2 * Copyright (C) 2016-2017 Red Hat, Inc. All rights reserved.
3 * Copyright (C) 2016-2017 Milan Broz
4 * Copyright (C) 2016-2017 Mikulas Patocka
6 * This file is released under the GPL.
9 #include "dm-bio-record.h"
11 #include <linux/compiler.h>
12 #include <linux/module.h>
13 #include <linux/device-mapper.h>
14 #include <linux/dm-io.h>
15 #include <linux/vmalloc.h>
16 #include <linux/sort.h>
17 #include <linux/rbtree.h>
18 #include <linux/delay.h>
19 #include <linux/random.h>
20 #include <linux/reboot.h>
21 #include <crypto/hash.h>
22 #include <crypto/skcipher.h>
23 #include <linux/async_tx.h>
24 #include <linux/dm-bufio.h>
26 #define DM_MSG_PREFIX "integrity"
28 #define DEFAULT_INTERLEAVE_SECTORS 32768
29 #define DEFAULT_JOURNAL_SIZE_FACTOR 7
30 #define DEFAULT_SECTORS_PER_BITMAP_BIT 32768
31 #define DEFAULT_BUFFER_SECTORS 128
32 #define DEFAULT_JOURNAL_WATERMARK 50
33 #define DEFAULT_SYNC_MSEC 10000
34 #define DEFAULT_MAX_JOURNAL_SECTORS (IS_ENABLED(CONFIG_64BIT) ? 131072 : 8192)
35 #define MIN_LOG2_INTERLEAVE_SECTORS 3
36 #define MAX_LOG2_INTERLEAVE_SECTORS 31
37 #define METADATA_WORKQUEUE_MAX_ACTIVE 16
38 #define RECALC_SECTORS (IS_ENABLED(CONFIG_64BIT) ? 32768 : 2048)
39 #define RECALC_WRITE_SUPER 16
40 #define BITMAP_BLOCK_SIZE 4096 /* don't change it */
41 #define BITMAP_FLUSH_INTERVAL (10 * HZ)
42 #define DISCARD_FILLER 0xf6
45 * Warning - DEBUG_PRINT prints security-sensitive data to the log,
46 * so it should not be enabled in the official kernel
49 //#define INTERNAL_VERIFY
55 #define SB_MAGIC "integrt"
56 #define SB_VERSION_1 1
57 #define SB_VERSION_2 2
58 #define SB_VERSION_3 3
59 #define SB_VERSION_4 4
61 #define MAX_SECTORS_PER_BLOCK 8
66 __u8 log2_interleave_sectors;
67 __u16 integrity_tag_size;
68 __u32 journal_sections;
69 __u64 provided_data_sectors; /* userspace uses this value */
71 __u8 log2_sectors_per_block;
72 __u8 log2_blocks_per_bitmap_bit;
77 #define SB_FLAG_HAVE_JOURNAL_MAC 0x1
78 #define SB_FLAG_RECALCULATING 0x2
79 #define SB_FLAG_DIRTY_BITMAP 0x4
80 #define SB_FLAG_FIXED_PADDING 0x8
82 #define JOURNAL_ENTRY_ROUNDUP 8
84 typedef __u64 commit_id_t;
85 #define JOURNAL_MAC_PER_SECTOR 8
87 struct journal_entry {
95 commit_id_t last_bytes[];
99 #define journal_entry_tag(ic, je) ((__u8 *)&(je)->last_bytes[(ic)->sectors_per_block])
101 #if BITS_PER_LONG == 64
102 #define journal_entry_set_sector(je, x) do { smp_wmb(); WRITE_ONCE((je)->u.sector, cpu_to_le64(x)); } while (0)
104 #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)
106 #define journal_entry_get_sector(je) le64_to_cpu((je)->u.sector)
107 #define journal_entry_is_unused(je) ((je)->u.s.sector_hi == cpu_to_le32(-1))
108 #define journal_entry_set_unused(je) do { ((je)->u.s.sector_hi = cpu_to_le32(-1)); } while (0)
109 #define journal_entry_is_inprogress(je) ((je)->u.s.sector_hi == cpu_to_le32(-2))
110 #define journal_entry_set_inprogress(je) do { ((je)->u.s.sector_hi = cpu_to_le32(-2)); } while (0)
112 #define JOURNAL_BLOCK_SECTORS 8
113 #define JOURNAL_SECTOR_DATA ((1 << SECTOR_SHIFT) - sizeof(commit_id_t))
114 #define JOURNAL_MAC_SIZE (JOURNAL_MAC_PER_SECTOR * JOURNAL_BLOCK_SECTORS)
116 struct journal_sector {
117 __u8 entries[JOURNAL_SECTOR_DATA - JOURNAL_MAC_PER_SECTOR];
118 __u8 mac[JOURNAL_MAC_PER_SECTOR];
119 commit_id_t commit_id;
122 #define MAX_TAG_SIZE (JOURNAL_SECTOR_DATA - JOURNAL_MAC_PER_SECTOR - offsetof(struct journal_entry, last_bytes[MAX_SECTORS_PER_BLOCK]))
124 #define METADATA_PADDING_SECTORS 8
126 #define N_COMMIT_IDS 4
128 static unsigned char prev_commit_seq(unsigned char seq)
130 return (seq + N_COMMIT_IDS - 1) % N_COMMIT_IDS;
133 static unsigned char next_commit_seq(unsigned char seq)
135 return (seq + 1) % N_COMMIT_IDS;
139 * In-memory structures
142 struct journal_node {
154 struct dm_integrity_c {
156 struct dm_dev *meta_dev;
160 mempool_t journal_io_mempool;
161 struct dm_io_client *io;
162 struct dm_bufio_client *bufio;
163 struct workqueue_struct *metadata_wq;
164 struct superblock *sb;
165 unsigned journal_pages;
166 unsigned n_bitmap_blocks;
168 struct page_list *journal;
169 struct page_list *journal_io;
170 struct page_list *journal_xor;
171 struct page_list *recalc_bitmap;
172 struct page_list *may_write_bitmap;
173 struct bitmap_block_status *bbs;
174 unsigned bitmap_flush_interval;
175 int synchronous_mode;
176 struct bio_list synchronous_bios;
177 struct delayed_work bitmap_flush_work;
179 struct crypto_skcipher *journal_crypt;
180 struct scatterlist **journal_scatterlist;
181 struct scatterlist **journal_io_scatterlist;
182 struct skcipher_request **sk_requests;
184 struct crypto_shash *journal_mac;
186 struct journal_node *journal_tree;
187 struct rb_root journal_tree_root;
189 sector_t provided_data_sectors;
191 unsigned short journal_entry_size;
192 unsigned char journal_entries_per_sector;
193 unsigned char journal_section_entries;
194 unsigned short journal_section_sectors;
195 unsigned journal_sections;
196 unsigned journal_entries;
197 sector_t data_device_sectors;
198 sector_t meta_device_sectors;
199 unsigned initial_sectors;
200 unsigned metadata_run;
201 __s8 log2_metadata_run;
202 __u8 log2_buffer_sectors;
203 __u8 sectors_per_block;
204 __u8 log2_blocks_per_bitmap_bit;
210 struct crypto_shash *internal_hash;
212 struct dm_target *ti;
214 /* these variables are locked with endio_wait.lock */
215 struct rb_root in_progress;
216 struct list_head wait_list;
217 wait_queue_head_t endio_wait;
218 struct workqueue_struct *wait_wq;
219 struct workqueue_struct *offload_wq;
221 unsigned char commit_seq;
222 commit_id_t commit_ids[N_COMMIT_IDS];
224 unsigned committed_section;
225 unsigned n_committed_sections;
227 unsigned uncommitted_section;
228 unsigned n_uncommitted_sections;
230 unsigned free_section;
231 unsigned char free_section_entry;
232 unsigned free_sectors;
234 unsigned free_sectors_threshold;
236 struct workqueue_struct *commit_wq;
237 struct work_struct commit_work;
239 struct workqueue_struct *writer_wq;
240 struct work_struct writer_work;
242 struct workqueue_struct *recalc_wq;
243 struct work_struct recalc_work;
247 struct bio_list flush_bio_list;
249 unsigned long autocommit_jiffies;
250 struct timer_list autocommit_timer;
251 unsigned autocommit_msec;
253 wait_queue_head_t copy_to_journal_wait;
255 struct completion crypto_backoff;
257 bool wrote_to_journal;
258 bool journal_uptodate;
260 bool recalculate_flag;
263 bool legacy_recalculate;
265 struct alg_spec internal_hash_alg;
266 struct alg_spec journal_crypt_alg;
267 struct alg_spec journal_mac_alg;
269 atomic64_t number_of_mismatches;
271 struct notifier_block reboot_notifier;
274 struct dm_integrity_range {
275 sector_t logical_sector;
281 struct task_struct *task;
282 struct list_head wait_entry;
287 struct dm_integrity_io {
288 struct work_struct work;
290 struct dm_integrity_c *ic;
294 struct dm_integrity_range range;
296 sector_t metadata_block;
297 unsigned metadata_offset;
300 blk_status_t bi_status;
302 struct completion *completion;
304 struct dm_bio_details bio_details;
307 struct journal_completion {
308 struct dm_integrity_c *ic;
310 struct completion comp;
314 struct dm_integrity_range range;
315 struct journal_completion *comp;
318 struct bitmap_block_status {
319 struct work_struct work;
320 struct dm_integrity_c *ic;
322 unsigned long *bitmap;
323 struct bio_list bio_queue;
324 spinlock_t bio_queue_lock;
328 static struct kmem_cache *journal_io_cache;
330 #define JOURNAL_IO_MEMPOOL 32
333 #define DEBUG_print(x, ...) printk(KERN_DEBUG x, ##__VA_ARGS__)
334 static void __DEBUG_bytes(__u8 *bytes, size_t len, const char *msg, ...)
343 pr_cont(" %02x", *bytes);
349 #define DEBUG_bytes(bytes, len, msg, ...) __DEBUG_bytes(bytes, len, KERN_DEBUG msg, ##__VA_ARGS__)
351 #define DEBUG_print(x, ...) do { } while (0)
352 #define DEBUG_bytes(bytes, len, msg, ...) do { } while (0)
355 static void dm_integrity_prepare(struct request *rq)
359 static void dm_integrity_complete(struct request *rq, unsigned int nr_bytes)
364 * DM Integrity profile, protection is performed layer above (dm-crypt)
366 static const struct blk_integrity_profile dm_integrity_profile = {
367 .name = "DM-DIF-EXT-TAG",
370 .prepare_fn = dm_integrity_prepare,
371 .complete_fn = dm_integrity_complete,
374 static void dm_integrity_map_continue(struct dm_integrity_io *dio, bool from_map);
375 static void integrity_bio_wait(struct work_struct *w);
376 static void dm_integrity_dtr(struct dm_target *ti);
378 static void dm_integrity_io_error(struct dm_integrity_c *ic, const char *msg, int err)
381 atomic64_inc(&ic->number_of_mismatches);
382 if (!cmpxchg(&ic->failed, 0, err))
383 DMERR("Error on %s: %d", msg, err);
386 static int dm_integrity_failed(struct dm_integrity_c *ic)
388 return READ_ONCE(ic->failed);
391 static bool dm_integrity_disable_recalculate(struct dm_integrity_c *ic)
393 if ((ic->internal_hash_alg.key || ic->journal_mac_alg.key) &&
394 !ic->legacy_recalculate)
399 static commit_id_t dm_integrity_commit_id(struct dm_integrity_c *ic, unsigned i,
400 unsigned j, unsigned char seq)
403 * Xor the number with section and sector, so that if a piece of
404 * journal is written at wrong place, it is detected.
406 return ic->commit_ids[seq] ^ cpu_to_le64(((__u64)i << 32) ^ j);
409 static void get_area_and_offset(struct dm_integrity_c *ic, sector_t data_sector,
410 sector_t *area, sector_t *offset)
413 __u8 log2_interleave_sectors = ic->sb->log2_interleave_sectors;
414 *area = data_sector >> log2_interleave_sectors;
415 *offset = (unsigned)data_sector & ((1U << log2_interleave_sectors) - 1);
418 *offset = data_sector;
422 #define sector_to_block(ic, n) \
424 BUG_ON((n) & (unsigned)((ic)->sectors_per_block - 1)); \
425 (n) >>= (ic)->sb->log2_sectors_per_block; \
428 static __u64 get_metadata_sector_and_offset(struct dm_integrity_c *ic, sector_t area,
429 sector_t offset, unsigned *metadata_offset)
434 ms = area << ic->sb->log2_interleave_sectors;
435 if (likely(ic->log2_metadata_run >= 0))
436 ms += area << ic->log2_metadata_run;
438 ms += area * ic->metadata_run;
439 ms >>= ic->log2_buffer_sectors;
441 sector_to_block(ic, offset);
443 if (likely(ic->log2_tag_size >= 0)) {
444 ms += offset >> (SECTOR_SHIFT + ic->log2_buffer_sectors - ic->log2_tag_size);
445 mo = (offset << ic->log2_tag_size) & ((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - 1);
447 ms += (__u64)offset * ic->tag_size >> (SECTOR_SHIFT + ic->log2_buffer_sectors);
448 mo = (offset * ic->tag_size) & ((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - 1);
450 *metadata_offset = mo;
454 static sector_t get_data_sector(struct dm_integrity_c *ic, sector_t area, sector_t offset)
461 result = area << ic->sb->log2_interleave_sectors;
462 if (likely(ic->log2_metadata_run >= 0))
463 result += (area + 1) << ic->log2_metadata_run;
465 result += (area + 1) * ic->metadata_run;
467 result += (sector_t)ic->initial_sectors + offset;
473 static void wraparound_section(struct dm_integrity_c *ic, unsigned *sec_ptr)
475 if (unlikely(*sec_ptr >= ic->journal_sections))
476 *sec_ptr -= ic->journal_sections;
479 static void sb_set_version(struct dm_integrity_c *ic)
481 if (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING))
482 ic->sb->version = SB_VERSION_4;
483 else if (ic->mode == 'B' || ic->sb->flags & cpu_to_le32(SB_FLAG_DIRTY_BITMAP))
484 ic->sb->version = SB_VERSION_3;
485 else if (ic->meta_dev || ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
486 ic->sb->version = SB_VERSION_2;
488 ic->sb->version = SB_VERSION_1;
491 static int sync_rw_sb(struct dm_integrity_c *ic, int op, int op_flags)
493 struct dm_io_request io_req;
494 struct dm_io_region io_loc;
497 io_req.bi_op_flags = op_flags;
498 io_req.mem.type = DM_IO_KMEM;
499 io_req.mem.ptr.addr = ic->sb;
500 io_req.notify.fn = NULL;
501 io_req.client = ic->io;
502 io_loc.bdev = ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev;
503 io_loc.sector = ic->start;
504 io_loc.count = SB_SECTORS;
506 if (op == REQ_OP_WRITE)
509 return dm_io(&io_req, 1, &io_loc, NULL);
512 #define BITMAP_OP_TEST_ALL_SET 0
513 #define BITMAP_OP_TEST_ALL_CLEAR 1
514 #define BITMAP_OP_SET 2
515 #define BITMAP_OP_CLEAR 3
517 static bool block_bitmap_op(struct dm_integrity_c *ic, struct page_list *bitmap,
518 sector_t sector, sector_t n_sectors, int mode)
520 unsigned long bit, end_bit, this_end_bit, page, end_page;
523 if (unlikely(((sector | n_sectors) & ((1 << ic->sb->log2_sectors_per_block) - 1)) != 0)) {
524 DMCRIT("invalid bitmap access (%llx,%llx,%d,%d,%d)",
527 ic->sb->log2_sectors_per_block,
528 ic->log2_blocks_per_bitmap_bit,
533 if (unlikely(!n_sectors))
536 bit = sector >> (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
537 end_bit = (sector + n_sectors - 1) >>
538 (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
540 page = bit / (PAGE_SIZE * 8);
541 bit %= PAGE_SIZE * 8;
543 end_page = end_bit / (PAGE_SIZE * 8);
544 end_bit %= PAGE_SIZE * 8;
547 if (page < end_page) {
548 this_end_bit = PAGE_SIZE * 8 - 1;
550 this_end_bit = end_bit;
553 data = lowmem_page_address(bitmap[page].page);
555 if (mode == BITMAP_OP_TEST_ALL_SET) {
556 while (bit <= this_end_bit) {
557 if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
559 if (data[bit / BITS_PER_LONG] != -1)
561 bit += BITS_PER_LONG;
562 } while (this_end_bit >= bit + BITS_PER_LONG - 1);
565 if (!test_bit(bit, data))
569 } else if (mode == BITMAP_OP_TEST_ALL_CLEAR) {
570 while (bit <= this_end_bit) {
571 if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
573 if (data[bit / BITS_PER_LONG] != 0)
575 bit += BITS_PER_LONG;
576 } while (this_end_bit >= bit + BITS_PER_LONG - 1);
579 if (test_bit(bit, data))
583 } else if (mode == BITMAP_OP_SET) {
584 while (bit <= this_end_bit) {
585 if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
587 data[bit / BITS_PER_LONG] = -1;
588 bit += BITS_PER_LONG;
589 } while (this_end_bit >= bit + BITS_PER_LONG - 1);
592 __set_bit(bit, data);
595 } else if (mode == BITMAP_OP_CLEAR) {
596 if (!bit && this_end_bit == PAGE_SIZE * 8 - 1)
598 else while (bit <= this_end_bit) {
599 if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
601 data[bit / BITS_PER_LONG] = 0;
602 bit += BITS_PER_LONG;
603 } while (this_end_bit >= bit + BITS_PER_LONG - 1);
606 __clear_bit(bit, data);
613 if (unlikely(page < end_page)) {
622 static void block_bitmap_copy(struct dm_integrity_c *ic, struct page_list *dst, struct page_list *src)
624 unsigned n_bitmap_pages = DIV_ROUND_UP(ic->n_bitmap_blocks, PAGE_SIZE / BITMAP_BLOCK_SIZE);
627 for (i = 0; i < n_bitmap_pages; i++) {
628 unsigned long *dst_data = lowmem_page_address(dst[i].page);
629 unsigned long *src_data = lowmem_page_address(src[i].page);
630 copy_page(dst_data, src_data);
634 static struct bitmap_block_status *sector_to_bitmap_block(struct dm_integrity_c *ic, sector_t sector)
636 unsigned bit = sector >> (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
637 unsigned bitmap_block = bit / (BITMAP_BLOCK_SIZE * 8);
639 BUG_ON(bitmap_block >= ic->n_bitmap_blocks);
640 return &ic->bbs[bitmap_block];
643 static void access_journal_check(struct dm_integrity_c *ic, unsigned section, unsigned offset,
644 bool e, const char *function)
646 #if defined(CONFIG_DM_DEBUG) || defined(INTERNAL_VERIFY)
647 unsigned limit = e ? ic->journal_section_entries : ic->journal_section_sectors;
649 if (unlikely(section >= ic->journal_sections) ||
650 unlikely(offset >= limit)) {
651 DMCRIT("%s: invalid access at (%u,%u), limit (%u,%u)",
652 function, section, offset, ic->journal_sections, limit);
658 static void page_list_location(struct dm_integrity_c *ic, unsigned section, unsigned offset,
659 unsigned *pl_index, unsigned *pl_offset)
663 access_journal_check(ic, section, offset, false, "page_list_location");
665 sector = section * ic->journal_section_sectors + offset;
667 *pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
668 *pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
671 static struct journal_sector *access_page_list(struct dm_integrity_c *ic, struct page_list *pl,
672 unsigned section, unsigned offset, unsigned *n_sectors)
674 unsigned pl_index, pl_offset;
677 page_list_location(ic, section, offset, &pl_index, &pl_offset);
680 *n_sectors = (PAGE_SIZE - pl_offset) >> SECTOR_SHIFT;
682 va = lowmem_page_address(pl[pl_index].page);
684 return (struct journal_sector *)(va + pl_offset);
687 static struct journal_sector *access_journal(struct dm_integrity_c *ic, unsigned section, unsigned offset)
689 return access_page_list(ic, ic->journal, section, offset, NULL);
692 static struct journal_entry *access_journal_entry(struct dm_integrity_c *ic, unsigned section, unsigned n)
694 unsigned rel_sector, offset;
695 struct journal_sector *js;
697 access_journal_check(ic, section, n, true, "access_journal_entry");
699 rel_sector = n % JOURNAL_BLOCK_SECTORS;
700 offset = n / JOURNAL_BLOCK_SECTORS;
702 js = access_journal(ic, section, rel_sector);
703 return (struct journal_entry *)((char *)js + offset * ic->journal_entry_size);
706 static struct journal_sector *access_journal_data(struct dm_integrity_c *ic, unsigned section, unsigned n)
708 n <<= ic->sb->log2_sectors_per_block;
710 n += JOURNAL_BLOCK_SECTORS;
712 access_journal_check(ic, section, n, false, "access_journal_data");
714 return access_journal(ic, section, n);
717 static void section_mac(struct dm_integrity_c *ic, unsigned section, __u8 result[JOURNAL_MAC_SIZE])
719 SHASH_DESC_ON_STACK(desc, ic->journal_mac);
723 desc->tfm = ic->journal_mac;
725 r = crypto_shash_init(desc);
727 dm_integrity_io_error(ic, "crypto_shash_init", r);
731 for (j = 0; j < ic->journal_section_entries; j++) {
732 struct journal_entry *je = access_journal_entry(ic, section, j);
733 r = crypto_shash_update(desc, (__u8 *)&je->u.sector, sizeof je->u.sector);
735 dm_integrity_io_error(ic, "crypto_shash_update", r);
740 size = crypto_shash_digestsize(ic->journal_mac);
742 if (likely(size <= JOURNAL_MAC_SIZE)) {
743 r = crypto_shash_final(desc, result);
745 dm_integrity_io_error(ic, "crypto_shash_final", r);
748 memset(result + size, 0, JOURNAL_MAC_SIZE - size);
750 __u8 digest[HASH_MAX_DIGESTSIZE];
752 if (WARN_ON(size > sizeof(digest))) {
753 dm_integrity_io_error(ic, "digest_size", -EINVAL);
756 r = crypto_shash_final(desc, digest);
758 dm_integrity_io_error(ic, "crypto_shash_final", r);
761 memcpy(result, digest, JOURNAL_MAC_SIZE);
766 memset(result, 0, JOURNAL_MAC_SIZE);
769 static void rw_section_mac(struct dm_integrity_c *ic, unsigned section, bool wr)
771 __u8 result[JOURNAL_MAC_SIZE];
774 if (!ic->journal_mac)
777 section_mac(ic, section, result);
779 for (j = 0; j < JOURNAL_BLOCK_SECTORS; j++) {
780 struct journal_sector *js = access_journal(ic, section, j);
783 memcpy(&js->mac, result + (j * JOURNAL_MAC_PER_SECTOR), JOURNAL_MAC_PER_SECTOR);
785 if (memcmp(&js->mac, result + (j * JOURNAL_MAC_PER_SECTOR), JOURNAL_MAC_PER_SECTOR))
786 dm_integrity_io_error(ic, "journal mac", -EILSEQ);
791 static void complete_journal_op(void *context)
793 struct journal_completion *comp = context;
794 BUG_ON(!atomic_read(&comp->in_flight));
795 if (likely(atomic_dec_and_test(&comp->in_flight)))
796 complete(&comp->comp);
799 static void xor_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
800 unsigned n_sections, struct journal_completion *comp)
802 struct async_submit_ctl submit;
803 size_t n_bytes = (size_t)(n_sections * ic->journal_section_sectors) << SECTOR_SHIFT;
804 unsigned pl_index, pl_offset, section_index;
805 struct page_list *source_pl, *target_pl;
807 if (likely(encrypt)) {
808 source_pl = ic->journal;
809 target_pl = ic->journal_io;
811 source_pl = ic->journal_io;
812 target_pl = ic->journal;
815 page_list_location(ic, section, 0, &pl_index, &pl_offset);
817 atomic_add(roundup(pl_offset + n_bytes, PAGE_SIZE) >> PAGE_SHIFT, &comp->in_flight);
819 init_async_submit(&submit, ASYNC_TX_XOR_ZERO_DST, NULL, complete_journal_op, comp, NULL);
821 section_index = pl_index;
825 struct page *src_pages[2];
826 struct page *dst_page;
828 while (unlikely(pl_index == section_index)) {
831 rw_section_mac(ic, section, true);
836 page_list_location(ic, section, 0, §ion_index, &dummy);
839 this_step = min(n_bytes, (size_t)PAGE_SIZE - pl_offset);
840 dst_page = target_pl[pl_index].page;
841 src_pages[0] = source_pl[pl_index].page;
842 src_pages[1] = ic->journal_xor[pl_index].page;
844 async_xor(dst_page, src_pages, pl_offset, 2, this_step, &submit);
848 n_bytes -= this_step;
853 async_tx_issue_pending_all();
856 static void complete_journal_encrypt(struct crypto_async_request *req, int err)
858 struct journal_completion *comp = req->data;
860 if (likely(err == -EINPROGRESS)) {
861 complete(&comp->ic->crypto_backoff);
864 dm_integrity_io_error(comp->ic, "asynchronous encrypt", err);
866 complete_journal_op(comp);
869 static bool do_crypt(bool encrypt, struct skcipher_request *req, struct journal_completion *comp)
872 skcipher_request_set_callback(req, CRYPTO_TFM_REQ_MAY_BACKLOG,
873 complete_journal_encrypt, comp);
875 r = crypto_skcipher_encrypt(req);
877 r = crypto_skcipher_decrypt(req);
880 if (likely(r == -EINPROGRESS))
882 if (likely(r == -EBUSY)) {
883 wait_for_completion(&comp->ic->crypto_backoff);
884 reinit_completion(&comp->ic->crypto_backoff);
887 dm_integrity_io_error(comp->ic, "encrypt", r);
891 static void crypt_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
892 unsigned n_sections, struct journal_completion *comp)
894 struct scatterlist **source_sg;
895 struct scatterlist **target_sg;
897 atomic_add(2, &comp->in_flight);
899 if (likely(encrypt)) {
900 source_sg = ic->journal_scatterlist;
901 target_sg = ic->journal_io_scatterlist;
903 source_sg = ic->journal_io_scatterlist;
904 target_sg = ic->journal_scatterlist;
908 struct skcipher_request *req;
913 rw_section_mac(ic, section, true);
915 req = ic->sk_requests[section];
916 ivsize = crypto_skcipher_ivsize(ic->journal_crypt);
919 memcpy(iv, iv + ivsize, ivsize);
921 req->src = source_sg[section];
922 req->dst = target_sg[section];
924 if (unlikely(do_crypt(encrypt, req, comp)))
925 atomic_inc(&comp->in_flight);
929 } while (n_sections);
931 atomic_dec(&comp->in_flight);
932 complete_journal_op(comp);
935 static void encrypt_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
936 unsigned n_sections, struct journal_completion *comp)
939 return xor_journal(ic, encrypt, section, n_sections, comp);
941 return crypt_journal(ic, encrypt, section, n_sections, comp);
944 static void complete_journal_io(unsigned long error, void *context)
946 struct journal_completion *comp = context;
947 if (unlikely(error != 0))
948 dm_integrity_io_error(comp->ic, "writing journal", -EIO);
949 complete_journal_op(comp);
952 static void rw_journal_sectors(struct dm_integrity_c *ic, int op, int op_flags,
953 unsigned sector, unsigned n_sectors, struct journal_completion *comp)
955 struct dm_io_request io_req;
956 struct dm_io_region io_loc;
957 unsigned pl_index, pl_offset;
960 if (unlikely(dm_integrity_failed(ic))) {
962 complete_journal_io(-1UL, comp);
966 pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
967 pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
970 io_req.bi_op_flags = op_flags;
971 io_req.mem.type = DM_IO_PAGE_LIST;
973 io_req.mem.ptr.pl = &ic->journal_io[pl_index];
975 io_req.mem.ptr.pl = &ic->journal[pl_index];
976 io_req.mem.offset = pl_offset;
977 if (likely(comp != NULL)) {
978 io_req.notify.fn = complete_journal_io;
979 io_req.notify.context = comp;
981 io_req.notify.fn = NULL;
983 io_req.client = ic->io;
984 io_loc.bdev = ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev;
985 io_loc.sector = ic->start + SB_SECTORS + sector;
986 io_loc.count = n_sectors;
988 r = dm_io(&io_req, 1, &io_loc, NULL);
990 dm_integrity_io_error(ic, op == REQ_OP_READ ? "reading journal" : "writing journal", r);
992 WARN_ONCE(1, "asynchronous dm_io failed: %d", r);
993 complete_journal_io(-1UL, comp);
998 static void rw_journal(struct dm_integrity_c *ic, int op, int op_flags, unsigned section,
999 unsigned n_sections, struct journal_completion *comp)
1001 unsigned sector, n_sectors;
1003 sector = section * ic->journal_section_sectors;
1004 n_sectors = n_sections * ic->journal_section_sectors;
1006 rw_journal_sectors(ic, op, op_flags, sector, n_sectors, comp);
1009 static void write_journal(struct dm_integrity_c *ic, unsigned commit_start, unsigned commit_sections)
1011 struct journal_completion io_comp;
1012 struct journal_completion crypt_comp_1;
1013 struct journal_completion crypt_comp_2;
1017 init_completion(&io_comp.comp);
1019 if (commit_start + commit_sections <= ic->journal_sections) {
1020 io_comp.in_flight = (atomic_t)ATOMIC_INIT(1);
1021 if (ic->journal_io) {
1022 crypt_comp_1.ic = ic;
1023 init_completion(&crypt_comp_1.comp);
1024 crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
1025 encrypt_journal(ic, true, commit_start, commit_sections, &crypt_comp_1);
1026 wait_for_completion_io(&crypt_comp_1.comp);
1028 for (i = 0; i < commit_sections; i++)
1029 rw_section_mac(ic, commit_start + i, true);
1031 rw_journal(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, commit_start,
1032 commit_sections, &io_comp);
1035 io_comp.in_flight = (atomic_t)ATOMIC_INIT(2);
1036 to_end = ic->journal_sections - commit_start;
1037 if (ic->journal_io) {
1038 crypt_comp_1.ic = ic;
1039 init_completion(&crypt_comp_1.comp);
1040 crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
1041 encrypt_journal(ic, true, commit_start, to_end, &crypt_comp_1);
1042 if (try_wait_for_completion(&crypt_comp_1.comp)) {
1043 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
1044 reinit_completion(&crypt_comp_1.comp);
1045 crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
1046 encrypt_journal(ic, true, 0, commit_sections - to_end, &crypt_comp_1);
1047 wait_for_completion_io(&crypt_comp_1.comp);
1049 crypt_comp_2.ic = ic;
1050 init_completion(&crypt_comp_2.comp);
1051 crypt_comp_2.in_flight = (atomic_t)ATOMIC_INIT(0);
1052 encrypt_journal(ic, true, 0, commit_sections - to_end, &crypt_comp_2);
1053 wait_for_completion_io(&crypt_comp_1.comp);
1054 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
1055 wait_for_completion_io(&crypt_comp_2.comp);
1058 for (i = 0; i < to_end; i++)
1059 rw_section_mac(ic, commit_start + i, true);
1060 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
1061 for (i = 0; i < commit_sections - to_end; i++)
1062 rw_section_mac(ic, i, true);
1064 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, 0, commit_sections - to_end, &io_comp);
1067 wait_for_completion_io(&io_comp.comp);
1070 static void copy_from_journal(struct dm_integrity_c *ic, unsigned section, unsigned offset,
1071 unsigned n_sectors, sector_t target, io_notify_fn fn, void *data)
1073 struct dm_io_request io_req;
1074 struct dm_io_region io_loc;
1076 unsigned sector, pl_index, pl_offset;
1078 BUG_ON((target | n_sectors | offset) & (unsigned)(ic->sectors_per_block - 1));
1080 if (unlikely(dm_integrity_failed(ic))) {
1085 sector = section * ic->journal_section_sectors + JOURNAL_BLOCK_SECTORS + offset;
1087 pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
1088 pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
1090 io_req.bi_op = REQ_OP_WRITE;
1091 io_req.bi_op_flags = 0;
1092 io_req.mem.type = DM_IO_PAGE_LIST;
1093 io_req.mem.ptr.pl = &ic->journal[pl_index];
1094 io_req.mem.offset = pl_offset;
1095 io_req.notify.fn = fn;
1096 io_req.notify.context = data;
1097 io_req.client = ic->io;
1098 io_loc.bdev = ic->dev->bdev;
1099 io_loc.sector = target;
1100 io_loc.count = n_sectors;
1102 r = dm_io(&io_req, 1, &io_loc, NULL);
1104 WARN_ONCE(1, "asynchronous dm_io failed: %d", r);
1109 static bool ranges_overlap(struct dm_integrity_range *range1, struct dm_integrity_range *range2)
1111 return range1->logical_sector < range2->logical_sector + range2->n_sectors &&
1112 range1->logical_sector + range1->n_sectors > range2->logical_sector;
1115 static bool add_new_range(struct dm_integrity_c *ic, struct dm_integrity_range *new_range, bool check_waiting)
1117 struct rb_node **n = &ic->in_progress.rb_node;
1118 struct rb_node *parent;
1120 BUG_ON((new_range->logical_sector | new_range->n_sectors) & (unsigned)(ic->sectors_per_block - 1));
1122 if (likely(check_waiting)) {
1123 struct dm_integrity_range *range;
1124 list_for_each_entry(range, &ic->wait_list, wait_entry) {
1125 if (unlikely(ranges_overlap(range, new_range)))
1133 struct dm_integrity_range *range = container_of(*n, struct dm_integrity_range, node);
1136 if (new_range->logical_sector + new_range->n_sectors <= range->logical_sector) {
1137 n = &range->node.rb_left;
1138 } else if (new_range->logical_sector >= range->logical_sector + range->n_sectors) {
1139 n = &range->node.rb_right;
1145 rb_link_node(&new_range->node, parent, n);
1146 rb_insert_color(&new_range->node, &ic->in_progress);
1151 static void remove_range_unlocked(struct dm_integrity_c *ic, struct dm_integrity_range *range)
1153 rb_erase(&range->node, &ic->in_progress);
1154 while (unlikely(!list_empty(&ic->wait_list))) {
1155 struct dm_integrity_range *last_range =
1156 list_first_entry(&ic->wait_list, struct dm_integrity_range, wait_entry);
1157 struct task_struct *last_range_task;
1158 last_range_task = last_range->task;
1159 list_del(&last_range->wait_entry);
1160 if (!add_new_range(ic, last_range, false)) {
1161 last_range->task = last_range_task;
1162 list_add(&last_range->wait_entry, &ic->wait_list);
1165 last_range->waiting = false;
1166 wake_up_process(last_range_task);
1170 static void remove_range(struct dm_integrity_c *ic, struct dm_integrity_range *range)
1172 unsigned long flags;
1174 spin_lock_irqsave(&ic->endio_wait.lock, flags);
1175 remove_range_unlocked(ic, range);
1176 spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
1179 static void wait_and_add_new_range(struct dm_integrity_c *ic, struct dm_integrity_range *new_range)
1181 new_range->waiting = true;
1182 list_add_tail(&new_range->wait_entry, &ic->wait_list);
1183 new_range->task = current;
1185 __set_current_state(TASK_UNINTERRUPTIBLE);
1186 spin_unlock_irq(&ic->endio_wait.lock);
1188 spin_lock_irq(&ic->endio_wait.lock);
1189 } while (unlikely(new_range->waiting));
1192 static void add_new_range_and_wait(struct dm_integrity_c *ic, struct dm_integrity_range *new_range)
1194 if (unlikely(!add_new_range(ic, new_range, true)))
1195 wait_and_add_new_range(ic, new_range);
1198 static void init_journal_node(struct journal_node *node)
1200 RB_CLEAR_NODE(&node->node);
1201 node->sector = (sector_t)-1;
1204 static void add_journal_node(struct dm_integrity_c *ic, struct journal_node *node, sector_t sector)
1206 struct rb_node **link;
1207 struct rb_node *parent;
1209 node->sector = sector;
1210 BUG_ON(!RB_EMPTY_NODE(&node->node));
1212 link = &ic->journal_tree_root.rb_node;
1216 struct journal_node *j;
1218 j = container_of(parent, struct journal_node, node);
1219 if (sector < j->sector)
1220 link = &j->node.rb_left;
1222 link = &j->node.rb_right;
1225 rb_link_node(&node->node, parent, link);
1226 rb_insert_color(&node->node, &ic->journal_tree_root);
1229 static void remove_journal_node(struct dm_integrity_c *ic, struct journal_node *node)
1231 BUG_ON(RB_EMPTY_NODE(&node->node));
1232 rb_erase(&node->node, &ic->journal_tree_root);
1233 init_journal_node(node);
1236 #define NOT_FOUND (-1U)
1238 static unsigned find_journal_node(struct dm_integrity_c *ic, sector_t sector, sector_t *next_sector)
1240 struct rb_node *n = ic->journal_tree_root.rb_node;
1241 unsigned found = NOT_FOUND;
1242 *next_sector = (sector_t)-1;
1244 struct journal_node *j = container_of(n, struct journal_node, node);
1245 if (sector == j->sector) {
1246 found = j - ic->journal_tree;
1248 if (sector < j->sector) {
1249 *next_sector = j->sector;
1250 n = j->node.rb_left;
1252 n = j->node.rb_right;
1259 static bool test_journal_node(struct dm_integrity_c *ic, unsigned pos, sector_t sector)
1261 struct journal_node *node, *next_node;
1262 struct rb_node *next;
1264 if (unlikely(pos >= ic->journal_entries))
1266 node = &ic->journal_tree[pos];
1267 if (unlikely(RB_EMPTY_NODE(&node->node)))
1269 if (unlikely(node->sector != sector))
1272 next = rb_next(&node->node);
1273 if (unlikely(!next))
1276 next_node = container_of(next, struct journal_node, node);
1277 return next_node->sector != sector;
1280 static bool find_newer_committed_node(struct dm_integrity_c *ic, struct journal_node *node)
1282 struct rb_node *next;
1283 struct journal_node *next_node;
1284 unsigned next_section;
1286 BUG_ON(RB_EMPTY_NODE(&node->node));
1288 next = rb_next(&node->node);
1289 if (unlikely(!next))
1292 next_node = container_of(next, struct journal_node, node);
1294 if (next_node->sector != node->sector)
1297 next_section = (unsigned)(next_node - ic->journal_tree) / ic->journal_section_entries;
1298 if (next_section >= ic->committed_section &&
1299 next_section < ic->committed_section + ic->n_committed_sections)
1301 if (next_section + ic->journal_sections < ic->committed_section + ic->n_committed_sections)
1311 static int dm_integrity_rw_tag(struct dm_integrity_c *ic, unsigned char *tag, sector_t *metadata_block,
1312 unsigned *metadata_offset, unsigned total_size, int op)
1314 #define MAY_BE_FILLER 1
1315 #define MAY_BE_HASH 2
1316 unsigned hash_offset = 0;
1317 unsigned may_be = MAY_BE_HASH | (ic->discard ? MAY_BE_FILLER : 0);
1320 unsigned char *data, *dp;
1321 struct dm_buffer *b;
1325 r = dm_integrity_failed(ic);
1329 data = dm_bufio_read(ic->bufio, *metadata_block, &b);
1331 return PTR_ERR(data);
1333 to_copy = min((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - *metadata_offset, total_size);
1334 dp = data + *metadata_offset;
1335 if (op == TAG_READ) {
1336 memcpy(tag, dp, to_copy);
1337 } else if (op == TAG_WRITE) {
1338 memcpy(dp, tag, to_copy);
1339 dm_bufio_mark_partial_buffer_dirty(b, *metadata_offset, *metadata_offset + to_copy);
1341 /* e.g.: op == TAG_CMP */
1343 if (likely(is_power_of_2(ic->tag_size))) {
1344 if (unlikely(memcmp(dp, tag, to_copy)))
1345 if (unlikely(!ic->discard) ||
1346 unlikely(memchr_inv(dp, DISCARD_FILLER, to_copy) != NULL)) {
1354 for (i = 0; i < to_copy; i++, ts--) {
1355 if (unlikely(dp[i] != tag[i]))
1356 may_be &= ~MAY_BE_HASH;
1357 if (likely(dp[i] != DISCARD_FILLER))
1358 may_be &= ~MAY_BE_FILLER;
1360 if (unlikely(hash_offset == ic->tag_size)) {
1361 if (unlikely(!may_be)) {
1362 dm_bufio_release(b);
1366 may_be = MAY_BE_HASH | (ic->discard ? MAY_BE_FILLER : 0);
1371 dm_bufio_release(b);
1374 *metadata_offset += to_copy;
1375 if (unlikely(*metadata_offset == 1U << SECTOR_SHIFT << ic->log2_buffer_sectors)) {
1376 (*metadata_block)++;
1377 *metadata_offset = 0;
1380 if (unlikely(!is_power_of_2(ic->tag_size))) {
1381 hash_offset = (hash_offset + to_copy) % ic->tag_size;
1384 total_size -= to_copy;
1385 } while (unlikely(total_size));
1388 #undef MAY_BE_FILLER
1392 struct flush_request {
1393 struct dm_io_request io_req;
1394 struct dm_io_region io_reg;
1395 struct dm_integrity_c *ic;
1396 struct completion comp;
1399 static void flush_notify(unsigned long error, void *fr_)
1401 struct flush_request *fr = fr_;
1402 if (unlikely(error != 0))
1403 dm_integrity_io_error(fr->ic, "flusing disk cache", -EIO);
1404 complete(&fr->comp);
1407 static void dm_integrity_flush_buffers(struct dm_integrity_c *ic, bool flush_data)
1411 struct flush_request fr;
1416 fr.io_req.bi_op = REQ_OP_WRITE,
1417 fr.io_req.bi_op_flags = REQ_PREFLUSH | REQ_SYNC,
1418 fr.io_req.mem.type = DM_IO_KMEM,
1419 fr.io_req.mem.ptr.addr = NULL,
1420 fr.io_req.notify.fn = flush_notify,
1421 fr.io_req.notify.context = &fr;
1422 fr.io_req.client = dm_bufio_get_dm_io_client(ic->bufio),
1423 fr.io_reg.bdev = ic->dev->bdev,
1424 fr.io_reg.sector = 0,
1425 fr.io_reg.count = 0,
1427 init_completion(&fr.comp);
1428 r = dm_io(&fr.io_req, 1, &fr.io_reg, NULL);
1432 r = dm_bufio_write_dirty_buffers(ic->bufio);
1434 dm_integrity_io_error(ic, "writing tags", r);
1437 wait_for_completion(&fr.comp);
1440 static void sleep_on_endio_wait(struct dm_integrity_c *ic)
1442 DECLARE_WAITQUEUE(wait, current);
1443 __add_wait_queue(&ic->endio_wait, &wait);
1444 __set_current_state(TASK_UNINTERRUPTIBLE);
1445 spin_unlock_irq(&ic->endio_wait.lock);
1447 spin_lock_irq(&ic->endio_wait.lock);
1448 __remove_wait_queue(&ic->endio_wait, &wait);
1451 static void autocommit_fn(struct timer_list *t)
1453 struct dm_integrity_c *ic = from_timer(ic, t, autocommit_timer);
1455 if (likely(!dm_integrity_failed(ic)))
1456 queue_work(ic->commit_wq, &ic->commit_work);
1459 static void schedule_autocommit(struct dm_integrity_c *ic)
1461 if (!timer_pending(&ic->autocommit_timer))
1462 mod_timer(&ic->autocommit_timer, jiffies + ic->autocommit_jiffies);
1465 static void submit_flush_bio(struct dm_integrity_c *ic, struct dm_integrity_io *dio)
1468 unsigned long flags;
1470 spin_lock_irqsave(&ic->endio_wait.lock, flags);
1471 bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1472 bio_list_add(&ic->flush_bio_list, bio);
1473 spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
1475 queue_work(ic->commit_wq, &ic->commit_work);
1478 static void do_endio(struct dm_integrity_c *ic, struct bio *bio)
1480 int r = dm_integrity_failed(ic);
1481 if (unlikely(r) && !bio->bi_status)
1482 bio->bi_status = errno_to_blk_status(r);
1483 if (unlikely(ic->synchronous_mode) && bio_op(bio) == REQ_OP_WRITE) {
1484 unsigned long flags;
1485 spin_lock_irqsave(&ic->endio_wait.lock, flags);
1486 bio_list_add(&ic->synchronous_bios, bio);
1487 queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, 0);
1488 spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
1494 static void do_endio_flush(struct dm_integrity_c *ic, struct dm_integrity_io *dio)
1496 struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1498 if (unlikely(dio->fua) && likely(!bio->bi_status) && likely(!dm_integrity_failed(ic)))
1499 submit_flush_bio(ic, dio);
1504 static void dec_in_flight(struct dm_integrity_io *dio)
1506 if (atomic_dec_and_test(&dio->in_flight)) {
1507 struct dm_integrity_c *ic = dio->ic;
1510 remove_range(ic, &dio->range);
1512 if (dio->op == REQ_OP_WRITE || unlikely(dio->op == REQ_OP_DISCARD))
1513 schedule_autocommit(ic);
1515 bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1517 if (unlikely(dio->bi_status) && !bio->bi_status)
1518 bio->bi_status = dio->bi_status;
1519 if (likely(!bio->bi_status) && unlikely(bio_sectors(bio) != dio->range.n_sectors)) {
1520 dio->range.logical_sector += dio->range.n_sectors;
1521 bio_advance(bio, dio->range.n_sectors << SECTOR_SHIFT);
1522 INIT_WORK(&dio->work, integrity_bio_wait);
1523 queue_work(ic->offload_wq, &dio->work);
1526 do_endio_flush(ic, dio);
1530 static void integrity_end_io(struct bio *bio)
1532 struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
1534 dm_bio_restore(&dio->bio_details, bio);
1535 if (bio->bi_integrity)
1536 bio->bi_opf |= REQ_INTEGRITY;
1538 if (dio->completion)
1539 complete(dio->completion);
1544 static void integrity_sector_checksum(struct dm_integrity_c *ic, sector_t sector,
1545 const char *data, char *result)
1547 __u64 sector_le = cpu_to_le64(sector);
1548 SHASH_DESC_ON_STACK(req, ic->internal_hash);
1550 unsigned digest_size;
1552 req->tfm = ic->internal_hash;
1554 r = crypto_shash_init(req);
1555 if (unlikely(r < 0)) {
1556 dm_integrity_io_error(ic, "crypto_shash_init", r);
1560 r = crypto_shash_update(req, (const __u8 *)§or_le, sizeof sector_le);
1561 if (unlikely(r < 0)) {
1562 dm_integrity_io_error(ic, "crypto_shash_update", r);
1566 r = crypto_shash_update(req, data, ic->sectors_per_block << SECTOR_SHIFT);
1567 if (unlikely(r < 0)) {
1568 dm_integrity_io_error(ic, "crypto_shash_update", r);
1572 r = crypto_shash_final(req, result);
1573 if (unlikely(r < 0)) {
1574 dm_integrity_io_error(ic, "crypto_shash_final", r);
1578 digest_size = crypto_shash_digestsize(ic->internal_hash);
1579 if (unlikely(digest_size < ic->tag_size))
1580 memset(result + digest_size, 0, ic->tag_size - digest_size);
1585 /* this shouldn't happen anyway, the hash functions have no reason to fail */
1586 get_random_bytes(result, ic->tag_size);
1589 static void integrity_metadata(struct work_struct *w)
1591 struct dm_integrity_io *dio = container_of(w, struct dm_integrity_io, work);
1592 struct dm_integrity_c *ic = dio->ic;
1596 if (ic->internal_hash) {
1597 struct bvec_iter iter;
1599 unsigned digest_size = crypto_shash_digestsize(ic->internal_hash);
1600 struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1602 unsigned extra_space = unlikely(digest_size > ic->tag_size) ? digest_size - ic->tag_size : 0;
1603 char checksums_onstack[max((size_t)HASH_MAX_DIGESTSIZE, MAX_TAG_SIZE)];
1605 unsigned sectors_to_process;
1607 if (unlikely(ic->mode == 'R'))
1610 if (likely(dio->op != REQ_OP_DISCARD))
1611 checksums = kmalloc((PAGE_SIZE >> SECTOR_SHIFT >> ic->sb->log2_sectors_per_block) * ic->tag_size + extra_space,
1612 GFP_NOIO | __GFP_NORETRY | __GFP_NOWARN);
1614 checksums = kmalloc(PAGE_SIZE, GFP_NOIO | __GFP_NORETRY | __GFP_NOWARN);
1616 checksums = checksums_onstack;
1617 if (WARN_ON(extra_space &&
1618 digest_size > sizeof(checksums_onstack))) {
1624 if (unlikely(dio->op == REQ_OP_DISCARD)) {
1625 sector_t bi_sector = dio->bio_details.bi_iter.bi_sector;
1626 unsigned bi_size = dio->bio_details.bi_iter.bi_size;
1627 unsigned max_size = likely(checksums != checksums_onstack) ? PAGE_SIZE : HASH_MAX_DIGESTSIZE;
1628 unsigned max_blocks = max_size / ic->tag_size;
1629 memset(checksums, DISCARD_FILLER, max_size);
1632 unsigned this_step_blocks = bi_size >> (SECTOR_SHIFT + ic->sb->log2_sectors_per_block);
1633 this_step_blocks = min(this_step_blocks, max_blocks);
1634 r = dm_integrity_rw_tag(ic, checksums, &dio->metadata_block, &dio->metadata_offset,
1635 this_step_blocks * ic->tag_size, TAG_WRITE);
1637 if (likely(checksums != checksums_onstack))
1642 /*if (bi_size < this_step_blocks << (SECTOR_SHIFT + ic->sb->log2_sectors_per_block)) {
1643 printk("BUGG: bi_sector: %llx, bi_size: %u\n", bi_sector, bi_size);
1644 printk("BUGG: this_step_blocks: %u\n", this_step_blocks);
1647 bi_size -= this_step_blocks << (SECTOR_SHIFT + ic->sb->log2_sectors_per_block);
1648 bi_sector += this_step_blocks << ic->sb->log2_sectors_per_block;
1651 if (likely(checksums != checksums_onstack))
1656 sector = dio->range.logical_sector;
1657 sectors_to_process = dio->range.n_sectors;
1659 __bio_for_each_segment(bv, bio, iter, dio->bio_details.bi_iter) {
1660 struct bio_vec bv_copy = bv;
1662 char *mem, *checksums_ptr;
1665 mem = (char *)kmap_atomic(bv_copy.bv_page) + bv_copy.bv_offset;
1667 checksums_ptr = checksums;
1669 integrity_sector_checksum(ic, sector, mem + pos, checksums_ptr);
1670 checksums_ptr += ic->tag_size;
1671 sectors_to_process -= ic->sectors_per_block;
1672 pos += ic->sectors_per_block << SECTOR_SHIFT;
1673 sector += ic->sectors_per_block;
1674 } while (pos < bv_copy.bv_len && sectors_to_process && checksums != checksums_onstack);
1677 r = dm_integrity_rw_tag(ic, checksums, &dio->metadata_block, &dio->metadata_offset,
1678 checksums_ptr - checksums, dio->op == REQ_OP_READ ? TAG_CMP : TAG_WRITE);
1681 char b[BDEVNAME_SIZE];
1682 DMERR_LIMIT("%s: Checksum failed at sector 0x%llx", bio_devname(bio, b),
1683 (sector - ((r + ic->tag_size - 1) / ic->tag_size)));
1685 atomic64_inc(&ic->number_of_mismatches);
1687 if (likely(checksums != checksums_onstack))
1692 if (!sectors_to_process)
1695 if (unlikely(pos < bv_copy.bv_len)) {
1696 bv_copy.bv_offset += pos;
1697 bv_copy.bv_len -= pos;
1702 if (likely(checksums != checksums_onstack))
1705 struct bio_integrity_payload *bip = dio->bio_details.bi_integrity;
1709 struct bvec_iter iter;
1710 unsigned data_to_process = dio->range.n_sectors;
1711 sector_to_block(ic, data_to_process);
1712 data_to_process *= ic->tag_size;
1714 bip_for_each_vec(biv, bip, iter) {
1718 BUG_ON(PageHighMem(biv.bv_page));
1719 tag = lowmem_page_address(biv.bv_page) + biv.bv_offset;
1720 this_len = min(biv.bv_len, data_to_process);
1721 r = dm_integrity_rw_tag(ic, tag, &dio->metadata_block, &dio->metadata_offset,
1722 this_len, dio->op == REQ_OP_READ ? TAG_READ : TAG_WRITE);
1725 data_to_process -= this_len;
1726 if (!data_to_process)
1735 dio->bi_status = errno_to_blk_status(r);
1739 static int dm_integrity_map(struct dm_target *ti, struct bio *bio)
1741 struct dm_integrity_c *ic = ti->private;
1742 struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
1743 struct bio_integrity_payload *bip;
1745 sector_t area, offset;
1749 dio->op = bio_op(bio);
1751 if (unlikely(dio->op == REQ_OP_DISCARD)) {
1752 if (ti->max_io_len) {
1753 sector_t sec = dm_target_offset(ti, bio->bi_iter.bi_sector);
1754 unsigned log2_max_io_len = __fls(ti->max_io_len);
1755 sector_t start_boundary = sec >> log2_max_io_len;
1756 sector_t end_boundary = (sec + bio_sectors(bio) - 1) >> log2_max_io_len;
1757 if (start_boundary < end_boundary) {
1758 sector_t len = ti->max_io_len - (sec & (ti->max_io_len - 1));
1759 dm_accept_partial_bio(bio, len);
1764 if (unlikely(bio->bi_opf & REQ_PREFLUSH)) {
1765 submit_flush_bio(ic, dio);
1766 return DM_MAPIO_SUBMITTED;
1769 dio->range.logical_sector = dm_target_offset(ti, bio->bi_iter.bi_sector);
1770 dio->fua = dio->op == REQ_OP_WRITE && bio->bi_opf & REQ_FUA;
1771 if (unlikely(dio->fua)) {
1773 * Don't pass down the FUA flag because we have to flush
1774 * disk cache anyway.
1776 bio->bi_opf &= ~REQ_FUA;
1778 if (unlikely(dio->range.logical_sector + bio_sectors(bio) > ic->provided_data_sectors)) {
1779 DMERR("Too big sector number: 0x%llx + 0x%x > 0x%llx",
1780 dio->range.logical_sector, bio_sectors(bio),
1781 ic->provided_data_sectors);
1782 return DM_MAPIO_KILL;
1784 if (unlikely((dio->range.logical_sector | bio_sectors(bio)) & (unsigned)(ic->sectors_per_block - 1))) {
1785 DMERR("Bio not aligned on %u sectors: 0x%llx, 0x%x",
1786 ic->sectors_per_block,
1787 dio->range.logical_sector, bio_sectors(bio));
1788 return DM_MAPIO_KILL;
1791 if (ic->sectors_per_block > 1 && likely(dio->op != REQ_OP_DISCARD)) {
1792 struct bvec_iter iter;
1794 bio_for_each_segment(bv, bio, iter) {
1795 if (unlikely(bv.bv_len & ((ic->sectors_per_block << SECTOR_SHIFT) - 1))) {
1796 DMERR("Bio vector (%u,%u) is not aligned on %u-sector boundary",
1797 bv.bv_offset, bv.bv_len, ic->sectors_per_block);
1798 return DM_MAPIO_KILL;
1803 bip = bio_integrity(bio);
1804 if (!ic->internal_hash) {
1806 unsigned wanted_tag_size = bio_sectors(bio) >> ic->sb->log2_sectors_per_block;
1807 if (ic->log2_tag_size >= 0)
1808 wanted_tag_size <<= ic->log2_tag_size;
1810 wanted_tag_size *= ic->tag_size;
1811 if (unlikely(wanted_tag_size != bip->bip_iter.bi_size)) {
1812 DMERR("Invalid integrity data size %u, expected %u",
1813 bip->bip_iter.bi_size, wanted_tag_size);
1814 return DM_MAPIO_KILL;
1818 if (unlikely(bip != NULL)) {
1819 DMERR("Unexpected integrity data when using internal hash");
1820 return DM_MAPIO_KILL;
1824 if (unlikely(ic->mode == 'R') && unlikely(dio->op != REQ_OP_READ))
1825 return DM_MAPIO_KILL;
1827 get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
1828 dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset, &dio->metadata_offset);
1829 bio->bi_iter.bi_sector = get_data_sector(ic, area, offset);
1831 dm_integrity_map_continue(dio, true);
1832 return DM_MAPIO_SUBMITTED;
1835 static bool __journal_read_write(struct dm_integrity_io *dio, struct bio *bio,
1836 unsigned journal_section, unsigned journal_entry)
1838 struct dm_integrity_c *ic = dio->ic;
1839 sector_t logical_sector;
1842 logical_sector = dio->range.logical_sector;
1843 n_sectors = dio->range.n_sectors;
1845 struct bio_vec bv = bio_iovec(bio);
1848 if (unlikely(bv.bv_len >> SECTOR_SHIFT > n_sectors))
1849 bv.bv_len = n_sectors << SECTOR_SHIFT;
1850 n_sectors -= bv.bv_len >> SECTOR_SHIFT;
1851 bio_advance_iter(bio, &bio->bi_iter, bv.bv_len);
1853 mem = kmap_atomic(bv.bv_page);
1854 if (likely(dio->op == REQ_OP_WRITE))
1855 flush_dcache_page(bv.bv_page);
1858 struct journal_entry *je = access_journal_entry(ic, journal_section, journal_entry);
1860 if (unlikely(dio->op == REQ_OP_READ)) {
1861 struct journal_sector *js;
1865 if (unlikely(journal_entry_is_inprogress(je))) {
1866 flush_dcache_page(bv.bv_page);
1869 __io_wait_event(ic->copy_to_journal_wait, !journal_entry_is_inprogress(je));
1873 BUG_ON(journal_entry_get_sector(je) != logical_sector);
1874 js = access_journal_data(ic, journal_section, journal_entry);
1875 mem_ptr = mem + bv.bv_offset;
1878 memcpy(mem_ptr, js, JOURNAL_SECTOR_DATA);
1879 *(commit_id_t *)(mem_ptr + JOURNAL_SECTOR_DATA) = je->last_bytes[s];
1881 mem_ptr += 1 << SECTOR_SHIFT;
1882 } while (++s < ic->sectors_per_block);
1883 #ifdef INTERNAL_VERIFY
1884 if (ic->internal_hash) {
1885 char checksums_onstack[max((size_t)HASH_MAX_DIGESTSIZE, MAX_TAG_SIZE)];
1887 integrity_sector_checksum(ic, logical_sector, mem + bv.bv_offset, checksums_onstack);
1888 if (unlikely(memcmp(checksums_onstack, journal_entry_tag(ic, je), ic->tag_size))) {
1889 DMERR_LIMIT("Checksum failed when reading from journal, at sector 0x%llx",
1896 if (!ic->internal_hash) {
1897 struct bio_integrity_payload *bip = bio_integrity(bio);
1898 unsigned tag_todo = ic->tag_size;
1899 char *tag_ptr = journal_entry_tag(ic, je);
1902 struct bio_vec biv = bvec_iter_bvec(bip->bip_vec, bip->bip_iter);
1903 unsigned tag_now = min(biv.bv_len, tag_todo);
1905 BUG_ON(PageHighMem(biv.bv_page));
1906 tag_addr = lowmem_page_address(biv.bv_page) + biv.bv_offset;
1907 if (likely(dio->op == REQ_OP_WRITE))
1908 memcpy(tag_ptr, tag_addr, tag_now);
1910 memcpy(tag_addr, tag_ptr, tag_now);
1911 bvec_iter_advance(bip->bip_vec, &bip->bip_iter, tag_now);
1913 tag_todo -= tag_now;
1914 } while (unlikely(tag_todo)); else {
1915 if (likely(dio->op == REQ_OP_WRITE))
1916 memset(tag_ptr, 0, tag_todo);
1920 if (likely(dio->op == REQ_OP_WRITE)) {
1921 struct journal_sector *js;
1924 js = access_journal_data(ic, journal_section, journal_entry);
1925 memcpy(js, mem + bv.bv_offset, ic->sectors_per_block << SECTOR_SHIFT);
1929 je->last_bytes[s] = js[s].commit_id;
1930 } while (++s < ic->sectors_per_block);
1932 if (ic->internal_hash) {
1933 unsigned digest_size = crypto_shash_digestsize(ic->internal_hash);
1934 if (unlikely(digest_size > ic->tag_size)) {
1935 char checksums_onstack[HASH_MAX_DIGESTSIZE];
1936 integrity_sector_checksum(ic, logical_sector, (char *)js, checksums_onstack);
1937 memcpy(journal_entry_tag(ic, je), checksums_onstack, ic->tag_size);
1939 integrity_sector_checksum(ic, logical_sector, (char *)js, journal_entry_tag(ic, je));
1942 journal_entry_set_sector(je, logical_sector);
1944 logical_sector += ic->sectors_per_block;
1947 if (unlikely(journal_entry == ic->journal_section_entries)) {
1950 wraparound_section(ic, &journal_section);
1953 bv.bv_offset += ic->sectors_per_block << SECTOR_SHIFT;
1954 } while (bv.bv_len -= ic->sectors_per_block << SECTOR_SHIFT);
1956 if (unlikely(dio->op == REQ_OP_READ))
1957 flush_dcache_page(bv.bv_page);
1959 } while (n_sectors);
1961 if (likely(dio->op == REQ_OP_WRITE)) {
1963 if (unlikely(waitqueue_active(&ic->copy_to_journal_wait)))
1964 wake_up(&ic->copy_to_journal_wait);
1965 if (READ_ONCE(ic->free_sectors) <= ic->free_sectors_threshold) {
1966 queue_work(ic->commit_wq, &ic->commit_work);
1968 schedule_autocommit(ic);
1971 remove_range(ic, &dio->range);
1974 if (unlikely(bio->bi_iter.bi_size)) {
1975 sector_t area, offset;
1977 dio->range.logical_sector = logical_sector;
1978 get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
1979 dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset, &dio->metadata_offset);
1986 static void dm_integrity_map_continue(struct dm_integrity_io *dio, bool from_map)
1988 struct dm_integrity_c *ic = dio->ic;
1989 struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1990 unsigned journal_section, journal_entry;
1991 unsigned journal_read_pos;
1992 struct completion read_comp;
1993 bool discard_retried = false;
1994 bool need_sync_io = ic->internal_hash && dio->op == REQ_OP_READ;
1995 if (unlikely(dio->op == REQ_OP_DISCARD) && ic->mode != 'D')
1996 need_sync_io = true;
1998 if (need_sync_io && from_map) {
1999 INIT_WORK(&dio->work, integrity_bio_wait);
2000 queue_work(ic->offload_wq, &dio->work);
2005 spin_lock_irq(&ic->endio_wait.lock);
2007 if (unlikely(dm_integrity_failed(ic))) {
2008 spin_unlock_irq(&ic->endio_wait.lock);
2012 dio->range.n_sectors = bio_sectors(bio);
2013 journal_read_pos = NOT_FOUND;
2014 if (ic->mode == 'J' && likely(dio->op != REQ_OP_DISCARD)) {
2015 if (dio->op == REQ_OP_WRITE) {
2016 unsigned next_entry, i, pos;
2017 unsigned ws, we, range_sectors;
2019 dio->range.n_sectors = min(dio->range.n_sectors,
2020 (sector_t)ic->free_sectors << ic->sb->log2_sectors_per_block);
2021 if (unlikely(!dio->range.n_sectors)) {
2023 goto offload_to_thread;
2024 sleep_on_endio_wait(ic);
2027 range_sectors = dio->range.n_sectors >> ic->sb->log2_sectors_per_block;
2028 ic->free_sectors -= range_sectors;
2029 journal_section = ic->free_section;
2030 journal_entry = ic->free_section_entry;
2032 next_entry = ic->free_section_entry + range_sectors;
2033 ic->free_section_entry = next_entry % ic->journal_section_entries;
2034 ic->free_section += next_entry / ic->journal_section_entries;
2035 ic->n_uncommitted_sections += next_entry / ic->journal_section_entries;
2036 wraparound_section(ic, &ic->free_section);
2038 pos = journal_section * ic->journal_section_entries + journal_entry;
2039 ws = journal_section;
2043 struct journal_entry *je;
2045 add_journal_node(ic, &ic->journal_tree[pos], dio->range.logical_sector + i);
2047 if (unlikely(pos >= ic->journal_entries))
2050 je = access_journal_entry(ic, ws, we);
2051 BUG_ON(!journal_entry_is_unused(je));
2052 journal_entry_set_inprogress(je);
2054 if (unlikely(we == ic->journal_section_entries)) {
2057 wraparound_section(ic, &ws);
2059 } while ((i += ic->sectors_per_block) < dio->range.n_sectors);
2061 spin_unlock_irq(&ic->endio_wait.lock);
2062 goto journal_read_write;
2064 sector_t next_sector;
2065 journal_read_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
2066 if (likely(journal_read_pos == NOT_FOUND)) {
2067 if (unlikely(dio->range.n_sectors > next_sector - dio->range.logical_sector))
2068 dio->range.n_sectors = next_sector - dio->range.logical_sector;
2071 unsigned jp = journal_read_pos + 1;
2072 for (i = ic->sectors_per_block; i < dio->range.n_sectors; i += ic->sectors_per_block, jp++) {
2073 if (!test_journal_node(ic, jp, dio->range.logical_sector + i))
2076 dio->range.n_sectors = i;
2080 if (unlikely(!add_new_range(ic, &dio->range, true))) {
2082 * We must not sleep in the request routine because it could
2083 * stall bios on current->bio_list.
2084 * So, we offload the bio to a workqueue if we have to sleep.
2088 spin_unlock_irq(&ic->endio_wait.lock);
2089 INIT_WORK(&dio->work, integrity_bio_wait);
2090 queue_work(ic->wait_wq, &dio->work);
2093 if (journal_read_pos != NOT_FOUND)
2094 dio->range.n_sectors = ic->sectors_per_block;
2095 wait_and_add_new_range(ic, &dio->range);
2097 * wait_and_add_new_range drops the spinlock, so the journal
2098 * may have been changed arbitrarily. We need to recheck.
2099 * To simplify the code, we restrict I/O size to just one block.
2101 if (journal_read_pos != NOT_FOUND) {
2102 sector_t next_sector;
2103 unsigned new_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
2104 if (unlikely(new_pos != journal_read_pos)) {
2105 remove_range_unlocked(ic, &dio->range);
2110 if (ic->mode == 'J' && likely(dio->op == REQ_OP_DISCARD) && !discard_retried) {
2111 sector_t next_sector;
2112 unsigned new_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
2113 if (unlikely(new_pos != NOT_FOUND) ||
2114 unlikely(next_sector < dio->range.logical_sector - dio->range.n_sectors)) {
2115 remove_range_unlocked(ic, &dio->range);
2116 spin_unlock_irq(&ic->endio_wait.lock);
2117 queue_work(ic->commit_wq, &ic->commit_work);
2118 flush_workqueue(ic->commit_wq);
2119 queue_work(ic->writer_wq, &ic->writer_work);
2120 flush_workqueue(ic->writer_wq);
2121 discard_retried = true;
2125 spin_unlock_irq(&ic->endio_wait.lock);
2127 if (unlikely(journal_read_pos != NOT_FOUND)) {
2128 journal_section = journal_read_pos / ic->journal_section_entries;
2129 journal_entry = journal_read_pos % ic->journal_section_entries;
2130 goto journal_read_write;
2133 if (ic->mode == 'B' && (dio->op == REQ_OP_WRITE || unlikely(dio->op == REQ_OP_DISCARD))) {
2134 if (!block_bitmap_op(ic, ic->may_write_bitmap, dio->range.logical_sector,
2135 dio->range.n_sectors, BITMAP_OP_TEST_ALL_SET)) {
2136 struct bitmap_block_status *bbs;
2138 bbs = sector_to_bitmap_block(ic, dio->range.logical_sector);
2139 spin_lock(&bbs->bio_queue_lock);
2140 bio_list_add(&bbs->bio_queue, bio);
2141 spin_unlock(&bbs->bio_queue_lock);
2142 queue_work(ic->writer_wq, &bbs->work);
2147 dio->in_flight = (atomic_t)ATOMIC_INIT(2);
2150 init_completion(&read_comp);
2151 dio->completion = &read_comp;
2153 dio->completion = NULL;
2155 dm_bio_record(&dio->bio_details, bio);
2156 bio_set_dev(bio, ic->dev->bdev);
2157 bio->bi_integrity = NULL;
2158 bio->bi_opf &= ~REQ_INTEGRITY;
2159 bio->bi_end_io = integrity_end_io;
2160 bio->bi_iter.bi_size = dio->range.n_sectors << SECTOR_SHIFT;
2162 if (unlikely(dio->op == REQ_OP_DISCARD) && likely(ic->mode != 'D')) {
2163 integrity_metadata(&dio->work);
2164 dm_integrity_flush_buffers(ic, false);
2166 dio->in_flight = (atomic_t)ATOMIC_INIT(1);
2167 dio->completion = NULL;
2169 submit_bio_noacct(bio);
2174 submit_bio_noacct(bio);
2177 wait_for_completion_io(&read_comp);
2178 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING) &&
2179 dio->range.logical_sector + dio->range.n_sectors > le64_to_cpu(ic->sb->recalc_sector))
2181 if (ic->mode == 'B') {
2182 if (!block_bitmap_op(ic, ic->recalc_bitmap, dio->range.logical_sector,
2183 dio->range.n_sectors, BITMAP_OP_TEST_ALL_CLEAR))
2187 if (likely(!bio->bi_status))
2188 integrity_metadata(&dio->work);
2194 INIT_WORK(&dio->work, integrity_metadata);
2195 queue_work(ic->metadata_wq, &dio->work);
2201 if (unlikely(__journal_read_write(dio, bio, journal_section, journal_entry)))
2204 do_endio_flush(ic, dio);
2208 static void integrity_bio_wait(struct work_struct *w)
2210 struct dm_integrity_io *dio = container_of(w, struct dm_integrity_io, work);
2212 dm_integrity_map_continue(dio, false);
2215 static void pad_uncommitted(struct dm_integrity_c *ic)
2217 if (ic->free_section_entry) {
2218 ic->free_sectors -= ic->journal_section_entries - ic->free_section_entry;
2219 ic->free_section_entry = 0;
2221 wraparound_section(ic, &ic->free_section);
2222 ic->n_uncommitted_sections++;
2224 if (WARN_ON(ic->journal_sections * ic->journal_section_entries !=
2225 (ic->n_uncommitted_sections + ic->n_committed_sections) *
2226 ic->journal_section_entries + ic->free_sectors)) {
2227 DMCRIT("journal_sections %u, journal_section_entries %u, "
2228 "n_uncommitted_sections %u, n_committed_sections %u, "
2229 "journal_section_entries %u, free_sectors %u",
2230 ic->journal_sections, ic->journal_section_entries,
2231 ic->n_uncommitted_sections, ic->n_committed_sections,
2232 ic->journal_section_entries, ic->free_sectors);
2236 static void integrity_commit(struct work_struct *w)
2238 struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, commit_work);
2239 unsigned commit_start, commit_sections;
2241 struct bio *flushes;
2243 del_timer(&ic->autocommit_timer);
2245 spin_lock_irq(&ic->endio_wait.lock);
2246 flushes = bio_list_get(&ic->flush_bio_list);
2247 if (unlikely(ic->mode != 'J')) {
2248 spin_unlock_irq(&ic->endio_wait.lock);
2249 dm_integrity_flush_buffers(ic, true);
2250 goto release_flush_bios;
2253 pad_uncommitted(ic);
2254 commit_start = ic->uncommitted_section;
2255 commit_sections = ic->n_uncommitted_sections;
2256 spin_unlock_irq(&ic->endio_wait.lock);
2258 if (!commit_sections)
2259 goto release_flush_bios;
2261 ic->wrote_to_journal = true;
2264 for (n = 0; n < commit_sections; n++) {
2265 for (j = 0; j < ic->journal_section_entries; j++) {
2266 struct journal_entry *je;
2267 je = access_journal_entry(ic, i, j);
2268 io_wait_event(ic->copy_to_journal_wait, !journal_entry_is_inprogress(je));
2270 for (j = 0; j < ic->journal_section_sectors; j++) {
2271 struct journal_sector *js;
2272 js = access_journal(ic, i, j);
2273 js->commit_id = dm_integrity_commit_id(ic, i, j, ic->commit_seq);
2276 if (unlikely(i >= ic->journal_sections))
2277 ic->commit_seq = next_commit_seq(ic->commit_seq);
2278 wraparound_section(ic, &i);
2282 write_journal(ic, commit_start, commit_sections);
2284 spin_lock_irq(&ic->endio_wait.lock);
2285 ic->uncommitted_section += commit_sections;
2286 wraparound_section(ic, &ic->uncommitted_section);
2287 ic->n_uncommitted_sections -= commit_sections;
2288 ic->n_committed_sections += commit_sections;
2289 spin_unlock_irq(&ic->endio_wait.lock);
2291 if (READ_ONCE(ic->free_sectors) <= ic->free_sectors_threshold)
2292 queue_work(ic->writer_wq, &ic->writer_work);
2296 struct bio *next = flushes->bi_next;
2297 flushes->bi_next = NULL;
2298 do_endio(ic, flushes);
2303 static void complete_copy_from_journal(unsigned long error, void *context)
2305 struct journal_io *io = context;
2306 struct journal_completion *comp = io->comp;
2307 struct dm_integrity_c *ic = comp->ic;
2308 remove_range(ic, &io->range);
2309 mempool_free(io, &ic->journal_io_mempool);
2310 if (unlikely(error != 0))
2311 dm_integrity_io_error(ic, "copying from journal", -EIO);
2312 complete_journal_op(comp);
2315 static void restore_last_bytes(struct dm_integrity_c *ic, struct journal_sector *js,
2316 struct journal_entry *je)
2320 js->commit_id = je->last_bytes[s];
2322 } while (++s < ic->sectors_per_block);
2325 static void do_journal_write(struct dm_integrity_c *ic, unsigned write_start,
2326 unsigned write_sections, bool from_replay)
2329 struct journal_completion comp;
2330 struct blk_plug plug;
2332 blk_start_plug(&plug);
2335 comp.in_flight = (atomic_t)ATOMIC_INIT(1);
2336 init_completion(&comp.comp);
2339 for (n = 0; n < write_sections; n++, i++, wraparound_section(ic, &i)) {
2340 #ifndef INTERNAL_VERIFY
2341 if (unlikely(from_replay))
2343 rw_section_mac(ic, i, false);
2344 for (j = 0; j < ic->journal_section_entries; j++) {
2345 struct journal_entry *je = access_journal_entry(ic, i, j);
2346 sector_t sec, area, offset;
2347 unsigned k, l, next_loop;
2348 sector_t metadata_block;
2349 unsigned metadata_offset;
2350 struct journal_io *io;
2352 if (journal_entry_is_unused(je))
2354 BUG_ON(unlikely(journal_entry_is_inprogress(je)) && !from_replay);
2355 sec = journal_entry_get_sector(je);
2356 if (unlikely(from_replay)) {
2357 if (unlikely(sec & (unsigned)(ic->sectors_per_block - 1))) {
2358 dm_integrity_io_error(ic, "invalid sector in journal", -EIO);
2359 sec &= ~(sector_t)(ic->sectors_per_block - 1);
2361 if (unlikely(sec >= ic->provided_data_sectors)) {
2362 journal_entry_set_unused(je);
2366 get_area_and_offset(ic, sec, &area, &offset);
2367 restore_last_bytes(ic, access_journal_data(ic, i, j), je);
2368 for (k = j + 1; k < ic->journal_section_entries; k++) {
2369 struct journal_entry *je2 = access_journal_entry(ic, i, k);
2370 sector_t sec2, area2, offset2;
2371 if (journal_entry_is_unused(je2))
2373 BUG_ON(unlikely(journal_entry_is_inprogress(je2)) && !from_replay);
2374 sec2 = journal_entry_get_sector(je2);
2375 if (unlikely(sec2 >= ic->provided_data_sectors))
2377 get_area_and_offset(ic, sec2, &area2, &offset2);
2378 if (area2 != area || offset2 != offset + ((k - j) << ic->sb->log2_sectors_per_block))
2380 restore_last_bytes(ic, access_journal_data(ic, i, k), je2);
2384 io = mempool_alloc(&ic->journal_io_mempool, GFP_NOIO);
2386 io->range.logical_sector = sec;
2387 io->range.n_sectors = (k - j) << ic->sb->log2_sectors_per_block;
2389 spin_lock_irq(&ic->endio_wait.lock);
2390 add_new_range_and_wait(ic, &io->range);
2392 if (likely(!from_replay)) {
2393 struct journal_node *section_node = &ic->journal_tree[i * ic->journal_section_entries];
2395 /* don't write if there is newer committed sector */
2396 while (j < k && find_newer_committed_node(ic, §ion_node[j])) {
2397 struct journal_entry *je2 = access_journal_entry(ic, i, j);
2399 journal_entry_set_unused(je2);
2400 remove_journal_node(ic, §ion_node[j]);
2402 sec += ic->sectors_per_block;
2403 offset += ic->sectors_per_block;
2405 while (j < k && find_newer_committed_node(ic, §ion_node[k - 1])) {
2406 struct journal_entry *je2 = access_journal_entry(ic, i, k - 1);
2408 journal_entry_set_unused(je2);
2409 remove_journal_node(ic, §ion_node[k - 1]);
2413 remove_range_unlocked(ic, &io->range);
2414 spin_unlock_irq(&ic->endio_wait.lock);
2415 mempool_free(io, &ic->journal_io_mempool);
2418 for (l = j; l < k; l++) {
2419 remove_journal_node(ic, §ion_node[l]);
2422 spin_unlock_irq(&ic->endio_wait.lock);
2424 metadata_block = get_metadata_sector_and_offset(ic, area, offset, &metadata_offset);
2425 for (l = j; l < k; l++) {
2427 struct journal_entry *je2 = access_journal_entry(ic, i, l);
2430 #ifndef INTERNAL_VERIFY
2431 unlikely(from_replay) &&
2433 ic->internal_hash) {
2434 char test_tag[max_t(size_t, HASH_MAX_DIGESTSIZE, MAX_TAG_SIZE)];
2436 integrity_sector_checksum(ic, sec + ((l - j) << ic->sb->log2_sectors_per_block),
2437 (char *)access_journal_data(ic, i, l), test_tag);
2438 if (unlikely(memcmp(test_tag, journal_entry_tag(ic, je2), ic->tag_size)))
2439 dm_integrity_io_error(ic, "tag mismatch when replaying journal", -EILSEQ);
2442 journal_entry_set_unused(je2);
2443 r = dm_integrity_rw_tag(ic, journal_entry_tag(ic, je2), &metadata_block, &metadata_offset,
2444 ic->tag_size, TAG_WRITE);
2446 dm_integrity_io_error(ic, "reading tags", r);
2450 atomic_inc(&comp.in_flight);
2451 copy_from_journal(ic, i, j << ic->sb->log2_sectors_per_block,
2452 (k - j) << ic->sb->log2_sectors_per_block,
2453 get_data_sector(ic, area, offset),
2454 complete_copy_from_journal, io);
2460 dm_bufio_write_dirty_buffers_async(ic->bufio);
2462 blk_finish_plug(&plug);
2464 complete_journal_op(&comp);
2465 wait_for_completion_io(&comp.comp);
2467 dm_integrity_flush_buffers(ic, true);
2470 static void integrity_writer(struct work_struct *w)
2472 struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, writer_work);
2473 unsigned write_start, write_sections;
2475 unsigned prev_free_sectors;
2477 spin_lock_irq(&ic->endio_wait.lock);
2478 write_start = ic->committed_section;
2479 write_sections = ic->n_committed_sections;
2480 spin_unlock_irq(&ic->endio_wait.lock);
2482 if (!write_sections)
2485 do_journal_write(ic, write_start, write_sections, false);
2487 spin_lock_irq(&ic->endio_wait.lock);
2489 ic->committed_section += write_sections;
2490 wraparound_section(ic, &ic->committed_section);
2491 ic->n_committed_sections -= write_sections;
2493 prev_free_sectors = ic->free_sectors;
2494 ic->free_sectors += write_sections * ic->journal_section_entries;
2495 if (unlikely(!prev_free_sectors))
2496 wake_up_locked(&ic->endio_wait);
2498 spin_unlock_irq(&ic->endio_wait.lock);
2501 static void recalc_write_super(struct dm_integrity_c *ic)
2505 dm_integrity_flush_buffers(ic, false);
2506 if (dm_integrity_failed(ic))
2509 r = sync_rw_sb(ic, REQ_OP_WRITE, 0);
2511 dm_integrity_io_error(ic, "writing superblock", r);
2514 static void integrity_recalc(struct work_struct *w)
2516 struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, recalc_work);
2517 struct dm_integrity_range range;
2518 struct dm_io_request io_req;
2519 struct dm_io_region io_loc;
2520 sector_t area, offset;
2521 sector_t metadata_block;
2522 unsigned metadata_offset;
2523 sector_t logical_sector, n_sectors;
2527 unsigned super_counter = 0;
2529 DEBUG_print("start recalculation... (position %llx)\n", le64_to_cpu(ic->sb->recalc_sector));
2531 spin_lock_irq(&ic->endio_wait.lock);
2535 if (unlikely(dm_post_suspending(ic->ti)))
2538 range.logical_sector = le64_to_cpu(ic->sb->recalc_sector);
2539 if (unlikely(range.logical_sector >= ic->provided_data_sectors)) {
2540 if (ic->mode == 'B') {
2541 block_bitmap_op(ic, ic->recalc_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_CLEAR);
2542 DEBUG_print("queue_delayed_work: bitmap_flush_work\n");
2543 queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, 0);
2548 get_area_and_offset(ic, range.logical_sector, &area, &offset);
2549 range.n_sectors = min((sector_t)RECALC_SECTORS, ic->provided_data_sectors - range.logical_sector);
2551 range.n_sectors = min(range.n_sectors, ((sector_t)1U << ic->sb->log2_interleave_sectors) - (unsigned)offset);
2553 add_new_range_and_wait(ic, &range);
2554 spin_unlock_irq(&ic->endio_wait.lock);
2555 logical_sector = range.logical_sector;
2556 n_sectors = range.n_sectors;
2558 if (ic->mode == 'B') {
2559 if (block_bitmap_op(ic, ic->recalc_bitmap, logical_sector, n_sectors, BITMAP_OP_TEST_ALL_CLEAR)) {
2560 goto advance_and_next;
2562 while (block_bitmap_op(ic, ic->recalc_bitmap, logical_sector,
2563 ic->sectors_per_block, BITMAP_OP_TEST_ALL_CLEAR)) {
2564 logical_sector += ic->sectors_per_block;
2565 n_sectors -= ic->sectors_per_block;
2568 while (block_bitmap_op(ic, ic->recalc_bitmap, logical_sector + n_sectors - ic->sectors_per_block,
2569 ic->sectors_per_block, BITMAP_OP_TEST_ALL_CLEAR)) {
2570 n_sectors -= ic->sectors_per_block;
2573 get_area_and_offset(ic, logical_sector, &area, &offset);
2576 DEBUG_print("recalculating: %llx, %llx\n", logical_sector, n_sectors);
2578 if (unlikely(++super_counter == RECALC_WRITE_SUPER)) {
2579 recalc_write_super(ic);
2580 if (ic->mode == 'B') {
2581 queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, ic->bitmap_flush_interval);
2586 if (unlikely(dm_integrity_failed(ic)))
2589 io_req.bi_op = REQ_OP_READ;
2590 io_req.bi_op_flags = 0;
2591 io_req.mem.type = DM_IO_VMA;
2592 io_req.mem.ptr.addr = ic->recalc_buffer;
2593 io_req.notify.fn = NULL;
2594 io_req.client = ic->io;
2595 io_loc.bdev = ic->dev->bdev;
2596 io_loc.sector = get_data_sector(ic, area, offset);
2597 io_loc.count = n_sectors;
2599 r = dm_io(&io_req, 1, &io_loc, NULL);
2601 dm_integrity_io_error(ic, "reading data", r);
2605 t = ic->recalc_tags;
2606 for (i = 0; i < n_sectors; i += ic->sectors_per_block) {
2607 integrity_sector_checksum(ic, logical_sector + i, ic->recalc_buffer + (i << SECTOR_SHIFT), t);
2611 metadata_block = get_metadata_sector_and_offset(ic, area, offset, &metadata_offset);
2613 r = dm_integrity_rw_tag(ic, ic->recalc_tags, &metadata_block, &metadata_offset, t - ic->recalc_tags, TAG_WRITE);
2615 dm_integrity_io_error(ic, "writing tags", r);
2619 if (ic->mode == 'B') {
2620 sector_t start, end;
2621 start = (range.logical_sector >>
2622 (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit)) <<
2623 (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
2624 end = ((range.logical_sector + range.n_sectors) >>
2625 (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit)) <<
2626 (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
2627 block_bitmap_op(ic, ic->recalc_bitmap, start, end - start, BITMAP_OP_CLEAR);
2633 spin_lock_irq(&ic->endio_wait.lock);
2634 remove_range_unlocked(ic, &range);
2635 ic->sb->recalc_sector = cpu_to_le64(range.logical_sector + range.n_sectors);
2639 remove_range(ic, &range);
2643 spin_unlock_irq(&ic->endio_wait.lock);
2645 recalc_write_super(ic);
2648 static void bitmap_block_work(struct work_struct *w)
2650 struct bitmap_block_status *bbs = container_of(w, struct bitmap_block_status, work);
2651 struct dm_integrity_c *ic = bbs->ic;
2653 struct bio_list bio_queue;
2654 struct bio_list waiting;
2656 bio_list_init(&waiting);
2658 spin_lock(&bbs->bio_queue_lock);
2659 bio_queue = bbs->bio_queue;
2660 bio_list_init(&bbs->bio_queue);
2661 spin_unlock(&bbs->bio_queue_lock);
2663 while ((bio = bio_list_pop(&bio_queue))) {
2664 struct dm_integrity_io *dio;
2666 dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
2668 if (block_bitmap_op(ic, ic->may_write_bitmap, dio->range.logical_sector,
2669 dio->range.n_sectors, BITMAP_OP_TEST_ALL_SET)) {
2670 remove_range(ic, &dio->range);
2671 INIT_WORK(&dio->work, integrity_bio_wait);
2672 queue_work(ic->offload_wq, &dio->work);
2674 block_bitmap_op(ic, ic->journal, dio->range.logical_sector,
2675 dio->range.n_sectors, BITMAP_OP_SET);
2676 bio_list_add(&waiting, bio);
2680 if (bio_list_empty(&waiting))
2683 rw_journal_sectors(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC,
2684 bbs->idx * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT),
2685 BITMAP_BLOCK_SIZE >> SECTOR_SHIFT, NULL);
2687 while ((bio = bio_list_pop(&waiting))) {
2688 struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
2690 block_bitmap_op(ic, ic->may_write_bitmap, dio->range.logical_sector,
2691 dio->range.n_sectors, BITMAP_OP_SET);
2693 remove_range(ic, &dio->range);
2694 INIT_WORK(&dio->work, integrity_bio_wait);
2695 queue_work(ic->offload_wq, &dio->work);
2698 queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, ic->bitmap_flush_interval);
2701 static void bitmap_flush_work(struct work_struct *work)
2703 struct dm_integrity_c *ic = container_of(work, struct dm_integrity_c, bitmap_flush_work.work);
2704 struct dm_integrity_range range;
2705 unsigned long limit;
2708 dm_integrity_flush_buffers(ic, false);
2710 range.logical_sector = 0;
2711 range.n_sectors = ic->provided_data_sectors;
2713 spin_lock_irq(&ic->endio_wait.lock);
2714 add_new_range_and_wait(ic, &range);
2715 spin_unlock_irq(&ic->endio_wait.lock);
2717 dm_integrity_flush_buffers(ic, true);
2719 limit = ic->provided_data_sectors;
2720 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
2721 limit = le64_to_cpu(ic->sb->recalc_sector)
2722 >> (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit)
2723 << (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
2725 /*DEBUG_print("zeroing journal\n");*/
2726 block_bitmap_op(ic, ic->journal, 0, limit, BITMAP_OP_CLEAR);
2727 block_bitmap_op(ic, ic->may_write_bitmap, 0, limit, BITMAP_OP_CLEAR);
2729 rw_journal_sectors(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, 0,
2730 ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
2732 spin_lock_irq(&ic->endio_wait.lock);
2733 remove_range_unlocked(ic, &range);
2734 while (unlikely((bio = bio_list_pop(&ic->synchronous_bios)) != NULL)) {
2736 spin_unlock_irq(&ic->endio_wait.lock);
2737 spin_lock_irq(&ic->endio_wait.lock);
2739 spin_unlock_irq(&ic->endio_wait.lock);
2743 static void init_journal(struct dm_integrity_c *ic, unsigned start_section,
2744 unsigned n_sections, unsigned char commit_seq)
2751 for (n = 0; n < n_sections; n++) {
2752 i = start_section + n;
2753 wraparound_section(ic, &i);
2754 for (j = 0; j < ic->journal_section_sectors; j++) {
2755 struct journal_sector *js = access_journal(ic, i, j);
2756 memset(&js->entries, 0, JOURNAL_SECTOR_DATA);
2757 js->commit_id = dm_integrity_commit_id(ic, i, j, commit_seq);
2759 for (j = 0; j < ic->journal_section_entries; j++) {
2760 struct journal_entry *je = access_journal_entry(ic, i, j);
2761 journal_entry_set_unused(je);
2765 write_journal(ic, start_section, n_sections);
2768 static int find_commit_seq(struct dm_integrity_c *ic, unsigned i, unsigned j, commit_id_t id)
2771 for (k = 0; k < N_COMMIT_IDS; k++) {
2772 if (dm_integrity_commit_id(ic, i, j, k) == id)
2775 dm_integrity_io_error(ic, "journal commit id", -EIO);
2779 static void replay_journal(struct dm_integrity_c *ic)
2782 bool used_commit_ids[N_COMMIT_IDS];
2783 unsigned max_commit_id_sections[N_COMMIT_IDS];
2784 unsigned write_start, write_sections;
2785 unsigned continue_section;
2787 unsigned char unused, last_used, want_commit_seq;
2789 if (ic->mode == 'R')
2792 if (ic->journal_uptodate)
2798 if (!ic->just_formatted) {
2799 DEBUG_print("reading journal\n");
2800 rw_journal(ic, REQ_OP_READ, 0, 0, ic->journal_sections, NULL);
2802 DEBUG_bytes(lowmem_page_address(ic->journal_io[0].page), 64, "read journal");
2803 if (ic->journal_io) {
2804 struct journal_completion crypt_comp;
2806 init_completion(&crypt_comp.comp);
2807 crypt_comp.in_flight = (atomic_t)ATOMIC_INIT(0);
2808 encrypt_journal(ic, false, 0, ic->journal_sections, &crypt_comp);
2809 wait_for_completion(&crypt_comp.comp);
2811 DEBUG_bytes(lowmem_page_address(ic->journal[0].page), 64, "decrypted journal");
2814 if (dm_integrity_failed(ic))
2817 journal_empty = true;
2818 memset(used_commit_ids, 0, sizeof used_commit_ids);
2819 memset(max_commit_id_sections, 0, sizeof max_commit_id_sections);
2820 for (i = 0; i < ic->journal_sections; i++) {
2821 for (j = 0; j < ic->journal_section_sectors; j++) {
2823 struct journal_sector *js = access_journal(ic, i, j);
2824 k = find_commit_seq(ic, i, j, js->commit_id);
2827 used_commit_ids[k] = true;
2828 max_commit_id_sections[k] = i;
2830 if (journal_empty) {
2831 for (j = 0; j < ic->journal_section_entries; j++) {
2832 struct journal_entry *je = access_journal_entry(ic, i, j);
2833 if (!journal_entry_is_unused(je)) {
2834 journal_empty = false;
2841 if (!used_commit_ids[N_COMMIT_IDS - 1]) {
2842 unused = N_COMMIT_IDS - 1;
2843 while (unused && !used_commit_ids[unused - 1])
2846 for (unused = 0; unused < N_COMMIT_IDS; unused++)
2847 if (!used_commit_ids[unused])
2849 if (unused == N_COMMIT_IDS) {
2850 dm_integrity_io_error(ic, "journal commit ids", -EIO);
2854 DEBUG_print("first unused commit seq %d [%d,%d,%d,%d]\n",
2855 unused, used_commit_ids[0], used_commit_ids[1],
2856 used_commit_ids[2], used_commit_ids[3]);
2858 last_used = prev_commit_seq(unused);
2859 want_commit_seq = prev_commit_seq(last_used);
2861 if (!used_commit_ids[want_commit_seq] && used_commit_ids[prev_commit_seq(want_commit_seq)])
2862 journal_empty = true;
2864 write_start = max_commit_id_sections[last_used] + 1;
2865 if (unlikely(write_start >= ic->journal_sections))
2866 want_commit_seq = next_commit_seq(want_commit_seq);
2867 wraparound_section(ic, &write_start);
2870 for (write_sections = 0; write_sections < ic->journal_sections; write_sections++) {
2871 for (j = 0; j < ic->journal_section_sectors; j++) {
2872 struct journal_sector *js = access_journal(ic, i, j);
2874 if (js->commit_id != dm_integrity_commit_id(ic, i, j, want_commit_seq)) {
2876 * This could be caused by crash during writing.
2877 * We won't replay the inconsistent part of the
2880 DEBUG_print("commit id mismatch at position (%u, %u): %d != %d\n",
2881 i, j, find_commit_seq(ic, i, j, js->commit_id), want_commit_seq);
2886 if (unlikely(i >= ic->journal_sections))
2887 want_commit_seq = next_commit_seq(want_commit_seq);
2888 wraparound_section(ic, &i);
2892 if (!journal_empty) {
2893 DEBUG_print("replaying %u sections, starting at %u, commit seq %d\n",
2894 write_sections, write_start, want_commit_seq);
2895 do_journal_write(ic, write_start, write_sections, true);
2898 if (write_sections == ic->journal_sections && (ic->mode == 'J' || journal_empty)) {
2899 continue_section = write_start;
2900 ic->commit_seq = want_commit_seq;
2901 DEBUG_print("continuing from section %u, commit seq %d\n", write_start, ic->commit_seq);
2904 unsigned char erase_seq;
2906 DEBUG_print("clearing journal\n");
2908 erase_seq = prev_commit_seq(prev_commit_seq(last_used));
2910 init_journal(ic, s, 1, erase_seq);
2912 wraparound_section(ic, &s);
2913 if (ic->journal_sections >= 2) {
2914 init_journal(ic, s, ic->journal_sections - 2, erase_seq);
2915 s += ic->journal_sections - 2;
2916 wraparound_section(ic, &s);
2917 init_journal(ic, s, 1, erase_seq);
2920 continue_section = 0;
2921 ic->commit_seq = next_commit_seq(erase_seq);
2924 ic->committed_section = continue_section;
2925 ic->n_committed_sections = 0;
2927 ic->uncommitted_section = continue_section;
2928 ic->n_uncommitted_sections = 0;
2930 ic->free_section = continue_section;
2931 ic->free_section_entry = 0;
2932 ic->free_sectors = ic->journal_entries;
2934 ic->journal_tree_root = RB_ROOT;
2935 for (i = 0; i < ic->journal_entries; i++)
2936 init_journal_node(&ic->journal_tree[i]);
2939 static void dm_integrity_enter_synchronous_mode(struct dm_integrity_c *ic)
2941 DEBUG_print("dm_integrity_enter_synchronous_mode\n");
2943 if (ic->mode == 'B') {
2944 ic->bitmap_flush_interval = msecs_to_jiffies(10) + 1;
2945 ic->synchronous_mode = 1;
2947 cancel_delayed_work_sync(&ic->bitmap_flush_work);
2948 queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, 0);
2949 flush_workqueue(ic->commit_wq);
2953 static int dm_integrity_reboot(struct notifier_block *n, unsigned long code, void *x)
2955 struct dm_integrity_c *ic = container_of(n, struct dm_integrity_c, reboot_notifier);
2957 DEBUG_print("dm_integrity_reboot\n");
2959 dm_integrity_enter_synchronous_mode(ic);
2964 static void dm_integrity_postsuspend(struct dm_target *ti)
2966 struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
2969 WARN_ON(unregister_reboot_notifier(&ic->reboot_notifier));
2971 del_timer_sync(&ic->autocommit_timer);
2974 drain_workqueue(ic->recalc_wq);
2976 if (ic->mode == 'B')
2977 cancel_delayed_work_sync(&ic->bitmap_flush_work);
2979 queue_work(ic->commit_wq, &ic->commit_work);
2980 drain_workqueue(ic->commit_wq);
2982 if (ic->mode == 'J') {
2983 queue_work(ic->writer_wq, &ic->writer_work);
2984 drain_workqueue(ic->writer_wq);
2985 dm_integrity_flush_buffers(ic, true);
2986 if (ic->wrote_to_journal) {
2987 init_journal(ic, ic->free_section,
2988 ic->journal_sections - ic->free_section, ic->commit_seq);
2989 if (ic->free_section) {
2990 init_journal(ic, 0, ic->free_section,
2991 next_commit_seq(ic->commit_seq));
2996 if (ic->mode == 'B') {
2997 dm_integrity_flush_buffers(ic, true);
2999 /* set to 0 to test bitmap replay code */
3000 init_journal(ic, 0, ic->journal_sections, 0);
3001 ic->sb->flags &= ~cpu_to_le32(SB_FLAG_DIRTY_BITMAP);
3002 r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
3004 dm_integrity_io_error(ic, "writing superblock", r);
3008 BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
3010 ic->journal_uptodate = true;
3013 static void dm_integrity_resume(struct dm_target *ti)
3015 struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
3016 __u64 old_provided_data_sectors = le64_to_cpu(ic->sb->provided_data_sectors);
3019 DEBUG_print("resume\n");
3021 ic->wrote_to_journal = false;
3023 if (ic->provided_data_sectors != old_provided_data_sectors) {
3024 if (ic->provided_data_sectors > old_provided_data_sectors &&
3026 ic->sb->log2_blocks_per_bitmap_bit == ic->log2_blocks_per_bitmap_bit) {
3027 rw_journal_sectors(ic, REQ_OP_READ, 0, 0,
3028 ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3029 block_bitmap_op(ic, ic->journal, old_provided_data_sectors,
3030 ic->provided_data_sectors - old_provided_data_sectors, BITMAP_OP_SET);
3031 rw_journal_sectors(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, 0,
3032 ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3035 ic->sb->provided_data_sectors = cpu_to_le64(ic->provided_data_sectors);
3036 r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
3038 dm_integrity_io_error(ic, "writing superblock", r);
3041 if (ic->sb->flags & cpu_to_le32(SB_FLAG_DIRTY_BITMAP)) {
3042 DEBUG_print("resume dirty_bitmap\n");
3043 rw_journal_sectors(ic, REQ_OP_READ, 0, 0,
3044 ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3045 if (ic->mode == 'B') {
3046 if (ic->sb->log2_blocks_per_bitmap_bit == ic->log2_blocks_per_bitmap_bit) {
3047 block_bitmap_copy(ic, ic->recalc_bitmap, ic->journal);
3048 block_bitmap_copy(ic, ic->may_write_bitmap, ic->journal);
3049 if (!block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors,
3050 BITMAP_OP_TEST_ALL_CLEAR)) {
3051 ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
3052 ic->sb->recalc_sector = cpu_to_le64(0);
3055 DEBUG_print("non-matching blocks_per_bitmap_bit: %u, %u\n",
3056 ic->sb->log2_blocks_per_bitmap_bit, ic->log2_blocks_per_bitmap_bit);
3057 ic->sb->log2_blocks_per_bitmap_bit = ic->log2_blocks_per_bitmap_bit;
3058 block_bitmap_op(ic, ic->recalc_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_SET);
3059 block_bitmap_op(ic, ic->may_write_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_SET);
3060 block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors, BITMAP_OP_SET);
3061 rw_journal_sectors(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, 0,
3062 ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3063 ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
3064 ic->sb->recalc_sector = cpu_to_le64(0);
3067 if (!(ic->sb->log2_blocks_per_bitmap_bit == ic->log2_blocks_per_bitmap_bit &&
3068 block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors, BITMAP_OP_TEST_ALL_CLEAR))) {
3069 ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
3070 ic->sb->recalc_sector = cpu_to_le64(0);
3072 init_journal(ic, 0, ic->journal_sections, 0);
3074 ic->sb->flags &= ~cpu_to_le32(SB_FLAG_DIRTY_BITMAP);
3076 r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
3078 dm_integrity_io_error(ic, "writing superblock", r);
3081 if (ic->mode == 'B') {
3082 ic->sb->flags |= cpu_to_le32(SB_FLAG_DIRTY_BITMAP);
3083 ic->sb->log2_blocks_per_bitmap_bit = ic->log2_blocks_per_bitmap_bit;
3084 r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
3086 dm_integrity_io_error(ic, "writing superblock", r);
3088 block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors, BITMAP_OP_CLEAR);
3089 block_bitmap_op(ic, ic->recalc_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_CLEAR);
3090 block_bitmap_op(ic, ic->may_write_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_CLEAR);
3091 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING) &&
3092 le64_to_cpu(ic->sb->recalc_sector) < ic->provided_data_sectors) {
3093 block_bitmap_op(ic, ic->journal, le64_to_cpu(ic->sb->recalc_sector),
3094 ic->provided_data_sectors - le64_to_cpu(ic->sb->recalc_sector), BITMAP_OP_SET);
3095 block_bitmap_op(ic, ic->recalc_bitmap, le64_to_cpu(ic->sb->recalc_sector),
3096 ic->provided_data_sectors - le64_to_cpu(ic->sb->recalc_sector), BITMAP_OP_SET);
3097 block_bitmap_op(ic, ic->may_write_bitmap, le64_to_cpu(ic->sb->recalc_sector),
3098 ic->provided_data_sectors - le64_to_cpu(ic->sb->recalc_sector), BITMAP_OP_SET);
3100 rw_journal_sectors(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, 0,
3101 ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3105 DEBUG_print("testing recalc: %x\n", ic->sb->flags);
3106 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
3107 __u64 recalc_pos = le64_to_cpu(ic->sb->recalc_sector);
3108 DEBUG_print("recalc pos: %llx / %llx\n", recalc_pos, ic->provided_data_sectors);
3109 if (recalc_pos < ic->provided_data_sectors) {
3110 queue_work(ic->recalc_wq, &ic->recalc_work);
3111 } else if (recalc_pos > ic->provided_data_sectors) {
3112 ic->sb->recalc_sector = cpu_to_le64(ic->provided_data_sectors);
3113 recalc_write_super(ic);
3117 ic->reboot_notifier.notifier_call = dm_integrity_reboot;
3118 ic->reboot_notifier.next = NULL;
3119 ic->reboot_notifier.priority = INT_MAX - 1; /* be notified after md and before hardware drivers */
3120 WARN_ON(register_reboot_notifier(&ic->reboot_notifier));
3123 /* set to 1 to stress test synchronous mode */
3124 dm_integrity_enter_synchronous_mode(ic);
3128 static void dm_integrity_status(struct dm_target *ti, status_type_t type,
3129 unsigned status_flags, char *result, unsigned maxlen)
3131 struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
3136 case STATUSTYPE_INFO:
3138 (unsigned long long)atomic64_read(&ic->number_of_mismatches),
3139 ic->provided_data_sectors);
3140 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
3141 DMEMIT(" %llu", le64_to_cpu(ic->sb->recalc_sector));
3146 case STATUSTYPE_TABLE: {
3147 __u64 watermark_percentage = (__u64)(ic->journal_entries - ic->free_sectors_threshold) * 100;
3148 watermark_percentage += ic->journal_entries / 2;
3149 do_div(watermark_percentage, ic->journal_entries);
3151 arg_count += !!ic->meta_dev;
3152 arg_count += ic->sectors_per_block != 1;
3153 arg_count += !!(ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING));
3154 arg_count += ic->discard;
3155 arg_count += ic->mode == 'J';
3156 arg_count += ic->mode == 'J';
3157 arg_count += ic->mode == 'B';
3158 arg_count += ic->mode == 'B';
3159 arg_count += !!ic->internal_hash_alg.alg_string;
3160 arg_count += !!ic->journal_crypt_alg.alg_string;
3161 arg_count += !!ic->journal_mac_alg.alg_string;
3162 arg_count += (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING)) != 0;
3163 arg_count += ic->legacy_recalculate;
3164 DMEMIT("%s %llu %u %c %u", ic->dev->name, ic->start,
3165 ic->tag_size, ic->mode, arg_count);
3167 DMEMIT(" meta_device:%s", ic->meta_dev->name);
3168 if (ic->sectors_per_block != 1)
3169 DMEMIT(" block_size:%u", ic->sectors_per_block << SECTOR_SHIFT);
3170 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
3171 DMEMIT(" recalculate");
3173 DMEMIT(" allow_discards");
3174 DMEMIT(" journal_sectors:%u", ic->initial_sectors - SB_SECTORS);
3175 DMEMIT(" interleave_sectors:%u", 1U << ic->sb->log2_interleave_sectors);
3176 DMEMIT(" buffer_sectors:%u", 1U << ic->log2_buffer_sectors);
3177 if (ic->mode == 'J') {
3178 DMEMIT(" journal_watermark:%u", (unsigned)watermark_percentage);
3179 DMEMIT(" commit_time:%u", ic->autocommit_msec);
3181 if (ic->mode == 'B') {
3182 DMEMIT(" sectors_per_bit:%llu", (sector_t)ic->sectors_per_block << ic->log2_blocks_per_bitmap_bit);
3183 DMEMIT(" bitmap_flush_interval:%u", jiffies_to_msecs(ic->bitmap_flush_interval));
3185 if ((ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING)) != 0)
3186 DMEMIT(" fix_padding");
3187 if (ic->legacy_recalculate)
3188 DMEMIT(" legacy_recalculate");
3190 #define EMIT_ALG(a, n) \
3192 if (ic->a.alg_string) { \
3193 DMEMIT(" %s:%s", n, ic->a.alg_string); \
3194 if (ic->a.key_string) \
3195 DMEMIT(":%s", ic->a.key_string);\
3198 EMIT_ALG(internal_hash_alg, "internal_hash");
3199 EMIT_ALG(journal_crypt_alg, "journal_crypt");
3200 EMIT_ALG(journal_mac_alg, "journal_mac");
3206 static int dm_integrity_iterate_devices(struct dm_target *ti,
3207 iterate_devices_callout_fn fn, void *data)
3209 struct dm_integrity_c *ic = ti->private;
3212 return fn(ti, ic->dev, ic->start + ic->initial_sectors + ic->metadata_run, ti->len, data);
3214 return fn(ti, ic->dev, 0, ti->len, data);
3217 static void dm_integrity_io_hints(struct dm_target *ti, struct queue_limits *limits)
3219 struct dm_integrity_c *ic = ti->private;
3221 if (ic->sectors_per_block > 1) {
3222 limits->logical_block_size = ic->sectors_per_block << SECTOR_SHIFT;
3223 limits->physical_block_size = ic->sectors_per_block << SECTOR_SHIFT;
3224 blk_limits_io_min(limits, ic->sectors_per_block << SECTOR_SHIFT);
3228 static void calculate_journal_section_size(struct dm_integrity_c *ic)
3230 unsigned sector_space = JOURNAL_SECTOR_DATA;
3232 ic->journal_sections = le32_to_cpu(ic->sb->journal_sections);
3233 ic->journal_entry_size = roundup(offsetof(struct journal_entry, last_bytes[ic->sectors_per_block]) + ic->tag_size,
3234 JOURNAL_ENTRY_ROUNDUP);
3236 if (ic->sb->flags & cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC))
3237 sector_space -= JOURNAL_MAC_PER_SECTOR;
3238 ic->journal_entries_per_sector = sector_space / ic->journal_entry_size;
3239 ic->journal_section_entries = ic->journal_entries_per_sector * JOURNAL_BLOCK_SECTORS;
3240 ic->journal_section_sectors = (ic->journal_section_entries << ic->sb->log2_sectors_per_block) + JOURNAL_BLOCK_SECTORS;
3241 ic->journal_entries = ic->journal_section_entries * ic->journal_sections;
3244 static int calculate_device_limits(struct dm_integrity_c *ic)
3246 __u64 initial_sectors;
3248 calculate_journal_section_size(ic);
3249 initial_sectors = SB_SECTORS + (__u64)ic->journal_section_sectors * ic->journal_sections;
3250 if (initial_sectors + METADATA_PADDING_SECTORS >= ic->meta_device_sectors || initial_sectors > UINT_MAX)
3252 ic->initial_sectors = initial_sectors;
3254 if (!ic->meta_dev) {
3255 sector_t last_sector, last_area, last_offset;
3257 /* we have to maintain excessive padding for compatibility with existing volumes */
3258 __u64 metadata_run_padding =
3259 ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING) ?
3260 (__u64)(METADATA_PADDING_SECTORS << SECTOR_SHIFT) :
3261 (__u64)(1 << SECTOR_SHIFT << METADATA_PADDING_SECTORS);
3263 ic->metadata_run = round_up((__u64)ic->tag_size << (ic->sb->log2_interleave_sectors - ic->sb->log2_sectors_per_block),
3264 metadata_run_padding) >> SECTOR_SHIFT;
3265 if (!(ic->metadata_run & (ic->metadata_run - 1)))
3266 ic->log2_metadata_run = __ffs(ic->metadata_run);
3268 ic->log2_metadata_run = -1;
3270 get_area_and_offset(ic, ic->provided_data_sectors - 1, &last_area, &last_offset);
3271 last_sector = get_data_sector(ic, last_area, last_offset);
3272 if (last_sector < ic->start || last_sector >= ic->meta_device_sectors)
3275 __u64 meta_size = (ic->provided_data_sectors >> ic->sb->log2_sectors_per_block) * ic->tag_size;
3276 meta_size = (meta_size + ((1U << (ic->log2_buffer_sectors + SECTOR_SHIFT)) - 1))
3277 >> (ic->log2_buffer_sectors + SECTOR_SHIFT);
3278 meta_size <<= ic->log2_buffer_sectors;
3279 if (ic->initial_sectors + meta_size < ic->initial_sectors ||
3280 ic->initial_sectors + meta_size > ic->meta_device_sectors)
3282 ic->metadata_run = 1;
3283 ic->log2_metadata_run = 0;
3289 static void get_provided_data_sectors(struct dm_integrity_c *ic)
3291 if (!ic->meta_dev) {
3293 ic->provided_data_sectors = 0;
3294 for (test_bit = fls64(ic->meta_device_sectors) - 1; test_bit >= 3; test_bit--) {
3295 __u64 prev_data_sectors = ic->provided_data_sectors;
3297 ic->provided_data_sectors |= (sector_t)1 << test_bit;
3298 if (calculate_device_limits(ic))
3299 ic->provided_data_sectors = prev_data_sectors;
3302 ic->provided_data_sectors = ic->data_device_sectors;
3303 ic->provided_data_sectors &= ~(sector_t)(ic->sectors_per_block - 1);
3307 static int initialize_superblock(struct dm_integrity_c *ic, unsigned journal_sectors, unsigned interleave_sectors)
3309 unsigned journal_sections;
3312 memset(ic->sb, 0, SB_SECTORS << SECTOR_SHIFT);
3313 memcpy(ic->sb->magic, SB_MAGIC, 8);
3314 ic->sb->integrity_tag_size = cpu_to_le16(ic->tag_size);
3315 ic->sb->log2_sectors_per_block = __ffs(ic->sectors_per_block);
3316 if (ic->journal_mac_alg.alg_string)
3317 ic->sb->flags |= cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC);
3319 calculate_journal_section_size(ic);
3320 journal_sections = journal_sectors / ic->journal_section_sectors;
3321 if (!journal_sections)
3322 journal_sections = 1;
3324 if (!ic->meta_dev) {
3325 if (ic->fix_padding)
3326 ic->sb->flags |= cpu_to_le32(SB_FLAG_FIXED_PADDING);
3327 ic->sb->journal_sections = cpu_to_le32(journal_sections);
3328 if (!interleave_sectors)
3329 interleave_sectors = DEFAULT_INTERLEAVE_SECTORS;
3330 ic->sb->log2_interleave_sectors = __fls(interleave_sectors);
3331 ic->sb->log2_interleave_sectors = max((__u8)MIN_LOG2_INTERLEAVE_SECTORS, ic->sb->log2_interleave_sectors);
3332 ic->sb->log2_interleave_sectors = min((__u8)MAX_LOG2_INTERLEAVE_SECTORS, ic->sb->log2_interleave_sectors);
3334 get_provided_data_sectors(ic);
3335 if (!ic->provided_data_sectors)
3338 ic->sb->log2_interleave_sectors = 0;
3340 get_provided_data_sectors(ic);
3341 if (!ic->provided_data_sectors)
3345 ic->sb->journal_sections = cpu_to_le32(0);
3346 for (test_bit = fls(journal_sections) - 1; test_bit >= 0; test_bit--) {
3347 __u32 prev_journal_sections = le32_to_cpu(ic->sb->journal_sections);
3348 __u32 test_journal_sections = prev_journal_sections | (1U << test_bit);
3349 if (test_journal_sections > journal_sections)
3351 ic->sb->journal_sections = cpu_to_le32(test_journal_sections);
3352 if (calculate_device_limits(ic))
3353 ic->sb->journal_sections = cpu_to_le32(prev_journal_sections);
3356 if (!le32_to_cpu(ic->sb->journal_sections)) {
3357 if (ic->log2_buffer_sectors > 3) {
3358 ic->log2_buffer_sectors--;
3359 goto try_smaller_buffer;
3365 ic->sb->provided_data_sectors = cpu_to_le64(ic->provided_data_sectors);
3372 static void dm_integrity_set(struct dm_target *ti, struct dm_integrity_c *ic)
3374 struct gendisk *disk = dm_disk(dm_table_get_md(ti->table));
3375 struct blk_integrity bi;
3377 memset(&bi, 0, sizeof(bi));
3378 bi.profile = &dm_integrity_profile;
3379 bi.tuple_size = ic->tag_size;
3380 bi.tag_size = bi.tuple_size;
3381 bi.interval_exp = ic->sb->log2_sectors_per_block + SECTOR_SHIFT;
3383 blk_integrity_register(disk, &bi);
3384 blk_queue_max_integrity_segments(disk->queue, UINT_MAX);
3387 static void dm_integrity_free_page_list(struct page_list *pl)
3393 for (i = 0; pl[i].page; i++)
3394 __free_page(pl[i].page);
3398 static struct page_list *dm_integrity_alloc_page_list(unsigned n_pages)
3400 struct page_list *pl;
3403 pl = kvmalloc_array(n_pages + 1, sizeof(struct page_list), GFP_KERNEL | __GFP_ZERO);
3407 for (i = 0; i < n_pages; i++) {
3408 pl[i].page = alloc_page(GFP_KERNEL);
3410 dm_integrity_free_page_list(pl);
3414 pl[i - 1].next = &pl[i];
3422 static void dm_integrity_free_journal_scatterlist(struct dm_integrity_c *ic, struct scatterlist **sl)
3425 for (i = 0; i < ic->journal_sections; i++)
3430 static struct scatterlist **dm_integrity_alloc_journal_scatterlist(struct dm_integrity_c *ic,
3431 struct page_list *pl)
3433 struct scatterlist **sl;
3436 sl = kvmalloc_array(ic->journal_sections,
3437 sizeof(struct scatterlist *),
3438 GFP_KERNEL | __GFP_ZERO);
3442 for (i = 0; i < ic->journal_sections; i++) {
3443 struct scatterlist *s;
3444 unsigned start_index, start_offset;
3445 unsigned end_index, end_offset;
3449 page_list_location(ic, i, 0, &start_index, &start_offset);
3450 page_list_location(ic, i, ic->journal_section_sectors - 1,
3451 &end_index, &end_offset);
3453 n_pages = (end_index - start_index + 1);
3455 s = kvmalloc_array(n_pages, sizeof(struct scatterlist),
3458 dm_integrity_free_journal_scatterlist(ic, sl);
3462 sg_init_table(s, n_pages);
3463 for (idx = start_index; idx <= end_index; idx++) {
3464 char *va = lowmem_page_address(pl[idx].page);
3465 unsigned start = 0, end = PAGE_SIZE;
3466 if (idx == start_index)
3467 start = start_offset;
3468 if (idx == end_index)
3469 end = end_offset + (1 << SECTOR_SHIFT);
3470 sg_set_buf(&s[idx - start_index], va + start, end - start);
3479 static void free_alg(struct alg_spec *a)
3481 kfree_sensitive(a->alg_string);
3482 kfree_sensitive(a->key);
3483 memset(a, 0, sizeof *a);
3486 static int get_alg_and_key(const char *arg, struct alg_spec *a, char **error, char *error_inval)
3492 a->alg_string = kstrdup(strchr(arg, ':') + 1, GFP_KERNEL);
3496 k = strchr(a->alg_string, ':');
3499 a->key_string = k + 1;
3500 if (strlen(a->key_string) & 1)
3503 a->key_size = strlen(a->key_string) / 2;
3504 a->key = kmalloc(a->key_size, GFP_KERNEL);
3507 if (hex2bin(a->key, a->key_string, a->key_size))
3513 *error = error_inval;
3516 *error = "Out of memory for an argument";
3520 static int get_mac(struct crypto_shash **hash, struct alg_spec *a, char **error,
3521 char *error_alg, char *error_key)
3525 if (a->alg_string) {
3526 *hash = crypto_alloc_shash(a->alg_string, 0, CRYPTO_ALG_ALLOCATES_MEMORY);
3527 if (IS_ERR(*hash)) {
3535 r = crypto_shash_setkey(*hash, a->key, a->key_size);
3540 } else if (crypto_shash_get_flags(*hash) & CRYPTO_TFM_NEED_KEY) {
3549 static int create_journal(struct dm_integrity_c *ic, char **error)
3553 __u64 journal_pages, journal_desc_size, journal_tree_size;
3554 unsigned char *crypt_data = NULL, *crypt_iv = NULL;
3555 struct skcipher_request *req = NULL;
3557 ic->commit_ids[0] = cpu_to_le64(0x1111111111111111ULL);
3558 ic->commit_ids[1] = cpu_to_le64(0x2222222222222222ULL);
3559 ic->commit_ids[2] = cpu_to_le64(0x3333333333333333ULL);
3560 ic->commit_ids[3] = cpu_to_le64(0x4444444444444444ULL);
3562 journal_pages = roundup((__u64)ic->journal_sections * ic->journal_section_sectors,
3563 PAGE_SIZE >> SECTOR_SHIFT) >> (PAGE_SHIFT - SECTOR_SHIFT);
3564 journal_desc_size = journal_pages * sizeof(struct page_list);
3565 if (journal_pages >= totalram_pages() - totalhigh_pages() || journal_desc_size > ULONG_MAX) {
3566 *error = "Journal doesn't fit into memory";
3570 ic->journal_pages = journal_pages;
3572 ic->journal = dm_integrity_alloc_page_list(ic->journal_pages);
3574 *error = "Could not allocate memory for journal";
3578 if (ic->journal_crypt_alg.alg_string) {
3579 unsigned ivsize, blocksize;
3580 struct journal_completion comp;
3583 ic->journal_crypt = crypto_alloc_skcipher(ic->journal_crypt_alg.alg_string, 0, CRYPTO_ALG_ALLOCATES_MEMORY);
3584 if (IS_ERR(ic->journal_crypt)) {
3585 *error = "Invalid journal cipher";
3586 r = PTR_ERR(ic->journal_crypt);
3587 ic->journal_crypt = NULL;
3590 ivsize = crypto_skcipher_ivsize(ic->journal_crypt);
3591 blocksize = crypto_skcipher_blocksize(ic->journal_crypt);
3593 if (ic->journal_crypt_alg.key) {
3594 r = crypto_skcipher_setkey(ic->journal_crypt, ic->journal_crypt_alg.key,
3595 ic->journal_crypt_alg.key_size);
3597 *error = "Error setting encryption key";
3601 DEBUG_print("cipher %s, block size %u iv size %u\n",
3602 ic->journal_crypt_alg.alg_string, blocksize, ivsize);
3604 ic->journal_io = dm_integrity_alloc_page_list(ic->journal_pages);
3605 if (!ic->journal_io) {
3606 *error = "Could not allocate memory for journal io";
3611 if (blocksize == 1) {
3612 struct scatterlist *sg;
3614 req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
3616 *error = "Could not allocate crypt request";
3621 crypt_iv = kzalloc(ivsize, GFP_KERNEL);
3623 *error = "Could not allocate iv";
3628 ic->journal_xor = dm_integrity_alloc_page_list(ic->journal_pages);
3629 if (!ic->journal_xor) {
3630 *error = "Could not allocate memory for journal xor";
3635 sg = kvmalloc_array(ic->journal_pages + 1,
3636 sizeof(struct scatterlist),
3639 *error = "Unable to allocate sg list";
3643 sg_init_table(sg, ic->journal_pages + 1);
3644 for (i = 0; i < ic->journal_pages; i++) {
3645 char *va = lowmem_page_address(ic->journal_xor[i].page);
3647 sg_set_buf(&sg[i], va, PAGE_SIZE);
3649 sg_set_buf(&sg[i], &ic->commit_ids, sizeof ic->commit_ids);
3651 skcipher_request_set_crypt(req, sg, sg,
3652 PAGE_SIZE * ic->journal_pages + sizeof ic->commit_ids, crypt_iv);
3653 init_completion(&comp.comp);
3654 comp.in_flight = (atomic_t)ATOMIC_INIT(1);
3655 if (do_crypt(true, req, &comp))
3656 wait_for_completion(&comp.comp);
3658 r = dm_integrity_failed(ic);
3660 *error = "Unable to encrypt journal";
3663 DEBUG_bytes(lowmem_page_address(ic->journal_xor[0].page), 64, "xor data");
3665 crypto_free_skcipher(ic->journal_crypt);
3666 ic->journal_crypt = NULL;
3668 unsigned crypt_len = roundup(ivsize, blocksize);
3670 req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
3672 *error = "Could not allocate crypt request";
3677 crypt_iv = kmalloc(ivsize, GFP_KERNEL);
3679 *error = "Could not allocate iv";
3684 crypt_data = kmalloc(crypt_len, GFP_KERNEL);
3686 *error = "Unable to allocate crypt data";
3691 ic->journal_scatterlist = dm_integrity_alloc_journal_scatterlist(ic, ic->journal);
3692 if (!ic->journal_scatterlist) {
3693 *error = "Unable to allocate sg list";
3697 ic->journal_io_scatterlist = dm_integrity_alloc_journal_scatterlist(ic, ic->journal_io);
3698 if (!ic->journal_io_scatterlist) {
3699 *error = "Unable to allocate sg list";
3703 ic->sk_requests = kvmalloc_array(ic->journal_sections,
3704 sizeof(struct skcipher_request *),
3705 GFP_KERNEL | __GFP_ZERO);
3706 if (!ic->sk_requests) {
3707 *error = "Unable to allocate sk requests";
3711 for (i = 0; i < ic->journal_sections; i++) {
3712 struct scatterlist sg;
3713 struct skcipher_request *section_req;
3714 __u32 section_le = cpu_to_le32(i);
3716 memset(crypt_iv, 0x00, ivsize);
3717 memset(crypt_data, 0x00, crypt_len);
3718 memcpy(crypt_data, §ion_le, min((size_t)crypt_len, sizeof(section_le)));
3720 sg_init_one(&sg, crypt_data, crypt_len);
3721 skcipher_request_set_crypt(req, &sg, &sg, crypt_len, crypt_iv);
3722 init_completion(&comp.comp);
3723 comp.in_flight = (atomic_t)ATOMIC_INIT(1);
3724 if (do_crypt(true, req, &comp))
3725 wait_for_completion(&comp.comp);
3727 r = dm_integrity_failed(ic);
3729 *error = "Unable to generate iv";
3733 section_req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
3735 *error = "Unable to allocate crypt request";
3739 section_req->iv = kmalloc_array(ivsize, 2,
3741 if (!section_req->iv) {
3742 skcipher_request_free(section_req);
3743 *error = "Unable to allocate iv";
3747 memcpy(section_req->iv + ivsize, crypt_data, ivsize);
3748 section_req->cryptlen = (size_t)ic->journal_section_sectors << SECTOR_SHIFT;
3749 ic->sk_requests[i] = section_req;
3750 DEBUG_bytes(crypt_data, ivsize, "iv(%u)", i);
3755 for (i = 0; i < N_COMMIT_IDS; i++) {
3758 for (j = 0; j < i; j++) {
3759 if (ic->commit_ids[j] == ic->commit_ids[i]) {
3760 ic->commit_ids[i] = cpu_to_le64(le64_to_cpu(ic->commit_ids[i]) + 1);
3761 goto retest_commit_id;
3764 DEBUG_print("commit id %u: %016llx\n", i, ic->commit_ids[i]);
3767 journal_tree_size = (__u64)ic->journal_entries * sizeof(struct journal_node);
3768 if (journal_tree_size > ULONG_MAX) {
3769 *error = "Journal doesn't fit into memory";
3773 ic->journal_tree = kvmalloc(journal_tree_size, GFP_KERNEL);
3774 if (!ic->journal_tree) {
3775 *error = "Could not allocate memory for journal tree";
3781 skcipher_request_free(req);
3787 * Construct a integrity mapping
3791 * offset from the start of the device
3793 * D - direct writes, J - journal writes, B - bitmap mode, R - recovery mode
3794 * number of optional arguments
3795 * optional arguments:
3797 * interleave_sectors
3804 * bitmap_flush_interval
3810 static int dm_integrity_ctr(struct dm_target *ti, unsigned argc, char **argv)
3812 struct dm_integrity_c *ic;
3815 unsigned extra_args;
3816 struct dm_arg_set as;
3817 static const struct dm_arg _args[] = {
3818 {0, 16, "Invalid number of feature args"},
3820 unsigned journal_sectors, interleave_sectors, buffer_sectors, journal_watermark, sync_msec;
3821 bool should_write_sb;
3823 unsigned long long start;
3824 __s8 log2_sectors_per_bitmap_bit = -1;
3825 __s8 log2_blocks_per_bitmap_bit;
3826 __u64 bits_in_journal;
3827 __u64 n_bitmap_bits;
3829 #define DIRECT_ARGUMENTS 4
3831 if (argc <= DIRECT_ARGUMENTS) {
3832 ti->error = "Invalid argument count";
3836 ic = kzalloc(sizeof(struct dm_integrity_c), GFP_KERNEL);
3838 ti->error = "Cannot allocate integrity context";
3842 ti->per_io_data_size = sizeof(struct dm_integrity_io);
3845 ic->in_progress = RB_ROOT;
3846 INIT_LIST_HEAD(&ic->wait_list);
3847 init_waitqueue_head(&ic->endio_wait);
3848 bio_list_init(&ic->flush_bio_list);
3849 init_waitqueue_head(&ic->copy_to_journal_wait);
3850 init_completion(&ic->crypto_backoff);
3851 atomic64_set(&ic->number_of_mismatches, 0);
3852 ic->bitmap_flush_interval = BITMAP_FLUSH_INTERVAL;
3854 r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &ic->dev);
3856 ti->error = "Device lookup failed";
3860 if (sscanf(argv[1], "%llu%c", &start, &dummy) != 1 || start != (sector_t)start) {
3861 ti->error = "Invalid starting offset";
3867 if (strcmp(argv[2], "-")) {
3868 if (sscanf(argv[2], "%u%c", &ic->tag_size, &dummy) != 1 || !ic->tag_size) {
3869 ti->error = "Invalid tag size";
3875 if (!strcmp(argv[3], "J") || !strcmp(argv[3], "B") ||
3876 !strcmp(argv[3], "D") || !strcmp(argv[3], "R")) {
3877 ic->mode = argv[3][0];
3879 ti->error = "Invalid mode (expecting J, B, D, R)";
3884 journal_sectors = 0;
3885 interleave_sectors = DEFAULT_INTERLEAVE_SECTORS;
3886 buffer_sectors = DEFAULT_BUFFER_SECTORS;
3887 journal_watermark = DEFAULT_JOURNAL_WATERMARK;
3888 sync_msec = DEFAULT_SYNC_MSEC;
3889 ic->sectors_per_block = 1;
3891 as.argc = argc - DIRECT_ARGUMENTS;
3892 as.argv = argv + DIRECT_ARGUMENTS;
3893 r = dm_read_arg_group(_args, &as, &extra_args, &ti->error);
3897 while (extra_args--) {
3898 const char *opt_string;
3900 unsigned long long llval;
3901 opt_string = dm_shift_arg(&as);
3904 ti->error = "Not enough feature arguments";
3907 if (sscanf(opt_string, "journal_sectors:%u%c", &val, &dummy) == 1)
3908 journal_sectors = val ? val : 1;
3909 else if (sscanf(opt_string, "interleave_sectors:%u%c", &val, &dummy) == 1)
3910 interleave_sectors = val;
3911 else if (sscanf(opt_string, "buffer_sectors:%u%c", &val, &dummy) == 1)
3912 buffer_sectors = val;
3913 else if (sscanf(opt_string, "journal_watermark:%u%c", &val, &dummy) == 1 && val <= 100)
3914 journal_watermark = val;
3915 else if (sscanf(opt_string, "commit_time:%u%c", &val, &dummy) == 1)
3917 else if (!strncmp(opt_string, "meta_device:", strlen("meta_device:"))) {
3919 dm_put_device(ti, ic->meta_dev);
3920 ic->meta_dev = NULL;
3922 r = dm_get_device(ti, strchr(opt_string, ':') + 1,
3923 dm_table_get_mode(ti->table), &ic->meta_dev);
3925 ti->error = "Device lookup failed";
3928 } else if (sscanf(opt_string, "block_size:%u%c", &val, &dummy) == 1) {
3929 if (val < 1 << SECTOR_SHIFT ||
3930 val > MAX_SECTORS_PER_BLOCK << SECTOR_SHIFT ||
3933 ti->error = "Invalid block_size argument";
3936 ic->sectors_per_block = val >> SECTOR_SHIFT;
3937 } else if (sscanf(opt_string, "sectors_per_bit:%llu%c", &llval, &dummy) == 1) {
3938 log2_sectors_per_bitmap_bit = !llval ? 0 : __ilog2_u64(llval);
3939 } else if (sscanf(opt_string, "bitmap_flush_interval:%u%c", &val, &dummy) == 1) {
3940 if ((uint64_t)val >= (uint64_t)UINT_MAX * 1000 / HZ) {
3942 ti->error = "Invalid bitmap_flush_interval argument";
3945 ic->bitmap_flush_interval = msecs_to_jiffies(val);
3946 } else if (!strncmp(opt_string, "internal_hash:", strlen("internal_hash:"))) {
3947 r = get_alg_and_key(opt_string, &ic->internal_hash_alg, &ti->error,
3948 "Invalid internal_hash argument");
3951 } else if (!strncmp(opt_string, "journal_crypt:", strlen("journal_crypt:"))) {
3952 r = get_alg_and_key(opt_string, &ic->journal_crypt_alg, &ti->error,
3953 "Invalid journal_crypt argument");
3956 } else if (!strncmp(opt_string, "journal_mac:", strlen("journal_mac:"))) {
3957 r = get_alg_and_key(opt_string, &ic->journal_mac_alg, &ti->error,
3958 "Invalid journal_mac argument");
3961 } else if (!strcmp(opt_string, "recalculate")) {
3962 ic->recalculate_flag = true;
3963 } else if (!strcmp(opt_string, "allow_discards")) {
3965 } else if (!strcmp(opt_string, "fix_padding")) {
3966 ic->fix_padding = true;
3967 } else if (!strcmp(opt_string, "legacy_recalculate")) {
3968 ic->legacy_recalculate = true;
3971 ti->error = "Invalid argument";
3976 ic->data_device_sectors = i_size_read(ic->dev->bdev->bd_inode) >> SECTOR_SHIFT;
3978 ic->meta_device_sectors = ic->data_device_sectors;
3980 ic->meta_device_sectors = i_size_read(ic->meta_dev->bdev->bd_inode) >> SECTOR_SHIFT;
3982 if (!journal_sectors) {
3983 journal_sectors = min((sector_t)DEFAULT_MAX_JOURNAL_SECTORS,
3984 ic->data_device_sectors >> DEFAULT_JOURNAL_SIZE_FACTOR);
3987 if (!buffer_sectors)
3989 ic->log2_buffer_sectors = min((int)__fls(buffer_sectors), 31 - SECTOR_SHIFT);
3991 r = get_mac(&ic->internal_hash, &ic->internal_hash_alg, &ti->error,
3992 "Invalid internal hash", "Error setting internal hash key");
3996 r = get_mac(&ic->journal_mac, &ic->journal_mac_alg, &ti->error,
3997 "Invalid journal mac", "Error setting journal mac key");
4001 if (!ic->tag_size) {
4002 if (!ic->internal_hash) {
4003 ti->error = "Unknown tag size";
4007 ic->tag_size = crypto_shash_digestsize(ic->internal_hash);
4009 if (ic->tag_size > MAX_TAG_SIZE) {
4010 ti->error = "Too big tag size";
4014 if (!(ic->tag_size & (ic->tag_size - 1)))
4015 ic->log2_tag_size = __ffs(ic->tag_size);
4017 ic->log2_tag_size = -1;
4019 if (ic->mode == 'B' && !ic->internal_hash) {
4021 ti->error = "Bitmap mode can be only used with internal hash";
4025 if (ic->discard && !ic->internal_hash) {
4027 ti->error = "Discard can be only used with internal hash";
4031 ic->autocommit_jiffies = msecs_to_jiffies(sync_msec);
4032 ic->autocommit_msec = sync_msec;
4033 timer_setup(&ic->autocommit_timer, autocommit_fn, 0);
4035 ic->io = dm_io_client_create();
4036 if (IS_ERR(ic->io)) {
4037 r = PTR_ERR(ic->io);
4039 ti->error = "Cannot allocate dm io";
4043 r = mempool_init_slab_pool(&ic->journal_io_mempool, JOURNAL_IO_MEMPOOL, journal_io_cache);
4045 ti->error = "Cannot allocate mempool";
4049 ic->metadata_wq = alloc_workqueue("dm-integrity-metadata",
4050 WQ_MEM_RECLAIM, METADATA_WORKQUEUE_MAX_ACTIVE);
4051 if (!ic->metadata_wq) {
4052 ti->error = "Cannot allocate workqueue";
4058 * If this workqueue were percpu, it would cause bio reordering
4059 * and reduced performance.
4061 ic->wait_wq = alloc_workqueue("dm-integrity-wait", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
4063 ti->error = "Cannot allocate workqueue";
4068 ic->offload_wq = alloc_workqueue("dm-integrity-offload", WQ_MEM_RECLAIM,
4069 METADATA_WORKQUEUE_MAX_ACTIVE);
4070 if (!ic->offload_wq) {
4071 ti->error = "Cannot allocate workqueue";
4076 ic->commit_wq = alloc_workqueue("dm-integrity-commit", WQ_MEM_RECLAIM, 1);
4077 if (!ic->commit_wq) {
4078 ti->error = "Cannot allocate workqueue";
4082 INIT_WORK(&ic->commit_work, integrity_commit);
4084 if (ic->mode == 'J' || ic->mode == 'B') {
4085 ic->writer_wq = alloc_workqueue("dm-integrity-writer", WQ_MEM_RECLAIM, 1);
4086 if (!ic->writer_wq) {
4087 ti->error = "Cannot allocate workqueue";
4091 INIT_WORK(&ic->writer_work, integrity_writer);
4094 ic->sb = alloc_pages_exact(SB_SECTORS << SECTOR_SHIFT, GFP_KERNEL);
4097 ti->error = "Cannot allocate superblock area";
4101 r = sync_rw_sb(ic, REQ_OP_READ, 0);
4103 ti->error = "Error reading superblock";
4106 should_write_sb = false;
4107 if (memcmp(ic->sb->magic, SB_MAGIC, 8)) {
4108 if (ic->mode != 'R') {
4109 if (memchr_inv(ic->sb, 0, SB_SECTORS << SECTOR_SHIFT)) {
4111 ti->error = "The device is not initialized";
4116 r = initialize_superblock(ic, journal_sectors, interleave_sectors);
4118 ti->error = "Could not initialize superblock";
4121 if (ic->mode != 'R')
4122 should_write_sb = true;
4125 if (!ic->sb->version || ic->sb->version > SB_VERSION_4) {
4127 ti->error = "Unknown version";
4130 if (le16_to_cpu(ic->sb->integrity_tag_size) != ic->tag_size) {
4132 ti->error = "Tag size doesn't match the information in superblock";
4135 if (ic->sb->log2_sectors_per_block != __ffs(ic->sectors_per_block)) {
4137 ti->error = "Block size doesn't match the information in superblock";
4140 if (!le32_to_cpu(ic->sb->journal_sections)) {
4142 ti->error = "Corrupted superblock, journal_sections is 0";
4145 /* make sure that ti->max_io_len doesn't overflow */
4146 if (!ic->meta_dev) {
4147 if (ic->sb->log2_interleave_sectors < MIN_LOG2_INTERLEAVE_SECTORS ||
4148 ic->sb->log2_interleave_sectors > MAX_LOG2_INTERLEAVE_SECTORS) {
4150 ti->error = "Invalid interleave_sectors in the superblock";
4154 if (ic->sb->log2_interleave_sectors) {
4156 ti->error = "Invalid interleave_sectors in the superblock";
4160 if (!!(ic->sb->flags & cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC)) != !!ic->journal_mac_alg.alg_string) {
4162 ti->error = "Journal mac mismatch";
4166 get_provided_data_sectors(ic);
4167 if (!ic->provided_data_sectors) {
4169 ti->error = "The device is too small";
4174 r = calculate_device_limits(ic);
4177 if (ic->log2_buffer_sectors > 3) {
4178 ic->log2_buffer_sectors--;
4179 goto try_smaller_buffer;
4182 ti->error = "The device is too small";
4186 if (log2_sectors_per_bitmap_bit < 0)
4187 log2_sectors_per_bitmap_bit = __fls(DEFAULT_SECTORS_PER_BITMAP_BIT);
4188 if (log2_sectors_per_bitmap_bit < ic->sb->log2_sectors_per_block)
4189 log2_sectors_per_bitmap_bit = ic->sb->log2_sectors_per_block;
4191 bits_in_journal = ((__u64)ic->journal_section_sectors * ic->journal_sections) << (SECTOR_SHIFT + 3);
4192 if (bits_in_journal > UINT_MAX)
4193 bits_in_journal = UINT_MAX;
4194 while (bits_in_journal < (ic->provided_data_sectors + ((sector_t)1 << log2_sectors_per_bitmap_bit) - 1) >> log2_sectors_per_bitmap_bit)
4195 log2_sectors_per_bitmap_bit++;
4197 log2_blocks_per_bitmap_bit = log2_sectors_per_bitmap_bit - ic->sb->log2_sectors_per_block;
4198 ic->log2_blocks_per_bitmap_bit = log2_blocks_per_bitmap_bit;
4199 if (should_write_sb) {
4200 ic->sb->log2_blocks_per_bitmap_bit = log2_blocks_per_bitmap_bit;
4202 n_bitmap_bits = ((ic->provided_data_sectors >> ic->sb->log2_sectors_per_block)
4203 + (((sector_t)1 << log2_blocks_per_bitmap_bit) - 1)) >> log2_blocks_per_bitmap_bit;
4204 ic->n_bitmap_blocks = DIV_ROUND_UP(n_bitmap_bits, BITMAP_BLOCK_SIZE * 8);
4207 ic->log2_buffer_sectors = min(ic->log2_buffer_sectors, (__u8)__ffs(ic->metadata_run));
4209 if (ti->len > ic->provided_data_sectors) {
4211 ti->error = "Not enough provided sectors for requested mapping size";
4216 threshold = (__u64)ic->journal_entries * (100 - journal_watermark);
4218 do_div(threshold, 100);
4219 ic->free_sectors_threshold = threshold;
4221 DEBUG_print("initialized:\n");
4222 DEBUG_print(" integrity_tag_size %u\n", le16_to_cpu(ic->sb->integrity_tag_size));
4223 DEBUG_print(" journal_entry_size %u\n", ic->journal_entry_size);
4224 DEBUG_print(" journal_entries_per_sector %u\n", ic->journal_entries_per_sector);
4225 DEBUG_print(" journal_section_entries %u\n", ic->journal_section_entries);
4226 DEBUG_print(" journal_section_sectors %u\n", ic->journal_section_sectors);
4227 DEBUG_print(" journal_sections %u\n", (unsigned)le32_to_cpu(ic->sb->journal_sections));
4228 DEBUG_print(" journal_entries %u\n", ic->journal_entries);
4229 DEBUG_print(" log2_interleave_sectors %d\n", ic->sb->log2_interleave_sectors);
4230 DEBUG_print(" data_device_sectors 0x%llx\n", i_size_read(ic->dev->bdev->bd_inode) >> SECTOR_SHIFT);
4231 DEBUG_print(" initial_sectors 0x%x\n", ic->initial_sectors);
4232 DEBUG_print(" metadata_run 0x%x\n", ic->metadata_run);
4233 DEBUG_print(" log2_metadata_run %d\n", ic->log2_metadata_run);
4234 DEBUG_print(" provided_data_sectors 0x%llx (%llu)\n", ic->provided_data_sectors, ic->provided_data_sectors);
4235 DEBUG_print(" log2_buffer_sectors %u\n", ic->log2_buffer_sectors);
4236 DEBUG_print(" bits_in_journal %llu\n", bits_in_journal);
4238 if (ic->recalculate_flag && !(ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))) {
4239 ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
4240 ic->sb->recalc_sector = cpu_to_le64(0);
4243 if (ic->internal_hash) {
4244 size_t recalc_tags_size;
4245 ic->recalc_wq = alloc_workqueue("dm-integrity-recalc", WQ_MEM_RECLAIM, 1);
4246 if (!ic->recalc_wq ) {
4247 ti->error = "Cannot allocate workqueue";
4251 INIT_WORK(&ic->recalc_work, integrity_recalc);
4252 ic->recalc_buffer = vmalloc(RECALC_SECTORS << SECTOR_SHIFT);
4253 if (!ic->recalc_buffer) {
4254 ti->error = "Cannot allocate buffer for recalculating";
4258 recalc_tags_size = (RECALC_SECTORS >> ic->sb->log2_sectors_per_block) * ic->tag_size;
4259 if (crypto_shash_digestsize(ic->internal_hash) > ic->tag_size)
4260 recalc_tags_size += crypto_shash_digestsize(ic->internal_hash) - ic->tag_size;
4261 ic->recalc_tags = kvmalloc(recalc_tags_size, GFP_KERNEL);
4262 if (!ic->recalc_tags) {
4263 ti->error = "Cannot allocate tags for recalculating";
4268 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
4269 ti->error = "Recalculate can only be specified with internal_hash";
4275 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING) &&
4276 le64_to_cpu(ic->sb->recalc_sector) < ic->provided_data_sectors &&
4277 dm_integrity_disable_recalculate(ic)) {
4278 ti->error = "Recalculating with HMAC is disabled for security reasons - if you really need it, use the argument \"legacy_recalculate\"";
4283 ic->bufio = dm_bufio_client_create(ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev,
4284 1U << (SECTOR_SHIFT + ic->log2_buffer_sectors), 1, 0, NULL, NULL);
4285 if (IS_ERR(ic->bufio)) {
4286 r = PTR_ERR(ic->bufio);
4287 ti->error = "Cannot initialize dm-bufio";
4291 dm_bufio_set_sector_offset(ic->bufio, ic->start + ic->initial_sectors);
4293 if (ic->mode != 'R') {
4294 r = create_journal(ic, &ti->error);
4300 if (ic->mode == 'B') {
4302 unsigned n_bitmap_pages = DIV_ROUND_UP(ic->n_bitmap_blocks, PAGE_SIZE / BITMAP_BLOCK_SIZE);
4304 ic->recalc_bitmap = dm_integrity_alloc_page_list(n_bitmap_pages);
4305 if (!ic->recalc_bitmap) {
4309 ic->may_write_bitmap = dm_integrity_alloc_page_list(n_bitmap_pages);
4310 if (!ic->may_write_bitmap) {
4314 ic->bbs = kvmalloc_array(ic->n_bitmap_blocks, sizeof(struct bitmap_block_status), GFP_KERNEL);
4319 INIT_DELAYED_WORK(&ic->bitmap_flush_work, bitmap_flush_work);
4320 for (i = 0; i < ic->n_bitmap_blocks; i++) {
4321 struct bitmap_block_status *bbs = &ic->bbs[i];
4322 unsigned sector, pl_index, pl_offset;
4324 INIT_WORK(&bbs->work, bitmap_block_work);
4327 bio_list_init(&bbs->bio_queue);
4328 spin_lock_init(&bbs->bio_queue_lock);
4330 sector = i * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT);
4331 pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
4332 pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
4334 bbs->bitmap = lowmem_page_address(ic->journal[pl_index].page) + pl_offset;
4338 if (should_write_sb) {
4339 init_journal(ic, 0, ic->journal_sections, 0);
4340 r = dm_integrity_failed(ic);
4342 ti->error = "Error initializing journal";
4345 r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
4347 ti->error = "Error initializing superblock";
4350 ic->just_formatted = true;
4353 if (!ic->meta_dev) {
4354 r = dm_set_target_max_io_len(ti, 1U << ic->sb->log2_interleave_sectors);
4358 if (ic->mode == 'B') {
4359 unsigned max_io_len = ((sector_t)ic->sectors_per_block << ic->log2_blocks_per_bitmap_bit) * (BITMAP_BLOCK_SIZE * 8);
4361 max_io_len = 1U << 31;
4362 DEBUG_print("max_io_len: old %u, new %u\n", ti->max_io_len, max_io_len);
4363 if (!ti->max_io_len || ti->max_io_len > max_io_len) {
4364 r = dm_set_target_max_io_len(ti, max_io_len);
4370 if (!ic->internal_hash)
4371 dm_integrity_set(ti, ic);
4373 ti->num_flush_bios = 1;
4374 ti->flush_supported = true;
4376 ti->num_discard_bios = 1;
4381 dm_integrity_dtr(ti);
4385 static void dm_integrity_dtr(struct dm_target *ti)
4387 struct dm_integrity_c *ic = ti->private;
4389 BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
4390 BUG_ON(!list_empty(&ic->wait_list));
4392 if (ic->mode == 'B')
4393 cancel_delayed_work_sync(&ic->bitmap_flush_work);
4394 if (ic->metadata_wq)
4395 destroy_workqueue(ic->metadata_wq);
4397 destroy_workqueue(ic->wait_wq);
4399 destroy_workqueue(ic->offload_wq);
4401 destroy_workqueue(ic->commit_wq);
4403 destroy_workqueue(ic->writer_wq);
4405 destroy_workqueue(ic->recalc_wq);
4406 vfree(ic->recalc_buffer);
4407 kvfree(ic->recalc_tags);
4410 dm_bufio_client_destroy(ic->bufio);
4411 mempool_exit(&ic->journal_io_mempool);
4413 dm_io_client_destroy(ic->io);
4415 dm_put_device(ti, ic->dev);
4417 dm_put_device(ti, ic->meta_dev);
4418 dm_integrity_free_page_list(ic->journal);
4419 dm_integrity_free_page_list(ic->journal_io);
4420 dm_integrity_free_page_list(ic->journal_xor);
4421 dm_integrity_free_page_list(ic->recalc_bitmap);
4422 dm_integrity_free_page_list(ic->may_write_bitmap);
4423 if (ic->journal_scatterlist)
4424 dm_integrity_free_journal_scatterlist(ic, ic->journal_scatterlist);
4425 if (ic->journal_io_scatterlist)
4426 dm_integrity_free_journal_scatterlist(ic, ic->journal_io_scatterlist);
4427 if (ic->sk_requests) {
4430 for (i = 0; i < ic->journal_sections; i++) {
4431 struct skcipher_request *req = ic->sk_requests[i];
4433 kfree_sensitive(req->iv);
4434 skcipher_request_free(req);
4437 kvfree(ic->sk_requests);
4439 kvfree(ic->journal_tree);
4441 free_pages_exact(ic->sb, SB_SECTORS << SECTOR_SHIFT);
4443 if (ic->internal_hash)
4444 crypto_free_shash(ic->internal_hash);
4445 free_alg(&ic->internal_hash_alg);
4447 if (ic->journal_crypt)
4448 crypto_free_skcipher(ic->journal_crypt);
4449 free_alg(&ic->journal_crypt_alg);
4451 if (ic->journal_mac)
4452 crypto_free_shash(ic->journal_mac);
4453 free_alg(&ic->journal_mac_alg);
4458 static struct target_type integrity_target = {
4459 .name = "integrity",
4460 .version = {1, 6, 0},
4461 .module = THIS_MODULE,
4462 .features = DM_TARGET_SINGLETON | DM_TARGET_INTEGRITY,
4463 .ctr = dm_integrity_ctr,
4464 .dtr = dm_integrity_dtr,
4465 .map = dm_integrity_map,
4466 .postsuspend = dm_integrity_postsuspend,
4467 .resume = dm_integrity_resume,
4468 .status = dm_integrity_status,
4469 .iterate_devices = dm_integrity_iterate_devices,
4470 .io_hints = dm_integrity_io_hints,
4473 static int __init dm_integrity_init(void)
4477 journal_io_cache = kmem_cache_create("integrity_journal_io",
4478 sizeof(struct journal_io), 0, 0, NULL);
4479 if (!journal_io_cache) {
4480 DMERR("can't allocate journal io cache");
4484 r = dm_register_target(&integrity_target);
4486 DMERR("register failed %d", r);
4487 kmem_cache_destroy(journal_io_cache);
4494 static void __exit dm_integrity_exit(void)
4496 dm_unregister_target(&integrity_target);
4497 kmem_cache_destroy(journal_io_cache);
4500 module_init(dm_integrity_init);
4501 module_exit(dm_integrity_exit);
4503 MODULE_AUTHOR("Milan Broz");
4504 MODULE_AUTHOR("Mikulas Patocka");
4505 MODULE_DESCRIPTION(DM_NAME " target for integrity tags extension");
4506 MODULE_LICENSE("GPL");