2 * Copyright (C) 2016-2017 Red Hat, Inc. All rights reserved.
3 * Copyright (C) 2016-2017 Milan Broz
4 * Copyright (C) 2016-2017 Mikulas Patocka
6 * This file is released under the GPL.
9 #include "dm-bio-record.h"
11 #include <linux/compiler.h>
12 #include <linux/module.h>
13 #include <linux/device-mapper.h>
14 #include <linux/dm-io.h>
15 #include <linux/vmalloc.h>
16 #include <linux/sort.h>
17 #include <linux/rbtree.h>
18 #include <linux/delay.h>
19 #include <linux/random.h>
20 #include <linux/reboot.h>
21 #include <crypto/hash.h>
22 #include <crypto/skcipher.h>
23 #include <linux/async_tx.h>
24 #include <linux/dm-bufio.h>
26 #define DM_MSG_PREFIX "integrity"
28 #define DEFAULT_INTERLEAVE_SECTORS 32768
29 #define DEFAULT_JOURNAL_SIZE_FACTOR 7
30 #define DEFAULT_SECTORS_PER_BITMAP_BIT 32768
31 #define DEFAULT_BUFFER_SECTORS 128
32 #define DEFAULT_JOURNAL_WATERMARK 50
33 #define DEFAULT_SYNC_MSEC 10000
34 #define DEFAULT_MAX_JOURNAL_SECTORS 131072
35 #define MIN_LOG2_INTERLEAVE_SECTORS 3
36 #define MAX_LOG2_INTERLEAVE_SECTORS 31
37 #define METADATA_WORKQUEUE_MAX_ACTIVE 16
38 #define RECALC_SECTORS 8192
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 journal_uptodate;
259 bool recalculate_flag;
262 bool legacy_recalculate;
264 struct alg_spec internal_hash_alg;
265 struct alg_spec journal_crypt_alg;
266 struct alg_spec journal_mac_alg;
268 atomic64_t number_of_mismatches;
270 struct notifier_block reboot_notifier;
273 struct dm_integrity_range {
274 sector_t logical_sector;
280 struct task_struct *task;
281 struct list_head wait_entry;
286 struct dm_integrity_io {
287 struct work_struct work;
289 struct dm_integrity_c *ic;
293 struct dm_integrity_range range;
295 sector_t metadata_block;
296 unsigned metadata_offset;
299 blk_status_t bi_status;
301 struct completion *completion;
303 struct dm_bio_details bio_details;
306 struct journal_completion {
307 struct dm_integrity_c *ic;
309 struct completion comp;
313 struct dm_integrity_range range;
314 struct journal_completion *comp;
317 struct bitmap_block_status {
318 struct work_struct work;
319 struct dm_integrity_c *ic;
321 unsigned long *bitmap;
322 struct bio_list bio_queue;
323 spinlock_t bio_queue_lock;
327 static struct kmem_cache *journal_io_cache;
329 #define JOURNAL_IO_MEMPOOL 32
332 #define DEBUG_print(x, ...) printk(KERN_DEBUG x, ##__VA_ARGS__)
333 static void __DEBUG_bytes(__u8 *bytes, size_t len, const char *msg, ...)
342 pr_cont(" %02x", *bytes);
348 #define DEBUG_bytes(bytes, len, msg, ...) __DEBUG_bytes(bytes, len, KERN_DEBUG msg, ##__VA_ARGS__)
350 #define DEBUG_print(x, ...) do { } while (0)
351 #define DEBUG_bytes(bytes, len, msg, ...) do { } while (0)
354 static void dm_integrity_prepare(struct request *rq)
358 static void dm_integrity_complete(struct request *rq, unsigned int nr_bytes)
363 * DM Integrity profile, protection is performed layer above (dm-crypt)
365 static const struct blk_integrity_profile dm_integrity_profile = {
366 .name = "DM-DIF-EXT-TAG",
369 .prepare_fn = dm_integrity_prepare,
370 .complete_fn = dm_integrity_complete,
373 static void dm_integrity_map_continue(struct dm_integrity_io *dio, bool from_map);
374 static void integrity_bio_wait(struct work_struct *w);
375 static void dm_integrity_dtr(struct dm_target *ti);
377 static void dm_integrity_io_error(struct dm_integrity_c *ic, const char *msg, int err)
380 atomic64_inc(&ic->number_of_mismatches);
381 if (!cmpxchg(&ic->failed, 0, err))
382 DMERR("Error on %s: %d", msg, err);
385 static int dm_integrity_failed(struct dm_integrity_c *ic)
387 return READ_ONCE(ic->failed);
390 static bool dm_integrity_disable_recalculate(struct dm_integrity_c *ic)
392 if ((ic->internal_hash_alg.key || ic->journal_mac_alg.key) &&
393 !ic->legacy_recalculate)
398 static commit_id_t dm_integrity_commit_id(struct dm_integrity_c *ic, unsigned i,
399 unsigned j, unsigned char seq)
402 * Xor the number with section and sector, so that if a piece of
403 * journal is written at wrong place, it is detected.
405 return ic->commit_ids[seq] ^ cpu_to_le64(((__u64)i << 32) ^ j);
408 static void get_area_and_offset(struct dm_integrity_c *ic, sector_t data_sector,
409 sector_t *area, sector_t *offset)
412 __u8 log2_interleave_sectors = ic->sb->log2_interleave_sectors;
413 *area = data_sector >> log2_interleave_sectors;
414 *offset = (unsigned)data_sector & ((1U << log2_interleave_sectors) - 1);
417 *offset = data_sector;
421 #define sector_to_block(ic, n) \
423 BUG_ON((n) & (unsigned)((ic)->sectors_per_block - 1)); \
424 (n) >>= (ic)->sb->log2_sectors_per_block; \
427 static __u64 get_metadata_sector_and_offset(struct dm_integrity_c *ic, sector_t area,
428 sector_t offset, unsigned *metadata_offset)
433 ms = area << ic->sb->log2_interleave_sectors;
434 if (likely(ic->log2_metadata_run >= 0))
435 ms += area << ic->log2_metadata_run;
437 ms += area * ic->metadata_run;
438 ms >>= ic->log2_buffer_sectors;
440 sector_to_block(ic, offset);
442 if (likely(ic->log2_tag_size >= 0)) {
443 ms += offset >> (SECTOR_SHIFT + ic->log2_buffer_sectors - ic->log2_tag_size);
444 mo = (offset << ic->log2_tag_size) & ((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - 1);
446 ms += (__u64)offset * ic->tag_size >> (SECTOR_SHIFT + ic->log2_buffer_sectors);
447 mo = (offset * ic->tag_size) & ((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - 1);
449 *metadata_offset = mo;
453 static sector_t get_data_sector(struct dm_integrity_c *ic, sector_t area, sector_t offset)
460 result = area << ic->sb->log2_interleave_sectors;
461 if (likely(ic->log2_metadata_run >= 0))
462 result += (area + 1) << ic->log2_metadata_run;
464 result += (area + 1) * ic->metadata_run;
466 result += (sector_t)ic->initial_sectors + offset;
472 static void wraparound_section(struct dm_integrity_c *ic, unsigned *sec_ptr)
474 if (unlikely(*sec_ptr >= ic->journal_sections))
475 *sec_ptr -= ic->journal_sections;
478 static void sb_set_version(struct dm_integrity_c *ic)
480 if (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING))
481 ic->sb->version = SB_VERSION_4;
482 else if (ic->mode == 'B' || ic->sb->flags & cpu_to_le32(SB_FLAG_DIRTY_BITMAP))
483 ic->sb->version = SB_VERSION_3;
484 else if (ic->meta_dev || ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
485 ic->sb->version = SB_VERSION_2;
487 ic->sb->version = SB_VERSION_1;
490 static int sync_rw_sb(struct dm_integrity_c *ic, int op, int op_flags)
492 struct dm_io_request io_req;
493 struct dm_io_region io_loc;
496 io_req.bi_op_flags = op_flags;
497 io_req.mem.type = DM_IO_KMEM;
498 io_req.mem.ptr.addr = ic->sb;
499 io_req.notify.fn = NULL;
500 io_req.client = ic->io;
501 io_loc.bdev = ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev;
502 io_loc.sector = ic->start;
503 io_loc.count = SB_SECTORS;
505 if (op == REQ_OP_WRITE)
508 return dm_io(&io_req, 1, &io_loc, NULL);
511 #define BITMAP_OP_TEST_ALL_SET 0
512 #define BITMAP_OP_TEST_ALL_CLEAR 1
513 #define BITMAP_OP_SET 2
514 #define BITMAP_OP_CLEAR 3
516 static bool block_bitmap_op(struct dm_integrity_c *ic, struct page_list *bitmap,
517 sector_t sector, sector_t n_sectors, int mode)
519 unsigned long bit, end_bit, this_end_bit, page, end_page;
522 if (unlikely(((sector | n_sectors) & ((1 << ic->sb->log2_sectors_per_block) - 1)) != 0)) {
523 DMCRIT("invalid bitmap access (%llx,%llx,%d,%d,%d)",
526 ic->sb->log2_sectors_per_block,
527 ic->log2_blocks_per_bitmap_bit,
532 if (unlikely(!n_sectors))
535 bit = sector >> (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
536 end_bit = (sector + n_sectors - 1) >>
537 (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
539 page = bit / (PAGE_SIZE * 8);
540 bit %= PAGE_SIZE * 8;
542 end_page = end_bit / (PAGE_SIZE * 8);
543 end_bit %= PAGE_SIZE * 8;
546 if (page < end_page) {
547 this_end_bit = PAGE_SIZE * 8 - 1;
549 this_end_bit = end_bit;
552 data = lowmem_page_address(bitmap[page].page);
554 if (mode == BITMAP_OP_TEST_ALL_SET) {
555 while (bit <= this_end_bit) {
556 if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
558 if (data[bit / BITS_PER_LONG] != -1)
560 bit += BITS_PER_LONG;
561 } while (this_end_bit >= bit + BITS_PER_LONG - 1);
564 if (!test_bit(bit, data))
568 } else if (mode == BITMAP_OP_TEST_ALL_CLEAR) {
569 while (bit <= this_end_bit) {
570 if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
572 if (data[bit / BITS_PER_LONG] != 0)
574 bit += BITS_PER_LONG;
575 } while (this_end_bit >= bit + BITS_PER_LONG - 1);
578 if (test_bit(bit, data))
582 } else if (mode == BITMAP_OP_SET) {
583 while (bit <= this_end_bit) {
584 if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
586 data[bit / BITS_PER_LONG] = -1;
587 bit += BITS_PER_LONG;
588 } while (this_end_bit >= bit + BITS_PER_LONG - 1);
591 __set_bit(bit, data);
594 } else if (mode == BITMAP_OP_CLEAR) {
595 if (!bit && this_end_bit == PAGE_SIZE * 8 - 1)
597 else while (bit <= this_end_bit) {
598 if (!(bit % BITS_PER_LONG) && this_end_bit >= bit + BITS_PER_LONG - 1) {
600 data[bit / BITS_PER_LONG] = 0;
601 bit += BITS_PER_LONG;
602 } while (this_end_bit >= bit + BITS_PER_LONG - 1);
605 __clear_bit(bit, data);
612 if (unlikely(page < end_page)) {
621 static void block_bitmap_copy(struct dm_integrity_c *ic, struct page_list *dst, struct page_list *src)
623 unsigned n_bitmap_pages = DIV_ROUND_UP(ic->n_bitmap_blocks, PAGE_SIZE / BITMAP_BLOCK_SIZE);
626 for (i = 0; i < n_bitmap_pages; i++) {
627 unsigned long *dst_data = lowmem_page_address(dst[i].page);
628 unsigned long *src_data = lowmem_page_address(src[i].page);
629 copy_page(dst_data, src_data);
633 static struct bitmap_block_status *sector_to_bitmap_block(struct dm_integrity_c *ic, sector_t sector)
635 unsigned bit = sector >> (ic->sb->log2_sectors_per_block + ic->log2_blocks_per_bitmap_bit);
636 unsigned bitmap_block = bit / (BITMAP_BLOCK_SIZE * 8);
638 BUG_ON(bitmap_block >= ic->n_bitmap_blocks);
639 return &ic->bbs[bitmap_block];
642 static void access_journal_check(struct dm_integrity_c *ic, unsigned section, unsigned offset,
643 bool e, const char *function)
645 #if defined(CONFIG_DM_DEBUG) || defined(INTERNAL_VERIFY)
646 unsigned limit = e ? ic->journal_section_entries : ic->journal_section_sectors;
648 if (unlikely(section >= ic->journal_sections) ||
649 unlikely(offset >= limit)) {
650 DMCRIT("%s: invalid access at (%u,%u), limit (%u,%u)",
651 function, section, offset, ic->journal_sections, limit);
657 static void page_list_location(struct dm_integrity_c *ic, unsigned section, unsigned offset,
658 unsigned *pl_index, unsigned *pl_offset)
662 access_journal_check(ic, section, offset, false, "page_list_location");
664 sector = section * ic->journal_section_sectors + offset;
666 *pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
667 *pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
670 static struct journal_sector *access_page_list(struct dm_integrity_c *ic, struct page_list *pl,
671 unsigned section, unsigned offset, unsigned *n_sectors)
673 unsigned pl_index, pl_offset;
676 page_list_location(ic, section, offset, &pl_index, &pl_offset);
679 *n_sectors = (PAGE_SIZE - pl_offset) >> SECTOR_SHIFT;
681 va = lowmem_page_address(pl[pl_index].page);
683 return (struct journal_sector *)(va + pl_offset);
686 static struct journal_sector *access_journal(struct dm_integrity_c *ic, unsigned section, unsigned offset)
688 return access_page_list(ic, ic->journal, section, offset, NULL);
691 static struct journal_entry *access_journal_entry(struct dm_integrity_c *ic, unsigned section, unsigned n)
693 unsigned rel_sector, offset;
694 struct journal_sector *js;
696 access_journal_check(ic, section, n, true, "access_journal_entry");
698 rel_sector = n % JOURNAL_BLOCK_SECTORS;
699 offset = n / JOURNAL_BLOCK_SECTORS;
701 js = access_journal(ic, section, rel_sector);
702 return (struct journal_entry *)((char *)js + offset * ic->journal_entry_size);
705 static struct journal_sector *access_journal_data(struct dm_integrity_c *ic, unsigned section, unsigned n)
707 n <<= ic->sb->log2_sectors_per_block;
709 n += JOURNAL_BLOCK_SECTORS;
711 access_journal_check(ic, section, n, false, "access_journal_data");
713 return access_journal(ic, section, n);
716 static void section_mac(struct dm_integrity_c *ic, unsigned section, __u8 result[JOURNAL_MAC_SIZE])
718 SHASH_DESC_ON_STACK(desc, ic->journal_mac);
722 desc->tfm = ic->journal_mac;
724 r = crypto_shash_init(desc);
726 dm_integrity_io_error(ic, "crypto_shash_init", r);
730 for (j = 0; j < ic->journal_section_entries; j++) {
731 struct journal_entry *je = access_journal_entry(ic, section, j);
732 r = crypto_shash_update(desc, (__u8 *)&je->u.sector, sizeof je->u.sector);
734 dm_integrity_io_error(ic, "crypto_shash_update", r);
739 size = crypto_shash_digestsize(ic->journal_mac);
741 if (likely(size <= JOURNAL_MAC_SIZE)) {
742 r = crypto_shash_final(desc, result);
744 dm_integrity_io_error(ic, "crypto_shash_final", r);
747 memset(result + size, 0, JOURNAL_MAC_SIZE - size);
749 __u8 digest[HASH_MAX_DIGESTSIZE];
751 if (WARN_ON(size > sizeof(digest))) {
752 dm_integrity_io_error(ic, "digest_size", -EINVAL);
755 r = crypto_shash_final(desc, digest);
757 dm_integrity_io_error(ic, "crypto_shash_final", r);
760 memcpy(result, digest, JOURNAL_MAC_SIZE);
765 memset(result, 0, JOURNAL_MAC_SIZE);
768 static void rw_section_mac(struct dm_integrity_c *ic, unsigned section, bool wr)
770 __u8 result[JOURNAL_MAC_SIZE];
773 if (!ic->journal_mac)
776 section_mac(ic, section, result);
778 for (j = 0; j < JOURNAL_BLOCK_SECTORS; j++) {
779 struct journal_sector *js = access_journal(ic, section, j);
782 memcpy(&js->mac, result + (j * JOURNAL_MAC_PER_SECTOR), JOURNAL_MAC_PER_SECTOR);
784 if (memcmp(&js->mac, result + (j * JOURNAL_MAC_PER_SECTOR), JOURNAL_MAC_PER_SECTOR))
785 dm_integrity_io_error(ic, "journal mac", -EILSEQ);
790 static void complete_journal_op(void *context)
792 struct journal_completion *comp = context;
793 BUG_ON(!atomic_read(&comp->in_flight));
794 if (likely(atomic_dec_and_test(&comp->in_flight)))
795 complete(&comp->comp);
798 static void xor_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
799 unsigned n_sections, struct journal_completion *comp)
801 struct async_submit_ctl submit;
802 size_t n_bytes = (size_t)(n_sections * ic->journal_section_sectors) << SECTOR_SHIFT;
803 unsigned pl_index, pl_offset, section_index;
804 struct page_list *source_pl, *target_pl;
806 if (likely(encrypt)) {
807 source_pl = ic->journal;
808 target_pl = ic->journal_io;
810 source_pl = ic->journal_io;
811 target_pl = ic->journal;
814 page_list_location(ic, section, 0, &pl_index, &pl_offset);
816 atomic_add(roundup(pl_offset + n_bytes, PAGE_SIZE) >> PAGE_SHIFT, &comp->in_flight);
818 init_async_submit(&submit, ASYNC_TX_XOR_ZERO_DST, NULL, complete_journal_op, comp, NULL);
820 section_index = pl_index;
824 struct page *src_pages[2];
825 struct page *dst_page;
827 while (unlikely(pl_index == section_index)) {
830 rw_section_mac(ic, section, true);
835 page_list_location(ic, section, 0, §ion_index, &dummy);
838 this_step = min(n_bytes, (size_t)PAGE_SIZE - pl_offset);
839 dst_page = target_pl[pl_index].page;
840 src_pages[0] = source_pl[pl_index].page;
841 src_pages[1] = ic->journal_xor[pl_index].page;
843 async_xor(dst_page, src_pages, pl_offset, 2, this_step, &submit);
847 n_bytes -= this_step;
852 async_tx_issue_pending_all();
855 static void complete_journal_encrypt(struct crypto_async_request *req, int err)
857 struct journal_completion *comp = req->data;
859 if (likely(err == -EINPROGRESS)) {
860 complete(&comp->ic->crypto_backoff);
863 dm_integrity_io_error(comp->ic, "asynchronous encrypt", err);
865 complete_journal_op(comp);
868 static bool do_crypt(bool encrypt, struct skcipher_request *req, struct journal_completion *comp)
871 skcipher_request_set_callback(req, CRYPTO_TFM_REQ_MAY_BACKLOG,
872 complete_journal_encrypt, comp);
874 r = crypto_skcipher_encrypt(req);
876 r = crypto_skcipher_decrypt(req);
879 if (likely(r == -EINPROGRESS))
881 if (likely(r == -EBUSY)) {
882 wait_for_completion(&comp->ic->crypto_backoff);
883 reinit_completion(&comp->ic->crypto_backoff);
886 dm_integrity_io_error(comp->ic, "encrypt", r);
890 static void crypt_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
891 unsigned n_sections, struct journal_completion *comp)
893 struct scatterlist **source_sg;
894 struct scatterlist **target_sg;
896 atomic_add(2, &comp->in_flight);
898 if (likely(encrypt)) {
899 source_sg = ic->journal_scatterlist;
900 target_sg = ic->journal_io_scatterlist;
902 source_sg = ic->journal_io_scatterlist;
903 target_sg = ic->journal_scatterlist;
907 struct skcipher_request *req;
912 rw_section_mac(ic, section, true);
914 req = ic->sk_requests[section];
915 ivsize = crypto_skcipher_ivsize(ic->journal_crypt);
918 memcpy(iv, iv + ivsize, ivsize);
920 req->src = source_sg[section];
921 req->dst = target_sg[section];
923 if (unlikely(do_crypt(encrypt, req, comp)))
924 atomic_inc(&comp->in_flight);
928 } while (n_sections);
930 atomic_dec(&comp->in_flight);
931 complete_journal_op(comp);
934 static void encrypt_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
935 unsigned n_sections, struct journal_completion *comp)
938 return xor_journal(ic, encrypt, section, n_sections, comp);
940 return crypt_journal(ic, encrypt, section, n_sections, comp);
943 static void complete_journal_io(unsigned long error, void *context)
945 struct journal_completion *comp = context;
946 if (unlikely(error != 0))
947 dm_integrity_io_error(comp->ic, "writing journal", -EIO);
948 complete_journal_op(comp);
951 static void rw_journal_sectors(struct dm_integrity_c *ic, int op, int op_flags,
952 unsigned sector, unsigned n_sectors, struct journal_completion *comp)
954 struct dm_io_request io_req;
955 struct dm_io_region io_loc;
956 unsigned pl_index, pl_offset;
959 if (unlikely(dm_integrity_failed(ic))) {
961 complete_journal_io(-1UL, comp);
965 pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
966 pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
969 io_req.bi_op_flags = op_flags;
970 io_req.mem.type = DM_IO_PAGE_LIST;
972 io_req.mem.ptr.pl = &ic->journal_io[pl_index];
974 io_req.mem.ptr.pl = &ic->journal[pl_index];
975 io_req.mem.offset = pl_offset;
976 if (likely(comp != NULL)) {
977 io_req.notify.fn = complete_journal_io;
978 io_req.notify.context = comp;
980 io_req.notify.fn = NULL;
982 io_req.client = ic->io;
983 io_loc.bdev = ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev;
984 io_loc.sector = ic->start + SB_SECTORS + sector;
985 io_loc.count = n_sectors;
987 r = dm_io(&io_req, 1, &io_loc, NULL);
989 dm_integrity_io_error(ic, op == REQ_OP_READ ? "reading journal" : "writing journal", r);
991 WARN_ONCE(1, "asynchronous dm_io failed: %d", r);
992 complete_journal_io(-1UL, comp);
997 static void rw_journal(struct dm_integrity_c *ic, int op, int op_flags, unsigned section,
998 unsigned n_sections, struct journal_completion *comp)
1000 unsigned sector, n_sectors;
1002 sector = section * ic->journal_section_sectors;
1003 n_sectors = n_sections * ic->journal_section_sectors;
1005 rw_journal_sectors(ic, op, op_flags, sector, n_sectors, comp);
1008 static void write_journal(struct dm_integrity_c *ic, unsigned commit_start, unsigned commit_sections)
1010 struct journal_completion io_comp;
1011 struct journal_completion crypt_comp_1;
1012 struct journal_completion crypt_comp_2;
1016 init_completion(&io_comp.comp);
1018 if (commit_start + commit_sections <= ic->journal_sections) {
1019 io_comp.in_flight = (atomic_t)ATOMIC_INIT(1);
1020 if (ic->journal_io) {
1021 crypt_comp_1.ic = ic;
1022 init_completion(&crypt_comp_1.comp);
1023 crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
1024 encrypt_journal(ic, true, commit_start, commit_sections, &crypt_comp_1);
1025 wait_for_completion_io(&crypt_comp_1.comp);
1027 for (i = 0; i < commit_sections; i++)
1028 rw_section_mac(ic, commit_start + i, true);
1030 rw_journal(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, commit_start,
1031 commit_sections, &io_comp);
1034 io_comp.in_flight = (atomic_t)ATOMIC_INIT(2);
1035 to_end = ic->journal_sections - commit_start;
1036 if (ic->journal_io) {
1037 crypt_comp_1.ic = ic;
1038 init_completion(&crypt_comp_1.comp);
1039 crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
1040 encrypt_journal(ic, true, commit_start, to_end, &crypt_comp_1);
1041 if (try_wait_for_completion(&crypt_comp_1.comp)) {
1042 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
1043 reinit_completion(&crypt_comp_1.comp);
1044 crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
1045 encrypt_journal(ic, true, 0, commit_sections - to_end, &crypt_comp_1);
1046 wait_for_completion_io(&crypt_comp_1.comp);
1048 crypt_comp_2.ic = ic;
1049 init_completion(&crypt_comp_2.comp);
1050 crypt_comp_2.in_flight = (atomic_t)ATOMIC_INIT(0);
1051 encrypt_journal(ic, true, 0, commit_sections - to_end, &crypt_comp_2);
1052 wait_for_completion_io(&crypt_comp_1.comp);
1053 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
1054 wait_for_completion_io(&crypt_comp_2.comp);
1057 for (i = 0; i < to_end; i++)
1058 rw_section_mac(ic, commit_start + i, true);
1059 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
1060 for (i = 0; i < commit_sections - to_end; i++)
1061 rw_section_mac(ic, i, true);
1063 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, 0, commit_sections - to_end, &io_comp);
1066 wait_for_completion_io(&io_comp.comp);
1069 static void copy_from_journal(struct dm_integrity_c *ic, unsigned section, unsigned offset,
1070 unsigned n_sectors, sector_t target, io_notify_fn fn, void *data)
1072 struct dm_io_request io_req;
1073 struct dm_io_region io_loc;
1075 unsigned sector, pl_index, pl_offset;
1077 BUG_ON((target | n_sectors | offset) & (unsigned)(ic->sectors_per_block - 1));
1079 if (unlikely(dm_integrity_failed(ic))) {
1084 sector = section * ic->journal_section_sectors + JOURNAL_BLOCK_SECTORS + offset;
1086 pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
1087 pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
1089 io_req.bi_op = REQ_OP_WRITE;
1090 io_req.bi_op_flags = 0;
1091 io_req.mem.type = DM_IO_PAGE_LIST;
1092 io_req.mem.ptr.pl = &ic->journal[pl_index];
1093 io_req.mem.offset = pl_offset;
1094 io_req.notify.fn = fn;
1095 io_req.notify.context = data;
1096 io_req.client = ic->io;
1097 io_loc.bdev = ic->dev->bdev;
1098 io_loc.sector = target;
1099 io_loc.count = n_sectors;
1101 r = dm_io(&io_req, 1, &io_loc, NULL);
1103 WARN_ONCE(1, "asynchronous dm_io failed: %d", r);
1108 static bool ranges_overlap(struct dm_integrity_range *range1, struct dm_integrity_range *range2)
1110 return range1->logical_sector < range2->logical_sector + range2->n_sectors &&
1111 range1->logical_sector + range1->n_sectors > range2->logical_sector;
1114 static bool add_new_range(struct dm_integrity_c *ic, struct dm_integrity_range *new_range, bool check_waiting)
1116 struct rb_node **n = &ic->in_progress.rb_node;
1117 struct rb_node *parent;
1119 BUG_ON((new_range->logical_sector | new_range->n_sectors) & (unsigned)(ic->sectors_per_block - 1));
1121 if (likely(check_waiting)) {
1122 struct dm_integrity_range *range;
1123 list_for_each_entry(range, &ic->wait_list, wait_entry) {
1124 if (unlikely(ranges_overlap(range, new_range)))
1132 struct dm_integrity_range *range = container_of(*n, struct dm_integrity_range, node);
1135 if (new_range->logical_sector + new_range->n_sectors <= range->logical_sector) {
1136 n = &range->node.rb_left;
1137 } else if (new_range->logical_sector >= range->logical_sector + range->n_sectors) {
1138 n = &range->node.rb_right;
1144 rb_link_node(&new_range->node, parent, n);
1145 rb_insert_color(&new_range->node, &ic->in_progress);
1150 static void remove_range_unlocked(struct dm_integrity_c *ic, struct dm_integrity_range *range)
1152 rb_erase(&range->node, &ic->in_progress);
1153 while (unlikely(!list_empty(&ic->wait_list))) {
1154 struct dm_integrity_range *last_range =
1155 list_first_entry(&ic->wait_list, struct dm_integrity_range, wait_entry);
1156 struct task_struct *last_range_task;
1157 last_range_task = last_range->task;
1158 list_del(&last_range->wait_entry);
1159 if (!add_new_range(ic, last_range, false)) {
1160 last_range->task = last_range_task;
1161 list_add(&last_range->wait_entry, &ic->wait_list);
1164 last_range->waiting = false;
1165 wake_up_process(last_range_task);
1169 static void remove_range(struct dm_integrity_c *ic, struct dm_integrity_range *range)
1171 unsigned long flags;
1173 spin_lock_irqsave(&ic->endio_wait.lock, flags);
1174 remove_range_unlocked(ic, range);
1175 spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
1178 static void wait_and_add_new_range(struct dm_integrity_c *ic, struct dm_integrity_range *new_range)
1180 new_range->waiting = true;
1181 list_add_tail(&new_range->wait_entry, &ic->wait_list);
1182 new_range->task = current;
1184 __set_current_state(TASK_UNINTERRUPTIBLE);
1185 spin_unlock_irq(&ic->endio_wait.lock);
1187 spin_lock_irq(&ic->endio_wait.lock);
1188 } while (unlikely(new_range->waiting));
1191 static void add_new_range_and_wait(struct dm_integrity_c *ic, struct dm_integrity_range *new_range)
1193 if (unlikely(!add_new_range(ic, new_range, true)))
1194 wait_and_add_new_range(ic, new_range);
1197 static void init_journal_node(struct journal_node *node)
1199 RB_CLEAR_NODE(&node->node);
1200 node->sector = (sector_t)-1;
1203 static void add_journal_node(struct dm_integrity_c *ic, struct journal_node *node, sector_t sector)
1205 struct rb_node **link;
1206 struct rb_node *parent;
1208 node->sector = sector;
1209 BUG_ON(!RB_EMPTY_NODE(&node->node));
1211 link = &ic->journal_tree_root.rb_node;
1215 struct journal_node *j;
1217 j = container_of(parent, struct journal_node, node);
1218 if (sector < j->sector)
1219 link = &j->node.rb_left;
1221 link = &j->node.rb_right;
1224 rb_link_node(&node->node, parent, link);
1225 rb_insert_color(&node->node, &ic->journal_tree_root);
1228 static void remove_journal_node(struct dm_integrity_c *ic, struct journal_node *node)
1230 BUG_ON(RB_EMPTY_NODE(&node->node));
1231 rb_erase(&node->node, &ic->journal_tree_root);
1232 init_journal_node(node);
1235 #define NOT_FOUND (-1U)
1237 static unsigned find_journal_node(struct dm_integrity_c *ic, sector_t sector, sector_t *next_sector)
1239 struct rb_node *n = ic->journal_tree_root.rb_node;
1240 unsigned found = NOT_FOUND;
1241 *next_sector = (sector_t)-1;
1243 struct journal_node *j = container_of(n, struct journal_node, node);
1244 if (sector == j->sector) {
1245 found = j - ic->journal_tree;
1247 if (sector < j->sector) {
1248 *next_sector = j->sector;
1249 n = j->node.rb_left;
1251 n = j->node.rb_right;
1258 static bool test_journal_node(struct dm_integrity_c *ic, unsigned pos, sector_t sector)
1260 struct journal_node *node, *next_node;
1261 struct rb_node *next;
1263 if (unlikely(pos >= ic->journal_entries))
1265 node = &ic->journal_tree[pos];
1266 if (unlikely(RB_EMPTY_NODE(&node->node)))
1268 if (unlikely(node->sector != sector))
1271 next = rb_next(&node->node);
1272 if (unlikely(!next))
1275 next_node = container_of(next, struct journal_node, node);
1276 return next_node->sector != sector;
1279 static bool find_newer_committed_node(struct dm_integrity_c *ic, struct journal_node *node)
1281 struct rb_node *next;
1282 struct journal_node *next_node;
1283 unsigned next_section;
1285 BUG_ON(RB_EMPTY_NODE(&node->node));
1287 next = rb_next(&node->node);
1288 if (unlikely(!next))
1291 next_node = container_of(next, struct journal_node, node);
1293 if (next_node->sector != node->sector)
1296 next_section = (unsigned)(next_node - ic->journal_tree) / ic->journal_section_entries;
1297 if (next_section >= ic->committed_section &&
1298 next_section < ic->committed_section + ic->n_committed_sections)
1300 if (next_section + ic->journal_sections < ic->committed_section + ic->n_committed_sections)
1310 static int dm_integrity_rw_tag(struct dm_integrity_c *ic, unsigned char *tag, sector_t *metadata_block,
1311 unsigned *metadata_offset, unsigned total_size, int op)
1313 #define MAY_BE_FILLER 1
1314 #define MAY_BE_HASH 2
1315 unsigned hash_offset = 0;
1316 unsigned may_be = MAY_BE_HASH | (ic->discard ? MAY_BE_FILLER : 0);
1319 unsigned char *data, *dp;
1320 struct dm_buffer *b;
1324 r = dm_integrity_failed(ic);
1328 data = dm_bufio_read(ic->bufio, *metadata_block, &b);
1330 return PTR_ERR(data);
1332 to_copy = min((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - *metadata_offset, total_size);
1333 dp = data + *metadata_offset;
1334 if (op == TAG_READ) {
1335 memcpy(tag, dp, to_copy);
1336 } else if (op == TAG_WRITE) {
1337 memcpy(dp, tag, to_copy);
1338 dm_bufio_mark_partial_buffer_dirty(b, *metadata_offset, *metadata_offset + to_copy);
1340 /* e.g.: op == TAG_CMP */
1342 if (likely(is_power_of_2(ic->tag_size))) {
1343 if (unlikely(memcmp(dp, tag, to_copy)))
1344 if (unlikely(!ic->discard) ||
1345 unlikely(memchr_inv(dp, DISCARD_FILLER, to_copy) != NULL)) {
1353 for (i = 0; i < to_copy; i++, ts--) {
1354 if (unlikely(dp[i] != tag[i]))
1355 may_be &= ~MAY_BE_HASH;
1356 if (likely(dp[i] != DISCARD_FILLER))
1357 may_be &= ~MAY_BE_FILLER;
1359 if (unlikely(hash_offset == ic->tag_size)) {
1360 if (unlikely(!may_be)) {
1361 dm_bufio_release(b);
1365 may_be = MAY_BE_HASH | (ic->discard ? MAY_BE_FILLER : 0);
1370 dm_bufio_release(b);
1373 *metadata_offset += to_copy;
1374 if (unlikely(*metadata_offset == 1U << SECTOR_SHIFT << ic->log2_buffer_sectors)) {
1375 (*metadata_block)++;
1376 *metadata_offset = 0;
1379 if (unlikely(!is_power_of_2(ic->tag_size))) {
1380 hash_offset = (hash_offset + to_copy) % ic->tag_size;
1383 total_size -= to_copy;
1384 } while (unlikely(total_size));
1387 #undef MAY_BE_FILLER
1391 struct flush_request {
1392 struct dm_io_request io_req;
1393 struct dm_io_region io_reg;
1394 struct dm_integrity_c *ic;
1395 struct completion comp;
1398 static void flush_notify(unsigned long error, void *fr_)
1400 struct flush_request *fr = fr_;
1401 if (unlikely(error != 0))
1402 dm_integrity_io_error(fr->ic, "flusing disk cache", -EIO);
1403 complete(&fr->comp);
1406 static void dm_integrity_flush_buffers(struct dm_integrity_c *ic, bool flush_data)
1410 struct flush_request fr;
1415 fr.io_req.bi_op = REQ_OP_WRITE,
1416 fr.io_req.bi_op_flags = REQ_PREFLUSH | REQ_SYNC,
1417 fr.io_req.mem.type = DM_IO_KMEM,
1418 fr.io_req.mem.ptr.addr = NULL,
1419 fr.io_req.notify.fn = flush_notify,
1420 fr.io_req.notify.context = &fr;
1421 fr.io_req.client = dm_bufio_get_dm_io_client(ic->bufio),
1422 fr.io_reg.bdev = ic->dev->bdev,
1423 fr.io_reg.sector = 0,
1424 fr.io_reg.count = 0,
1426 init_completion(&fr.comp);
1427 r = dm_io(&fr.io_req, 1, &fr.io_reg, NULL);
1431 r = dm_bufio_write_dirty_buffers(ic->bufio);
1433 dm_integrity_io_error(ic, "writing tags", r);
1436 wait_for_completion(&fr.comp);
1439 static void sleep_on_endio_wait(struct dm_integrity_c *ic)
1441 DECLARE_WAITQUEUE(wait, current);
1442 __add_wait_queue(&ic->endio_wait, &wait);
1443 __set_current_state(TASK_UNINTERRUPTIBLE);
1444 spin_unlock_irq(&ic->endio_wait.lock);
1446 spin_lock_irq(&ic->endio_wait.lock);
1447 __remove_wait_queue(&ic->endio_wait, &wait);
1450 static void autocommit_fn(struct timer_list *t)
1452 struct dm_integrity_c *ic = from_timer(ic, t, autocommit_timer);
1454 if (likely(!dm_integrity_failed(ic)))
1455 queue_work(ic->commit_wq, &ic->commit_work);
1458 static void schedule_autocommit(struct dm_integrity_c *ic)
1460 if (!timer_pending(&ic->autocommit_timer))
1461 mod_timer(&ic->autocommit_timer, jiffies + ic->autocommit_jiffies);
1464 static void submit_flush_bio(struct dm_integrity_c *ic, struct dm_integrity_io *dio)
1467 unsigned long flags;
1469 spin_lock_irqsave(&ic->endio_wait.lock, flags);
1470 bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1471 bio_list_add(&ic->flush_bio_list, bio);
1472 spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
1474 queue_work(ic->commit_wq, &ic->commit_work);
1477 static void do_endio(struct dm_integrity_c *ic, struct bio *bio)
1479 int r = dm_integrity_failed(ic);
1480 if (unlikely(r) && !bio->bi_status)
1481 bio->bi_status = errno_to_blk_status(r);
1482 if (unlikely(ic->synchronous_mode) && bio_op(bio) == REQ_OP_WRITE) {
1483 unsigned long flags;
1484 spin_lock_irqsave(&ic->endio_wait.lock, flags);
1485 bio_list_add(&ic->synchronous_bios, bio);
1486 queue_delayed_work(ic->commit_wq, &ic->bitmap_flush_work, 0);
1487 spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
1493 static void do_endio_flush(struct dm_integrity_c *ic, struct dm_integrity_io *dio)
1495 struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1497 if (unlikely(dio->fua) && likely(!bio->bi_status) && likely(!dm_integrity_failed(ic)))
1498 submit_flush_bio(ic, dio);
1503 static void dec_in_flight(struct dm_integrity_io *dio)
1505 if (atomic_dec_and_test(&dio->in_flight)) {
1506 struct dm_integrity_c *ic = dio->ic;
1509 remove_range(ic, &dio->range);
1511 if (dio->op == REQ_OP_WRITE || unlikely(dio->op == REQ_OP_DISCARD))
1512 schedule_autocommit(ic);
1514 bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1516 if (unlikely(dio->bi_status) && !bio->bi_status)
1517 bio->bi_status = dio->bi_status;
1518 if (likely(!bio->bi_status) && unlikely(bio_sectors(bio) != dio->range.n_sectors)) {
1519 dio->range.logical_sector += dio->range.n_sectors;
1520 bio_advance(bio, dio->range.n_sectors << SECTOR_SHIFT);
1521 INIT_WORK(&dio->work, integrity_bio_wait);
1522 queue_work(ic->offload_wq, &dio->work);
1525 do_endio_flush(ic, dio);
1529 static void integrity_end_io(struct bio *bio)
1531 struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
1533 dm_bio_restore(&dio->bio_details, bio);
1534 if (bio->bi_integrity)
1535 bio->bi_opf |= REQ_INTEGRITY;
1537 if (dio->completion)
1538 complete(dio->completion);
1543 static void integrity_sector_checksum(struct dm_integrity_c *ic, sector_t sector,
1544 const char *data, char *result)
1546 __u64 sector_le = cpu_to_le64(sector);
1547 SHASH_DESC_ON_STACK(req, ic->internal_hash);
1549 unsigned digest_size;
1551 req->tfm = ic->internal_hash;
1553 r = crypto_shash_init(req);
1554 if (unlikely(r < 0)) {
1555 dm_integrity_io_error(ic, "crypto_shash_init", r);
1559 r = crypto_shash_update(req, (const __u8 *)§or_le, sizeof sector_le);
1560 if (unlikely(r < 0)) {
1561 dm_integrity_io_error(ic, "crypto_shash_update", r);
1565 r = crypto_shash_update(req, data, ic->sectors_per_block << SECTOR_SHIFT);
1566 if (unlikely(r < 0)) {
1567 dm_integrity_io_error(ic, "crypto_shash_update", r);
1571 r = crypto_shash_final(req, result);
1572 if (unlikely(r < 0)) {
1573 dm_integrity_io_error(ic, "crypto_shash_final", r);
1577 digest_size = crypto_shash_digestsize(ic->internal_hash);
1578 if (unlikely(digest_size < ic->tag_size))
1579 memset(result + digest_size, 0, ic->tag_size - digest_size);
1584 /* this shouldn't happen anyway, the hash functions have no reason to fail */
1585 get_random_bytes(result, ic->tag_size);
1588 static void integrity_metadata(struct work_struct *w)
1590 struct dm_integrity_io *dio = container_of(w, struct dm_integrity_io, work);
1591 struct dm_integrity_c *ic = dio->ic;
1595 if (ic->internal_hash) {
1596 struct bvec_iter iter;
1598 unsigned digest_size = crypto_shash_digestsize(ic->internal_hash);
1599 struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1601 unsigned extra_space = unlikely(digest_size > ic->tag_size) ? digest_size - ic->tag_size : 0;
1602 char checksums_onstack[max((size_t)HASH_MAX_DIGESTSIZE, MAX_TAG_SIZE)];
1604 unsigned sectors_to_process;
1606 if (unlikely(ic->mode == 'R'))
1609 if (likely(dio->op != REQ_OP_DISCARD))
1610 checksums = kmalloc((PAGE_SIZE >> SECTOR_SHIFT >> ic->sb->log2_sectors_per_block) * ic->tag_size + extra_space,
1611 GFP_NOIO | __GFP_NORETRY | __GFP_NOWARN);
1613 checksums = kmalloc(PAGE_SIZE, GFP_NOIO | __GFP_NORETRY | __GFP_NOWARN);
1615 checksums = checksums_onstack;
1616 if (WARN_ON(extra_space &&
1617 digest_size > sizeof(checksums_onstack))) {
1623 if (unlikely(dio->op == REQ_OP_DISCARD)) {
1624 sector_t bi_sector = dio->bio_details.bi_iter.bi_sector;
1625 unsigned bi_size = dio->bio_details.bi_iter.bi_size;
1626 unsigned max_size = likely(checksums != checksums_onstack) ? PAGE_SIZE : HASH_MAX_DIGESTSIZE;
1627 unsigned max_blocks = max_size / ic->tag_size;
1628 memset(checksums, DISCARD_FILLER, max_size);
1631 unsigned this_step_blocks = bi_size >> (SECTOR_SHIFT + ic->sb->log2_sectors_per_block);
1632 this_step_blocks = min(this_step_blocks, max_blocks);
1633 r = dm_integrity_rw_tag(ic, checksums, &dio->metadata_block, &dio->metadata_offset,
1634 this_step_blocks * ic->tag_size, TAG_WRITE);
1636 if (likely(checksums != checksums_onstack))
1641 /*if (bi_size < this_step_blocks << (SECTOR_SHIFT + ic->sb->log2_sectors_per_block)) {
1642 printk("BUGG: bi_sector: %llx, bi_size: %u\n", bi_sector, bi_size);
1643 printk("BUGG: this_step_blocks: %u\n", this_step_blocks);
1646 bi_size -= this_step_blocks << (SECTOR_SHIFT + ic->sb->log2_sectors_per_block);
1647 bi_sector += this_step_blocks << ic->sb->log2_sectors_per_block;
1650 if (likely(checksums != checksums_onstack))
1655 sector = dio->range.logical_sector;
1656 sectors_to_process = dio->range.n_sectors;
1658 __bio_for_each_segment(bv, bio, iter, dio->bio_details.bi_iter) {
1660 char *mem, *checksums_ptr;
1663 mem = (char *)kmap_atomic(bv.bv_page) + bv.bv_offset;
1665 checksums_ptr = checksums;
1667 integrity_sector_checksum(ic, sector, mem + pos, checksums_ptr);
1668 checksums_ptr += ic->tag_size;
1669 sectors_to_process -= ic->sectors_per_block;
1670 pos += ic->sectors_per_block << SECTOR_SHIFT;
1671 sector += ic->sectors_per_block;
1672 } while (pos < bv.bv_len && sectors_to_process && checksums != checksums_onstack);
1675 r = dm_integrity_rw_tag(ic, checksums, &dio->metadata_block, &dio->metadata_offset,
1676 checksums_ptr - checksums, dio->op == REQ_OP_READ ? TAG_CMP : TAG_WRITE);
1679 char b[BDEVNAME_SIZE];
1680 DMERR_LIMIT("%s: Checksum failed at sector 0x%llx", bio_devname(bio, b),
1681 (sector - ((r + ic->tag_size - 1) / ic->tag_size)));
1683 atomic64_inc(&ic->number_of_mismatches);
1685 if (likely(checksums != checksums_onstack))
1690 if (!sectors_to_process)
1693 if (unlikely(pos < bv.bv_len)) {
1694 bv.bv_offset += pos;
1700 if (likely(checksums != checksums_onstack))
1703 struct bio_integrity_payload *bip = dio->bio_details.bi_integrity;
1707 struct bvec_iter iter;
1708 unsigned data_to_process = dio->range.n_sectors;
1709 sector_to_block(ic, data_to_process);
1710 data_to_process *= ic->tag_size;
1712 bip_for_each_vec(biv, bip, iter) {
1716 BUG_ON(PageHighMem(biv.bv_page));
1717 tag = lowmem_page_address(biv.bv_page) + biv.bv_offset;
1718 this_len = min(biv.bv_len, data_to_process);
1719 r = dm_integrity_rw_tag(ic, tag, &dio->metadata_block, &dio->metadata_offset,
1720 this_len, dio->op == REQ_OP_READ ? TAG_READ : TAG_WRITE);
1723 data_to_process -= this_len;
1724 if (!data_to_process)
1733 dio->bi_status = errno_to_blk_status(r);
1737 static int dm_integrity_map(struct dm_target *ti, struct bio *bio)
1739 struct dm_integrity_c *ic = ti->private;
1740 struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
1741 struct bio_integrity_payload *bip;
1743 sector_t area, offset;
1747 dio->op = bio_op(bio);
1749 if (unlikely(dio->op == REQ_OP_DISCARD)) {
1750 if (ti->max_io_len) {
1751 sector_t sec = dm_target_offset(ti, bio->bi_iter.bi_sector);
1752 unsigned log2_max_io_len = __fls(ti->max_io_len);
1753 sector_t start_boundary = sec >> log2_max_io_len;
1754 sector_t end_boundary = (sec + bio_sectors(bio) - 1) >> log2_max_io_len;
1755 if (start_boundary < end_boundary) {
1756 sector_t len = ti->max_io_len - (sec & (ti->max_io_len - 1));
1757 dm_accept_partial_bio(bio, len);
1762 if (unlikely(bio->bi_opf & REQ_PREFLUSH)) {
1763 submit_flush_bio(ic, dio);
1764 return DM_MAPIO_SUBMITTED;
1767 dio->range.logical_sector = dm_target_offset(ti, bio->bi_iter.bi_sector);
1768 dio->fua = dio->op == REQ_OP_WRITE && bio->bi_opf & REQ_FUA;
1769 if (unlikely(dio->fua)) {
1771 * Don't pass down the FUA flag because we have to flush
1772 * disk cache anyway.
1774 bio->bi_opf &= ~REQ_FUA;
1776 if (unlikely(dio->range.logical_sector + bio_sectors(bio) > ic->provided_data_sectors)) {
1777 DMERR("Too big sector number: 0x%llx + 0x%x > 0x%llx",
1778 dio->range.logical_sector, bio_sectors(bio),
1779 ic->provided_data_sectors);
1780 return DM_MAPIO_KILL;
1782 if (unlikely((dio->range.logical_sector | bio_sectors(bio)) & (unsigned)(ic->sectors_per_block - 1))) {
1783 DMERR("Bio not aligned on %u sectors: 0x%llx, 0x%x",
1784 ic->sectors_per_block,
1785 dio->range.logical_sector, bio_sectors(bio));
1786 return DM_MAPIO_KILL;
1789 if (ic->sectors_per_block > 1 && likely(dio->op != REQ_OP_DISCARD)) {
1790 struct bvec_iter iter;
1792 bio_for_each_segment(bv, bio, iter) {
1793 if (unlikely(bv.bv_len & ((ic->sectors_per_block << SECTOR_SHIFT) - 1))) {
1794 DMERR("Bio vector (%u,%u) is not aligned on %u-sector boundary",
1795 bv.bv_offset, bv.bv_len, ic->sectors_per_block);
1796 return DM_MAPIO_KILL;
1801 bip = bio_integrity(bio);
1802 if (!ic->internal_hash) {
1804 unsigned wanted_tag_size = bio_sectors(bio) >> ic->sb->log2_sectors_per_block;
1805 if (ic->log2_tag_size >= 0)
1806 wanted_tag_size <<= ic->log2_tag_size;
1808 wanted_tag_size *= ic->tag_size;
1809 if (unlikely(wanted_tag_size != bip->bip_iter.bi_size)) {
1810 DMERR("Invalid integrity data size %u, expected %u",
1811 bip->bip_iter.bi_size, wanted_tag_size);
1812 return DM_MAPIO_KILL;
1816 if (unlikely(bip != NULL)) {
1817 DMERR("Unexpected integrity data when using internal hash");
1818 return DM_MAPIO_KILL;
1822 if (unlikely(ic->mode == 'R') && unlikely(dio->op != REQ_OP_READ))
1823 return DM_MAPIO_KILL;
1825 get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
1826 dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset, &dio->metadata_offset);
1827 bio->bi_iter.bi_sector = get_data_sector(ic, area, offset);
1829 dm_integrity_map_continue(dio, true);
1830 return DM_MAPIO_SUBMITTED;
1833 static bool __journal_read_write(struct dm_integrity_io *dio, struct bio *bio,
1834 unsigned journal_section, unsigned journal_entry)
1836 struct dm_integrity_c *ic = dio->ic;
1837 sector_t logical_sector;
1840 logical_sector = dio->range.logical_sector;
1841 n_sectors = dio->range.n_sectors;
1843 struct bio_vec bv = bio_iovec(bio);
1846 if (unlikely(bv.bv_len >> SECTOR_SHIFT > n_sectors))
1847 bv.bv_len = n_sectors << SECTOR_SHIFT;
1848 n_sectors -= bv.bv_len >> SECTOR_SHIFT;
1849 bio_advance_iter(bio, &bio->bi_iter, bv.bv_len);
1851 mem = kmap_atomic(bv.bv_page);
1852 if (likely(dio->op == REQ_OP_WRITE))
1853 flush_dcache_page(bv.bv_page);
1856 struct journal_entry *je = access_journal_entry(ic, journal_section, journal_entry);
1858 if (unlikely(dio->op == REQ_OP_READ)) {
1859 struct journal_sector *js;
1863 if (unlikely(journal_entry_is_inprogress(je))) {
1864 flush_dcache_page(bv.bv_page);
1867 __io_wait_event(ic->copy_to_journal_wait, !journal_entry_is_inprogress(je));
1871 BUG_ON(journal_entry_get_sector(je) != logical_sector);
1872 js = access_journal_data(ic, journal_section, journal_entry);
1873 mem_ptr = mem + bv.bv_offset;
1876 memcpy(mem_ptr, js, JOURNAL_SECTOR_DATA);
1877 *(commit_id_t *)(mem_ptr + JOURNAL_SECTOR_DATA) = je->last_bytes[s];
1879 mem_ptr += 1 << SECTOR_SHIFT;
1880 } while (++s < ic->sectors_per_block);
1881 #ifdef INTERNAL_VERIFY
1882 if (ic->internal_hash) {
1883 char checksums_onstack[max((size_t)HASH_MAX_DIGESTSIZE, MAX_TAG_SIZE)];
1885 integrity_sector_checksum(ic, logical_sector, mem + bv.bv_offset, checksums_onstack);
1886 if (unlikely(memcmp(checksums_onstack, journal_entry_tag(ic, je), ic->tag_size))) {
1887 DMERR_LIMIT("Checksum failed when reading from journal, at sector 0x%llx",
1894 if (!ic->internal_hash) {
1895 struct bio_integrity_payload *bip = bio_integrity(bio);
1896 unsigned tag_todo = ic->tag_size;
1897 char *tag_ptr = journal_entry_tag(ic, je);
1900 struct bio_vec biv = bvec_iter_bvec(bip->bip_vec, bip->bip_iter);
1901 unsigned tag_now = min(biv.bv_len, tag_todo);
1903 BUG_ON(PageHighMem(biv.bv_page));
1904 tag_addr = lowmem_page_address(biv.bv_page) + biv.bv_offset;
1905 if (likely(dio->op == REQ_OP_WRITE))
1906 memcpy(tag_ptr, tag_addr, tag_now);
1908 memcpy(tag_addr, tag_ptr, tag_now);
1909 bvec_iter_advance(bip->bip_vec, &bip->bip_iter, tag_now);
1911 tag_todo -= tag_now;
1912 } while (unlikely(tag_todo)); else {
1913 if (likely(dio->op == REQ_OP_WRITE))
1914 memset(tag_ptr, 0, tag_todo);
1918 if (likely(dio->op == REQ_OP_WRITE)) {
1919 struct journal_sector *js;
1922 js = access_journal_data(ic, journal_section, journal_entry);
1923 memcpy(js, mem + bv.bv_offset, ic->sectors_per_block << SECTOR_SHIFT);
1927 je->last_bytes[s] = js[s].commit_id;
1928 } while (++s < ic->sectors_per_block);
1930 if (ic->internal_hash) {
1931 unsigned digest_size = crypto_shash_digestsize(ic->internal_hash);
1932 if (unlikely(digest_size > ic->tag_size)) {
1933 char checksums_onstack[HASH_MAX_DIGESTSIZE];
1934 integrity_sector_checksum(ic, logical_sector, (char *)js, checksums_onstack);
1935 memcpy(journal_entry_tag(ic, je), checksums_onstack, ic->tag_size);
1937 integrity_sector_checksum(ic, logical_sector, (char *)js, journal_entry_tag(ic, je));
1940 journal_entry_set_sector(je, logical_sector);
1942 logical_sector += ic->sectors_per_block;
1945 if (unlikely(journal_entry == ic->journal_section_entries)) {
1948 wraparound_section(ic, &journal_section);
1951 bv.bv_offset += ic->sectors_per_block << SECTOR_SHIFT;
1952 } while (bv.bv_len -= ic->sectors_per_block << SECTOR_SHIFT);
1954 if (unlikely(dio->op == REQ_OP_READ))
1955 flush_dcache_page(bv.bv_page);
1957 } while (n_sectors);
1959 if (likely(dio->op == REQ_OP_WRITE)) {
1961 if (unlikely(waitqueue_active(&ic->copy_to_journal_wait)))
1962 wake_up(&ic->copy_to_journal_wait);
1963 if (READ_ONCE(ic->free_sectors) <= ic->free_sectors_threshold) {
1964 queue_work(ic->commit_wq, &ic->commit_work);
1966 schedule_autocommit(ic);
1969 remove_range(ic, &dio->range);
1972 if (unlikely(bio->bi_iter.bi_size)) {
1973 sector_t area, offset;
1975 dio->range.logical_sector = logical_sector;
1976 get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
1977 dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset, &dio->metadata_offset);
1984 static void dm_integrity_map_continue(struct dm_integrity_io *dio, bool from_map)
1986 struct dm_integrity_c *ic = dio->ic;
1987 struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1988 unsigned journal_section, journal_entry;
1989 unsigned journal_read_pos;
1990 struct completion read_comp;
1991 bool discard_retried = false;
1992 bool need_sync_io = ic->internal_hash && dio->op == REQ_OP_READ;
1993 if (unlikely(dio->op == REQ_OP_DISCARD) && ic->mode != 'D')
1994 need_sync_io = true;
1996 if (need_sync_io && from_map) {
1997 INIT_WORK(&dio->work, integrity_bio_wait);
1998 queue_work(ic->offload_wq, &dio->work);
2003 spin_lock_irq(&ic->endio_wait.lock);
2005 if (unlikely(dm_integrity_failed(ic))) {
2006 spin_unlock_irq(&ic->endio_wait.lock);
2010 dio->range.n_sectors = bio_sectors(bio);
2011 journal_read_pos = NOT_FOUND;
2012 if (ic->mode == 'J' && likely(dio->op != REQ_OP_DISCARD)) {
2013 if (dio->op == REQ_OP_WRITE) {
2014 unsigned next_entry, i, pos;
2015 unsigned ws, we, range_sectors;
2017 dio->range.n_sectors = min(dio->range.n_sectors,
2018 (sector_t)ic->free_sectors << ic->sb->log2_sectors_per_block);
2019 if (unlikely(!dio->range.n_sectors)) {
2021 goto offload_to_thread;
2022 sleep_on_endio_wait(ic);
2025 range_sectors = dio->range.n_sectors >> ic->sb->log2_sectors_per_block;
2026 ic->free_sectors -= range_sectors;
2027 journal_section = ic->free_section;
2028 journal_entry = ic->free_section_entry;
2030 next_entry = ic->free_section_entry + range_sectors;
2031 ic->free_section_entry = next_entry % ic->journal_section_entries;
2032 ic->free_section += next_entry / ic->journal_section_entries;
2033 ic->n_uncommitted_sections += next_entry / ic->journal_section_entries;
2034 wraparound_section(ic, &ic->free_section);
2036 pos = journal_section * ic->journal_section_entries + journal_entry;
2037 ws = journal_section;
2041 struct journal_entry *je;
2043 add_journal_node(ic, &ic->journal_tree[pos], dio->range.logical_sector + i);
2045 if (unlikely(pos >= ic->journal_entries))
2048 je = access_journal_entry(ic, ws, we);
2049 BUG_ON(!journal_entry_is_unused(je));
2050 journal_entry_set_inprogress(je);
2052 if (unlikely(we == ic->journal_section_entries)) {
2055 wraparound_section(ic, &ws);
2057 } while ((i += ic->sectors_per_block) < dio->range.n_sectors);
2059 spin_unlock_irq(&ic->endio_wait.lock);
2060 goto journal_read_write;
2062 sector_t next_sector;
2063 journal_read_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
2064 if (likely(journal_read_pos == NOT_FOUND)) {
2065 if (unlikely(dio->range.n_sectors > next_sector - dio->range.logical_sector))
2066 dio->range.n_sectors = next_sector - dio->range.logical_sector;
2069 unsigned jp = journal_read_pos + 1;
2070 for (i = ic->sectors_per_block; i < dio->range.n_sectors; i += ic->sectors_per_block, jp++) {
2071 if (!test_journal_node(ic, jp, dio->range.logical_sector + i))
2074 dio->range.n_sectors = i;
2078 if (unlikely(!add_new_range(ic, &dio->range, true))) {
2080 * We must not sleep in the request routine because it could
2081 * stall bios on current->bio_list.
2082 * So, we offload the bio to a workqueue if we have to sleep.
2086 spin_unlock_irq(&ic->endio_wait.lock);
2087 INIT_WORK(&dio->work, integrity_bio_wait);
2088 queue_work(ic->wait_wq, &dio->work);
2091 if (journal_read_pos != NOT_FOUND)
2092 dio->range.n_sectors = ic->sectors_per_block;
2093 wait_and_add_new_range(ic, &dio->range);
2095 * wait_and_add_new_range drops the spinlock, so the journal
2096 * may have been changed arbitrarily. We need to recheck.
2097 * To simplify the code, we restrict I/O size to just one block.
2099 if (journal_read_pos != NOT_FOUND) {
2100 sector_t next_sector;
2101 unsigned new_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
2102 if (unlikely(new_pos != journal_read_pos)) {
2103 remove_range_unlocked(ic, &dio->range);
2108 if (ic->mode == 'J' && likely(dio->op == REQ_OP_DISCARD) && !discard_retried) {
2109 sector_t next_sector;
2110 unsigned new_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
2111 if (unlikely(new_pos != NOT_FOUND) ||
2112 unlikely(next_sector < dio->range.logical_sector - dio->range.n_sectors)) {
2113 remove_range_unlocked(ic, &dio->range);
2114 spin_unlock_irq(&ic->endio_wait.lock);
2115 queue_work(ic->commit_wq, &ic->commit_work);
2116 flush_workqueue(ic->commit_wq);
2117 queue_work(ic->writer_wq, &ic->writer_work);
2118 flush_workqueue(ic->writer_wq);
2119 discard_retried = true;
2123 spin_unlock_irq(&ic->endio_wait.lock);
2125 if (unlikely(journal_read_pos != NOT_FOUND)) {
2126 journal_section = journal_read_pos / ic->journal_section_entries;
2127 journal_entry = journal_read_pos % ic->journal_section_entries;
2128 goto journal_read_write;
2131 if (ic->mode == 'B' && (dio->op == REQ_OP_WRITE || unlikely(dio->op == REQ_OP_DISCARD))) {
2132 if (!block_bitmap_op(ic, ic->may_write_bitmap, dio->range.logical_sector,
2133 dio->range.n_sectors, BITMAP_OP_TEST_ALL_SET)) {
2134 struct bitmap_block_status *bbs;
2136 bbs = sector_to_bitmap_block(ic, dio->range.logical_sector);
2137 spin_lock(&bbs->bio_queue_lock);
2138 bio_list_add(&bbs->bio_queue, bio);
2139 spin_unlock(&bbs->bio_queue_lock);
2140 queue_work(ic->writer_wq, &bbs->work);
2145 dio->in_flight = (atomic_t)ATOMIC_INIT(2);
2148 init_completion(&read_comp);
2149 dio->completion = &read_comp;
2151 dio->completion = NULL;
2153 dm_bio_record(&dio->bio_details, bio);
2154 bio_set_dev(bio, ic->dev->bdev);
2155 bio->bi_integrity = NULL;
2156 bio->bi_opf &= ~REQ_INTEGRITY;
2157 bio->bi_end_io = integrity_end_io;
2158 bio->bi_iter.bi_size = dio->range.n_sectors << SECTOR_SHIFT;
2160 if (unlikely(dio->op == REQ_OP_DISCARD) && likely(ic->mode != 'D')) {
2161 integrity_metadata(&dio->work);
2162 dm_integrity_flush_buffers(ic, false);
2164 dio->in_flight = (atomic_t)ATOMIC_INIT(1);
2165 dio->completion = NULL;
2167 submit_bio_noacct(bio);
2172 submit_bio_noacct(bio);
2175 wait_for_completion_io(&read_comp);
2176 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING) &&
2177 dio->range.logical_sector + dio->range.n_sectors > le64_to_cpu(ic->sb->recalc_sector))
2179 if (ic->mode == 'B') {
2180 if (!block_bitmap_op(ic, ic->recalc_bitmap, dio->range.logical_sector,
2181 dio->range.n_sectors, BITMAP_OP_TEST_ALL_CLEAR))
2185 if (likely(!bio->bi_status))
2186 integrity_metadata(&dio->work);
2192 INIT_WORK(&dio->work, integrity_metadata);
2193 queue_work(ic->metadata_wq, &dio->work);
2199 if (unlikely(__journal_read_write(dio, bio, journal_section, journal_entry)))
2202 do_endio_flush(ic, dio);
2206 static void integrity_bio_wait(struct work_struct *w)
2208 struct dm_integrity_io *dio = container_of(w, struct dm_integrity_io, work);
2210 dm_integrity_map_continue(dio, false);
2213 static void pad_uncommitted(struct dm_integrity_c *ic)
2215 if (ic->free_section_entry) {
2216 ic->free_sectors -= ic->journal_section_entries - ic->free_section_entry;
2217 ic->free_section_entry = 0;
2219 wraparound_section(ic, &ic->free_section);
2220 ic->n_uncommitted_sections++;
2222 if (WARN_ON(ic->journal_sections * ic->journal_section_entries !=
2223 (ic->n_uncommitted_sections + ic->n_committed_sections) *
2224 ic->journal_section_entries + ic->free_sectors)) {
2225 DMCRIT("journal_sections %u, journal_section_entries %u, "
2226 "n_uncommitted_sections %u, n_committed_sections %u, "
2227 "journal_section_entries %u, free_sectors %u",
2228 ic->journal_sections, ic->journal_section_entries,
2229 ic->n_uncommitted_sections, ic->n_committed_sections,
2230 ic->journal_section_entries, ic->free_sectors);
2234 static void integrity_commit(struct work_struct *w)
2236 struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, commit_work);
2237 unsigned commit_start, commit_sections;
2239 struct bio *flushes;
2241 del_timer(&ic->autocommit_timer);
2243 spin_lock_irq(&ic->endio_wait.lock);
2244 flushes = bio_list_get(&ic->flush_bio_list);
2245 if (unlikely(ic->mode != 'J')) {
2246 spin_unlock_irq(&ic->endio_wait.lock);
2247 dm_integrity_flush_buffers(ic, true);
2248 goto release_flush_bios;
2251 pad_uncommitted(ic);
2252 commit_start = ic->uncommitted_section;
2253 commit_sections = ic->n_uncommitted_sections;
2254 spin_unlock_irq(&ic->endio_wait.lock);
2256 if (!commit_sections)
2257 goto release_flush_bios;
2260 for (n = 0; n < commit_sections; n++) {
2261 for (j = 0; j < ic->journal_section_entries; j++) {
2262 struct journal_entry *je;
2263 je = access_journal_entry(ic, i, j);
2264 io_wait_event(ic->copy_to_journal_wait, !journal_entry_is_inprogress(je));
2266 for (j = 0; j < ic->journal_section_sectors; j++) {
2267 struct journal_sector *js;
2268 js = access_journal(ic, i, j);
2269 js->commit_id = dm_integrity_commit_id(ic, i, j, ic->commit_seq);
2272 if (unlikely(i >= ic->journal_sections))
2273 ic->commit_seq = next_commit_seq(ic->commit_seq);
2274 wraparound_section(ic, &i);
2278 write_journal(ic, commit_start, commit_sections);
2280 spin_lock_irq(&ic->endio_wait.lock);
2281 ic->uncommitted_section += commit_sections;
2282 wraparound_section(ic, &ic->uncommitted_section);
2283 ic->n_uncommitted_sections -= commit_sections;
2284 ic->n_committed_sections += commit_sections;
2285 spin_unlock_irq(&ic->endio_wait.lock);
2287 if (READ_ONCE(ic->free_sectors) <= ic->free_sectors_threshold)
2288 queue_work(ic->writer_wq, &ic->writer_work);
2292 struct bio *next = flushes->bi_next;
2293 flushes->bi_next = NULL;
2294 do_endio(ic, flushes);
2299 static void complete_copy_from_journal(unsigned long error, void *context)
2301 struct journal_io *io = context;
2302 struct journal_completion *comp = io->comp;
2303 struct dm_integrity_c *ic = comp->ic;
2304 remove_range(ic, &io->range);
2305 mempool_free(io, &ic->journal_io_mempool);
2306 if (unlikely(error != 0))
2307 dm_integrity_io_error(ic, "copying from journal", -EIO);
2308 complete_journal_op(comp);
2311 static void restore_last_bytes(struct dm_integrity_c *ic, struct journal_sector *js,
2312 struct journal_entry *je)
2316 js->commit_id = je->last_bytes[s];
2318 } while (++s < ic->sectors_per_block);
2321 static void do_journal_write(struct dm_integrity_c *ic, unsigned write_start,
2322 unsigned write_sections, bool from_replay)
2325 struct journal_completion comp;
2326 struct blk_plug plug;
2328 blk_start_plug(&plug);
2331 comp.in_flight = (atomic_t)ATOMIC_INIT(1);
2332 init_completion(&comp.comp);
2335 for (n = 0; n < write_sections; n++, i++, wraparound_section(ic, &i)) {
2336 #ifndef INTERNAL_VERIFY
2337 if (unlikely(from_replay))
2339 rw_section_mac(ic, i, false);
2340 for (j = 0; j < ic->journal_section_entries; j++) {
2341 struct journal_entry *je = access_journal_entry(ic, i, j);
2342 sector_t sec, area, offset;
2343 unsigned k, l, next_loop;
2344 sector_t metadata_block;
2345 unsigned metadata_offset;
2346 struct journal_io *io;
2348 if (journal_entry_is_unused(je))
2350 BUG_ON(unlikely(journal_entry_is_inprogress(je)) && !from_replay);
2351 sec = journal_entry_get_sector(je);
2352 if (unlikely(from_replay)) {
2353 if (unlikely(sec & (unsigned)(ic->sectors_per_block - 1))) {
2354 dm_integrity_io_error(ic, "invalid sector in journal", -EIO);
2355 sec &= ~(sector_t)(ic->sectors_per_block - 1);
2357 if (unlikely(sec >= ic->provided_data_sectors)) {
2358 journal_entry_set_unused(je);
2362 get_area_and_offset(ic, sec, &area, &offset);
2363 restore_last_bytes(ic, access_journal_data(ic, i, j), je);
2364 for (k = j + 1; k < ic->journal_section_entries; k++) {
2365 struct journal_entry *je2 = access_journal_entry(ic, i, k);
2366 sector_t sec2, area2, offset2;
2367 if (journal_entry_is_unused(je2))
2369 BUG_ON(unlikely(journal_entry_is_inprogress(je2)) && !from_replay);
2370 sec2 = journal_entry_get_sector(je2);
2371 if (unlikely(sec2 >= ic->provided_data_sectors))
2373 get_area_and_offset(ic, sec2, &area2, &offset2);
2374 if (area2 != area || offset2 != offset + ((k - j) << ic->sb->log2_sectors_per_block))
2376 restore_last_bytes(ic, access_journal_data(ic, i, k), je2);
2380 io = mempool_alloc(&ic->journal_io_mempool, GFP_NOIO);
2382 io->range.logical_sector = sec;
2383 io->range.n_sectors = (k - j) << ic->sb->log2_sectors_per_block;
2385 spin_lock_irq(&ic->endio_wait.lock);
2386 add_new_range_and_wait(ic, &io->range);
2388 if (likely(!from_replay)) {
2389 struct journal_node *section_node = &ic->journal_tree[i * ic->journal_section_entries];
2391 /* don't write if there is newer committed sector */
2392 while (j < k && find_newer_committed_node(ic, §ion_node[j])) {
2393 struct journal_entry *je2 = access_journal_entry(ic, i, j);
2395 journal_entry_set_unused(je2);
2396 remove_journal_node(ic, §ion_node[j]);
2398 sec += ic->sectors_per_block;
2399 offset += ic->sectors_per_block;
2401 while (j < k && find_newer_committed_node(ic, §ion_node[k - 1])) {
2402 struct journal_entry *je2 = access_journal_entry(ic, i, k - 1);
2404 journal_entry_set_unused(je2);
2405 remove_journal_node(ic, §ion_node[k - 1]);
2409 remove_range_unlocked(ic, &io->range);
2410 spin_unlock_irq(&ic->endio_wait.lock);
2411 mempool_free(io, &ic->journal_io_mempool);
2414 for (l = j; l < k; l++) {
2415 remove_journal_node(ic, §ion_node[l]);
2418 spin_unlock_irq(&ic->endio_wait.lock);
2420 metadata_block = get_metadata_sector_and_offset(ic, area, offset, &metadata_offset);
2421 for (l = j; l < k; l++) {
2423 struct journal_entry *je2 = access_journal_entry(ic, i, l);
2426 #ifndef INTERNAL_VERIFY
2427 unlikely(from_replay) &&
2429 ic->internal_hash) {
2430 char test_tag[max_t(size_t, HASH_MAX_DIGESTSIZE, MAX_TAG_SIZE)];
2432 integrity_sector_checksum(ic, sec + ((l - j) << ic->sb->log2_sectors_per_block),
2433 (char *)access_journal_data(ic, i, l), test_tag);
2434 if (unlikely(memcmp(test_tag, journal_entry_tag(ic, je2), ic->tag_size)))
2435 dm_integrity_io_error(ic, "tag mismatch when replaying journal", -EILSEQ);
2438 journal_entry_set_unused(je2);
2439 r = dm_integrity_rw_tag(ic, journal_entry_tag(ic, je2), &metadata_block, &metadata_offset,
2440 ic->tag_size, TAG_WRITE);
2442 dm_integrity_io_error(ic, "reading tags", r);
2446 atomic_inc(&comp.in_flight);
2447 copy_from_journal(ic, i, j << ic->sb->log2_sectors_per_block,
2448 (k - j) << ic->sb->log2_sectors_per_block,
2449 get_data_sector(ic, area, offset),
2450 complete_copy_from_journal, io);
2456 dm_bufio_write_dirty_buffers_async(ic->bufio);
2458 blk_finish_plug(&plug);
2460 complete_journal_op(&comp);
2461 wait_for_completion_io(&comp.comp);
2463 dm_integrity_flush_buffers(ic, true);
2466 static void integrity_writer(struct work_struct *w)
2468 struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, writer_work);
2469 unsigned write_start, write_sections;
2471 unsigned prev_free_sectors;
2473 /* the following test is not needed, but it tests the replay code */
2474 if (unlikely(dm_post_suspending(ic->ti)) && !ic->meta_dev)
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') {
2984 queue_work(ic->writer_wq, &ic->writer_work);
2985 drain_workqueue(ic->writer_wq);
2986 dm_integrity_flush_buffers(ic, true);
2989 if (ic->mode == 'B') {
2990 dm_integrity_flush_buffers(ic, true);
2992 /* set to 0 to test bitmap replay code */
2993 init_journal(ic, 0, ic->journal_sections, 0);
2994 ic->sb->flags &= ~cpu_to_le32(SB_FLAG_DIRTY_BITMAP);
2995 r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
2997 dm_integrity_io_error(ic, "writing superblock", r);
3001 BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
3003 ic->journal_uptodate = true;
3006 static void dm_integrity_resume(struct dm_target *ti)
3008 struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
3009 __u64 old_provided_data_sectors = le64_to_cpu(ic->sb->provided_data_sectors);
3012 DEBUG_print("resume\n");
3014 if (ic->provided_data_sectors != old_provided_data_sectors) {
3015 if (ic->provided_data_sectors > old_provided_data_sectors &&
3017 ic->sb->log2_blocks_per_bitmap_bit == ic->log2_blocks_per_bitmap_bit) {
3018 rw_journal_sectors(ic, REQ_OP_READ, 0, 0,
3019 ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3020 block_bitmap_op(ic, ic->journal, old_provided_data_sectors,
3021 ic->provided_data_sectors - old_provided_data_sectors, BITMAP_OP_SET);
3022 rw_journal_sectors(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, 0,
3023 ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3026 ic->sb->provided_data_sectors = cpu_to_le64(ic->provided_data_sectors);
3027 r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
3029 dm_integrity_io_error(ic, "writing superblock", r);
3032 if (ic->sb->flags & cpu_to_le32(SB_FLAG_DIRTY_BITMAP)) {
3033 DEBUG_print("resume dirty_bitmap\n");
3034 rw_journal_sectors(ic, REQ_OP_READ, 0, 0,
3035 ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3036 if (ic->mode == 'B') {
3037 if (ic->sb->log2_blocks_per_bitmap_bit == ic->log2_blocks_per_bitmap_bit) {
3038 block_bitmap_copy(ic, ic->recalc_bitmap, ic->journal);
3039 block_bitmap_copy(ic, ic->may_write_bitmap, ic->journal);
3040 if (!block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors,
3041 BITMAP_OP_TEST_ALL_CLEAR)) {
3042 ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
3043 ic->sb->recalc_sector = cpu_to_le64(0);
3046 DEBUG_print("non-matching blocks_per_bitmap_bit: %u, %u\n",
3047 ic->sb->log2_blocks_per_bitmap_bit, ic->log2_blocks_per_bitmap_bit);
3048 ic->sb->log2_blocks_per_bitmap_bit = ic->log2_blocks_per_bitmap_bit;
3049 block_bitmap_op(ic, ic->recalc_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_SET);
3050 block_bitmap_op(ic, ic->may_write_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_SET);
3051 block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors, BITMAP_OP_SET);
3052 rw_journal_sectors(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, 0,
3053 ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3054 ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
3055 ic->sb->recalc_sector = cpu_to_le64(0);
3058 if (!(ic->sb->log2_blocks_per_bitmap_bit == ic->log2_blocks_per_bitmap_bit &&
3059 block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors, BITMAP_OP_TEST_ALL_CLEAR))) {
3060 ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
3061 ic->sb->recalc_sector = cpu_to_le64(0);
3063 init_journal(ic, 0, ic->journal_sections, 0);
3065 ic->sb->flags &= ~cpu_to_le32(SB_FLAG_DIRTY_BITMAP);
3067 r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
3069 dm_integrity_io_error(ic, "writing superblock", r);
3072 if (ic->mode == 'B') {
3073 ic->sb->flags |= cpu_to_le32(SB_FLAG_DIRTY_BITMAP);
3074 ic->sb->log2_blocks_per_bitmap_bit = ic->log2_blocks_per_bitmap_bit;
3075 r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
3077 dm_integrity_io_error(ic, "writing superblock", r);
3079 block_bitmap_op(ic, ic->journal, 0, ic->provided_data_sectors, BITMAP_OP_CLEAR);
3080 block_bitmap_op(ic, ic->recalc_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_CLEAR);
3081 block_bitmap_op(ic, ic->may_write_bitmap, 0, ic->provided_data_sectors, BITMAP_OP_CLEAR);
3082 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING) &&
3083 le64_to_cpu(ic->sb->recalc_sector) < ic->provided_data_sectors) {
3084 block_bitmap_op(ic, ic->journal, le64_to_cpu(ic->sb->recalc_sector),
3085 ic->provided_data_sectors - le64_to_cpu(ic->sb->recalc_sector), BITMAP_OP_SET);
3086 block_bitmap_op(ic, ic->recalc_bitmap, le64_to_cpu(ic->sb->recalc_sector),
3087 ic->provided_data_sectors - le64_to_cpu(ic->sb->recalc_sector), BITMAP_OP_SET);
3088 block_bitmap_op(ic, ic->may_write_bitmap, le64_to_cpu(ic->sb->recalc_sector),
3089 ic->provided_data_sectors - le64_to_cpu(ic->sb->recalc_sector), BITMAP_OP_SET);
3091 rw_journal_sectors(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, 0,
3092 ic->n_bitmap_blocks * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT), NULL);
3096 DEBUG_print("testing recalc: %x\n", ic->sb->flags);
3097 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
3098 __u64 recalc_pos = le64_to_cpu(ic->sb->recalc_sector);
3099 DEBUG_print("recalc pos: %llx / %llx\n", recalc_pos, ic->provided_data_sectors);
3100 if (recalc_pos < ic->provided_data_sectors) {
3101 queue_work(ic->recalc_wq, &ic->recalc_work);
3102 } else if (recalc_pos > ic->provided_data_sectors) {
3103 ic->sb->recalc_sector = cpu_to_le64(ic->provided_data_sectors);
3104 recalc_write_super(ic);
3108 ic->reboot_notifier.notifier_call = dm_integrity_reboot;
3109 ic->reboot_notifier.next = NULL;
3110 ic->reboot_notifier.priority = INT_MAX - 1; /* be notified after md and before hardware drivers */
3111 WARN_ON(register_reboot_notifier(&ic->reboot_notifier));
3114 /* set to 1 to stress test synchronous mode */
3115 dm_integrity_enter_synchronous_mode(ic);
3119 static void dm_integrity_status(struct dm_target *ti, status_type_t type,
3120 unsigned status_flags, char *result, unsigned maxlen)
3122 struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
3127 case STATUSTYPE_INFO:
3129 (unsigned long long)atomic64_read(&ic->number_of_mismatches),
3130 ic->provided_data_sectors);
3131 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
3132 DMEMIT(" %llu", le64_to_cpu(ic->sb->recalc_sector));
3137 case STATUSTYPE_TABLE: {
3138 __u64 watermark_percentage = (__u64)(ic->journal_entries - ic->free_sectors_threshold) * 100;
3139 watermark_percentage += ic->journal_entries / 2;
3140 do_div(watermark_percentage, ic->journal_entries);
3142 arg_count += !!ic->meta_dev;
3143 arg_count += ic->sectors_per_block != 1;
3144 arg_count += !!(ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING));
3145 arg_count += ic->discard;
3146 arg_count += ic->mode == 'J';
3147 arg_count += ic->mode == 'J';
3148 arg_count += ic->mode == 'B';
3149 arg_count += ic->mode == 'B';
3150 arg_count += !!ic->internal_hash_alg.alg_string;
3151 arg_count += !!ic->journal_crypt_alg.alg_string;
3152 arg_count += !!ic->journal_mac_alg.alg_string;
3153 arg_count += (ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING)) != 0;
3154 arg_count += ic->legacy_recalculate;
3155 DMEMIT("%s %llu %u %c %u", ic->dev->name, ic->start,
3156 ic->tag_size, ic->mode, arg_count);
3158 DMEMIT(" meta_device:%s", ic->meta_dev->name);
3159 if (ic->sectors_per_block != 1)
3160 DMEMIT(" block_size:%u", ic->sectors_per_block << SECTOR_SHIFT);
3161 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
3162 DMEMIT(" recalculate");
3164 DMEMIT(" allow_discards");
3165 DMEMIT(" journal_sectors:%u", ic->initial_sectors - SB_SECTORS);
3166 DMEMIT(" interleave_sectors:%u", 1U << ic->sb->log2_interleave_sectors);
3167 DMEMIT(" buffer_sectors:%u", 1U << ic->log2_buffer_sectors);
3168 if (ic->mode == 'J') {
3169 DMEMIT(" journal_watermark:%u", (unsigned)watermark_percentage);
3170 DMEMIT(" commit_time:%u", ic->autocommit_msec);
3172 if (ic->mode == 'B') {
3173 DMEMIT(" sectors_per_bit:%llu", (sector_t)ic->sectors_per_block << ic->log2_blocks_per_bitmap_bit);
3174 DMEMIT(" bitmap_flush_interval:%u", jiffies_to_msecs(ic->bitmap_flush_interval));
3176 if ((ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING)) != 0)
3177 DMEMIT(" fix_padding");
3178 if (ic->legacy_recalculate)
3179 DMEMIT(" legacy_recalculate");
3181 #define EMIT_ALG(a, n) \
3183 if (ic->a.alg_string) { \
3184 DMEMIT(" %s:%s", n, ic->a.alg_string); \
3185 if (ic->a.key_string) \
3186 DMEMIT(":%s", ic->a.key_string);\
3189 EMIT_ALG(internal_hash_alg, "internal_hash");
3190 EMIT_ALG(journal_crypt_alg, "journal_crypt");
3191 EMIT_ALG(journal_mac_alg, "journal_mac");
3197 static int dm_integrity_iterate_devices(struct dm_target *ti,
3198 iterate_devices_callout_fn fn, void *data)
3200 struct dm_integrity_c *ic = ti->private;
3203 return fn(ti, ic->dev, ic->start + ic->initial_sectors + ic->metadata_run, ti->len, data);
3205 return fn(ti, ic->dev, 0, ti->len, data);
3208 static void dm_integrity_io_hints(struct dm_target *ti, struct queue_limits *limits)
3210 struct dm_integrity_c *ic = ti->private;
3212 if (ic->sectors_per_block > 1) {
3213 limits->logical_block_size = ic->sectors_per_block << SECTOR_SHIFT;
3214 limits->physical_block_size = ic->sectors_per_block << SECTOR_SHIFT;
3215 blk_limits_io_min(limits, ic->sectors_per_block << SECTOR_SHIFT);
3219 static void calculate_journal_section_size(struct dm_integrity_c *ic)
3221 unsigned sector_space = JOURNAL_SECTOR_DATA;
3223 ic->journal_sections = le32_to_cpu(ic->sb->journal_sections);
3224 ic->journal_entry_size = roundup(offsetof(struct journal_entry, last_bytes[ic->sectors_per_block]) + ic->tag_size,
3225 JOURNAL_ENTRY_ROUNDUP);
3227 if (ic->sb->flags & cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC))
3228 sector_space -= JOURNAL_MAC_PER_SECTOR;
3229 ic->journal_entries_per_sector = sector_space / ic->journal_entry_size;
3230 ic->journal_section_entries = ic->journal_entries_per_sector * JOURNAL_BLOCK_SECTORS;
3231 ic->journal_section_sectors = (ic->journal_section_entries << ic->sb->log2_sectors_per_block) + JOURNAL_BLOCK_SECTORS;
3232 ic->journal_entries = ic->journal_section_entries * ic->journal_sections;
3235 static int calculate_device_limits(struct dm_integrity_c *ic)
3237 __u64 initial_sectors;
3239 calculate_journal_section_size(ic);
3240 initial_sectors = SB_SECTORS + (__u64)ic->journal_section_sectors * ic->journal_sections;
3241 if (initial_sectors + METADATA_PADDING_SECTORS >= ic->meta_device_sectors || initial_sectors > UINT_MAX)
3243 ic->initial_sectors = initial_sectors;
3245 if (!ic->meta_dev) {
3246 sector_t last_sector, last_area, last_offset;
3248 /* we have to maintain excessive padding for compatibility with existing volumes */
3249 __u64 metadata_run_padding =
3250 ic->sb->flags & cpu_to_le32(SB_FLAG_FIXED_PADDING) ?
3251 (__u64)(METADATA_PADDING_SECTORS << SECTOR_SHIFT) :
3252 (__u64)(1 << SECTOR_SHIFT << METADATA_PADDING_SECTORS);
3254 ic->metadata_run = round_up((__u64)ic->tag_size << (ic->sb->log2_interleave_sectors - ic->sb->log2_sectors_per_block),
3255 metadata_run_padding) >> SECTOR_SHIFT;
3256 if (!(ic->metadata_run & (ic->metadata_run - 1)))
3257 ic->log2_metadata_run = __ffs(ic->metadata_run);
3259 ic->log2_metadata_run = -1;
3261 get_area_and_offset(ic, ic->provided_data_sectors - 1, &last_area, &last_offset);
3262 last_sector = get_data_sector(ic, last_area, last_offset);
3263 if (last_sector < ic->start || last_sector >= ic->meta_device_sectors)
3266 __u64 meta_size = (ic->provided_data_sectors >> ic->sb->log2_sectors_per_block) * ic->tag_size;
3267 meta_size = (meta_size + ((1U << (ic->log2_buffer_sectors + SECTOR_SHIFT)) - 1))
3268 >> (ic->log2_buffer_sectors + SECTOR_SHIFT);
3269 meta_size <<= ic->log2_buffer_sectors;
3270 if (ic->initial_sectors + meta_size < ic->initial_sectors ||
3271 ic->initial_sectors + meta_size > ic->meta_device_sectors)
3273 ic->metadata_run = 1;
3274 ic->log2_metadata_run = 0;
3280 static void get_provided_data_sectors(struct dm_integrity_c *ic)
3282 if (!ic->meta_dev) {
3284 ic->provided_data_sectors = 0;
3285 for (test_bit = fls64(ic->meta_device_sectors) - 1; test_bit >= 3; test_bit--) {
3286 __u64 prev_data_sectors = ic->provided_data_sectors;
3288 ic->provided_data_sectors |= (sector_t)1 << test_bit;
3289 if (calculate_device_limits(ic))
3290 ic->provided_data_sectors = prev_data_sectors;
3293 ic->provided_data_sectors = ic->data_device_sectors;
3294 ic->provided_data_sectors &= ~(sector_t)(ic->sectors_per_block - 1);
3298 static int initialize_superblock(struct dm_integrity_c *ic, unsigned journal_sectors, unsigned interleave_sectors)
3300 unsigned journal_sections;
3303 memset(ic->sb, 0, SB_SECTORS << SECTOR_SHIFT);
3304 memcpy(ic->sb->magic, SB_MAGIC, 8);
3305 ic->sb->integrity_tag_size = cpu_to_le16(ic->tag_size);
3306 ic->sb->log2_sectors_per_block = __ffs(ic->sectors_per_block);
3307 if (ic->journal_mac_alg.alg_string)
3308 ic->sb->flags |= cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC);
3310 calculate_journal_section_size(ic);
3311 journal_sections = journal_sectors / ic->journal_section_sectors;
3312 if (!journal_sections)
3313 journal_sections = 1;
3315 if (!ic->meta_dev) {
3316 if (ic->fix_padding)
3317 ic->sb->flags |= cpu_to_le32(SB_FLAG_FIXED_PADDING);
3318 ic->sb->journal_sections = cpu_to_le32(journal_sections);
3319 if (!interleave_sectors)
3320 interleave_sectors = DEFAULT_INTERLEAVE_SECTORS;
3321 ic->sb->log2_interleave_sectors = __fls(interleave_sectors);
3322 ic->sb->log2_interleave_sectors = max((__u8)MIN_LOG2_INTERLEAVE_SECTORS, ic->sb->log2_interleave_sectors);
3323 ic->sb->log2_interleave_sectors = min((__u8)MAX_LOG2_INTERLEAVE_SECTORS, ic->sb->log2_interleave_sectors);
3325 get_provided_data_sectors(ic);
3326 if (!ic->provided_data_sectors)
3329 ic->sb->log2_interleave_sectors = 0;
3331 get_provided_data_sectors(ic);
3332 if (!ic->provided_data_sectors)
3336 ic->sb->journal_sections = cpu_to_le32(0);
3337 for (test_bit = fls(journal_sections) - 1; test_bit >= 0; test_bit--) {
3338 __u32 prev_journal_sections = le32_to_cpu(ic->sb->journal_sections);
3339 __u32 test_journal_sections = prev_journal_sections | (1U << test_bit);
3340 if (test_journal_sections > journal_sections)
3342 ic->sb->journal_sections = cpu_to_le32(test_journal_sections);
3343 if (calculate_device_limits(ic))
3344 ic->sb->journal_sections = cpu_to_le32(prev_journal_sections);
3347 if (!le32_to_cpu(ic->sb->journal_sections)) {
3348 if (ic->log2_buffer_sectors > 3) {
3349 ic->log2_buffer_sectors--;
3350 goto try_smaller_buffer;
3356 ic->sb->provided_data_sectors = cpu_to_le64(ic->provided_data_sectors);
3363 static void dm_integrity_set(struct dm_target *ti, struct dm_integrity_c *ic)
3365 struct gendisk *disk = dm_disk(dm_table_get_md(ti->table));
3366 struct blk_integrity bi;
3368 memset(&bi, 0, sizeof(bi));
3369 bi.profile = &dm_integrity_profile;
3370 bi.tuple_size = ic->tag_size;
3371 bi.tag_size = bi.tuple_size;
3372 bi.interval_exp = ic->sb->log2_sectors_per_block + SECTOR_SHIFT;
3374 blk_integrity_register(disk, &bi);
3375 blk_queue_max_integrity_segments(disk->queue, UINT_MAX);
3378 static void dm_integrity_free_page_list(struct page_list *pl)
3384 for (i = 0; pl[i].page; i++)
3385 __free_page(pl[i].page);
3389 static struct page_list *dm_integrity_alloc_page_list(unsigned n_pages)
3391 struct page_list *pl;
3394 pl = kvmalloc_array(n_pages + 1, sizeof(struct page_list), GFP_KERNEL | __GFP_ZERO);
3398 for (i = 0; i < n_pages; i++) {
3399 pl[i].page = alloc_page(GFP_KERNEL);
3401 dm_integrity_free_page_list(pl);
3405 pl[i - 1].next = &pl[i];
3413 static void dm_integrity_free_journal_scatterlist(struct dm_integrity_c *ic, struct scatterlist **sl)
3416 for (i = 0; i < ic->journal_sections; i++)
3421 static struct scatterlist **dm_integrity_alloc_journal_scatterlist(struct dm_integrity_c *ic,
3422 struct page_list *pl)
3424 struct scatterlist **sl;
3427 sl = kvmalloc_array(ic->journal_sections,
3428 sizeof(struct scatterlist *),
3429 GFP_KERNEL | __GFP_ZERO);
3433 for (i = 0; i < ic->journal_sections; i++) {
3434 struct scatterlist *s;
3435 unsigned start_index, start_offset;
3436 unsigned end_index, end_offset;
3440 page_list_location(ic, i, 0, &start_index, &start_offset);
3441 page_list_location(ic, i, ic->journal_section_sectors - 1,
3442 &end_index, &end_offset);
3444 n_pages = (end_index - start_index + 1);
3446 s = kvmalloc_array(n_pages, sizeof(struct scatterlist),
3449 dm_integrity_free_journal_scatterlist(ic, sl);
3453 sg_init_table(s, n_pages);
3454 for (idx = start_index; idx <= end_index; idx++) {
3455 char *va = lowmem_page_address(pl[idx].page);
3456 unsigned start = 0, end = PAGE_SIZE;
3457 if (idx == start_index)
3458 start = start_offset;
3459 if (idx == end_index)
3460 end = end_offset + (1 << SECTOR_SHIFT);
3461 sg_set_buf(&s[idx - start_index], va + start, end - start);
3470 static void free_alg(struct alg_spec *a)
3472 kfree_sensitive(a->alg_string);
3473 kfree_sensitive(a->key);
3474 memset(a, 0, sizeof *a);
3477 static int get_alg_and_key(const char *arg, struct alg_spec *a, char **error, char *error_inval)
3483 a->alg_string = kstrdup(strchr(arg, ':') + 1, GFP_KERNEL);
3487 k = strchr(a->alg_string, ':');
3490 a->key_string = k + 1;
3491 if (strlen(a->key_string) & 1)
3494 a->key_size = strlen(a->key_string) / 2;
3495 a->key = kmalloc(a->key_size, GFP_KERNEL);
3498 if (hex2bin(a->key, a->key_string, a->key_size))
3504 *error = error_inval;
3507 *error = "Out of memory for an argument";
3511 static int get_mac(struct crypto_shash **hash, struct alg_spec *a, char **error,
3512 char *error_alg, char *error_key)
3516 if (a->alg_string) {
3517 *hash = crypto_alloc_shash(a->alg_string, 0, CRYPTO_ALG_ALLOCATES_MEMORY);
3518 if (IS_ERR(*hash)) {
3526 r = crypto_shash_setkey(*hash, a->key, a->key_size);
3531 } else if (crypto_shash_get_flags(*hash) & CRYPTO_TFM_NEED_KEY) {
3540 static int create_journal(struct dm_integrity_c *ic, char **error)
3544 __u64 journal_pages, journal_desc_size, journal_tree_size;
3545 unsigned char *crypt_data = NULL, *crypt_iv = NULL;
3546 struct skcipher_request *req = NULL;
3548 ic->commit_ids[0] = cpu_to_le64(0x1111111111111111ULL);
3549 ic->commit_ids[1] = cpu_to_le64(0x2222222222222222ULL);
3550 ic->commit_ids[2] = cpu_to_le64(0x3333333333333333ULL);
3551 ic->commit_ids[3] = cpu_to_le64(0x4444444444444444ULL);
3553 journal_pages = roundup((__u64)ic->journal_sections * ic->journal_section_sectors,
3554 PAGE_SIZE >> SECTOR_SHIFT) >> (PAGE_SHIFT - SECTOR_SHIFT);
3555 journal_desc_size = journal_pages * sizeof(struct page_list);
3556 if (journal_pages >= totalram_pages() - totalhigh_pages() || journal_desc_size > ULONG_MAX) {
3557 *error = "Journal doesn't fit into memory";
3561 ic->journal_pages = journal_pages;
3563 ic->journal = dm_integrity_alloc_page_list(ic->journal_pages);
3565 *error = "Could not allocate memory for journal";
3569 if (ic->journal_crypt_alg.alg_string) {
3570 unsigned ivsize, blocksize;
3571 struct journal_completion comp;
3574 ic->journal_crypt = crypto_alloc_skcipher(ic->journal_crypt_alg.alg_string, 0, CRYPTO_ALG_ALLOCATES_MEMORY);
3575 if (IS_ERR(ic->journal_crypt)) {
3576 *error = "Invalid journal cipher";
3577 r = PTR_ERR(ic->journal_crypt);
3578 ic->journal_crypt = NULL;
3581 ivsize = crypto_skcipher_ivsize(ic->journal_crypt);
3582 blocksize = crypto_skcipher_blocksize(ic->journal_crypt);
3584 if (ic->journal_crypt_alg.key) {
3585 r = crypto_skcipher_setkey(ic->journal_crypt, ic->journal_crypt_alg.key,
3586 ic->journal_crypt_alg.key_size);
3588 *error = "Error setting encryption key";
3592 DEBUG_print("cipher %s, block size %u iv size %u\n",
3593 ic->journal_crypt_alg.alg_string, blocksize, ivsize);
3595 ic->journal_io = dm_integrity_alloc_page_list(ic->journal_pages);
3596 if (!ic->journal_io) {
3597 *error = "Could not allocate memory for journal io";
3602 if (blocksize == 1) {
3603 struct scatterlist *sg;
3605 req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
3607 *error = "Could not allocate crypt request";
3612 crypt_iv = kzalloc(ivsize, GFP_KERNEL);
3614 *error = "Could not allocate iv";
3619 ic->journal_xor = dm_integrity_alloc_page_list(ic->journal_pages);
3620 if (!ic->journal_xor) {
3621 *error = "Could not allocate memory for journal xor";
3626 sg = kvmalloc_array(ic->journal_pages + 1,
3627 sizeof(struct scatterlist),
3630 *error = "Unable to allocate sg list";
3634 sg_init_table(sg, ic->journal_pages + 1);
3635 for (i = 0; i < ic->journal_pages; i++) {
3636 char *va = lowmem_page_address(ic->journal_xor[i].page);
3638 sg_set_buf(&sg[i], va, PAGE_SIZE);
3640 sg_set_buf(&sg[i], &ic->commit_ids, sizeof ic->commit_ids);
3642 skcipher_request_set_crypt(req, sg, sg,
3643 PAGE_SIZE * ic->journal_pages + sizeof ic->commit_ids, crypt_iv);
3644 init_completion(&comp.comp);
3645 comp.in_flight = (atomic_t)ATOMIC_INIT(1);
3646 if (do_crypt(true, req, &comp))
3647 wait_for_completion(&comp.comp);
3649 r = dm_integrity_failed(ic);
3651 *error = "Unable to encrypt journal";
3654 DEBUG_bytes(lowmem_page_address(ic->journal_xor[0].page), 64, "xor data");
3656 crypto_free_skcipher(ic->journal_crypt);
3657 ic->journal_crypt = NULL;
3659 unsigned crypt_len = roundup(ivsize, blocksize);
3661 req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
3663 *error = "Could not allocate crypt request";
3668 crypt_iv = kmalloc(ivsize, GFP_KERNEL);
3670 *error = "Could not allocate iv";
3675 crypt_data = kmalloc(crypt_len, GFP_KERNEL);
3677 *error = "Unable to allocate crypt data";
3682 ic->journal_scatterlist = dm_integrity_alloc_journal_scatterlist(ic, ic->journal);
3683 if (!ic->journal_scatterlist) {
3684 *error = "Unable to allocate sg list";
3688 ic->journal_io_scatterlist = dm_integrity_alloc_journal_scatterlist(ic, ic->journal_io);
3689 if (!ic->journal_io_scatterlist) {
3690 *error = "Unable to allocate sg list";
3694 ic->sk_requests = kvmalloc_array(ic->journal_sections,
3695 sizeof(struct skcipher_request *),
3696 GFP_KERNEL | __GFP_ZERO);
3697 if (!ic->sk_requests) {
3698 *error = "Unable to allocate sk requests";
3702 for (i = 0; i < ic->journal_sections; i++) {
3703 struct scatterlist sg;
3704 struct skcipher_request *section_req;
3705 __u32 section_le = cpu_to_le32(i);
3707 memset(crypt_iv, 0x00, ivsize);
3708 memset(crypt_data, 0x00, crypt_len);
3709 memcpy(crypt_data, §ion_le, min((size_t)crypt_len, sizeof(section_le)));
3711 sg_init_one(&sg, crypt_data, crypt_len);
3712 skcipher_request_set_crypt(req, &sg, &sg, crypt_len, crypt_iv);
3713 init_completion(&comp.comp);
3714 comp.in_flight = (atomic_t)ATOMIC_INIT(1);
3715 if (do_crypt(true, req, &comp))
3716 wait_for_completion(&comp.comp);
3718 r = dm_integrity_failed(ic);
3720 *error = "Unable to generate iv";
3724 section_req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
3726 *error = "Unable to allocate crypt request";
3730 section_req->iv = kmalloc_array(ivsize, 2,
3732 if (!section_req->iv) {
3733 skcipher_request_free(section_req);
3734 *error = "Unable to allocate iv";
3738 memcpy(section_req->iv + ivsize, crypt_data, ivsize);
3739 section_req->cryptlen = (size_t)ic->journal_section_sectors << SECTOR_SHIFT;
3740 ic->sk_requests[i] = section_req;
3741 DEBUG_bytes(crypt_data, ivsize, "iv(%u)", i);
3746 for (i = 0; i < N_COMMIT_IDS; i++) {
3749 for (j = 0; j < i; j++) {
3750 if (ic->commit_ids[j] == ic->commit_ids[i]) {
3751 ic->commit_ids[i] = cpu_to_le64(le64_to_cpu(ic->commit_ids[i]) + 1);
3752 goto retest_commit_id;
3755 DEBUG_print("commit id %u: %016llx\n", i, ic->commit_ids[i]);
3758 journal_tree_size = (__u64)ic->journal_entries * sizeof(struct journal_node);
3759 if (journal_tree_size > ULONG_MAX) {
3760 *error = "Journal doesn't fit into memory";
3764 ic->journal_tree = kvmalloc(journal_tree_size, GFP_KERNEL);
3765 if (!ic->journal_tree) {
3766 *error = "Could not allocate memory for journal tree";
3772 skcipher_request_free(req);
3778 * Construct a integrity mapping
3782 * offset from the start of the device
3784 * D - direct writes, J - journal writes, B - bitmap mode, R - recovery mode
3785 * number of optional arguments
3786 * optional arguments:
3788 * interleave_sectors
3795 * bitmap_flush_interval
3801 static int dm_integrity_ctr(struct dm_target *ti, unsigned argc, char **argv)
3803 struct dm_integrity_c *ic;
3806 unsigned extra_args;
3807 struct dm_arg_set as;
3808 static const struct dm_arg _args[] = {
3809 {0, 16, "Invalid number of feature args"},
3811 unsigned journal_sectors, interleave_sectors, buffer_sectors, journal_watermark, sync_msec;
3812 bool should_write_sb;
3814 unsigned long long start;
3815 __s8 log2_sectors_per_bitmap_bit = -1;
3816 __s8 log2_blocks_per_bitmap_bit;
3817 __u64 bits_in_journal;
3818 __u64 n_bitmap_bits;
3820 #define DIRECT_ARGUMENTS 4
3822 if (argc <= DIRECT_ARGUMENTS) {
3823 ti->error = "Invalid argument count";
3827 ic = kzalloc(sizeof(struct dm_integrity_c), GFP_KERNEL);
3829 ti->error = "Cannot allocate integrity context";
3833 ti->per_io_data_size = sizeof(struct dm_integrity_io);
3836 ic->in_progress = RB_ROOT;
3837 INIT_LIST_HEAD(&ic->wait_list);
3838 init_waitqueue_head(&ic->endio_wait);
3839 bio_list_init(&ic->flush_bio_list);
3840 init_waitqueue_head(&ic->copy_to_journal_wait);
3841 init_completion(&ic->crypto_backoff);
3842 atomic64_set(&ic->number_of_mismatches, 0);
3843 ic->bitmap_flush_interval = BITMAP_FLUSH_INTERVAL;
3845 r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &ic->dev);
3847 ti->error = "Device lookup failed";
3851 if (sscanf(argv[1], "%llu%c", &start, &dummy) != 1 || start != (sector_t)start) {
3852 ti->error = "Invalid starting offset";
3858 if (strcmp(argv[2], "-")) {
3859 if (sscanf(argv[2], "%u%c", &ic->tag_size, &dummy) != 1 || !ic->tag_size) {
3860 ti->error = "Invalid tag size";
3866 if (!strcmp(argv[3], "J") || !strcmp(argv[3], "B") ||
3867 !strcmp(argv[3], "D") || !strcmp(argv[3], "R")) {
3868 ic->mode = argv[3][0];
3870 ti->error = "Invalid mode (expecting J, B, D, R)";
3875 journal_sectors = 0;
3876 interleave_sectors = DEFAULT_INTERLEAVE_SECTORS;
3877 buffer_sectors = DEFAULT_BUFFER_SECTORS;
3878 journal_watermark = DEFAULT_JOURNAL_WATERMARK;
3879 sync_msec = DEFAULT_SYNC_MSEC;
3880 ic->sectors_per_block = 1;
3882 as.argc = argc - DIRECT_ARGUMENTS;
3883 as.argv = argv + DIRECT_ARGUMENTS;
3884 r = dm_read_arg_group(_args, &as, &extra_args, &ti->error);
3888 while (extra_args--) {
3889 const char *opt_string;
3891 unsigned long long llval;
3892 opt_string = dm_shift_arg(&as);
3895 ti->error = "Not enough feature arguments";
3898 if (sscanf(opt_string, "journal_sectors:%u%c", &val, &dummy) == 1)
3899 journal_sectors = val ? val : 1;
3900 else if (sscanf(opt_string, "interleave_sectors:%u%c", &val, &dummy) == 1)
3901 interleave_sectors = val;
3902 else if (sscanf(opt_string, "buffer_sectors:%u%c", &val, &dummy) == 1)
3903 buffer_sectors = val;
3904 else if (sscanf(opt_string, "journal_watermark:%u%c", &val, &dummy) == 1 && val <= 100)
3905 journal_watermark = val;
3906 else if (sscanf(opt_string, "commit_time:%u%c", &val, &dummy) == 1)
3908 else if (!strncmp(opt_string, "meta_device:", strlen("meta_device:"))) {
3910 dm_put_device(ti, ic->meta_dev);
3911 ic->meta_dev = NULL;
3913 r = dm_get_device(ti, strchr(opt_string, ':') + 1,
3914 dm_table_get_mode(ti->table), &ic->meta_dev);
3916 ti->error = "Device lookup failed";
3919 } else if (sscanf(opt_string, "block_size:%u%c", &val, &dummy) == 1) {
3920 if (val < 1 << SECTOR_SHIFT ||
3921 val > MAX_SECTORS_PER_BLOCK << SECTOR_SHIFT ||
3924 ti->error = "Invalid block_size argument";
3927 ic->sectors_per_block = val >> SECTOR_SHIFT;
3928 } else if (sscanf(opt_string, "sectors_per_bit:%llu%c", &llval, &dummy) == 1) {
3929 log2_sectors_per_bitmap_bit = !llval ? 0 : __ilog2_u64(llval);
3930 } else if (sscanf(opt_string, "bitmap_flush_interval:%u%c", &val, &dummy) == 1) {
3931 if (val >= (uint64_t)UINT_MAX * 1000 / HZ) {
3933 ti->error = "Invalid bitmap_flush_interval argument";
3936 ic->bitmap_flush_interval = msecs_to_jiffies(val);
3937 } else if (!strncmp(opt_string, "internal_hash:", strlen("internal_hash:"))) {
3938 r = get_alg_and_key(opt_string, &ic->internal_hash_alg, &ti->error,
3939 "Invalid internal_hash argument");
3942 } else if (!strncmp(opt_string, "journal_crypt:", strlen("journal_crypt:"))) {
3943 r = get_alg_and_key(opt_string, &ic->journal_crypt_alg, &ti->error,
3944 "Invalid journal_crypt argument");
3947 } else if (!strncmp(opt_string, "journal_mac:", strlen("journal_mac:"))) {
3948 r = get_alg_and_key(opt_string, &ic->journal_mac_alg, &ti->error,
3949 "Invalid journal_mac argument");
3952 } else if (!strcmp(opt_string, "recalculate")) {
3953 ic->recalculate_flag = true;
3954 } else if (!strcmp(opt_string, "allow_discards")) {
3956 } else if (!strcmp(opt_string, "fix_padding")) {
3957 ic->fix_padding = true;
3958 } else if (!strcmp(opt_string, "legacy_recalculate")) {
3959 ic->legacy_recalculate = true;
3962 ti->error = "Invalid argument";
3967 ic->data_device_sectors = i_size_read(ic->dev->bdev->bd_inode) >> SECTOR_SHIFT;
3969 ic->meta_device_sectors = ic->data_device_sectors;
3971 ic->meta_device_sectors = i_size_read(ic->meta_dev->bdev->bd_inode) >> SECTOR_SHIFT;
3973 if (!journal_sectors) {
3974 journal_sectors = min((sector_t)DEFAULT_MAX_JOURNAL_SECTORS,
3975 ic->data_device_sectors >> DEFAULT_JOURNAL_SIZE_FACTOR);
3978 if (!buffer_sectors)
3980 ic->log2_buffer_sectors = min((int)__fls(buffer_sectors), 31 - SECTOR_SHIFT);
3982 r = get_mac(&ic->internal_hash, &ic->internal_hash_alg, &ti->error,
3983 "Invalid internal hash", "Error setting internal hash key");
3987 r = get_mac(&ic->journal_mac, &ic->journal_mac_alg, &ti->error,
3988 "Invalid journal mac", "Error setting journal mac key");
3992 if (!ic->tag_size) {
3993 if (!ic->internal_hash) {
3994 ti->error = "Unknown tag size";
3998 ic->tag_size = crypto_shash_digestsize(ic->internal_hash);
4000 if (ic->tag_size > MAX_TAG_SIZE) {
4001 ti->error = "Too big tag size";
4005 if (!(ic->tag_size & (ic->tag_size - 1)))
4006 ic->log2_tag_size = __ffs(ic->tag_size);
4008 ic->log2_tag_size = -1;
4010 if (ic->mode == 'B' && !ic->internal_hash) {
4012 ti->error = "Bitmap mode can be only used with internal hash";
4016 if (ic->discard && !ic->internal_hash) {
4018 ti->error = "Discard can be only used with internal hash";
4022 ic->autocommit_jiffies = msecs_to_jiffies(sync_msec);
4023 ic->autocommit_msec = sync_msec;
4024 timer_setup(&ic->autocommit_timer, autocommit_fn, 0);
4026 ic->io = dm_io_client_create();
4027 if (IS_ERR(ic->io)) {
4028 r = PTR_ERR(ic->io);
4030 ti->error = "Cannot allocate dm io";
4034 r = mempool_init_slab_pool(&ic->journal_io_mempool, JOURNAL_IO_MEMPOOL, journal_io_cache);
4036 ti->error = "Cannot allocate mempool";
4040 ic->metadata_wq = alloc_workqueue("dm-integrity-metadata",
4041 WQ_MEM_RECLAIM, METADATA_WORKQUEUE_MAX_ACTIVE);
4042 if (!ic->metadata_wq) {
4043 ti->error = "Cannot allocate workqueue";
4049 * If this workqueue were percpu, it would cause bio reordering
4050 * and reduced performance.
4052 ic->wait_wq = alloc_workqueue("dm-integrity-wait", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
4054 ti->error = "Cannot allocate workqueue";
4059 ic->offload_wq = alloc_workqueue("dm-integrity-offload", WQ_MEM_RECLAIM,
4060 METADATA_WORKQUEUE_MAX_ACTIVE);
4061 if (!ic->offload_wq) {
4062 ti->error = "Cannot allocate workqueue";
4067 ic->commit_wq = alloc_workqueue("dm-integrity-commit", WQ_MEM_RECLAIM, 1);
4068 if (!ic->commit_wq) {
4069 ti->error = "Cannot allocate workqueue";
4073 INIT_WORK(&ic->commit_work, integrity_commit);
4075 if (ic->mode == 'J' || ic->mode == 'B') {
4076 ic->writer_wq = alloc_workqueue("dm-integrity-writer", WQ_MEM_RECLAIM, 1);
4077 if (!ic->writer_wq) {
4078 ti->error = "Cannot allocate workqueue";
4082 INIT_WORK(&ic->writer_work, integrity_writer);
4085 ic->sb = alloc_pages_exact(SB_SECTORS << SECTOR_SHIFT, GFP_KERNEL);
4088 ti->error = "Cannot allocate superblock area";
4092 r = sync_rw_sb(ic, REQ_OP_READ, 0);
4094 ti->error = "Error reading superblock";
4097 should_write_sb = false;
4098 if (memcmp(ic->sb->magic, SB_MAGIC, 8)) {
4099 if (ic->mode != 'R') {
4100 if (memchr_inv(ic->sb, 0, SB_SECTORS << SECTOR_SHIFT)) {
4102 ti->error = "The device is not initialized";
4107 r = initialize_superblock(ic, journal_sectors, interleave_sectors);
4109 ti->error = "Could not initialize superblock";
4112 if (ic->mode != 'R')
4113 should_write_sb = true;
4116 if (!ic->sb->version || ic->sb->version > SB_VERSION_4) {
4118 ti->error = "Unknown version";
4121 if (le16_to_cpu(ic->sb->integrity_tag_size) != ic->tag_size) {
4123 ti->error = "Tag size doesn't match the information in superblock";
4126 if (ic->sb->log2_sectors_per_block != __ffs(ic->sectors_per_block)) {
4128 ti->error = "Block size doesn't match the information in superblock";
4131 if (!le32_to_cpu(ic->sb->journal_sections)) {
4133 ti->error = "Corrupted superblock, journal_sections is 0";
4136 /* make sure that ti->max_io_len doesn't overflow */
4137 if (!ic->meta_dev) {
4138 if (ic->sb->log2_interleave_sectors < MIN_LOG2_INTERLEAVE_SECTORS ||
4139 ic->sb->log2_interleave_sectors > MAX_LOG2_INTERLEAVE_SECTORS) {
4141 ti->error = "Invalid interleave_sectors in the superblock";
4145 if (ic->sb->log2_interleave_sectors) {
4147 ti->error = "Invalid interleave_sectors in the superblock";
4151 if (!!(ic->sb->flags & cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC)) != !!ic->journal_mac_alg.alg_string) {
4153 ti->error = "Journal mac mismatch";
4157 get_provided_data_sectors(ic);
4158 if (!ic->provided_data_sectors) {
4160 ti->error = "The device is too small";
4165 r = calculate_device_limits(ic);
4168 if (ic->log2_buffer_sectors > 3) {
4169 ic->log2_buffer_sectors--;
4170 goto try_smaller_buffer;
4173 ti->error = "The device is too small";
4177 if (log2_sectors_per_bitmap_bit < 0)
4178 log2_sectors_per_bitmap_bit = __fls(DEFAULT_SECTORS_PER_BITMAP_BIT);
4179 if (log2_sectors_per_bitmap_bit < ic->sb->log2_sectors_per_block)
4180 log2_sectors_per_bitmap_bit = ic->sb->log2_sectors_per_block;
4182 bits_in_journal = ((__u64)ic->journal_section_sectors * ic->journal_sections) << (SECTOR_SHIFT + 3);
4183 if (bits_in_journal > UINT_MAX)
4184 bits_in_journal = UINT_MAX;
4185 while (bits_in_journal < (ic->provided_data_sectors + ((sector_t)1 << log2_sectors_per_bitmap_bit) - 1) >> log2_sectors_per_bitmap_bit)
4186 log2_sectors_per_bitmap_bit++;
4188 log2_blocks_per_bitmap_bit = log2_sectors_per_bitmap_bit - ic->sb->log2_sectors_per_block;
4189 ic->log2_blocks_per_bitmap_bit = log2_blocks_per_bitmap_bit;
4190 if (should_write_sb) {
4191 ic->sb->log2_blocks_per_bitmap_bit = log2_blocks_per_bitmap_bit;
4193 n_bitmap_bits = ((ic->provided_data_sectors >> ic->sb->log2_sectors_per_block)
4194 + (((sector_t)1 << log2_blocks_per_bitmap_bit) - 1)) >> log2_blocks_per_bitmap_bit;
4195 ic->n_bitmap_blocks = DIV_ROUND_UP(n_bitmap_bits, BITMAP_BLOCK_SIZE * 8);
4198 ic->log2_buffer_sectors = min(ic->log2_buffer_sectors, (__u8)__ffs(ic->metadata_run));
4200 if (ti->len > ic->provided_data_sectors) {
4202 ti->error = "Not enough provided sectors for requested mapping size";
4207 threshold = (__u64)ic->journal_entries * (100 - journal_watermark);
4209 do_div(threshold, 100);
4210 ic->free_sectors_threshold = threshold;
4212 DEBUG_print("initialized:\n");
4213 DEBUG_print(" integrity_tag_size %u\n", le16_to_cpu(ic->sb->integrity_tag_size));
4214 DEBUG_print(" journal_entry_size %u\n", ic->journal_entry_size);
4215 DEBUG_print(" journal_entries_per_sector %u\n", ic->journal_entries_per_sector);
4216 DEBUG_print(" journal_section_entries %u\n", ic->journal_section_entries);
4217 DEBUG_print(" journal_section_sectors %u\n", ic->journal_section_sectors);
4218 DEBUG_print(" journal_sections %u\n", (unsigned)le32_to_cpu(ic->sb->journal_sections));
4219 DEBUG_print(" journal_entries %u\n", ic->journal_entries);
4220 DEBUG_print(" log2_interleave_sectors %d\n", ic->sb->log2_interleave_sectors);
4221 DEBUG_print(" data_device_sectors 0x%llx\n", i_size_read(ic->dev->bdev->bd_inode) >> SECTOR_SHIFT);
4222 DEBUG_print(" initial_sectors 0x%x\n", ic->initial_sectors);
4223 DEBUG_print(" metadata_run 0x%x\n", ic->metadata_run);
4224 DEBUG_print(" log2_metadata_run %d\n", ic->log2_metadata_run);
4225 DEBUG_print(" provided_data_sectors 0x%llx (%llu)\n", ic->provided_data_sectors, ic->provided_data_sectors);
4226 DEBUG_print(" log2_buffer_sectors %u\n", ic->log2_buffer_sectors);
4227 DEBUG_print(" bits_in_journal %llu\n", bits_in_journal);
4229 if (ic->recalculate_flag && !(ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))) {
4230 ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
4231 ic->sb->recalc_sector = cpu_to_le64(0);
4234 if (ic->internal_hash) {
4235 size_t recalc_tags_size;
4236 ic->recalc_wq = alloc_workqueue("dm-integrity-recalc", WQ_MEM_RECLAIM, 1);
4237 if (!ic->recalc_wq ) {
4238 ti->error = "Cannot allocate workqueue";
4242 INIT_WORK(&ic->recalc_work, integrity_recalc);
4243 ic->recalc_buffer = vmalloc(RECALC_SECTORS << SECTOR_SHIFT);
4244 if (!ic->recalc_buffer) {
4245 ti->error = "Cannot allocate buffer for recalculating";
4249 recalc_tags_size = (RECALC_SECTORS >> ic->sb->log2_sectors_per_block) * ic->tag_size;
4250 if (crypto_shash_digestsize(ic->internal_hash) > ic->tag_size)
4251 recalc_tags_size += crypto_shash_digestsize(ic->internal_hash) - ic->tag_size;
4252 ic->recalc_tags = kvmalloc(recalc_tags_size, GFP_KERNEL);
4253 if (!ic->recalc_tags) {
4254 ti->error = "Cannot allocate tags for recalculating";
4259 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
4260 ti->error = "Recalculate can only be specified with internal_hash";
4266 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING) &&
4267 le64_to_cpu(ic->sb->recalc_sector) < ic->provided_data_sectors &&
4268 dm_integrity_disable_recalculate(ic)) {
4269 ti->error = "Recalculating with HMAC is disabled for security reasons - if you really need it, use the argument \"legacy_recalculate\"";
4274 ic->bufio = dm_bufio_client_create(ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev,
4275 1U << (SECTOR_SHIFT + ic->log2_buffer_sectors), 1, 0, NULL, NULL);
4276 if (IS_ERR(ic->bufio)) {
4277 r = PTR_ERR(ic->bufio);
4278 ti->error = "Cannot initialize dm-bufio";
4282 dm_bufio_set_sector_offset(ic->bufio, ic->start + ic->initial_sectors);
4284 if (ic->mode != 'R') {
4285 r = create_journal(ic, &ti->error);
4291 if (ic->mode == 'B') {
4293 unsigned n_bitmap_pages = DIV_ROUND_UP(ic->n_bitmap_blocks, PAGE_SIZE / BITMAP_BLOCK_SIZE);
4295 ic->recalc_bitmap = dm_integrity_alloc_page_list(n_bitmap_pages);
4296 if (!ic->recalc_bitmap) {
4300 ic->may_write_bitmap = dm_integrity_alloc_page_list(n_bitmap_pages);
4301 if (!ic->may_write_bitmap) {
4305 ic->bbs = kvmalloc_array(ic->n_bitmap_blocks, sizeof(struct bitmap_block_status), GFP_KERNEL);
4310 INIT_DELAYED_WORK(&ic->bitmap_flush_work, bitmap_flush_work);
4311 for (i = 0; i < ic->n_bitmap_blocks; i++) {
4312 struct bitmap_block_status *bbs = &ic->bbs[i];
4313 unsigned sector, pl_index, pl_offset;
4315 INIT_WORK(&bbs->work, bitmap_block_work);
4318 bio_list_init(&bbs->bio_queue);
4319 spin_lock_init(&bbs->bio_queue_lock);
4321 sector = i * (BITMAP_BLOCK_SIZE >> SECTOR_SHIFT);
4322 pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
4323 pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
4325 bbs->bitmap = lowmem_page_address(ic->journal[pl_index].page) + pl_offset;
4329 if (should_write_sb) {
4330 init_journal(ic, 0, ic->journal_sections, 0);
4331 r = dm_integrity_failed(ic);
4333 ti->error = "Error initializing journal";
4336 r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
4338 ti->error = "Error initializing superblock";
4341 ic->just_formatted = true;
4344 if (!ic->meta_dev) {
4345 r = dm_set_target_max_io_len(ti, 1U << ic->sb->log2_interleave_sectors);
4349 if (ic->mode == 'B') {
4350 unsigned max_io_len = ((sector_t)ic->sectors_per_block << ic->log2_blocks_per_bitmap_bit) * (BITMAP_BLOCK_SIZE * 8);
4352 max_io_len = 1U << 31;
4353 DEBUG_print("max_io_len: old %u, new %u\n", ti->max_io_len, max_io_len);
4354 if (!ti->max_io_len || ti->max_io_len > max_io_len) {
4355 r = dm_set_target_max_io_len(ti, max_io_len);
4361 if (!ic->internal_hash)
4362 dm_integrity_set(ti, ic);
4364 ti->num_flush_bios = 1;
4365 ti->flush_supported = true;
4367 ti->num_discard_bios = 1;
4372 dm_integrity_dtr(ti);
4376 static void dm_integrity_dtr(struct dm_target *ti)
4378 struct dm_integrity_c *ic = ti->private;
4380 BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
4381 BUG_ON(!list_empty(&ic->wait_list));
4383 if (ic->metadata_wq)
4384 destroy_workqueue(ic->metadata_wq);
4386 destroy_workqueue(ic->wait_wq);
4388 destroy_workqueue(ic->offload_wq);
4390 destroy_workqueue(ic->commit_wq);
4392 destroy_workqueue(ic->writer_wq);
4394 destroy_workqueue(ic->recalc_wq);
4395 vfree(ic->recalc_buffer);
4396 kvfree(ic->recalc_tags);
4399 dm_bufio_client_destroy(ic->bufio);
4400 mempool_exit(&ic->journal_io_mempool);
4402 dm_io_client_destroy(ic->io);
4404 dm_put_device(ti, ic->dev);
4406 dm_put_device(ti, ic->meta_dev);
4407 dm_integrity_free_page_list(ic->journal);
4408 dm_integrity_free_page_list(ic->journal_io);
4409 dm_integrity_free_page_list(ic->journal_xor);
4410 dm_integrity_free_page_list(ic->recalc_bitmap);
4411 dm_integrity_free_page_list(ic->may_write_bitmap);
4412 if (ic->journal_scatterlist)
4413 dm_integrity_free_journal_scatterlist(ic, ic->journal_scatterlist);
4414 if (ic->journal_io_scatterlist)
4415 dm_integrity_free_journal_scatterlist(ic, ic->journal_io_scatterlist);
4416 if (ic->sk_requests) {
4419 for (i = 0; i < ic->journal_sections; i++) {
4420 struct skcipher_request *req = ic->sk_requests[i];
4422 kfree_sensitive(req->iv);
4423 skcipher_request_free(req);
4426 kvfree(ic->sk_requests);
4428 kvfree(ic->journal_tree);
4430 free_pages_exact(ic->sb, SB_SECTORS << SECTOR_SHIFT);
4432 if (ic->internal_hash)
4433 crypto_free_shash(ic->internal_hash);
4434 free_alg(&ic->internal_hash_alg);
4436 if (ic->journal_crypt)
4437 crypto_free_skcipher(ic->journal_crypt);
4438 free_alg(&ic->journal_crypt_alg);
4440 if (ic->journal_mac)
4441 crypto_free_shash(ic->journal_mac);
4442 free_alg(&ic->journal_mac_alg);
4447 static struct target_type integrity_target = {
4448 .name = "integrity",
4449 .version = {1, 6, 0},
4450 .module = THIS_MODULE,
4451 .features = DM_TARGET_SINGLETON | DM_TARGET_INTEGRITY,
4452 .ctr = dm_integrity_ctr,
4453 .dtr = dm_integrity_dtr,
4454 .map = dm_integrity_map,
4455 .postsuspend = dm_integrity_postsuspend,
4456 .resume = dm_integrity_resume,
4457 .status = dm_integrity_status,
4458 .iterate_devices = dm_integrity_iterate_devices,
4459 .io_hints = dm_integrity_io_hints,
4462 static int __init dm_integrity_init(void)
4466 journal_io_cache = kmem_cache_create("integrity_journal_io",
4467 sizeof(struct journal_io), 0, 0, NULL);
4468 if (!journal_io_cache) {
4469 DMERR("can't allocate journal io cache");
4473 r = dm_register_target(&integrity_target);
4476 DMERR("register failed %d", r);
4481 static void __exit dm_integrity_exit(void)
4483 dm_unregister_target(&integrity_target);
4484 kmem_cache_destroy(journal_io_cache);
4487 module_init(dm_integrity_init);
4488 module_exit(dm_integrity_exit);
4490 MODULE_AUTHOR("Milan Broz");
4491 MODULE_AUTHOR("Mikulas Patocka");
4492 MODULE_DESCRIPTION(DM_NAME " target for integrity tags extension");
4493 MODULE_LICENSE("GPL");