mac(ic, i, true);
868 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, 0, commit_sections - to_end, &io_comp);
871 wait_for_completion_io(&io_comp.comp);
874 static void copy_from_journal(struct dm_integrity_c *ic, unsigned section, unsigned offset,
875 unsigned n_sectors, sector_t target, io_notify_fn fn, void *data)
877 struct dm_io_request io_req;
878 struct dm_io_region io_loc;
880 unsigned sector, pl_index, pl_offset;
882 BUG_ON((target | n_sectors | offset) & (unsigned)(ic->sectors_per_block - 1));
884 if (unlikely(dm_integrity_failed(ic))) {
889 sector = section * ic->journal_section_sectors + JOURNAL_BLOCK_SECTORS + offset;
891 pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
892 pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
894 io_req.bi_op = REQ_OP_WRITE;
895 io_req.bi_op_flags = 0;
896 io_req.mem.type = DM_IO_PAGE_LIST;
897 io_req.mem.ptr.pl = &ic->journal[pl_index];
898 io_req.mem.offset = pl_offset;
899 io_req.notify.fn = fn;
900 io_req.notify.context = data;
901 io_req.client = ic->io;
902 io_loc.bdev = ic->dev->bdev;
903 io_loc.sector = target;
904 io_loc.count = n_sectors;
906 r = dm_io(&io_req, 1, &io_loc, NULL);
908 WARN_ONCE(1, "asynchronous dm_io failed: %d", r);
913 static bool ranges_overlap(struct dm_integrity_range *range1, struct dm_integrity_range *range2)
915 return range1->logical_sector < range2->logical_sector + range2->n_sectors &&
916 range1->logical_sector + range1->n_sectors > range2->logical_sector;
919 static bool add_new_range(struct dm_integrity_c *ic, struct dm_integrity_range *new_range, bool check_waiting)
921 struct rb_node **n = &ic->in_progress.rb_node;
922 struct rb_node *parent;
924 BUG_ON((new_range->logical_sector | new_range->n_sectors) & (unsigned)(ic->sectors_per_block - 1));
926 if (likely(check_waiting)) {
927 struct dm_integrity_range *range;
928 list_for_each_entry(range, &ic->wait_list, wait_entry) {
929 if (unlikely(ranges_overlap(range, new_range)))
937 struct dm_integrity_range *range = container_of(*n, struct dm_integrity_range, node);
940 if (new_range->logical_sector + new_range->n_sectors <= range->logical_sector) {
941 n = &range->node.rb_left;
942 } else if (new_range->logical_sector >= range->logical_sector + range->n_sectors) {
943 n = &range->node.rb_right;
949 rb_link_node(&new_range->node, parent, n);
950 rb_insert_color(&new_range->node, &ic->in_progress);
955 static void remove_range_unlocked(struct dm_integrity_c *ic, struct dm_integrity_range *range)
957 rb_erase(&range->node, &ic->in_progress);
958 while (unlikely(!list_empty(&ic->wait_list))) {
959 struct dm_integrity_range *last_range =
960 list_first_entry(&ic->wait_list, struct dm_integrity_range, wait_entry);
961 struct task_struct *last_range_task;
962 last_range_task = last_range->task;
963 list_del(&last_range->wait_entry);
964 if (!add_new_range(ic, last_range, false)) {
965 last_range->task = last_range_task;
966 list_add(&last_range->wait_entry, &ic->wait_list);
969 last_range->waiting = false;
970 wake_up_process(last_range_task);
974 static void remove_range(struct dm_integrity_c *ic, struct dm_integrity_range *range)
978 spin_lock_irqsave(&ic->endio_wait.lock, flags);
979 remove_range_unlocked(ic, range);
980 spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
983 static void wait_and_add_new_range(struct dm_integrity_c *ic, struct dm_integrity_range *new_range)
985 new_range->waiting = true;
986 list_add_tail(&new_range->wait_entry, &ic->wait_list);
987 new_range->task = current;
989 __set_current_state(TASK_UNINTERRUPTIBLE);
990 spin_unlock_irq(&ic->endio_wait.lock);
992 spin_lock_irq(&ic->endio_wait.lock);
993 } while (unlikely(new_range->waiting));
996 static void init_journal_node(struct journal_node *node)
998 RB_CLEAR_NODE(&node->node);
999 node->sector = (sector_t)-1;
1002 static void add_journal_node(struct dm_integrity_c *ic, struct journal_node *node, sector_t sector)
1004 struct rb_node **link;
1005 struct rb_node *parent;
1007 node->sector = sector;
1008 BUG_ON(!RB_EMPTY_NODE(&node->node));
1010 link = &ic->journal_tree_root.rb_node;
1014 struct journal_node *j;
1016 j = container_of(parent, struct journal_node, node);
1017 if (sector < j->sector)
1018 link = &j->node.rb_left;
1020 link = &j->node.rb_right;
1023 rb_link_node(&node->node, parent, link);
1024 rb_insert_color(&node->node, &ic->journal_tree_root);
1027 static void remove_journal_node(struct dm_integrity_c *ic, struct journal_node *node)
1029 BUG_ON(RB_EMPTY_NODE(&node->node));
1030 rb_erase(&node->node, &ic->journal_tree_root);
1031 init_journal_node(node);
1034 #define NOT_FOUND (-1U)
1036 static unsigned find_journal_node(struct dm_integrity_c *ic, sector_t sector, sector_t *next_sector)
1038 struct rb_node *n = ic->journal_tree_root.rb_node;
1039 unsigned found = NOT_FOUND;
1040 *next_sector = (sector_t)-1;
1042 struct journal_node *j = container_of(n, struct journal_node, node);
1043 if (sector == j->sector) {
1044 found = j - ic->journal_tree;
1046 if (sector < j->sector) {
1047 *next_sector = j->sector;
1048 n = j->node.rb_left;
1050 n = j->node.rb_right;
1057 static bool test_journal_node(struct dm_integrity_c *ic, unsigned pos, sector_t sector)
1059 struct journal_node *node, *next_node;
1060 struct rb_node *next;
1062 if (unlikely(pos >= ic->journal_entries))
1064 node = &ic->journal_tree[pos];
1065 if (unlikely(RB_EMPTY_NODE(&node->node)))
1067 if (unlikely(node->sector != sector))
1070 next = rb_next(&node->node);
1071 if (unlikely(!next))
1074 next_node = container_of(next, struct journal_node, node);
1075 return next_node->sector != sector;
1078 static bool find_newer_committed_node(struct dm_integrity_c *ic, struct journal_node *node)
1080 struct rb_node *next;
1081 struct journal_node *next_node;
1082 unsigned next_section;
1084 BUG_ON(RB_EMPTY_NODE(&node->node));
1086 next = rb_next(&node->node);
1087 if (unlikely(!next))
1090 next_node = container_of(next, struct journal_node, node);
1092 if (next_node->sector != node->sector)
1095 next_section = (unsigned)(next_node - ic->journal_tree) / ic->journal_section_entries;
1096 if (next_section >= ic->committed_section &&
1097 next_section < ic->committed_section + ic->n_committed_sections)
1099 if (next_section + ic->journal_sections < ic->committed_section + ic->n_committed_sections)
1109 static int dm_integrity_rw_tag(struct dm_integrity_c *ic, unsigned char *tag, sector_t *metadata_block,
1110 unsigned *metadata_offset, unsigned total_size, int op)
1113 unsigned char *data, *dp;
1114 struct dm_buffer *b;
1118 r = dm_integrity_failed(ic);
1122 data = dm_bufio_read(ic->bufio, *metadata_block, &b);
1124 return PTR_ERR(data);
1126 to_copy = min((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - *metadata_offset, total_size);
1127 dp = data + *metadata_offset;
1128 if (op == TAG_READ) {
1129 memcpy(tag, dp, to_copy);
1130 } else if (op == TAG_WRITE) {
1131 memcpy(dp, tag, to_copy);
1132 dm_bufio_mark_partial_buffer_dirty(b, *metadata_offset, *metadata_offset + to_copy);
1134 /* e.g.: op == TAG_CMP */
1135 if (unlikely(memcmp(dp, tag, to_copy))) {
1138 for (i = 0; i < to_copy; i++) {
1139 if (dp[i] != tag[i])
1143 dm_bufio_release(b);
1147 dm_bufio_release(b);
1150 *metadata_offset += to_copy;
1151 if (unlikely(*metadata_offset == 1U << SECTOR_SHIFT << ic->log2_buffer_sectors)) {
1152 (*metadata_block)++;
1153 *metadata_offset = 0;
1155 total_size -= to_copy;
1156 } while (unlikely(total_size));
1161 static void dm_integrity_flush_buffers(struct dm_integrity_c *ic)
1164 r = dm_bufio_write_dirty_buffers(ic->bufio);
1166 dm_integrity_io_error(ic, "writing tags", r);
1169 static void sleep_on_endio_wait(struct dm_integrity_c *ic)
1171 DECLARE_WAITQUEUE(wait, current);
1172 __add_wait_queue(&ic->endio_wait, &wait);
1173 __set_current_state(TASK_UNINTERRUPTIBLE);
1174 spin_unlock_irq(&ic->endio_wait.lock);
1176 spin_lock_irq(&ic->endio_wait.lock);
1177 __remove_wait_queue(&ic->endio_wait, &wait);
1180 static void autocommit_fn(struct timer_list *t)
1182 struct dm_integrity_c *ic = from_timer(ic, t, autocommit_timer);
1184 if (likely(!dm_integrity_failed(ic)))
1185 queue_work(ic->commit_wq, &ic->commit_work);
1188 static void schedule_autocommit(struct dm_integrity_c *ic)
1190 if (!timer_pending(&ic->autocommit_timer))
1191 mod_timer(&ic->autocommit_timer, jiffies + ic->autocommit_jiffies);
1194 static void submit_flush_bio(struct dm_integrity_c *ic, struct dm_integrity_io *dio)
1197 unsigned long flags;
1199 spin_lock_irqsave(&ic->endio_wait.lock, flags);
1200 bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1201 bio_list_add(&ic->flush_bio_list, bio);
1202 spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
1204 queue_work(ic->commit_wq, &ic->commit_work);
1207 static void do_endio(struct dm_integrity_c *ic, struct bio *bio)
1209 int r = dm_integrity_failed(ic);
1210 if (unlikely(r) && !bio->bi_status)
1211 bio->bi_status = errno_to_blk_status(r);
1215 static void do_endio_flush(struct dm_integrity_c *ic, struct dm_integrity_io *dio)
1217 struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1219 if (unlikely(dio->fua) && likely(!bio->bi_status) && likely(!dm_integrity_failed(ic)))
1220 submit_flush_bio(ic, dio);
1225 static void dec_in_flight(struct dm_integrity_io *dio)
1227 if (atomic_dec_and_test(&dio->in_flight)) {
1228 struct dm_integrity_c *ic = dio->ic;
1231 remove_range(ic, &dio->range);
1233 if (unlikely(dio->write))
1234 schedule_autocommit(ic);
1236 bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1238 if (unlikely(dio->bi_status) && !bio->bi_status)
1239 bio->bi_status = dio->bi_status;
1240 if (likely(!bio->bi_status) && unlikely(bio_sectors(bio) != dio->range.n_sectors)) {
1241 dio->range.logical_sector += dio->range.n_sectors;
1242 bio_advance(bio, dio->range.n_sectors << SECTOR_SHIFT);
1243 INIT_WORK(&dio->work, integrity_bio_wait);
1244 queue_work(ic->wait_wq, &dio->work);
1247 do_endio_flush(ic, dio);
1251 static void integrity_end_io(struct bio *bio)
1253 struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
1255 bio->bi_iter = dio->orig_bi_iter;
1256 bio->bi_disk = dio->orig_bi_disk;
1257 bio->bi_partno = dio->orig_bi_partno;
1258 if (dio->orig_bi_integrity) {
1259 bio->bi_integrity = dio->orig_bi_integrity;
1260 bio->bi_opf |= REQ_INTEGRITY;
1262 bio->bi_end_io = dio->orig_bi_end_io;
1264 if (dio->completion)
1265 complete(dio->completion);
1270 static void integrity_sector_checksum(struct dm_integrity_c *ic, sector_t sector,
1271 const char *data, char *result)
1273 __u64 sector_le = cpu_to_le64(sector);
1274 SHASH_DESC_ON_STACK(req, ic->internal_hash);
1276 unsigned digest_size;
1278 req->tfm = ic->internal_hash;
1281 r = crypto_shash_init(req);
1282 if (unlikely(r < 0)) {
1283 dm_integrity_io_error(ic, "crypto_shash_init", r);
1287 r = crypto_shash_update(req, (const __u8 *)§or_le, sizeof sector_le);
1288 if (unlikely(r < 0)) {
1289 dm_integrity_io_error(ic, "crypto_shash_update", r);
1293 r = crypto_shash_update(req, data, ic->sectors_per_block << SECTOR_SHIFT);
1294 if (unlikely(r < 0)) {
1295 dm_integrity_io_error(ic, "crypto_shash_update", r);
1299 r = crypto_shash_final(req, result);
1300 if (unlikely(r < 0)) {
1301 dm_integrity_io_error(ic, "crypto_shash_final", r);
1305 digest_size = crypto_shash_digestsize(ic->internal_hash);
1306 if (unlikely(digest_size < ic->tag_size))
1307 memset(result + digest_size, 0, ic->tag_size - digest_size);
1312 /* this shouldn't happen anyway, the hash functions have no reason to fail */
1313 get_random_bytes(result, ic->tag_size);
1316 static void integrity_metadata(struct work_struct *w)
1318 struct dm_integrity_io *dio = container_of(w, struct dm_integrity_io, work);
1319 struct dm_integrity_c *ic = dio->ic;
1323 if (ic->internal_hash) {
1324 struct bvec_iter iter;
1326 unsigned digest_size = crypto_shash_digestsize(ic->internal_hash);
1327 struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1329 unsigned extra_space = unlikely(digest_size > ic->tag_size) ? digest_size - ic->tag_size : 0;
1330 char checksums_onstack[HASH_MAX_DIGESTSIZE];
1331 unsigned sectors_to_process = dio->range.n_sectors;
1332 sector_t sector = dio->range.logical_sector;
1334 if (unlikely(ic->mode == 'R'))
1337 checksums = kmalloc((PAGE_SIZE >> SECTOR_SHIFT >> ic->sb->log2_sectors_per_block) * ic->tag_size + extra_space,
1338 GFP_NOIO | __GFP_NORETRY | __GFP_NOWARN);
1340 checksums = checksums_onstack;
1341 if (WARN_ON(extra_space &&
1342 digest_size > sizeof(checksums_onstack))) {
1348 __bio_for_each_segment(bv, bio, iter, dio->orig_bi_iter) {
1350 char *mem, *checksums_ptr;
1353 mem = (char *)kmap_atomic(bv.bv_page) + bv.bv_offset;
1355 checksums_ptr = checksums;
1357 integrity_sector_checksum(ic, sector, mem + pos, checksums_ptr);
1358 checksums_ptr += ic->tag_size;
1359 sectors_to_process -= ic->sectors_per_block;
1360 pos += ic->sectors_per_block << SECTOR_SHIFT;
1361 sector += ic->sectors_per_block;
1362 } while (pos < bv.bv_len && sectors_to_process && checksums != checksums_onstack);
1365 r = dm_integrity_rw_tag(ic, checksums, &dio->metadata_block, &dio->metadata_offset,
1366 checksums_ptr - checksums, !dio->write ? TAG_CMP : TAG_WRITE);
1369 DMERR_LIMIT("Checksum failed at sector 0x%llx",
1370 (unsigned long long)(sector - ((r + ic->tag_size - 1) / ic->tag_size)));
1372 atomic64_inc(&ic->number_of_mismatches);
1374 if (likely(checksums != checksums_onstack))
1379 if (!sectors_to_process)
1382 if (unlikely(pos < bv.bv_len)) {
1383 bv.bv_offset += pos;
1389 if (likely(checksums != checksums_onstack))
1392 struct bio_integrity_payload *bip = dio->orig_bi_integrity;
1396 struct bvec_iter iter;
1397 unsigned data_to_process = dio->range.n_sectors;
1398 sector_to_block(ic, data_to_process);
1399 data_to_process *= ic->tag_size;
1401 bip_for_each_vec(biv, bip, iter) {
1405 BUG_ON(PageHighMem(biv.bv_page));
1406 tag = lowmem_page_address(biv.bv_page) + biv.bv_offset;
1407 this_len = min(biv.bv_len, data_to_process);
1408 r = dm_integrity_rw_tag(ic, tag, &dio->metadata_block, &dio->metadata_offset,
1409 this_len, !dio->write ? TAG_READ : TAG_WRITE);
1412 data_to_process -= this_len;
1413 if (!data_to_process)
1422 dio->bi_status = errno_to_blk_status(r);
1426 static int dm_integrity_map(struct dm_target *ti, struct bio *bio)
1428 struct dm_integrity_c *ic = ti->private;
1429 struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
1430 struct bio_integrity_payload *bip;
1432 sector_t area, offset;
1437 if (unlikely(bio->bi_opf & REQ_PREFLUSH)) {
1438 submit_flush_bio(ic, dio);
1439 return DM_MAPIO_SUBMITTED;
1442 dio->range.logical_sector = dm_target_offset(ti, bio->bi_iter.bi_sector);
1443 dio->write = bio_op(bio) == REQ_OP_WRITE;
1444 dio->fua = dio->write && bio->bi_opf & REQ_FUA;
1445 if (unlikely(dio->fua)) {
1447 * Don't pass down the FUA flag because we have to flush
1448 * disk cache anyway.
1450 bio->bi_opf &= ~REQ_FUA;
1452 if (unlikely(dio->range.logical_sector + bio_sectors(bio) > ic->provided_data_sectors)) {
1453 DMERR("Too big sector number: 0x%llx + 0x%x > 0x%llx",
1454 (unsigned long long)dio->range.logical_sector, bio_sectors(bio),
1455 (unsigned long long)ic->provided_data_sectors);
1456 return DM_MAPIO_KILL;
1458 if (unlikely((dio->range.logical_sector | bio_sectors(bio)) & (unsigned)(ic->sectors_per_block - 1))) {
1459 DMERR("Bio not aligned on %u sectors: 0x%llx, 0x%x",
1460 ic->sectors_per_block,
1461 (unsigned long long)dio->range.logical_sector, bio_sectors(bio));
1462 return DM_MAPIO_KILL;
1465 if (ic->sectors_per_block > 1) {
1466 struct bvec_iter iter;
1468 bio_for_each_segment(bv, bio, iter) {
1469 if (unlikely(bv.bv_len & ((ic->sectors_per_block << SECTOR_SHIFT) - 1))) {
1470 DMERR("Bio vector (%u,%u) is not aligned on %u-sector boundary",
1471 bv.bv_offset, bv.bv_len, ic->sectors_per_block);
1472 return DM_MAPIO_KILL;
1477 bip = bio_integrity(bio);
1478 if (!ic->internal_hash) {
1480 unsigned wanted_tag_size = bio_sectors(bio) >> ic->sb->log2_sectors_per_block;
1481 if (ic->log2_tag_size >= 0)
1482 wanted_tag_size <<= ic->log2_tag_size;
1484 wanted_tag_size *= ic->tag_size;
1485 if (unlikely(wanted_tag_size != bip->bip_iter.bi_size)) {
1486 DMERR("Invalid integrity data size %u, expected %u", bip->bip_iter.bi_size, wanted_tag_size);
1487 return DM_MAPIO_KILL;
1491 if (unlikely(bip != NULL)) {
1492 DMERR("Unexpected integrity data when using internal hash");
1493 return DM_MAPIO_KILL;
1497 if (unlikely(ic->mode == 'R') && unlikely(dio->write))
1498 return DM_MAPIO_KILL;
1500 get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
1501 dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset, &dio->metadata_offset);
1502 bio->bi_iter.bi_sector = get_data_sector(ic, area, offset);
1504 dm_integrity_map_continue(dio, true);
1505 return DM_MAPIO_SUBMITTED;
1508 static bool __journal_read_write(struct dm_integrity_io *dio, struct bio *bio,
1509 unsigned journal_section, unsigned journal_entry)
1511 struct dm_integrity_c *ic = dio->ic;
1512 sector_t logical_sector;
1515 logical_sector = dio->range.logical_sector;
1516 n_sectors = dio->range.n_sectors;
1518 struct bio_vec bv = bio_iovec(bio);
1521 if (unlikely(bv.bv_len >> SECTOR_SHIFT > n_sectors))
1522 bv.bv_len = n_sectors << SECTOR_SHIFT;
1523 n_sectors -= bv.bv_len >> SECTOR_SHIFT;
1524 bio_advance_iter(bio, &bio->bi_iter, bv.bv_len);
1526 mem = kmap_atomic(bv.bv_page);
1527 if (likely(dio->write))
1528 flush_dcache_page(bv.bv_page);
1531 struct journal_entry *je = access_journal_entry(ic, journal_section, journal_entry);
1533 if (unlikely(!dio->write)) {
1534 struct journal_sector *js;
1538 if (unlikely(journal_entry_is_inprogress(je))) {
1539 flush_dcache_page(bv.bv_page);
1542 __io_wait_event(ic->copy_to_journal_wait, !journal_entry_is_inprogress(je));
1546 BUG_ON(journal_entry_get_sector(je) != logical_sector);
1547 js = access_journal_data(ic, journal_section, journal_entry);
1548 mem_ptr = mem + bv.bv_offset;
1551 memcpy(mem_ptr, js, JOURNAL_SECTOR_DATA);
1552 *(commit_id_t *)(mem_ptr + JOURNAL_SECTOR_DATA) = je->last_bytes[s];
1554 mem_ptr += 1 << SECTOR_SHIFT;
1555 } while (++s < ic->sectors_per_block);
1556 #ifdef INTERNAL_VERIFY
1557 if (ic->internal_hash) {
1558 char checksums_onstack[max(HASH_MAX_DIGESTSIZE, MAX_TAG_SIZE)];
1560 integrity_sector_checksum(ic, logical_sector, mem + bv.bv_offset, checksums_onstack);
1561 if (unlikely(memcmp(checksums_onstack, journal_entry_tag(ic, je), ic->tag_size))) {
1562 DMERR_LIMIT("Checksum failed when reading from journal, at sector 0x%llx",
1563 (unsigned long long)logical_sector);
1569 if (!ic->internal_hash) {
1570 struct bio_integrity_payload *bip = bio_integrity(bio);
1571 unsigned tag_todo = ic->tag_size;
1572 char *tag_ptr = journal_entry_tag(ic, je);
1575 struct bio_vec biv = bvec_iter_bvec(bip->bip_vec, bip->bip_iter);
1576 unsigned tag_now = min(biv.bv_len, tag_todo);
1578 BUG_ON(PageHighMem(biv.bv_page));
1579 tag_addr = lowmem_page_address(biv.bv_page) + biv.bv_offset;
1580 if (likely(dio->write))
1581 memcpy(tag_ptr, tag_addr, tag_now);
1583 memcpy(tag_addr, tag_ptr, tag_now);
1584 bvec_iter_advance(bip->bip_vec, &bip->bip_iter, tag_now);
1586 tag_todo -= tag_now;
1587 } while (unlikely(tag_todo)); else {
1588 if (likely(dio->write))
1589 memset(tag_ptr, 0, tag_todo);
1593 if (likely(dio->write)) {
1594 struct journal_sector *js;
1597 js = access_journal_data(ic, journal_section, journal_entry);
1598 memcpy(js, mem + bv.bv_offset, ic->sectors_per_block << SECTOR_SHIFT);
1602 je->last_bytes[s] = js[s].commit_id;
1603 } while (++s < ic->sectors_per_block);
1605 if (ic->internal_hash) {
1606 unsigned digest_size = crypto_shash_digestsize(ic->internal_hash);
1607 if (unlikely(digest_size > ic->tag_size)) {
1608 char checksums_onstack[HASH_MAX_DIGESTSIZE];
1609 integrity_sector_checksum(ic, logical_sector, (char *)js, checksums_onstack);
1610 memcpy(journal_entry_tag(ic, je), checksums_onstack, ic->tag_size);
1612 integrity_sector_checksum(ic, logical_sector, (char *)js, journal_entry_tag(ic, je));
1615 journal_entry_set_sector(je, logical_sector);
1617 logical_sector += ic->sectors_per_block;
1620 if (unlikely(journal_entry == ic->journal_section_entries)) {
1623 wraparound_section(ic, &journal_section);
1626 bv.bv_offset += ic->sectors_per_block << SECTOR_SHIFT;
1627 } while (bv.bv_len -= ic->sectors_per_block << SECTOR_SHIFT);
1629 if (unlikely(!dio->write))
1630 flush_dcache_page(bv.bv_page);
1632 } while (n_sectors);
1634 if (likely(dio->write)) {
1636 if (unlikely(waitqueue_active(&ic->copy_to_journal_wait)))
1637 wake_up(&ic->copy_to_journal_wait);
1638 if (READ_ONCE(ic->free_sectors) <= ic->free_sectors_threshold) {
1639 queue_work(ic->commit_wq, &ic->commit_work);
1641 schedule_autocommit(ic);
1644 remove_range(ic, &dio->range);
1647 if (unlikely(bio->bi_iter.bi_size)) {
1648 sector_t area, offset;
1650 dio->range.logical_sector = logical_sector;
1651 get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
1652 dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset, &dio->metadata_offset);
1659 static void dm_integrity_map_continue(struct dm_integrity_io *dio, bool from_map)
1661 struct dm_integrity_c *ic = dio->ic;
1662 struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1663 unsigned journal_section, journal_entry;
1664 unsigned journal_read_pos;
1665 struct completion read_comp;
1666 bool need_sync_io = ic->internal_hash && !dio->write;
1668 if (need_sync_io && from_map) {
1669 INIT_WORK(&dio->work, integrity_bio_wait);
1670 queue_work(ic->metadata_wq, &dio->work);
1675 spin_lock_irq(&ic->endio_wait.lock);
1677 if (unlikely(dm_integrity_failed(ic))) {
1678 spin_unlock_irq(&ic->endio_wait.lock);
1682 dio->range.n_sectors = bio_sectors(bio);
1683 journal_read_pos = NOT_FOUND;
1684 if (likely(ic->mode == 'J')) {
1686 unsigned next_entry, i, pos;
1687 unsigned ws, we, range_sectors;
1689 dio->range.n_sectors = min(dio->range.n_sectors,
1690 ic->free_sectors << ic->sb->log2_sectors_per_block);
1691 if (unlikely(!dio->range.n_sectors)) {
1693 goto offload_to_thread;
1694 sleep_on_endio_wait(ic);
1697 range_sectors = dio->range.n_sectors >> ic->sb->log2_sectors_per_block;
1698 ic->free_sectors -= range_sectors;
1699 journal_section = ic->free_section;
1700 journal_entry = ic->free_section_entry;
1702 next_entry = ic->free_section_entry + range_sectors;
1703 ic->free_section_entry = next_entry % ic->journal_section_entries;
1704 ic->free_section += next_entry / ic->journal_section_entries;
1705 ic->n_uncommitted_sections += next_entry / ic->journal_section_entries;
1706 wraparound_section(ic, &ic->free_section);
1708 pos = journal_section * ic->journal_section_entries + journal_entry;
1709 ws = journal_section;
1713 struct journal_entry *je;
1715 add_journal_node(ic, &ic->journal_tree[pos], dio->range.logical_sector + i);
1717 if (unlikely(pos >= ic->journal_entries))
1720 je = access_journal_entry(ic, ws, we);
1721 BUG_ON(!journal_entry_is_unused(je));
1722 journal_entry_set_inprogress(je);
1724 if (unlikely(we == ic->journal_section_entries)) {
1727 wraparound_section(ic, &ws);
1729 } while ((i += ic->sectors_per_block) < dio->range.n_sectors);
1731 spin_unlock_irq(&ic->endio_wait.lock);
1732 goto journal_read_write;
1734 sector_t next_sector;
1735 journal_read_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
1736 if (likely(journal_read_pos == NOT_FOUND)) {
1737 if (unlikely(dio->range.n_sectors > next_sector - dio->range.logical_sector))
1738 dio->range.n_sectors = next_sector - dio->range.logical_sector;
1741 unsigned jp = journal_read_pos + 1;
1742 for (i = ic->sectors_per_block; i < dio->range.n_sectors; i += ic->sectors_per_block, jp++) {
1743 if (!test_journal_node(ic, jp, dio->range.logical_sector + i))
1746 dio->range.n_sectors = i;
1750 if (unlikely(!add_new_range(ic, &dio->range, true))) {
1752 * We must not sleep in the request routine because it could
1753 * stall bios on current->bio_list.
1754 * So, we offload the bio to a workqueue if we have to sleep.
1758 spin_unlock_irq(&ic->endio_wait.lock);
1759 INIT_WORK(&dio->work, integrity_bio_wait);
1760 queue_work(ic->wait_wq, &dio->work);
1763 wait_and_add_new_range(ic, &dio->range);
1765 spin_unlock_irq(&ic->endio_wait.lock);
1767 if (unlikely(journal_read_pos != NOT_FOUND)) {
1768 journal_section = journal_read_pos / ic->journal_section_entries;
1769 journal_entry = journal_read_pos % ic->journal_section_entries;
1770 goto journal_read_write;
1773 dio->in_flight = (atomic_t)ATOMIC_INIT(2);
1776 init_completion(&read_comp);
1777 dio->completion = &read_comp;
1779 dio->completion = NULL;
1781 dio->orig_bi_iter = bio->bi_iter;
1783 dio->orig_bi_disk = bio->bi_disk;
1784 dio->orig_bi_partno = bio->bi_partno;
1785 bio_set_dev(bio, ic->dev->bdev);
1787 dio->orig_bi_integrity = bio_integrity(bio);
1788 bio->bi_integrity = NULL;
1789 bio->bi_opf &= ~REQ_INTEGRITY;
1791 dio->orig_bi_end_io = bio->bi_end_io;
1792 bio->bi_end_io = integrity_end_io;
1794 bio->bi_iter.bi_size = dio->range.n_sectors << SECTOR_SHIFT;
1795 generic_make_request(bio);
1798 wait_for_completion_io(&read_comp);
1799 if (unlikely(ic->recalc_wq != NULL) &&
1800 ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING) &&
1801 dio->range.logical_sector + dio->range.n_sectors > le64_to_cpu(ic->sb->recalc_sector))
1803 if (likely(!bio->bi_status))
1804 integrity_metadata(&dio->work);
1810 INIT_WORK(&dio->work, integrity_metadata);
1811 queue_work(ic->metadata_wq, &dio->work);
1817 if (unlikely(__journal_read_write(dio, bio, journal_section, journal_entry)))
1820 do_endio_flush(ic, dio);
1824 static void integrity_bio_wait(struct work_struct *w)
1826 struct dm_integrity_io *dio = container_of(w, struct dm_integrity_io, work);
1828 dm_integrity_map_continue(dio, false);
1831 static void pad_uncommitted(struct dm_integrity_c *ic)
1833 if (ic->free_section_entry) {
1834 ic->free_sectors -= ic->journal_section_entries - ic->free_section_entry;
1835 ic->free_section_entry = 0;
1837 wraparound_section(ic, &ic->free_section);
1838 ic->n_uncommitted_sections++;
1840 WARN_ON(ic->journal_sections * ic->journal_section_entries !=
1841 (ic->n_uncommitted_sections + ic->n_committed_sections) * ic->journal_section_entries + ic->free_sectors);
1844 static void integrity_commit(struct work_struct *w)
1846 struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, commit_work);
1847 unsigned commit_start, commit_sections;
1849 struct bio *flushes;
1851 del_timer(&ic->autocommit_timer);
1853 spin_lock_irq(&ic->endio_wait.lock);
1854 flushes = bio_list_get(&ic->flush_bio_list);
1855 if (unlikely(ic->mode != 'J')) {
1856 spin_unlock_irq(&ic->endio_wait.lock);
1857 dm_integrity_flush_buffers(ic);
1858 goto release_flush_bios;
1861 pad_uncommitted(ic);
1862 commit_start = ic->uncommitted_section;
1863 commit_sections = ic->n_uncommitted_sections;
1864 spin_unlock_irq(&ic->endio_wait.lock);
1866 if (!commit_sections)
1867 goto release_flush_bios;
1870 for (n = 0; n < commit_sections; n++) {
1871 for (j = 0; j < ic->journal_section_entries; j++) {
1872 struct journal_entry *je;
1873 je = access_journal_entry(ic, i, j);
1874 io_wait_event(ic->copy_to_journal_wait, !journal_entry_is_inprogress(je));
1876 for (j = 0; j < ic->journal_section_sectors; j++) {
1877 struct journal_sector *js;
1878 js = access_journal(ic, i, j);
1879 js->commit_id = dm_integrity_commit_id(ic, i, j, ic->commit_seq);
1882 if (unlikely(i >= ic->journal_sections))
1883 ic->commit_seq = next_commit_seq(ic->commit_seq);
1884 wraparound_section(ic, &i);
1888 write_journal(ic, commit_start, commit_sections);
1890 spin_lock_irq(&ic->endio_wait.lock);
1891 ic->uncommitted_section += commit_sections;
1892 wraparound_section(ic, &ic->uncommitted_section);
1893 ic->n_uncommitted_sections -= commit_sections;
1894 ic->n_committed_sections += commit_sections;
1895 spin_unlock_irq(&ic->endio_wait.lock);
1897 if (READ_ONCE(ic->free_sectors) <= ic->free_sectors_threshold)
1898 queue_work(ic->writer_wq, &ic->writer_work);
1902 struct bio *next = flushes->bi_next;
1903 flushes->bi_next = NULL;
1904 do_endio(ic, flushes);
1909 static void complete_copy_from_journal(unsigned long error, void *context)
1911 struct journal_io *io = context;
1912 struct journal_completion *comp = io->comp;
1913 struct dm_integrity_c *ic = comp->ic;
1914 remove_range(ic, &io->range);
1915 mempool_free(io, &ic->journal_io_mempool);
1916 if (unlikely(error != 0))
1917 dm_integrity_io_error(ic, "copying from journal", -EIO);
1918 complete_journal_op(comp);
1921 static void restore_last_bytes(struct dm_integrity_c *ic, struct journal_sector *js,
1922 struct journal_entry *je)
1926 js->commit_id = je->last_bytes[s];
1928 } while (++s < ic->sectors_per_block);
1931 static void do_journal_write(struct dm_integrity_c *ic, unsigned write_start,
1932 unsigned write_sections, bool from_replay)
1935 struct journal_completion comp;
1936 struct blk_plug plug;
1938 blk_start_plug(&plug);
1941 comp.in_flight = (atomic_t)ATOMIC_INIT(1);
1942 init_completion(&comp.comp);
1945 for (n = 0; n < write_sections; n++, i++, wraparound_section(ic, &i)) {
1946 #ifndef INTERNAL_VERIFY
1947 if (unlikely(from_replay))
1949 rw_section_mac(ic, i, false);
1950 for (j = 0; j < ic->journal_section_entries; j++) {
1951 struct journal_entry *je = access_journal_entry(ic, i, j);
1952 sector_t sec, area, offset;
1953 unsigned k, l, next_loop;
1954 sector_t metadata_block;
1955 unsigned metadata_offset;
1956 struct journal_io *io;
1958 if (journal_entry_is_unused(je))
1960 BUG_ON(unlikely(journal_entry_is_inprogress(je)) && !from_replay);
1961 sec = journal_entry_get_sector(je);
1962 if (unlikely(from_replay)) {
1963 if (unlikely(sec & (unsigned)(ic->sectors_per_block - 1))) {
1964 dm_integrity_io_error(ic, "invalid sector in journal", -EIO);
1965 sec &= ~(sector_t)(ic->sectors_per_block - 1);
1968 get_area_and_offset(ic, sec, &area, &offset);
1969 restore_last_bytes(ic, access_journal_data(ic, i, j), je);
1970 for (k = j + 1; k < ic->journal_section_entries; k++) {
1971 struct journal_entry *je2 = access_journal_entry(ic, i, k);
1972 sector_t sec2, area2, offset2;
1973 if (journal_entry_is_unused(je2))
1975 BUG_ON(unlikely(journal_entry_is_inprogress(je2)) && !from_replay);
1976 sec2 = journal_entry_get_sector(je2);
1977 get_area_and_offset(ic, sec2, &area2, &offset2);
1978 if (area2 != area || offset2 != offset + ((k - j) << ic->sb->log2_sectors_per_block))
1980 restore_last_bytes(ic, access_journal_data(ic, i, k), je2);
1984 io = mempool_alloc(&ic->journal_io_mempool, GFP_NOIO);
1986 io->range.logical_sector = sec;
1987 io->range.n_sectors = (k - j) << ic->sb->log2_sectors_per_block;
1989 spin_lock_irq(&ic->endio_wait.lock);
1990 if (unlikely(!add_new_range(ic, &io->range, true)))
1991 wait_and_add_new_range(ic, &io->range);
1993 if (likely(!from_replay)) {
1994 struct journal_node *section_node = &ic->journal_tree[i * ic->journal_section_entries];
1996 /* don't write if there is newer committed sector */
1997 while (j < k && find_newer_committed_node(ic, §ion_node[j])) {
1998 struct journal_entry *je2 = access_journal_entry(ic, i, j);
2000 journal_entry_set_unused(je2);
2001 remove_journal_node(ic, §ion_node[j]);
2003 sec += ic->sectors_per_block;
2004 offset += ic->sectors_per_block;
2006 while (j < k && find_newer_committed_node(ic, §ion_node[k - 1])) {
2007 struct journal_entry *je2 = access_journal_entry(ic, i, k - 1);
2009 journal_entry_set_unused(je2);
2010 remove_journal_node(ic, §ion_node[k - 1]);
2014 remove_range_unlocked(ic, &io->range);
2015 spin_unlock_irq(&ic->endio_wait.lock);
2016 mempool_free(io, &ic->journal_io_mempool);
2019 for (l = j; l < k; l++) {
2020 remove_journal_node(ic, §ion_node[l]);
2023 spin_unlock_irq(&ic->endio_wait.lock);
2025 metadata_block = get_metadata_sector_and_offset(ic, area, offset, &metadata_offset);
2026 for (l = j; l < k; l++) {
2028 struct journal_entry *je2 = access_journal_entry(ic, i, l);
2031 #ifndef INTERNAL_VERIFY
2032 unlikely(from_replay) &&
2034 ic->internal_hash) {
2035 char test_tag[max_t(size_t, HASH_MAX_DIGESTSIZE, MAX_TAG_SIZE)];
2037 integrity_sector_checksum(ic, sec + ((l - j) << ic->sb->log2_sectors_per_block),
2038 (char *)access_journal_data(ic, i, l), test_tag);
2039 if (unlikely(memcmp(test_tag, journal_entry_tag(ic, je2), ic->tag_size)))
2040 dm_integrity_io_error(ic, "tag mismatch when replaying journal", -EILSEQ);
2043 journal_entry_set_unused(je2);
2044 r = dm_integrity_rw_tag(ic, journal_entry_tag(ic, je2), &metadata_block, &metadata_offset,
2045 ic->tag_size, TAG_WRITE);
2047 dm_integrity_io_error(ic, "reading tags", r);
2051 atomic_inc(&comp.in_flight);
2052 copy_from_journal(ic, i, j << ic->sb->log2_sectors_per_block,
2053 (k - j) << ic->sb->log2_sectors_per_block,
2054 get_data_sector(ic, area, offset),
2055 complete_copy_from_journal, io);
2061 dm_bufio_write_dirty_buffers_async(ic->bufio);
2063 blk_finish_plug(&plug);
2065 complete_journal_op(&comp);
2066 wait_for_completion_io(&comp.comp);
2068 dm_integrity_flush_buffers(ic);
2071 static void integrity_writer(struct work_struct *w)
2073 struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, writer_work);
2074 unsigned write_start, write_sections;
2076 unsigned prev_free_sectors;
2078 /* the following test is not needed, but it tests the replay code */
2079 if (READ_ONCE(ic->suspending) && !ic->meta_dev)
2082 spin_lock_irq(&ic->endio_wait.lock);
2083 write_start = ic->committed_section;
2084 write_sections = ic->n_committed_sections;
2085 spin_unlock_irq(&ic->endio_wait.lock);
2087 if (!write_sections)
2090 do_journal_write(ic, write_start, write_sections, false);
2092 spin_lock_irq(&ic->endio_wait.lock);
2094 ic->committed_section += write_sections;
2095 wraparound_section(ic, &ic->committed_section);
2096 ic->n_committed_sections -= write_sections;
2098 prev_free_sectors = ic->free_sectors;
2099 ic->free_sectors += write_sections * ic->journal_section_entries;
2100 if (unlikely(!prev_free_sectors))
2101 wake_up_locked(&ic->endio_wait);
2103 spin_unlock_irq(&ic->endio_wait.lock);
2106 static void recalc_write_super(struct dm_integrity_c *ic)
2110 dm_integrity_flush_buffers(ic);
2111 if (dm_integrity_failed(ic))
2115 r = sync_rw_sb(ic, REQ_OP_WRITE, 0);
2117 dm_integrity_io_error(ic, "writing superblock", r);
2120 static void integrity_recalc(struct work_struct *w)
2122 struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, recalc_work);
2123 struct dm_integrity_range range;
2124 struct dm_io_request io_req;
2125 struct dm_io_region io_loc;
2126 sector_t area, offset;
2127 sector_t metadata_block;
2128 unsigned metadata_offset;
2132 unsigned super_counter = 0;
2134 spin_lock_irq(&ic->endio_wait.lock);
2138 if (unlikely(READ_ONCE(ic->suspending)))
2141 range.logical_sector = le64_to_cpu(ic->sb->recalc_sector);
2142 if (unlikely(range.logical_sector >= ic->provided_data_sectors))
2145 get_area_and_offset(ic, range.logical_sector, &area, &offset);
2146 range.n_sectors = min((sector_t)RECALC_SECTORS, ic->provided_data_sectors - range.logical_sector);
2148 range.n_sectors = min(range.n_sectors, (1U << ic->sb->log2_interleave_sectors) - (unsigned)offset);
2150 if (unlikely(!add_new_range(ic, &range, true)))
2151 wait_and_add_new_range(ic, &range);
2153 spin_unlock_irq(&ic->endio_wait.lock);
2155 if (unlikely(++super_counter == RECALC_WRITE_SUPER)) {
2156 recalc_write_super(ic);
2160 if (unlikely(dm_integrity_failed(ic)))
2163 io_req.bi_op = REQ_OP_READ;
2164 io_req.bi_op_flags = 0;
2165 io_req.mem.type = DM_IO_VMA;
2166 io_req.mem.ptr.addr = ic->recalc_buffer;
2167 io_req.notify.fn = NULL;
2168 io_req.client = ic->io;
2169 io_loc.bdev = ic->dev->bdev;
2170 io_loc.sector = get_data_sector(ic, area, offset);
2171 io_loc.count = range.n_sectors;
2173 r = dm_io(&io_req, 1, &io_loc, NULL);
2175 dm_integrity_io_error(ic, "reading data", r);
2179 t = ic->recalc_tags;
2180 for (i = 0; i < range.n_sectors; i += ic->sectors_per_block) {
2181 integrity_sector_checksum(ic, range.logical_sector + i, ic->recalc_buffer + (i << SECTOR_SHIFT), t);
2185 metadata_block = get_metadata_sector_and_offset(ic, area, offset, &metadata_offset);
2187 r = dm_integrity_rw_tag(ic, ic->recalc_tags, &metadata_block, &metadata_offset, t - ic->recalc_tags, TAG_WRITE);
2189 dm_integrity_io_error(ic, "writing tags", r);
2193 spin_lock_irq(&ic->endio_wait.lock);
2194 remove_range_unlocked(ic, &range);
2195 ic->sb->recalc_sector = cpu_to_le64(range.logical_sector + range.n_sectors);
2199 remove_range(ic, &range);
2203 spin_unlock_irq(&ic->endio_wait.lock);
2205 recalc_write_super(ic);
2208 static void init_journal(struct dm_integrity_c *ic, unsigned start_section,
2209 unsigned n_sections, unsigned char commit_seq)
2216 for (n = 0; n < n_sections; n++) {
2217 i = start_section + n;
2218 wraparound_section(ic, &i);
2219 for (j = 0; j < ic->journal_section_sectors; j++) {
2220 struct journal_sector *js = access_journal(ic, i, j);
2221 memset(&js->entries, 0, JOURNAL_SECTOR_DATA);
2222 js->commit_id = dm_integrity_commit_id(ic, i, j, commit_seq);
2224 for (j = 0; j < ic->journal_section_entries; j++) {
2225 struct journal_entry *je = access_journal_entry(ic, i, j);
2226 journal_entry_set_unused(je);
2230 write_journal(ic, start_section, n_sections);
2233 static int find_commit_seq(struct dm_integrity_c *ic, unsigned i, unsigned j, commit_id_t id)
2236 for (k = 0; k < N_COMMIT_IDS; k++) {
2237 if (dm_integrity_commit_id(ic, i, j, k) == id)
2240 dm_integrity_io_error(ic, "journal commit id", -EIO);
2244 static void replay_journal(struct dm_integrity_c *ic)
2247 bool used_commit_ids[N_COMMIT_IDS];
2248 unsigned max_commit_id_sections[N_COMMIT_IDS];
2249 unsigned write_start, write_sections;
2250 unsigned continue_section;
2252 unsigned char unused, last_used, want_commit_seq;
2254 if (ic->mode == 'R')
2257 if (ic->journal_uptodate)
2263 if (!ic->just_formatted) {
2264 DEBUG_print("reading journal\n");
2265 rw_journal(ic, REQ_OP_READ, 0, 0, ic->journal_sections, NULL);
2267 DEBUG_bytes(lowmem_page_address(ic->journal_io[0].page), 64, "read journal");
2268 if (ic->journal_io) {
2269 struct journal_completion crypt_comp;
2271 init_completion(&crypt_comp.comp);
2272 crypt_comp.in_flight = (atomic_t)ATOMIC_INIT(0);
2273 encrypt_journal(ic, false, 0, ic->journal_sections, &crypt_comp);
2274 wait_for_completion(&crypt_comp.comp);
2276 DEBUG_bytes(lowmem_page_address(ic->journal[0].page), 64, "decrypted journal");
2279 if (dm_integrity_failed(ic))
2282 journal_empty = true;
2283 memset(used_commit_ids, 0, sizeof used_commit_ids);
2284 memset(max_commit_id_sections, 0, sizeof max_commit_id_sections);
2285 for (i = 0; i < ic->journal_sections; i++) {
2286 for (j = 0; j < ic->journal_section_sectors; j++) {
2288 struct journal_sector *js = access_journal(ic, i, j);
2289 k = find_commit_seq(ic, i, j, js->commit_id);
2292 used_commit_ids[k] = true;
2293 max_commit_id_sections[k] = i;
2295 if (journal_empty) {
2296 for (j = 0; j < ic->journal_section_entries; j++) {
2297 struct journal_entry *je = access_journal_entry(ic, i, j);
2298 if (!journal_entry_is_unused(je)) {
2299 journal_empty = false;
2306 if (!used_commit_ids[N_COMMIT_IDS - 1]) {
2307 unused = N_COMMIT_IDS - 1;
2308 while (unused && !used_commit_ids[unused - 1])
2311 for (unused = 0; unused < N_COMMIT_IDS; unused++)
2312 if (!used_commit_ids[unused])
2314 if (unused == N_COMMIT_IDS) {
2315 dm_integrity_io_error(ic, "journal commit ids", -EIO);
2319 DEBUG_print("first unused commit seq %d [%d,%d,%d,%d]\n",
2320 unused, used_commit_ids[0], used_commit_ids[1],
2321 used_commit_ids[2], used_commit_ids[3]);
2323 last_used = prev_commit_seq(unused);
2324 want_commit_seq = prev_commit_seq(last_used);
2326 if (!used_commit_ids[want_commit_seq] && used_commit_ids[prev_commit_seq(want_commit_seq)])
2327 journal_empty = true;
2329 write_start = max_commit_id_sections[last_used] + 1;
2330 if (unlikely(write_start >= ic->journal_sections))
2331 want_commit_seq = next_commit_seq(want_commit_seq);
2332 wraparound_section(ic, &write_start);
2335 for (write_sections = 0; write_sections < ic->journal_sections; write_sections++) {
2336 for (j = 0; j < ic->journal_section_sectors; j++) {
2337 struct journal_sector *js = access_journal(ic, i, j);
2339 if (js->commit_id != dm_integrity_commit_id(ic, i, j, want_commit_seq)) {
2341 * This could be caused by crash during writing.
2342 * We won't replay the inconsistent part of the
2345 DEBUG_print("commit id mismatch at position (%u, %u): %d != %d\n",
2346 i, j, find_commit_seq(ic, i, j, js->commit_id), want_commit_seq);
2351 if (unlikely(i >= ic->journal_sections))
2352 want_commit_seq = next_commit_seq(want_commit_seq);
2353 wraparound_section(ic, &i);
2357 if (!journal_empty) {
2358 DEBUG_print("replaying %u sections, starting at %u, commit seq %d\n",
2359 write_sections, write_start, want_commit_seq);
2360 do_journal_write(ic, write_start, write_sections, true);
2363 if (write_sections == ic->journal_sections && (ic->mode == 'J' || journal_empty)) {
2364 continue_section = write_start;
2365 ic->commit_seq = want_commit_seq;
2366 DEBUG_print("continuing from section %u, commit seq %d\n", write_start, ic->commit_seq);
2369 unsigned char erase_seq;
2371 DEBUG_print("clearing journal\n");
2373 erase_seq = prev_commit_seq(prev_commit_seq(last_used));
2375 init_journal(ic, s, 1, erase_seq);
2377 wraparound_section(ic, &s);
2378 if (ic->journal_sections >= 2) {
2379 init_journal(ic, s, ic->journal_sections - 2, erase_seq);
2380 s += ic->journal_sections - 2;
2381 wraparound_section(ic, &s);
2382 init_journal(ic, s, 1, erase_seq);
2385 continue_section = 0;
2386 ic->commit_seq = next_commit_seq(erase_seq);
2389 ic->committed_section = continue_section;
2390 ic->n_committed_sections = 0;
2392 ic->uncommitted_section = continue_section;
2393 ic->n_uncommitted_sections = 0;
2395 ic->free_section = continue_section;
2396 ic->free_section_entry = 0;
2397 ic->free_sectors = ic->journal_entries;
2399 ic->journal_tree_root = RB_ROOT;
2400 for (i = 0; i < ic->journal_entries; i++)
2401 init_journal_node(&ic->journal_tree[i]);
2404 static void dm_integrity_postsuspend(struct dm_target *ti)
2406 struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
2408 del_timer_sync(&ic->autocommit_timer);
2410 WRITE_ONCE(ic->suspending, 1);
2413 drain_workqueue(ic->recalc_wq);
2415 queue_work(ic->commit_wq, &ic->commit_work);
2416 drain_workqueue(ic->commit_wq);
2418 if (ic->mode == 'J') {
2420 queue_work(ic->writer_wq, &ic->writer_work);
2421 drain_workqueue(ic->writer_wq);
2422 dm_integrity_flush_buffers(ic);
2425 WRITE_ONCE(ic->suspending, 0);
2427 BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
2429 ic->journal_uptodate = true;
2432 static void dm_integrity_resume(struct dm_target *ti)
2434 struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
2438 if (ic->recalc_wq && ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
2439 __u64 recalc_pos = le64_to_cpu(ic->sb->recalc_sector);
2440 if (recalc_pos < ic->provided_data_sectors) {
2441 queue_work(ic->recalc_wq, &ic->recalc_work);
2442 } else if (recalc_pos > ic->provided_data_sectors) {
2443 ic->sb->recalc_sector = cpu_to_le64(ic->provided_data_sectors);
2444 recalc_write_super(ic);
2449 static void dm_integrity_status(struct dm_target *ti, status_type_t type,
2450 unsigned status_flags, char *result, unsigned maxlen)
2452 struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
2457 case STATUSTYPE_INFO:
2459 (unsigned long long)atomic64_read(&ic->number_of_mismatches),
2460 (unsigned long long)ic->provided_data_sectors);
2461 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
2462 DMEMIT(" %llu", (unsigned long long)le64_to_cpu(ic->sb->recalc_sector));
2467 case STATUSTYPE_TABLE: {
2468 __u64 watermark_percentage = (__u64)(ic->journal_entries - ic->free_sectors_threshold) * 100;
2469 watermark_percentage += ic->journal_entries / 2;
2470 do_div(watermark_percentage, ic->journal_entries);
2472 arg_count += !!ic->meta_dev;
2473 arg_count += ic->sectors_per_block != 1;
2474 arg_count += !!(ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING));
2475 arg_count += ic->mode == 'J';
2476 arg_count += ic->mode == 'J';
2477 arg_count += !!ic->internal_hash_alg.alg_string;
2478 arg_count += !!ic->journal_crypt_alg.alg_string;
2479 arg_count += !!ic->journal_mac_alg.alg_string;
2480 DMEMIT("%s %llu %u %c %u", ic->dev->name, (unsigned long long)ic->start,
2481 ic->tag_size, ic->mode, arg_count);
2483 DMEMIT(" meta_device:%s", ic->meta_dev->name);
2484 if (ic->sectors_per_block != 1)
2485 DMEMIT(" block_size:%u", ic->sectors_per_block << SECTOR_SHIFT);
2486 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
2487 DMEMIT(" recalculate");
2488 DMEMIT(" journal_sectors:%u", ic->initial_sectors - SB_SECTORS);
2489 DMEMIT(" interleave_sectors:%u", 1U << ic->sb->log2_interleave_sectors);
2490 DMEMIT(" buffer_sectors:%u", 1U << ic->log2_buffer_sectors);
2491 if (ic->mode == 'J') {
2492 DMEMIT(" journal_watermark:%u", (unsigned)watermark_percentage);
2493 DMEMIT(" commit_time:%u", ic->autocommit_msec);
2496 #define EMIT_ALG(a, n) \
2498 if (ic->a.alg_string) { \
2499 DMEMIT(" %s:%s", n, ic->a.alg_string); \
2500 if (ic->a.key_string) \
2501 DMEMIT(":%s", ic->a.key_string);\
2504 EMIT_ALG(internal_hash_alg, "internal_hash");
2505 EMIT_ALG(journal_crypt_alg, "journal_crypt");
2506 EMIT_ALG(journal_mac_alg, "journal_mac");
2512 static int dm_integrity_iterate_devices(struct dm_target *ti,
2513 iterate_devices_callout_fn fn, void *data)
2515 struct dm_integrity_c *ic = ti->private;
2518 return fn(ti, ic->dev, ic->start + ic->initial_sectors + ic->metadata_run, ti->len, data);
2520 return fn(ti, ic->dev, 0, ti->len, data);
2523 static void dm_integrity_io_hints(struct dm_target *ti, struct queue_limits *limits)
2525 struct dm_integrity_c *ic = ti->private;
2527 if (ic->sectors_per_block > 1) {
2528 limits->logical_block_size = ic->sectors_per_block << SECTOR_SHIFT;
2529 limits->physical_block_size = ic->sectors_per_block << SECTOR_SHIFT;
2530 blk_limits_io_min(limits, ic->sectors_per_block << SECTOR_SHIFT);
2534 static void calculate_journal_section_size(struct dm_integrity_c *ic)
2536 unsigned sector_space = JOURNAL_SECTOR_DATA;
2538 ic->journal_sections = le32_to_cpu(ic->sb->journal_sections);
2539 ic->journal_entry_size = roundup(offsetof(struct journal_entry, last_bytes[ic->sectors_per_block]) + ic->tag_size,
2540 JOURNAL_ENTRY_ROUNDUP);
2542 if (ic->sb->flags & cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC))
2543 sector_space -= JOURNAL_MAC_PER_SECTOR;
2544 ic->journal_entries_per_sector = sector_space / ic->journal_entry_size;
2545 ic->journal_section_entries = ic->journal_entries_per_sector * JOURNAL_BLOCK_SECTORS;
2546 ic->journal_section_sectors = (ic->journal_section_entries << ic->sb->log2_sectors_per_block) + JOURNAL_BLOCK_SECTORS;
2547 ic->journal_entries = ic->journal_section_entries * ic->journal_sections;
2550 static int calculate_device_limits(struct dm_integrity_c *ic)
2552 __u64 initial_sectors;
2554 calculate_journal_section_size(ic);
2555 initial_sectors = SB_SECTORS + (__u64)ic->journal_section_sectors * ic->journal_sections;
2556 if (initial_sectors + METADATA_PADDING_SECTORS >= ic->meta_device_sectors || initial_sectors > UINT_MAX)
2558 ic->initial_sectors = initial_sectors;
2560 if (!ic->meta_dev) {
2561 sector_t last_sector, last_area, last_offset;
2563 ic->metadata_run = roundup((__u64)ic->tag_size << (ic->sb->log2_interleave_sectors - ic->sb->log2_sectors_per_block),
2564 (__u64)(1 << SECTOR_SHIFT << METADATA_PADDING_SECTORS)) >> SECTOR_SHIFT;
2565 if (!(ic->metadata_run & (ic->metadata_run - 1)))
2566 ic->log2_metadata_run = __ffs(ic->metadata_run);
2568 ic->log2_metadata_run = -1;
2570 get_area_and_offset(ic, ic->provided_data_sectors - 1, &last_area, &last_offset);
2571 last_sector = get_data_sector(ic, last_area, last_offset);
2572 if (last_sector < ic->start || last_sector >= ic->meta_device_sectors)
2575 __u64 meta_size = (ic->provided_data_sectors >> ic->sb->log2_sectors_per_block) * ic->tag_size;
2576 meta_size = (meta_size + ((1U << (ic->log2_buffer_sectors + SECTOR_SHIFT)) - 1))
2577 >> (ic->log2_buffer_sectors + SECTOR_SHIFT);
2578 meta_size <<= ic->log2_buffer_sectors;
2579 if (ic->initial_sectors + meta_size < ic->initial_sectors ||
2580 ic->initial_sectors + meta_size > ic->meta_device_sectors)
2582 ic->metadata_run = 1;
2583 ic->log2_metadata_run = 0;
2589 static int initialize_superblock(struct dm_integrity_c *ic, unsigned journal_sectors, unsigned interleave_sectors)
2591 unsigned journal_sections;
2594 memset(ic->sb, 0, SB_SECTORS << SECTOR_SHIFT);
2595 memcpy(ic->sb->magic, SB_MAGIC, 8);
2596 ic->sb->integrity_tag_size = cpu_to_le16(ic->tag_size);
2597 ic->sb->log2_sectors_per_block = __ffs(ic->sectors_per_block);
2598 if (ic->journal_mac_alg.alg_string)
2599 ic->sb->flags |= cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC);
2601 calculate_journal_section_size(ic);
2602 journal_sections = journal_sectors / ic->journal_section_sectors;
2603 if (!journal_sections)
2604 journal_sections = 1;
2606 if (!ic->meta_dev) {
2607 ic->sb->journal_sections = cpu_to_le32(journal_sections);
2608 if (!interleave_sectors)
2609 interleave_sectors = DEFAULT_INTERLEAVE_SECTORS;
2610 ic->sb->log2_interleave_sectors = __fls(interleave_sectors);
2611 ic->sb->log2_interleave_sectors = max((__u8)MIN_LOG2_INTERLEAVE_SECTORS, ic->sb->log2_interleave_sectors);
2612 ic->sb->log2_interleave_sectors = min((__u8)MAX_LOG2_INTERLEAVE_SECTORS, ic->sb->log2_interleave_sectors);
2614 ic->provided_data_sectors = 0;
2615 for (test_bit = fls64(ic->meta_device_sectors) - 1; test_bit >= 3; test_bit--) {
2616 __u64 prev_data_sectors = ic->provided_data_sectors;
2618 ic->provided_data_sectors |= (sector_t)1 << test_bit;
2619 if (calculate_device_limits(ic))
2620 ic->provided_data_sectors = prev_data_sectors;
2622 if (!ic->provided_data_sectors)
2625 ic->sb->log2_interleave_sectors = 0;
2626 ic->provided_data_sectors = ic->data_device_sectors;
2627 ic->provided_data_sectors &= ~(sector_t)(ic->sectors_per_block - 1);
2630 ic->sb->journal_sections = cpu_to_le32(0);
2631 for (test_bit = fls(journal_sections) - 1; test_bit >= 0; test_bit--) {
2632 __u32 prev_journal_sections = le32_to_cpu(ic->sb->journal_sections);
2633 __u32 test_journal_sections = prev_journal_sections | (1U << test_bit);
2634 if (test_journal_sections > journal_sections)
2636 ic->sb->journal_sections = cpu_to_le32(test_journal_sections);
2637 if (calculate_device_limits(ic))
2638 ic->sb->journal_sections = cpu_to_le32(prev_journal_sections);
2641 if (!le32_to_cpu(ic->sb->journal_sections)) {
2642 if (ic->log2_buffer_sectors > 3) {
2643 ic->log2_buffer_sectors--;
2644 goto try_smaller_buffer;
2650 ic->sb->provided_data_sectors = cpu_to_le64(ic->provided_data_sectors);
2657 static void dm_integrity_set(struct dm_target *ti, struct dm_integrity_c *ic)
2659 struct gendisk *disk = dm_disk(dm_table_get_md(ti->table));
2660 struct blk_integrity bi;
2662 memset(&bi, 0, sizeof(bi));
2663 bi.profile = &dm_integrity_profile;
2664 bi.tuple_size = ic->tag_size;
2665 bi.tag_size = bi.tuple_size;
2666 bi.interval_exp = ic->sb->log2_sectors_per_block + SECTOR_SHIFT;
2668 blk_integrity_register(disk, &bi);
2669 blk_queue_max_integrity_segments(disk->queue, UINT_MAX);
2672 static void dm_integrity_free_page_list(struct dm_integrity_c *ic, struct page_list *pl)
2678 for (i = 0; i < ic->journal_pages; i++)
2680 __free_page(pl[i].page);
2684 static struct page_list *dm_integrity_alloc_page_list(struct dm_integrity_c *ic)
2686 size_t page_list_desc_size = ic->journal_pages * sizeof(struct page_list);
2687 struct page_list *pl;
2690 pl = kvmalloc(page_list_desc_size, GFP_KERNEL | __GFP_ZERO);
2694 for (i = 0; i < ic->journal_pages; i++) {
2695 pl[i].page = alloc_page(GFP_KERNEL);
2697 dm_integrity_free_page_list(ic, pl);
2701 pl[i - 1].next = &pl[i];
2707 static void dm_integrity_free_journal_scatterlist(struct dm_integrity_c *ic, struct scatterlist **sl)
2710 for (i = 0; i < ic->journal_sections; i++)
2715 static struct scatterlist **dm_integrity_alloc_journal_scatterlist(struct dm_integrity_c *ic, struct page_list *pl)
2717 struct scatterlist **sl;
2720 sl = kvmalloc_array(ic->journal_sections,
2721 sizeof(struct scatterlist *),
2722 GFP_KERNEL | __GFP_ZERO);
2726 for (i = 0; i < ic->journal_sections; i++) {
2727 struct scatterlist *s;
2728 unsigned start_index, start_offset;
2729 unsigned end_index, end_offset;
2733 page_list_location(ic, i, 0, &start_index, &start_offset);
2734 page_list_location(ic, i, ic->journal_section_sectors - 1, &end_index, &end_offset);
2736 n_pages = (end_index - start_index + 1);
2738 s = kvmalloc_array(n_pages, sizeof(struct scatterlist),
2741 dm_integrity_free_journal_scatterlist(ic, sl);
2745 sg_init_table(s, n_pages);
2746 for (idx = start_index; idx <= end_index; idx++) {
2747 char *va = lowmem_page_address(pl[idx].page);
2748 unsigned start = 0, end = PAGE_SIZE;
2749 if (idx == start_index)
2750 start = start_offset;
2751 if (idx == end_index)
2752 end = end_offset + (1 << SECTOR_SHIFT);
2753 sg_set_buf(&s[idx - start_index], va + start, end - start);
2762 static void free_alg(struct alg_spec *a)
2764 kzfree(a->alg_string);
2766 memset(a, 0, sizeof *a);
2769 static int get_alg_and_key(const char *arg, struct alg_spec *a, char **error, char *error_inval)
2775 a->alg_string = kstrdup(strchr(arg, ':') + 1, GFP_KERNEL);
2779 k = strchr(a->alg_string, ':');
2782 a->key_string = k + 1;
2783 if (strlen(a->key_string) & 1)
2786 a->key_size = strlen(a->key_string) / 2;
2787 a->key = kmalloc(a->key_size, GFP_KERNEL);
2790 if (hex2bin(a->key, a->key_string, a->key_size))
2796 *error = error_inval;
2799 *error = "Out of memory for an argument";
2803 static int get_mac(struct crypto_shash **hash, struct alg_spec *a, char **error,
2804 char *error_alg, char *error_key)
2808 if (a->alg_string) {
2809 *hash = crypto_alloc_shash(a->alg_string, 0, 0);
2810 if (IS_ERR(*hash)) {
2818 r = crypto_shash_setkey(*hash, a->key, a->key_size);
2823 } else if (crypto_shash_get_flags(*hash) & CRYPTO_TFM_NEED_KEY) {
2832 static int create_journal(struct dm_integrity_c *ic, char **error)
2836 __u64 journal_pages, journal_desc_size, journal_tree_size;
2837 unsigned char *crypt_data = NULL, *crypt_iv = NULL;
2838 struct skcipher_request *req = NULL;
2840 ic->commit_ids[0] = cpu_to_le64(0x1111111111111111ULL);
2841 ic->commit_ids[1] = cpu_to_le64(0x2222222222222222ULL);
2842 ic->commit_ids[2] = cpu_to_le64(0x3333333333333333ULL);
2843 ic->commit_ids[3] = cpu_to_le64(0x4444444444444444ULL);
2845 journal_pages = roundup((__u64)ic->journal_sections * ic->journal_section_sectors,
2846 PAGE_SIZE >> SECTOR_SHIFT) >> (PAGE_SHIFT - SECTOR_SHIFT);
2847 journal_desc_size = journal_pages * sizeof(struct page_list);
2848 if (journal_pages >= totalram_pages() - totalhigh_pages() || journal_desc_size > ULONG_MAX) {
2849 *error = "Journal doesn't fit into memory";
2853 ic->journal_pages = journal_pages;
2855 ic->journal = dm_integrity_alloc_page_list(ic);
2857 *error = "Could not allocate memory for journal";
2861 if (ic->journal_crypt_alg.alg_string) {
2862 unsigned ivsize, blocksize;
2863 struct journal_completion comp;
2866 ic->journal_crypt = crypto_alloc_skcipher(ic->journal_crypt_alg.alg_string, 0, 0);
2867 if (IS_ERR(ic->journal_crypt)) {
2868 *error = "Invalid journal cipher";
2869 r = PTR_ERR(ic->journal_crypt);
2870 ic->journal_crypt = NULL;
2873 ivsize = crypto_skcipher_ivsize(ic->journal_crypt);
2874 blocksize = crypto_skcipher_blocksize(ic->journal_crypt);
2876 if (ic->journal_crypt_alg.key) {
2877 r = crypto_skcipher_setkey(ic->journal_crypt, ic->journal_crypt_alg.key,
2878 ic->journal_crypt_alg.key_size);
2880 *error = "Error setting encryption key";
2884 DEBUG_print("cipher %s, block size %u iv size %u\n",
2885 ic->journal_crypt_alg.alg_string, blocksize, ivsize);
2887 ic->journal_io = dm_integrity_alloc_page_list(ic);
2888 if (!ic->journal_io) {
2889 *error = "Could not allocate memory for journal io";
2894 if (blocksize == 1) {
2895 struct scatterlist *sg;
2897 req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
2899 *error = "Could not allocate crypt request";
2904 crypt_iv = kmalloc(ivsize, GFP_KERNEL);
2906 *error = "Could not allocate iv";
2911 ic->journal_xor = dm_integrity_alloc_page_list(ic);
2912 if (!ic->journal_xor) {
2913 *error = "Could not allocate memory for journal xor";
2918 sg = kvmalloc_array(ic->journal_pages + 1,
2919 sizeof(struct scatterlist),
2922 *error = "Unable to allocate sg list";
2926 sg_init_table(sg, ic->journal_pages + 1);
2927 for (i = 0; i < ic->journal_pages; i++) {
2928 char *va = lowmem_page_address(ic->journal_xor[i].page);
2930 sg_set_buf(&sg[i], va, PAGE_SIZE);
2932 sg_set_buf(&sg[i], &ic->commit_ids, sizeof ic->commit_ids);
2933 memset(crypt_iv, 0x00, ivsize);
2935 skcipher_request_set_crypt(req, sg, sg, PAGE_SIZE * ic->journal_pages + sizeof ic->commit_ids, crypt_iv);
2936 init_completion(&comp.comp);
2937 comp.in_flight = (atomic_t)ATOMIC_INIT(1);
2938 if (do_crypt(true, req, &comp))
2939 wait_for_completion(&comp.comp);
2941 r = dm_integrity_failed(ic);
2943 *error = "Unable to encrypt journal";
2946 DEBUG_bytes(lowmem_page_address(ic->journal_xor[0].page), 64, "xor data");
2948 crypto_free_skcipher(ic->journal_crypt);
2949 ic->journal_crypt = NULL;
2951 unsigned crypt_len = roundup(ivsize, blocksize);
2953 req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
2955 *error = "Could not allocate crypt request";
2960 crypt_iv = kmalloc(ivsize, GFP_KERNEL);
2962 *error = "Could not allocate iv";
2967 crypt_data = kmalloc(crypt_len, GFP_KERNEL);
2969 *error = "Unable to allocate crypt data";
2974 ic->journal_scatterlist = dm_integrity_alloc_journal_scatterlist(ic, ic->journal);
2975 if (!ic->journal_scatterlist) {
2976 *error = "Unable to allocate sg list";
2980 ic->journal_io_scatterlist = dm_integrity_alloc_journal_scatterlist(ic, ic->journal_io);
2981 if (!ic->journal_io_scatterlist) {
2982 *error = "Unable to allocate sg list";
2986 ic->sk_requests = kvmalloc_array(ic->journal_sections,
2987 sizeof(struct skcipher_request *),
2988 GFP_KERNEL | __GFP_ZERO);
2989 if (!ic->sk_requests) {
2990 *error = "Unable to allocate sk requests";
2994 for (i = 0; i < ic->journal_sections; i++) {
2995 struct scatterlist sg;
2996 struct skcipher_request *section_req;
2997 __u32 section_le = cpu_to_le32(i);
2999 memset(crypt_iv, 0x00, ivsize);
3000 memset(crypt_data, 0x00, crypt_len);
3001 memcpy(crypt_data, §ion_le, min((size_t)crypt_len, sizeof(section_le)));
3003 sg_init_one(&sg, crypt_data, crypt_len);
3004 skcipher_request_set_crypt(req, &sg, &sg, crypt_len, crypt_iv);
3005 init_completion(&comp.comp);
3006 comp.in_flight = (atomic_t)ATOMIC_INIT(1);
3007 if (do_crypt(true, req, &comp))
3008 wait_for_completion(&comp.comp);
3010 r = dm_integrity_failed(ic);
3012 *error = "Unable to generate iv";
3016 section_req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
3018 *error = "Unable to allocate crypt request";
3022 section_req->iv = kmalloc_array(ivsize, 2,
3024 if (!section_req->iv) {
3025 skcipher_request_free(section_req);
3026 *error = "Unable to allocate iv";
3030 memcpy(section_req->iv + ivsize, crypt_data, ivsize);
3031 section_req->cryptlen = (size_t)ic->journal_section_sectors << SECTOR_SHIFT;
3032 ic->sk_requests[i] = section_req;
3033 DEBUG_bytes(crypt_data, ivsize, "iv(%u)", i);
3038 for (i = 0; i < N_COMMIT_IDS; i++) {
3041 for (j = 0; j < i; j++) {
3042 if (ic->commit_ids[j] == ic->commit_ids[i]) {
3043 ic->commit_ids[i] = cpu_to_le64(le64_to_cpu(ic->commit_ids[i]) + 1);
3044 goto retest_commit_id;
3047 DEBUG_print("commit id %u: %016llx\n", i, ic->commit_ids[i]);
3050 journal_tree_size = (__u64)ic->journal_entries * sizeof(struct journal_node);
3051 if (journal_tree_size > ULONG_MAX) {
3052 *error = "Journal doesn't fit into memory";
3056 ic->journal_tree = kvmalloc(journal_tree_size, GFP_KERNEL);
3057 if (!ic->journal_tree) {
3058 *error = "Could not allocate memory for journal tree";
3064 skcipher_request_free(req);
3070 * Construct a integrity mapping
3074 * offset from the start of the device
3076 * D - direct writes, J - journal writes, R - recovery mode
3077 * number of optional arguments
3078 * optional arguments:
3080 * interleave_sectors
3091 static int dm_integrity_ctr(struct dm_target *ti, unsigned argc, char **argv)
3093 struct dm_integrity_c *ic;
3096 unsigned extra_args;
3097 struct dm_arg_set as;
3098 static const struct dm_arg _args[] = {
3099 {0, 9, "Invalid number of feature args"},
3101 unsigned journal_sectors, interleave_sectors, buffer_sectors, journal_watermark, sync_msec;
3103 bool should_write_sb;
3105 unsigned long long start;
3107 #define DIRECT_ARGUMENTS 4
3109 if (argc <= DIRECT_ARGUMENTS) {
3110 ti->error = "Invalid argument count";
3114 ic = kzalloc(sizeof(struct dm_integrity_c), GFP_KERNEL);
3116 ti->error = "Cannot allocate integrity context";
3120 ti->per_io_data_size = sizeof(struct dm_integrity_io);
3122 ic->in_progress = RB_ROOT;
3123 INIT_LIST_HEAD(&ic->wait_list);
3124 init_waitqueue_head(&ic->endio_wait);
3125 bio_list_init(&ic->flush_bio_list);
3126 init_waitqueue_head(&ic->copy_to_journal_wait);
3127 init_completion(&ic->crypto_backoff);
3128 atomic64_set(&ic->number_of_mismatches, 0);
3130 r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &ic->dev);
3132 ti->error = "Device lookup failed";
3136 if (sscanf(argv[1], "%llu%c", &start, &dummy) != 1 || start != (sector_t)start) {
3137 ti->error = "Invalid starting offset";
3143 if (strcmp(argv[2], "-")) {
3144 if (sscanf(argv[2], "%u%c", &ic->tag_size, &dummy) != 1 || !ic->tag_size) {
3145 ti->error = "Invalid tag size";
3151 if (!strcmp(argv[3], "J") || !strcmp(argv[3], "D") || !strcmp(argv[3], "R"))
3152 ic->mode = argv[3][0];
3154 ti->error = "Invalid mode (expecting J, D, R)";
3159 journal_sectors = 0;
3160 interleave_sectors = DEFAULT_INTERLEAVE_SECTORS;
3161 buffer_sectors = DEFAULT_BUFFER_SECTORS;
3162 journal_watermark = DEFAULT_JOURNAL_WATERMARK;
3163 sync_msec = DEFAULT_SYNC_MSEC;
3164 recalculate = false;
3165 ic->sectors_per_block = 1;
3167 as.argc = argc - DIRECT_ARGUMENTS;
3168 as.argv = argv + DIRECT_ARGUMENTS;
3169 r = dm_read_arg_group(_args, &as, &extra_args, &ti->error);
3173 while (extra_args--) {
3174 const char *opt_string;
3176 opt_string = dm_shift_arg(&as);
3179 ti->error = "Not enough feature arguments";
3182 if (sscanf(opt_string, "journal_sectors:%u%c", &val, &dummy) == 1)
3183 journal_sectors = val ? val : 1;
3184 else if (sscanf(opt_string, "interleave_sectors:%u%c", &val, &dummy) == 1)
3185 interleave_sectors = val;
3186 else if (sscanf(opt_string, "buffer_sectors:%u%c", &val, &dummy) == 1)
3187 buffer_sectors = val;
3188 else if (sscanf(opt_string, "journal_watermark:%u%c", &val, &dummy) == 1 && val <= 100)
3189 journal_watermark = val;
3190 else if (sscanf(opt_string, "commit_time:%u%c", &val, &dummy) == 1)
3192 else if (!strncmp(opt_string, "meta_device:", strlen("meta_device:"))) {
3194 dm_put_device(ti, ic->meta_dev);
3195 ic->meta_dev = NULL;
3197 r = dm_get_device(ti, strchr(opt_string, ':') + 1, dm_table_get_mode(ti->table), &ic->meta_dev);
3199 ti->error = "Device lookup failed";
3202 } else if (sscanf(opt_string, "block_size:%u%c", &val, &dummy) == 1) {
3203 if (val < 1 << SECTOR_SHIFT ||
3204 val > MAX_SECTORS_PER_BLOCK << SECTOR_SHIFT ||
3207 ti->error = "Invalid block_size argument";
3210 ic->sectors_per_block = val >> SECTOR_SHIFT;
3211 } else if (!strncmp(opt_string, "internal_hash:", strlen("internal_hash:"))) {
3212 r = get_alg_and_key(opt_string, &ic->internal_hash_alg, &ti->error,
3213 "Invalid internal_hash argument");
3216 } else if (!strncmp(opt_string, "journal_crypt:", strlen("journal_crypt:"))) {
3217 r = get_alg_and_key(opt_string, &ic->journal_crypt_alg, &ti->error,
3218 "Invalid journal_crypt argument");
3221 } else if (!strncmp(opt_string, "journal_mac:", strlen("journal_mac:"))) {
3222 r = get_alg_and_key(opt_string, &ic->journal_mac_alg, &ti->error,
3223 "Invalid journal_mac argument");
3226 } else if (!strcmp(opt_string, "recalculate")) {
3230 ti->error = "Invalid argument";
3235 ic->data_device_sectors = i_size_read(ic->dev->bdev->bd_inode) >> SECTOR_SHIFT;
3237 ic->meta_device_sectors = ic->data_device_sectors;
3239 ic->meta_device_sectors = i_size_read(ic->meta_dev->bdev->bd_inode) >> SECTOR_SHIFT;
3241 if (!journal_sectors) {
3242 journal_sectors = min((sector_t)DEFAULT_MAX_JOURNAL_SECTORS,
3243 ic->data_device_sectors >> DEFAULT_JOURNAL_SIZE_FACTOR);
3246 if (!buffer_sectors)
3248 ic->log2_buffer_sectors = min((int)__fls(buffer_sectors), 31 - SECTOR_SHIFT);
3250 r = get_mac(&ic->internal_hash, &ic->internal_hash_alg, &ti->error,
3251 "Invalid internal hash", "Error setting internal hash key");
3255 r = get_mac(&ic->journal_mac, &ic->journal_mac_alg, &ti->error,
3256 "Invalid journal mac", "Error setting journal mac key");
3260 if (!ic->tag_size) {
3261 if (!ic->internal_hash) {
3262 ti->error = "Unknown tag size";
3266 ic->tag_size = crypto_shash_digestsize(ic->internal_hash);
3268 if (ic->tag_size > MAX_TAG_SIZE) {
3269 ti->error = "Too big tag size";
3273 if (!(ic->tag_size & (ic->tag_size - 1)))
3274 ic->log2_tag_size = __ffs(ic->tag_size);
3276 ic->log2_tag_size = -1;
3278 ic->autocommit_jiffies = msecs_to_jiffies(sync_msec);
3279 ic->autocommit_msec = sync_msec;
3280 timer_setup(&ic->autocommit_timer, autocommit_fn, 0);
3282 ic->io = dm_io_client_create();
3283 if (IS_ERR(ic->io)) {
3284 r = PTR_ERR(ic->io);
3286 ti->error = "Cannot allocate dm io";
3290 r = mempool_init_slab_pool(&ic->journal_io_mempool, JOURNAL_IO_MEMPOOL, journal_io_cache);
3292 ti->error = "Cannot allocate mempool";
3296 ic->metadata_wq = alloc_workqueue("dm-integrity-metadata",
3297 WQ_MEM_RECLAIM, METADATA_WORKQUEUE_MAX_ACTIVE);
3298 if (!ic->metadata_wq) {
3299 ti->error = "Cannot allocate workqueue";
3305 * If this workqueue were percpu, it would cause bio reordering
3306 * and reduced performance.
3308 ic->wait_wq = alloc_workqueue("dm-integrity-wait", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3310 ti->error = "Cannot allocate workqueue";
3315 ic->commit_wq = alloc_workqueue("dm-integrity-commit", WQ_MEM_RECLAIM, 1);
3316 if (!ic->commit_wq) {
3317 ti->error = "Cannot allocate workqueue";
3321 INIT_WORK(&ic->commit_work, integrity_commit);
3323 if (ic->mode == 'J') {
3324 ic->writer_wq = alloc_workqueue("dm-integrity-writer", WQ_MEM_RECLAIM, 1);
3325 if (!ic->writer_wq) {
3326 ti->error = "Cannot allocate workqueue";
3330 INIT_WORK(&ic->writer_work, integrity_writer);
3333 ic->sb = alloc_pages_exact(SB_SECTORS << SECTOR_SHIFT, GFP_KERNEL);
3336 ti->error = "Cannot allocate superblock area";
3340 r = sync_rw_sb(ic, REQ_OP_READ, 0);
3342 ti->error = "Error reading superblock";
3345 should_write_sb = false;
3346 if (memcmp(ic->sb->magic, SB_MAGIC, 8)) {
3347 if (ic->mode != 'R') {
3348 if (memchr_inv(ic->sb, 0, SB_SECTORS << SECTOR_SHIFT)) {
3350 ti->error = "The device is not initialized";
3355 r = initialize_superblock(ic, journal_sectors, interleave_sectors);
3357 ti->error = "Could not initialize superblock";
3360 if (ic->mode != 'R')
3361 should_write_sb = true;
3364 if (!ic->sb->version || ic->sb->version > SB_VERSION_2) {
3366 ti->error = "Unknown version";
3369 if (le16_to_cpu(ic->sb->integrity_tag_size) != ic->tag_size) {
3371 ti->error = "Tag size doesn't match the information in superblock";
3374 if (ic->sb->log2_sectors_per_block != __ffs(ic->sectors_per_block)) {
3376 ti->error = "Block size doesn't match the information in superblock";
3379 if (!le32_to_cpu(ic->sb->journal_sections)) {
3381 ti->error = "Corrupted superblock, journal_sections is 0";
3384 /* make sure that ti->max_io_len doesn't overflow */
3385 if (!ic->meta_dev) {
3386 if (ic->sb->log2_interleave_sectors < MIN_LOG2_INTERLEAVE_SECTORS ||
3387 ic->sb->log2_interleave_sectors > MAX_LOG2_INTERLEAVE_SECTORS) {
3389 ti->error = "Invalid interleave_sectors in the superblock";
3393 if (ic->sb->log2_interleave_sectors) {
3395 ti->error = "Invalid interleave_sectors in the superblock";
3399 ic->provided_data_sectors = le64_to_cpu(ic->sb->provided_data_sectors);
3400 if (ic->provided_data_sectors != le64_to_cpu(ic->sb->provided_data_sectors)) {
3401 /* test for overflow */
3403 ti->error = "The superblock has 64-bit device size, but the kernel was compiled with 32-bit sectors";
3406 if (!!(ic->sb->flags & cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC)) != !!ic->journal_mac_alg.alg_string) {
3408 ti->error = "Journal mac mismatch";
3413 r = calculate_device_limits(ic);
3416 if (ic->log2_buffer_sectors > 3) {
3417 ic->log2_buffer_sectors--;
3418 goto try_smaller_buffer;
3421 ti->error = "The device is too small";
3425 ic->log2_buffer_sectors = min(ic->log2_buffer_sectors, (__u8)__ffs(ic->metadata_run));
3427 if (ti->len > ic->provided_data_sectors) {
3429 ti->error = "Not enough provided sectors for requested mapping size";
3434 threshold = (__u64)ic->journal_entries * (100 - journal_watermark);
3436 do_div(threshold, 100);
3437 ic->free_sectors_threshold = threshold;
3439 DEBUG_print("initialized:\n");
3440 DEBUG_print(" integrity_tag_size %u\n", le16_to_cpu(ic->sb->integrity_tag_size));
3441 DEBUG_print(" journal_entry_size %u\n", ic->journal_entry_size);
3442 DEBUG_print(" journal_entries_per_sector %u\n", ic->journal_entries_per_sector);
3443 DEBUG_print(" journal_section_entries %u\n", ic->journal_section_entries);
3444 DEBUG_print(" journal_section_sectors %u\n", ic->journal_section_sectors);
3445 DEBUG_print(" journal_sections %u\n", (unsigned)le32_to_cpu(ic->sb->journal_sections));
3446 DEBUG_print(" journal_entries %u\n", ic->journal_entries);
3447 DEBUG_print(" log2_interleave_sectors %d\n", ic->sb->log2_interleave_sectors);
3448 DEBUG_print(" data_device_sectors 0x%llx\n", (unsigned long long)ic->data_device_sectors);
3449 DEBUG_print(" initial_sectors 0x%x\n", ic->initial_sectors);
3450 DEBUG_print(" metadata_run 0x%x\n", ic->metadata_run);
3451 DEBUG_print(" log2_metadata_run %d\n", ic->log2_metadata_run);
3452 DEBUG_print(" provided_data_sectors 0x%llx (%llu)\n", (unsigned long long)ic->provided_data_sectors,
3453 (unsigned long long)ic->provided_data_sectors);
3454 DEBUG_print(" log2_buffer_sectors %u\n", ic->log2_buffer_sectors);
3456 if (recalculate && !(ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))) {
3457 ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
3458 ic->sb->recalc_sector = cpu_to_le64(0);
3461 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
3462 if (!ic->internal_hash) {
3464 ti->error = "Recalculate is only valid with internal hash";
3467 ic->recalc_wq = alloc_workqueue("dm-integrity-recalc", WQ_MEM_RECLAIM, 1);
3468 if (!ic->recalc_wq ) {
3469 ti->error = "Cannot allocate workqueue";
3473 INIT_WORK(&ic->recalc_work, integrity_recalc);
3474 ic->recalc_buffer = vmalloc(RECALC_SECTORS << SECTOR_SHIFT);
3475 if (!ic->recalc_buffer) {
3476 ti->error = "Cannot allocate buffer for recalculating";
3480 ic->recalc_tags = kvmalloc_array(RECALC_SECTORS >> ic->sb->log2_sectors_per_block,
3481 ic->tag_size, GFP_KERNEL);
3482 if (!ic->recalc_tags) {
3483 ti->error = "Cannot allocate tags for recalculating";
3489 ic->bufio = dm_bufio_client_create(ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev,
3490 1U << (SECTOR_SHIFT + ic->log2_buffer_sectors), 1, 0, NULL, NULL);
3491 if (IS_ERR(ic->bufio)) {
3492 r = PTR_ERR(ic->bufio);
3493 ti->error = "Cannot initialize dm-bufio";
3497 dm_bufio_set_sector_offset(ic->bufio, ic->start + ic->initial_sectors);
3499 if (ic->mode != 'R') {
3500 r = create_journal(ic, &ti->error);
3505 if (should_write_sb) {
3508 init_journal(ic, 0, ic->journal_sections, 0);
3509 r = dm_integrity_failed(ic);
3511 ti->error = "Error initializing journal";
3514 r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
3516 ti->error = "Error initializing superblock";
3519 ic->just_formatted = true;
3522 if (!ic->meta_dev) {
3523 r = dm_set_target_max_io_len(ti, 1U << ic->sb->log2_interleave_sectors);
3528 if (!ic->internal_hash)
3529 dm_integrity_set(ti, ic);
3531 ti->num_flush_bios = 1;
3532 ti->flush_supported = true;
3536 dm_integrity_dtr(ti);
3540 static void dm_integrity_dtr(struct dm_target *ti)
3542 struct dm_integrity_c *ic = ti->private;
3544 BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
3545 BUG_ON(!list_empty(&ic->wait_list));
3547 if (ic->metadata_wq)
3548 destroy_workqueue(ic->metadata_wq);
3550 destroy_workqueue(ic->wait_wq);
3552 destroy_workqueue(ic->commit_wq);
3554 destroy_workqueue(ic->writer_wq);
3556 destroy_workqueue(ic->recalc_wq);
3557 vfree(ic->recalc_buffer);
3558 kvfree(ic->recalc_tags);
3560 dm_bufio_client_destroy(ic->bufio);
3561 mempool_exit(&ic->journal_io_mempool);
3563 dm_io_client_destroy(ic->io);
3565 dm_put_device(ti, ic->dev);
3567 dm_put_device(ti, ic->meta_dev);
3568 dm_integrity_free_page_list(ic, ic->journal);
3569 dm_integrity_free_page_list(ic, ic->journal_io);
3570 dm_integrity_free_page_list(ic, ic->journal_xor);
3571 if (ic->journal_scatterlist)
3572 dm_integrity_free_journal_scatterlist(ic, ic->journal_scatterlist);
3573 if (ic->journal_io_scatterlist)
3574 dm_integrity_free_journal_scatterlist(ic, ic->journal_io_scatterlist);
3575 if (ic->sk_requests) {
3578 for (i = 0; i < ic->journal_sections; i++) {
3579 struct skcipher_request *req = ic->sk_requests[i];
3582 skcipher_request_free(req);
3585 kvfree(ic->sk_requests);
3587 kvfree(ic->journal_tree);
3589 free_pages_exact(ic->sb, SB_SECTORS << SECTOR_SHIFT);
3591 if (ic->internal_hash)
3592 crypto_free_shash(ic->internal_hash);
3593 free_alg(&ic->internal_hash_alg);
3595 if (ic->journal_crypt)
3596 crypto_free_skcipher(ic->journal_crypt);
3597 free_alg(&ic->journal_crypt_alg);
3599 if (ic->journal_mac)
3600 crypto_free_shash(ic->journal_mac);
3601 free_alg(&ic->journal_mac_alg);
3606 static struct target_type integrity_target = {
3607 .name = "integrity",
3608 .version = {1, 2, 0},
3609 .module = THIS_MODULE,
3610 .features = DM_TARGET_SINGLETON | DM_TARGET_INTEGRITY,
3611 .ctr = dm_integrity_ctr,
3612 .dtr = dm_integrity_dtr,
3613 .map = dm_integrity_map,
3614 .postsuspend = dm_integrity_postsuspend,
3615 .resume = dm_integrity_resume,
3616 .status = dm_integrity_status,
3617 .iterate_devices = dm_integrity_iterate_devices,
3618 .io_hints = dm_integrity_io_hints,
3621 static int __init dm_integrity_init(void)
3625 journal_io_cache = kmem_cache_create("integrity_journal_io",
3626 sizeof(struct journal_io), 0, 0, NULL);
3627 if (!journal_io_cache) {
3628 DMERR("can't allocate journal io cache");
3632 r = dm_register_target(&integrity_target);
3635 DMERR("register failed %d", r);
3640 static void __exit dm_integrity_exit(void)
3642 dm_unregister_target(&integrity_target);
3643 kmem_cache_destroy(journal_io_cache);
3646 module_init(dm_integrity_init);
3647 module_exit(dm_integrity_exit);
3649 MODULE_AUTHOR("Milan Broz");
3650 MODULE_AUTHOR("Mikulas Patocka");
3651 MODULE_DESCRIPTION(DM_NAME " target for integrity tags extension");
3652 MODULE_LICENSE("GPL");