1 /*
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
5 *
6 * This file is released under the GPL.
7 */
9 #include <linux/compiler.h>
10 #include <linux/module.h>
11 #include <linux/device-mapper.h>
12 #include <linux/dm-io.h>
13 #include <linux/vmalloc.h>
14 #include <linux/sort.h>
15 #include <linux/rbtree.h>
16 #include <linux/delay.h>
17 #include <linux/random.h>
18 #include <crypto/hash.h>
19 #include <crypto/skcipher.h>
20 #include <linux/async_tx.h>
21 #include <linux/dm-bufio.h>
23 #define DM_MSG_PREFIX "integrity"
25 #define DEFAULT_INTERLEAVE_SECTORS 32768
26 #define DEFAULT_JOURNAL_SIZE_FACTOR 7
27 #define DEFAULT_BUFFER_SECTORS 128
28 #define DEFAULT_JOURNAL_WATERMARK 50
29 #define DEFAULT_SYNC_MSEC 10000
30 #define DEFAULT_MAX_JOURNAL_SECTORS 131072
31 #define MIN_LOG2_INTERLEAVE_SECTORS 3
32 #define MAX_LOG2_INTERLEAVE_SECTORS 31
33 #define METADATA_WORKQUEUE_MAX_ACTIVE 16
34 #define RECALC_SECTORS 8192
35 #define RECALC_WRITE_SUPER 16
37 /*
38 * Warning - DEBUG_PRINT prints security-sensitive data to the log,
39 * so it should not be enabled in the official kernel
40 */
41 //#define DEBUG_PRINT
42 //#define INTERNAL_VERIFY
44 /*
45 * On disk structures
46 */
48 #define SB_MAGIC "integrt"
49 #define SB_VERSION_1 1
50 #define SB_VERSION_2 2
51 #define SB_SECTORS 8
52 #define MAX_SECTORS_PER_BLOCK 8
54 struct superblock {
55 __u8 magic[8];
56 __u8 version;
57 __u8 log2_interleave_sectors;
58 __u16 integrity_tag_size;
59 __u32 journal_sections;
60 __u64 provided_data_sectors; /* userspace uses this value */
61 __u32 flags;
62 __u8 log2_sectors_per_block;
63 __u8 pad[3];
64 __u64 recalc_sector;
65 };
67 #define SB_FLAG_HAVE_JOURNAL_MAC 0x1
68 #define SB_FLAG_RECALCULATING 0x2
70 #define JOURNAL_ENTRY_ROUNDUP 8
72 typedef __u64 commit_id_t;
73 #define JOURNAL_MAC_PER_SECTOR 8
75 struct journal_entry {
76 union {
77 struct {
78 __u32 sector_lo;
79 __u32 sector_hi;
80 } s;
81 __u64 sector;
82 } u;
83 commit_id_t last_bytes[0];
84 /* __u8 tag[0]; */
85 };
87 #define journal_entry_tag(ic, je) ((__u8 *)&(je)->last_bytes[(ic)->sectors_per_block])
89 #if BITS_PER_LONG == 64
90 #define journal_entry_set_sector(je, x) do { smp_wmb(); WRITE_ONCE((je)->u.sector, cpu_to_le64(x)); } while (0)
91 #define journal_entry_get_sector(je) le64_to_cpu((je)->u.sector)
92 #elif defined(CONFIG_LBDAF)
93 #define journal_entry_set_sector(je, x) do { (je)->u.s.sector_lo = cpu_to_le32(x); smp_wmb(); WRITE_ONCE((je)->u.s.sector_hi, cpu_to_le32((x) >> 32)); } while (0)
94 #define journal_entry_get_sector(je) le64_to_cpu((je)->u.sector)
95 #else
96 #define journal_entry_set_sector(je, x) do { (je)->u.s.sector_lo = cpu_to_le32(x); smp_wmb(); WRITE_ONCE((je)->u.s.sector_hi, cpu_to_le32(0)); } while (0)
97 #define journal_entry_get_sector(je) le32_to_cpu((je)->u.s.sector_lo)
98 #endif
99 #define journal_entry_is_unused(je) ((je)->u.s.sector_hi == cpu_to_le32(-1))
100 #define journal_entry_set_unused(je) do { ((je)->u.s.sector_hi = cpu_to_le32(-1)); } while (0)
101 #define journal_entry_is_inprogress(je) ((je)->u.s.sector_hi == cpu_to_le32(-2))
102 #define journal_entry_set_inprogress(je) do { ((je)->u.s.sector_hi = cpu_to_le32(-2)); } while (0)
104 #define JOURNAL_BLOCK_SECTORS 8
105 #define JOURNAL_SECTOR_DATA ((1 << SECTOR_SHIFT) - sizeof(commit_id_t))
106 #define JOURNAL_MAC_SIZE (JOURNAL_MAC_PER_SECTOR * JOURNAL_BLOCK_SECTORS)
108 struct journal_sector {
109 __u8 entries[JOURNAL_SECTOR_DATA - JOURNAL_MAC_PER_SECTOR];
110 __u8 mac[JOURNAL_MAC_PER_SECTOR];
111 commit_id_t commit_id;
112 };
114 #define MAX_TAG_SIZE (JOURNAL_SECTOR_DATA - JOURNAL_MAC_PER_SECTOR - offsetof(struct journal_entry, last_bytes[MAX_SECTORS_PER_BLOCK]))
116 #define METADATA_PADDING_SECTORS 8
118 #define N_COMMIT_IDS 4
120 static unsigned char prev_commit_seq(unsigned char seq)
121 {
122 return (seq + N_COMMIT_IDS - 1) % N_COMMIT_IDS;
123 }
125 static unsigned char next_commit_seq(unsigned char seq)
126 {
127 return (seq + 1) % N_COMMIT_IDS;
128 }
130 /*
131 * In-memory structures
132 */
134 struct journal_node {
135 struct rb_node node;
136 sector_t sector;
137 };
139 struct alg_spec {
140 char *alg_string;
141 char *key_string;
142 __u8 *key;
143 unsigned key_size;
144 };
146 struct dm_integrity_c {
147 struct dm_dev *dev;
148 struct dm_dev *meta_dev;
149 unsigned tag_size;
150 __s8 log2_tag_size;
151 sector_t start;
152 mempool_t journal_io_mempool;
153 struct dm_io_client *io;
154 struct dm_bufio_client *bufio;
155 struct workqueue_struct *metadata_wq;
156 struct superblock *sb;
157 unsigned journal_pages;
158 struct page_list *journal;
159 struct page_list *journal_io;
160 struct page_list *journal_xor;
162 struct crypto_skcipher *journal_crypt;
163 struct scatterlist **journal_scatterlist;
164 struct scatterlist **journal_io_scatterlist;
165 struct skcipher_request **sk_requests;
167 struct crypto_shash *journal_mac;
169 struct journal_node *journal_tree;
170 struct rb_root journal_tree_root;
172 sector_t provided_data_sectors;
174 unsigned short journal_entry_size;
175 unsigned char journal_entries_per_sector;
176 unsigned char journal_section_entries;
177 unsigned short journal_section_sectors;
178 unsigned journal_sections;
179 unsigned journal_entries;
180 sector_t data_device_sectors;
181 sector_t meta_device_sectors;
182 unsigned initial_sectors;
183 unsigned metadata_run;
184 __s8 log2_metadata_run;
185 __u8 log2_buffer_sectors;
186 __u8 sectors_per_block;
188 unsigned char mode;
189 int suspending;
191 int failed;
193 struct crypto_shash *internal_hash;
195 /* these variables are locked with endio_wait.lock */
196 struct rb_root in_progress;
197 struct list_head wait_list;
198 wait_queue_head_t endio_wait;
199 struct workqueue_struct *wait_wq;
201 unsigned char commit_seq;
202 commit_id_t commit_ids[N_COMMIT_IDS];
204 unsigned committed_section;
205 unsigned n_committed_sections;
207 unsigned uncommitted_section;
208 unsigned n_uncommitted_sections;
210 unsigned free_section;
211 unsigned char free_section_entry;
212 unsigned free_sectors;
214 unsigned free_sectors_threshold;
216 struct workqueue_struct *commit_wq;
217 struct work_struct commit_work;
219 struct workqueue_struct *writer_wq;
220 struct work_struct writer_work;
222 struct workqueue_struct *recalc_wq;
223 struct work_struct recalc_work;
224 u8 *recalc_buffer;
225 u8 *recalc_tags;
227 struct bio_list flush_bio_list;
229 unsigned long autocommit_jiffies;
230 struct timer_list autocommit_timer;
231 unsigned autocommit_msec;
233 wait_queue_head_t copy_to_journal_wait;
235 struct completion crypto_backoff;
237 bool journal_uptodate;
238 bool just_formatted;
240 struct alg_spec internal_hash_alg;
241 struct alg_spec journal_crypt_alg;
242 struct alg_spec journal_mac_alg;
244 atomic64_t number_of_mismatches;
245 };
247 struct dm_integrity_range {
248 sector_t logical_sector;
249 unsigned n_sectors;
250 bool waiting;
251 union {
252 struct rb_node node;
253 struct {
254 struct task_struct *task;
255 struct list_head wait_entry;
256 };
257 };
258 };
260 struct dm_integrity_io {
261 struct work_struct work;
263 struct dm_integrity_c *ic;
264 bool write;
265 bool fua;
267 struct dm_integrity_range range;
269 sector_t metadata_block;
270 unsigned metadata_offset;
272 atomic_t in_flight;
273 blk_status_t bi_status;
275 struct completion *completion;
277 struct gendisk *orig_bi_disk;
278 u8 orig_bi_partno;
279 bio_end_io_t *orig_bi_end_io;
280 struct bio_integrity_payload *orig_bi_integrity;
281 struct bvec_iter orig_bi_iter;
282 };
284 struct journal_completion {
285 struct dm_integrity_c *ic;
286 atomic_t in_flight;
287 struct completion comp;
288 };
290 struct journal_io {
291 struct dm_integrity_range range;
292 struct journal_completion *comp;
293 };
295 static struct kmem_cache *journal_io_cache;
297 #define JOURNAL_IO_MEMPOOL 32
299 #ifdef DEBUG_PRINT
300 #define DEBUG_print(x, ...) printk(KERN_DEBUG x, ##__VA_ARGS__)
301 static void __DEBUG_bytes(__u8 *bytes, size_t len, const char *msg, ...)
302 {
303 va_list args;
304 va_start(args, msg);
305 vprintk(msg, args);
306 va_end(args);
307 if (len)
308 pr_cont(":");
309 while (len) {
310 pr_cont(" %02x", *bytes);
311 bytes++;
312 len--;
313 }
314 pr_cont("\n");
315 }
316 #define DEBUG_bytes(bytes, len, msg, ...) __DEBUG_bytes(bytes, len, KERN_DEBUG msg, ##__VA_ARGS__)
317 #else
318 #define DEBUG_print(x, ...) do { } while (0)
319 #define DEBUG_bytes(bytes, len, msg, ...) do { } while (0)
320 #endif
322 /*
323 * DM Integrity profile, protection is performed layer above (dm-crypt)
324 */
325 static const struct blk_integrity_profile dm_integrity_profile = {
326 .name = "DM-DIF-EXT-TAG",
327 .generate_fn = NULL,
328 .verify_fn = NULL,
329 };
331 static void dm_integrity_map_continue(struct dm_integrity_io *dio, bool from_map);
332 static void integrity_bio_wait(struct work_struct *w);
333 static void dm_integrity_dtr(struct dm_target *ti);
335 static void dm_integrity_io_error(struct dm_integrity_c *ic, const char *msg, int err)
336 {
337 if (err == -EILSEQ)
338 atomic64_inc(&ic->number_of_mismatches);
339 if (!cmpxchg(&ic->failed, 0, err))
340 DMERR("Error on %s: %d", msg, err);
341 }
343 static int dm_integrity_failed(struct dm_integrity_c *ic)
344 {
345 return READ_ONCE(ic->failed);
346 }
348 static commit_id_t dm_integrity_commit_id(struct dm_integrity_c *ic, unsigned i,
349 unsigned j, unsigned char seq)
350 {
351 /*
352 * Xor the number with section and sector, so that if a piece of
353 * journal is written at wrong place, it is detected.
354 */
355 return ic->commit_ids[seq] ^ cpu_to_le64(((__u64)i << 32) ^ j);
356 }
358 static void get_area_and_offset(struct dm_integrity_c *ic, sector_t data_sector,
359 sector_t *area, sector_t *offset)
360 {
361 if (!ic->meta_dev) {
362 __u8 log2_interleave_sectors = ic->sb->log2_interleave_sectors;
363 *area = data_sector >> log2_interleave_sectors;
364 *offset = (unsigned)data_sector & ((1U << log2_interleave_sectors) - 1);
365 } else {
366 *area = 0;
367 *offset = data_sector;
368 }
369 }
371 #define sector_to_block(ic, n) \
372 do { \
373 BUG_ON((n) & (unsigned)((ic)->sectors_per_block - 1)); \
374 (n) >>= (ic)->sb->log2_sectors_per_block; \
375 } while (0)
377 static __u64 get_metadata_sector_and_offset(struct dm_integrity_c *ic, sector_t area,
378 sector_t offset, unsigned *metadata_offset)
379 {
380 __u64 ms;
381 unsigned mo;
383 ms = area << ic->sb->log2_interleave_sectors;
384 if (likely(ic->log2_metadata_run >= 0))
385 ms += area << ic->log2_metadata_run;
386 else
387 ms += area * ic->metadata_run;
388 ms >>= ic->log2_buffer_sectors;
390 sector_to_block(ic, offset);
392 if (likely(ic->log2_tag_size >= 0)) {
393 ms += offset >> (SECTOR_SHIFT + ic->log2_buffer_sectors - ic->log2_tag_size);
394 mo = (offset << ic->log2_tag_size) & ((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - 1);
395 } else {
396 ms += (__u64)offset * ic->tag_size >> (SECTOR_SHIFT + ic->log2_buffer_sectors);
397 mo = (offset * ic->tag_size) & ((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - 1);
398 }
399 *metadata_offset = mo;
400 return ms;
401 }
403 static sector_t get_data_sector(struct dm_integrity_c *ic, sector_t area, sector_t offset)
404 {
405 sector_t result;
407 if (ic->meta_dev)
408 return offset;
410 result = area << ic->sb->log2_interleave_sectors;
411 if (likely(ic->log2_metadata_run >= 0))
412 result += (area + 1) << ic->log2_metadata_run;
413 else
414 result += (area + 1) * ic->metadata_run;
416 result += (sector_t)ic->initial_sectors + offset;
417 result += ic->start;
419 return result;
420 }
422 static void wraparound_section(struct dm_integrity_c *ic, unsigned *sec_ptr)
423 {
424 if (unlikely(*sec_ptr >= ic->journal_sections))
425 *sec_ptr -= ic->journal_sections;
426 }
428 static void sb_set_version(struct dm_integrity_c *ic)
429 {
430 if (ic->meta_dev || ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
431 ic->sb->version = SB_VERSION_2;
432 else
433 ic->sb->version = SB_VERSION_1;
434 }
436 static int sync_rw_sb(struct dm_integrity_c *ic, int op, int op_flags)
437 {
438 struct dm_io_request io_req;
439 struct dm_io_region io_loc;
441 io_req.bi_op = op;
442 io_req.bi_op_flags = op_flags;
443 io_req.mem.type = DM_IO_KMEM;
444 io_req.mem.ptr.addr = ic->sb;
445 io_req.notify.fn = NULL;
446 io_req.client = ic->io;
447 io_loc.bdev = ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev;
448 io_loc.sector = ic->start;
449 io_loc.count = SB_SECTORS;
451 return dm_io(&io_req, 1, &io_loc, NULL);
452 }
454 static void access_journal_check(struct dm_integrity_c *ic, unsigned section, unsigned offset,
455 bool e, const char *function)
456 {
457 #if defined(CONFIG_DM_DEBUG) || defined(INTERNAL_VERIFY)
458 unsigned limit = e ? ic->journal_section_entries : ic->journal_section_sectors;
460 if (unlikely(section >= ic->journal_sections) ||
461 unlikely(offset >= limit)) {
462 printk(KERN_CRIT "%s: invalid access at (%u,%u), limit (%u,%u)\n",
463 function, section, offset, ic->journal_sections, limit);
464 BUG();
465 }
466 #endif
467 }
469 static void page_list_location(struct dm_integrity_c *ic, unsigned section, unsigned offset,
470 unsigned *pl_index, unsigned *pl_offset)
471 {
472 unsigned sector;
474 access_journal_check(ic, section, offset, false, "page_list_location");
476 sector = section * ic->journal_section_sectors + offset;
478 *pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
479 *pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
480 }
482 static struct journal_sector *access_page_list(struct dm_integrity_c *ic, struct page_list *pl,
483 unsigned section, unsigned offset, unsigned *n_sectors)
484 {
485 unsigned pl_index, pl_offset;
486 char *va;
488 page_list_location(ic, section, offset, &pl_index, &pl_offset);
490 if (n_sectors)
491 *n_sectors = (PAGE_SIZE - pl_offset) >> SECTOR_SHIFT;
493 va = lowmem_page_address(pl[pl_index].page);
495 return (struct journal_sector *)(va + pl_offset);
496 }
498 static struct journal_sector *access_journal(struct dm_integrity_c *ic, unsigned section, unsigned offset)
499 {
500 return access_page_list(ic, ic->journal, section, offset, NULL);
501 }
503 static struct journal_entry *access_journal_entry(struct dm_integrity_c *ic, unsigned section, unsigned n)
504 {
505 unsigned rel_sector, offset;
506 struct journal_sector *js;
508 access_journal_check(ic, section, n, true, "access_journal_entry");
510 rel_sector = n % JOURNAL_BLOCK_SECTORS;
511 offset = n / JOURNAL_BLOCK_SECTORS;
513 js = access_journal(ic, section, rel_sector);
514 return (struct journal_entry *)((char *)js + offset * ic->journal_entry_size);
515 }
517 static struct journal_sector *access_journal_data(struct dm_integrity_c *ic, unsigned section, unsigned n)
518 {
519 n <<= ic->sb->log2_sectors_per_block;
521 n += JOURNAL_BLOCK_SECTORS;
523 access_journal_check(ic, section, n, false, "access_journal_data");
525 return access_journal(ic, section, n);
526 }
528 static void section_mac(struct dm_integrity_c *ic, unsigned section, __u8 result[JOURNAL_MAC_SIZE])
529 {
530 SHASH_DESC_ON_STACK(desc, ic->journal_mac);
531 int r;
532 unsigned j, size;
534 desc->tfm = ic->journal_mac;
535 desc->flags = 0;
537 r = crypto_shash_init(desc);
538 if (unlikely(r)) {
539 dm_integrity_io_error(ic, "crypto_shash_init", r);
540 goto err;
541 }
543 for (j = 0; j < ic->journal_section_entries; j++) {
544 struct journal_entry *je = access_journal_entry(ic, section, j);
545 r = crypto_shash_update(desc, (__u8 *)&je->u.sector, sizeof je->u.sector);
546 if (unlikely(r)) {
547 dm_integrity_io_error(ic, "crypto_shash_update", r);
548 goto err;
549 }
550 }
552 size = crypto_shash_digestsize(ic->journal_mac);
554 if (likely(size <= JOURNAL_MAC_SIZE)) {
555 r = crypto_shash_final(desc, result);
556 if (unlikely(r)) {
557 dm_integrity_io_error(ic, "crypto_shash_final", r);
558 goto err;
559 }
560 memset(result + size, 0, JOURNAL_MAC_SIZE - size);
561 } else {
562 __u8 digest[size];
563 r = crypto_shash_final(desc, digest);
564 if (unlikely(r)) {
565 dm_integrity_io_error(ic, "crypto_shash_final", r);
566 goto err;
567 }
568 memcpy(result, digest, JOURNAL_MAC_SIZE);
569 }
571 return;
572 err:
573 memset(result, 0, JOURNAL_MAC_SIZE);
574 }
576 static void rw_section_mac(struct dm_integrity_c *ic, unsigned section, bool wr)
577 {
578 __u8 result[JOURNAL_MAC_SIZE];
579 unsigned j;
581 if (!ic->journal_mac)
582 return;
584 section_mac(ic, section, result);
586 for (j = 0; j < JOURNAL_BLOCK_SECTORS; j++) {
587 struct journal_sector *js = access_journal(ic, section, j);
589 if (likely(wr))
590 memcpy(&js->mac, result + (j * JOURNAL_MAC_PER_SECTOR), JOURNAL_MAC_PER_SECTOR);
591 else {
592 if (memcmp(&js->mac, result + (j * JOURNAL_MAC_PER_SECTOR), JOURNAL_MAC_PER_SECTOR))
593 dm_integrity_io_error(ic, "journal mac", -EILSEQ);
594 }
595 }
596 }
598 static void complete_journal_op(void *context)
599 {
600 struct journal_completion *comp = context;
601 BUG_ON(!atomic_read(&comp->in_flight));
602 if (likely(atomic_dec_and_test(&comp->in_flight)))
603 complete(&comp->comp);
604 }
606 static void xor_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
607 unsigned n_sections, struct journal_completion *comp)
608 {
609 struct async_submit_ctl submit;
610 size_t n_bytes = (size_t)(n_sections * ic->journal_section_sectors) << SECTOR_SHIFT;
611 unsigned pl_index, pl_offset, section_index;
612 struct page_list *source_pl, *target_pl;
614 if (likely(encrypt)) {
615 source_pl = ic->journal;
616 target_pl = ic->journal_io;
617 } else {
618 source_pl = ic->journal_io;
619 target_pl = ic->journal;
620 }
622 page_list_location(ic, section, 0, &pl_index, &pl_offset);
624 atomic_add(roundup(pl_offset + n_bytes, PAGE_SIZE) >> PAGE_SHIFT, &comp->in_flight);
626 init_async_submit(&submit, ASYNC_TX_XOR_ZERO_DST, NULL, complete_journal_op, comp, NULL);
628 section_index = pl_index;
630 do {
631 size_t this_step;
632 struct page *src_pages[2];
633 struct page *dst_page;
635 while (unlikely(pl_index == section_index)) {
636 unsigned dummy;
637 if (likely(encrypt))
638 rw_section_mac(ic, section, true);
639 section++;
640 n_sections--;
641 if (!n_sections)
642 break;
643 page_list_location(ic, section, 0, §ion_index, &dummy);
644 }
646 this_step = min(n_bytes, (size_t)PAGE_SIZE - pl_offset);
647 dst_page = target_pl[pl_index].page;
648 src_pages[0] = source_pl[pl_index].page;
649 src_pages[1] = ic->journal_xor[pl_index].page;
651 async_xor(dst_page, src_pages, pl_offset, 2, this_step, &submit);
653 pl_index++;
654 pl_offset = 0;
655 n_bytes -= this_step;
656 } while (n_bytes);
658 BUG_ON(n_sections);
660 async_tx_issue_pending_all();
661 }
663 static void complete_journal_encrypt(struct crypto_async_request *req, int err)
664 {
665 struct journal_completion *comp = req->data;
666 if (unlikely(err)) {
667 if (likely(err == -EINPROGRESS)) {
668 complete(&comp->ic->crypto_backoff);
669 return;
670 }
671 dm_integrity_io_error(comp->ic, "asynchronous encrypt", err);
672 }
673 complete_journal_op(comp);
674 }
676 static bool do_crypt(bool encrypt, struct skcipher_request *req, struct journal_completion *comp)
677 {
678 int r;
679 skcipher_request_set_callback(req, CRYPTO_TFM_REQ_MAY_BACKLOG,
680 complete_journal_encrypt, comp);
681 if (likely(encrypt))
682 r = crypto_skcipher_encrypt(req);
683 else
684 r = crypto_skcipher_decrypt(req);
685 if (likely(!r))
686 return false;
687 if (likely(r == -EINPROGRESS))
688 return true;
689 if (likely(r == -EBUSY)) {
690 wait_for_completion(&comp->ic->crypto_backoff);
691 reinit_completion(&comp->ic->crypto_backoff);
692 return true;
693 }
694 dm_integrity_io_error(comp->ic, "encrypt", r);
695 return false;
696 }
698 static void crypt_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
699 unsigned n_sections, struct journal_completion *comp)
700 {
701 struct scatterlist **source_sg;
702 struct scatterlist **target_sg;
704 atomic_add(2, &comp->in_flight);
706 if (likely(encrypt)) {
707 source_sg = ic->journal_scatterlist;
708 target_sg = ic->journal_io_scatterlist;
709 } else {
710 source_sg = ic->journal_io_scatterlist;
711 target_sg = ic->journal_scatterlist;
712 }
714 do {
715 struct skcipher_request *req;
716 unsigned ivsize;
717 char *iv;
719 if (likely(encrypt))
720 rw_section_mac(ic, section, true);
722 req = ic->sk_requests[section];
723 ivsize = crypto_skcipher_ivsize(ic->journal_crypt);
724 iv = req->iv;
726 memcpy(iv, iv + ivsize, ivsize);
728 req->src = source_sg[section];
729 req->dst = target_sg[section];
731 if (unlikely(do_crypt(encrypt, req, comp)))
732 atomic_inc(&comp->in_flight);
734 section++;
735 n_sections--;
736 } while (n_sections);
738 atomic_dec(&comp->in_flight);
739 complete_journal_op(comp);
740 }
742 static void encrypt_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
743 unsigned n_sections, struct journal_completion *comp)
744 {
745 if (ic->journal_xor)
746 return xor_journal(ic, encrypt, section, n_sections, comp);
747 else
748 return crypt_journal(ic, encrypt, section, n_sections, comp);
749 }
751 static void complete_journal_io(unsigned long error, void *context)
752 {
753 struct journal_completion *comp = context;
754 if (unlikely(error != 0))
755 dm_integrity_io_error(comp->ic, "writing journal", -EIO);
756 complete_journal_op(comp);
757 }
759 static void rw_journal(struct dm_integrity_c *ic, int op, int op_flags, unsigned section,
760 unsigned n_sections, struct journal_completion *comp)
761 {
762 struct dm_io_request io_req;
763 struct dm_io_region io_loc;
764 unsigned sector, n_sectors, pl_index, pl_offset;
765 int r;
767 if (unlikely(dm_integrity_failed(ic))) {
768 if (comp)
769 complete_journal_io(-1UL, comp);
770 return;
771 }
773 sector = section * ic->journal_section_sectors;
774 n_sectors = n_sections * ic->journal_section_sectors;
776 pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
777 pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
779 io_req.bi_op = op;
780 io_req.bi_op_flags = op_flags;
781 io_req.mem.type = DM_IO_PAGE_LIST;
782 if (ic->journal_io)
783 io_req.mem.ptr.pl = &ic->journal_io[pl_index];
784 else
785 io_req.mem.ptr.pl = &ic->journal[pl_index];
786 io_req.mem.offset = pl_offset;
787 if (likely(comp != NULL)) {
788 io_req.notify.fn = complete_journal_io;
789 io_req.notify.context = comp;
790 } else {
791 io_req.notify.fn = NULL;
792 }
793 io_req.client = ic->io;
794 io_loc.bdev = ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev;
795 io_loc.sector = ic->start + SB_SECTORS + sector;
796 io_loc.count = n_sectors;
798 r = dm_io(&io_req, 1, &io_loc, NULL);
799 if (unlikely(r)) {
800 dm_integrity_io_error(ic, op == REQ_OP_READ ? "reading journal" : "writing journal", r);
801 if (comp) {
802 WARN_ONCE(1, "asynchronous dm_io failed: %d", r);
803 complete_journal_io(-1UL, comp);
804 }
805 }
806 }
808 static void write_journal(struct dm_integrity_c *ic, unsigned commit_start, unsigned commit_sections)
809 {
810 struct journal_completion io_comp;
811 struct journal_completion crypt_comp_1;
812 struct journal_completion crypt_comp_2;
813 unsigned i;
815 io_comp.ic = ic;
816 init_completion(&io_comp.comp);
818 if (commit_start + commit_sections <= ic->journal_sections) {
819 io_comp.in_flight = (atomic_t)ATOMIC_INIT(1);
820 if (ic->journal_io) {
821 crypt_comp_1.ic = ic;
822 init_completion(&crypt_comp_1.comp);
823 crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
824 encrypt_journal(ic, true, commit_start, commit_sections, &crypt_comp_1);
825 wait_for_completion_io(&crypt_comp_1.comp);
826 } else {
827 for (i = 0; i < commit_sections; i++)
828 rw_section_mac(ic, commit_start + i, true);
829 }
830 rw_journal(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, commit_start,
831 commit_sections, &io_comp);
832 } else {
833 unsigned to_end;
834 io_comp.in_flight = (atomic_t)ATOMIC_INIT(2);
835 to_end = ic->journal_sections - commit_start;
836 if (ic->journal_io) {
837 crypt_comp_1.ic = ic;
838 init_completion(&crypt_comp_1.comp);
839 crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
840 encrypt_journal(ic, true, commit_start, to_end, &crypt_comp_1);
841 if (try_wait_for_completion(&crypt_comp_1.comp)) {
842 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
843 reinit_completion(&crypt_comp_1.comp);
844 crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
845 encrypt_journal(ic, true, 0, commit_sections - to_end, &crypt_comp_1);
846 wait_for_completion_io(&crypt_comp_1.comp);
847 } else {
848 crypt_comp_2.ic = ic;
849 init_completion(&crypt_comp_2.comp);
850 crypt_comp_2.in_flight = (atomic_t)ATOMIC_INIT(0);
851 encrypt_journal(ic, true, 0, commit_sections - to_end, &crypt_comp_2);
852 wait_for_completion_io(&crypt_comp_1.comp);
853 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
854 wait_for_completion_io(&crypt_comp_2.comp);
855 }
856 } else {
857 for (i = 0; i < to_end; i++)
858 rw_section_mac(ic, commit_start + i, true);
859 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
860 for (i = 0; i < commit_sections - to_end; i++)
861 rw_section_mac(ic, i, true);
862 }
863 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, 0, commit_sections - to_end, &io_comp);
864 }
866 wait_for_completion_io(&io_comp.comp);
867 }
869 static void copy_from_journal(struct dm_integrity_c *ic, unsigned section, unsigned offset,
870 unsigned n_sectors, sector_t target, io_notify_fn fn, void *data)
871 {
872 struct dm_io_request io_req;
873 struct dm_io_region io_loc;
874 int r;
875 unsigned sector, pl_index, pl_offset;
877 BUG_ON((target | n_sectors | offset) & (unsigned)(ic->sectors_per_block - 1));
879 if (unlikely(dm_integrity_failed(ic))) {
880 fn(-1UL, data);
881 return;
882 }
884 sector = section * ic->journal_section_sectors + JOURNAL_BLOCK_SECTORS + offset;
886 pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
887 pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
889 io_req.bi_op = REQ_OP_WRITE;
890 io_req.bi_op_flags = 0;
891 io_req.mem.type = DM_IO_PAGE_LIST;
892 io_req.mem.ptr.pl = &ic->journal[pl_index];
893 io_req.mem.offset = pl_offset;
894 io_req.notify.fn = fn;
895 io_req.notify.context = data;
896 io_req.client = ic->io;
897 io_loc.bdev = ic->dev->bdev;
898 io_loc.sector = target;
899 io_loc.count = n_sectors;
901 r = dm_io(&io_req, 1, &io_loc, NULL);
902 if (unlikely(r)) {
903 WARN_ONCE(1, "asynchronous dm_io failed: %d", r);
904 fn(-1UL, data);
905 }
906 }
908 static bool ranges_overlap(struct dm_integrity_range *range1, struct dm_integrity_range *range2)
909 {
910 return range1->logical_sector < range2->logical_sector + range2->n_sectors &&
911 range2->logical_sector + range2->n_sectors > range2->logical_sector;
912 }
914 static bool add_new_range(struct dm_integrity_c *ic, struct dm_integrity_range *new_range, bool check_waiting)
915 {
916 struct rb_node **n = &ic->in_progress.rb_node;
917 struct rb_node *parent;
919 BUG_ON((new_range->logical_sector | new_range->n_sectors) & (unsigned)(ic->sectors_per_block - 1));
921 if (likely(check_waiting)) {
922 struct dm_integrity_range *range;
923 list_for_each_entry(range, &ic->wait_list, wait_entry) {
924 if (unlikely(ranges_overlap(range, new_range)))
925 return false;
926 }
927 }
929 parent = NULL;
931 while (*n) {
932 struct dm_integrity_range *range = container_of(*n, struct dm_integrity_range, node);
934 parent = *n;
935 if (new_range->logical_sector + new_range->n_sectors <= range->logical_sector) {
936 n = &range->node.rb_left;
937 } else if (new_range->logical_sector >= range->logical_sector + range->n_sectors) {
938 n = &range->node.rb_right;
939 } else {
940 return false;
941 }
942 }
944 rb_link_node(&new_range->node, parent, n);
945 rb_insert_color(&new_range->node, &ic->in_progress);
947 return true;
948 }
950 static void remove_range_unlocked(struct dm_integrity_c *ic, struct dm_integrity_range *range)
951 {
952 rb_erase(&range->node, &ic->in_progress);
953 while (unlikely(!list_empty(&ic->wait_list))) {
954 struct dm_integrity_range *last_range =
955 list_first_entry(&ic->wait_list, struct dm_integrity_range, wait_entry);
956 struct task_struct *last_range_task;
957 if (!ranges_overlap(range, last_range))
958 break;
959 last_range_task = last_range->task;
960 list_del(&last_range->wait_entry);
961 if (!add_new_range(ic, last_range, false)) {
962 last_range->task = last_range_task;
963 list_add(&last_range->wait_entry, &ic->wait_list);
964 break;
965 }
966 last_range->waiting = false;
967 wake_up_process(last_range_task);
968 }
969 }
971 static void remove_range(struct dm_integrity_c *ic, struct dm_integrity_range *range)
972 {
973 unsigned long flags;
975 spin_lock_irqsave(&ic->endio_wait.lock, flags);
976 remove_range_unlocked(ic, range);
977 spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
978 }
980 static void wait_and_add_new_range(struct dm_integrity_c *ic, struct dm_integrity_range *new_range)
981 {
982 new_range->waiting = true;
983 list_add_tail(&new_range->wait_entry, &ic->wait_list);
984 new_range->task = current;
985 do {
986 __set_current_state(TASK_UNINTERRUPTIBLE);
987 spin_unlock_irq(&ic->endio_wait.lock);
988 io_schedule();
989 spin_lock_irq(&ic->endio_wait.lock);
990 } while (unlikely(new_range->waiting));
991 }
993 static void init_journal_node(struct journal_node *node)
994 {
995 RB_CLEAR_NODE(&node->node);
996 node->sector = (sector_t)-1;
997 }
999 static void add_journal_node(struct dm_integrity_c *ic, struct journal_node *node, sector_t sector)
1000 {
1001 struct rb_node **link;
1002 struct rb_node *parent;
1004 node->sector = sector;
1005 BUG_ON(!RB_EMPTY_NODE(&node->node));
1007 link = &ic->journal_tree_root.rb_node;
1008 parent = NULL;
1010 while (*link) {
1011 struct journal_node *j;
1012 parent = *link;
1013 j = container_of(parent, struct journal_node, node);
1014 if (sector < j->sector)
1015 link = &j->node.rb_left;
1016 else
1017 link = &j->node.rb_right;
1018 }
1020 rb_link_node(&node->node, parent, link);
1021 rb_insert_color(&node->node, &ic->journal_tree_root);
1022 }
1024 static void remove_journal_node(struct dm_integrity_c *ic, struct journal_node *node)
1025 {
1026 BUG_ON(RB_EMPTY_NODE(&node->node));
1027 rb_erase(&node->node, &ic->journal_tree_root);
1028 init_journal_node(node);
1029 }
1031 #define NOT_FOUND (-1U)
1033 static unsigned find_journal_node(struct dm_integrity_c *ic, sector_t sector, sector_t *next_sector)
1034 {
1035 struct rb_node *n = ic->journal_tree_root.rb_node;
1036 unsigned found = NOT_FOUND;
1037 *next_sector = (sector_t)-1;
1038 while (n) {
1039 struct journal_node *j = container_of(n, struct journal_node, node);
1040 if (sector == j->sector) {
1041 found = j - ic->journal_tree;
1042 }
1043 if (sector < j->sector) {
1044 *next_sector = j->sector;
1045 n = j->node.rb_left;
1046 } else {
1047 n = j->node.rb_right;
1048 }
1049 }
1051 return found;
1052 }
1054 static bool test_journal_node(struct dm_integrity_c *ic, unsigned pos, sector_t sector)
1055 {
1056 struct journal_node *node, *next_node;
1057 struct rb_node *next;
1059 if (unlikely(pos >= ic->journal_entries))
1060 return false;
1061 node = &ic->journal_tree[pos];
1062 if (unlikely(RB_EMPTY_NODE(&node->node)))
1063 return false;
1064 if (unlikely(node->sector != sector))
1065 return false;
1067 next = rb_next(&node->node);
1068 if (unlikely(!next))
1069 return true;
1071 next_node = container_of(next, struct journal_node, node);
1072 return next_node->sector != sector;
1073 }
1075 static bool find_newer_committed_node(struct dm_integrity_c *ic, struct journal_node *node)
1076 {
1077 struct rb_node *next;
1078 struct journal_node *next_node;
1079 unsigned next_section;
1081 BUG_ON(RB_EMPTY_NODE(&node->node));
1083 next = rb_next(&node->node);
1084 if (unlikely(!next))
1085 return false;
1087 next_node = container_of(next, struct journal_node, node);
1089 if (next_node->sector != node->sector)
1090 return false;
1092 next_section = (unsigned)(next_node - ic->journal_tree) / ic->journal_section_entries;
1093 if (next_section >= ic->committed_section &&
1094 next_section < ic->committed_section + ic->n_committed_sections)
1095 return true;
1096 if (next_section + ic->journal_sections < ic->committed_section + ic->n_committed_sections)
1097 return true;
1099 return false;
1100 }
1102 #define TAG_READ 0
1103 #define TAG_WRITE 1
1104 #define TAG_CMP 2
1106 static int dm_integrity_rw_tag(struct dm_integrity_c *ic, unsigned char *tag, sector_t *metadata_block,
1107 unsigned *metadata_offset, unsigned total_size, int op)
1108 {
1109 do {
1110 unsigned char *data, *dp;
1111 struct dm_buffer *b;
1112 unsigned to_copy;
1113 int r;
1115 r = dm_integrity_failed(ic);
1116 if (unlikely(r))
1117 return r;
1119 data = dm_bufio_read(ic->bufio, *metadata_block, &b);
1120 if (unlikely(IS_ERR(data)))
1121 return PTR_ERR(data);
1123 to_copy = min((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - *metadata_offset, total_size);
1124 dp = data + *metadata_offset;
1125 if (op == TAG_READ) {
1126 memcpy(tag, dp, to_copy);
1127 } else if (op == TAG_WRITE) {
1128 memcpy(dp, tag, to_copy);
1129 dm_bufio_mark_partial_buffer_dirty(b, *metadata_offset, *metadata_offset + to_copy);
1130 } else {
1131 /* e.g.: op == TAG_CMP */
1132 if (unlikely(memcmp(dp, tag, to_copy))) {
1133 unsigned i;
1135 for (i = 0; i < to_copy; i++) {
1136 if (dp[i] != tag[i])
1137 break;
1138 total_size--;
1139 }
1140 dm_bufio_release(b);
1141 return total_size;
1142 }
1143 }
1144 dm_bufio_release(b);
1146 tag += to_copy;
1147 *metadata_offset += to_copy;
1148 if (unlikely(*metadata_offset == 1U << SECTOR_SHIFT << ic->log2_buffer_sectors)) {
1149 (*metadata_block)++;
1150 *metadata_offset = 0;
1151 }
1152 total_size -= to_copy;
1153 } while (unlikely(total_size));
1155 return 0;
1156 }
1158 static void dm_integrity_flush_buffers(struct dm_integrity_c *ic)
1159 {
1160 int r;
1161 r = dm_bufio_write_dirty_buffers(ic->bufio);
1162 if (unlikely(r))
1163 dm_integrity_io_error(ic, "writing tags", r);
1164 }
1166 static void sleep_on_endio_wait(struct dm_integrity_c *ic)
1167 {
1168 DECLARE_WAITQUEUE(wait, current);
1169 __add_wait_queue(&ic->endio_wait, &wait);
1170 __set_current_state(TASK_UNINTERRUPTIBLE);
1171 spin_unlock_irq(&ic->endio_wait.lock);
1172 io_schedule();
1173 spin_lock_irq(&ic->endio_wait.lock);
1174 __remove_wait_queue(&ic->endio_wait, &wait);
1175 }
1177 static void autocommit_fn(struct timer_list *t)
1178 {
1179 struct dm_integrity_c *ic = from_timer(ic, t, autocommit_timer);
1181 if (likely(!dm_integrity_failed(ic)))
1182 queue_work(ic->commit_wq, &ic->commit_work);
1183 }
1185 static void schedule_autocommit(struct dm_integrity_c *ic)
1186 {
1187 if (!timer_pending(&ic->autocommit_timer))
1188 mod_timer(&ic->autocommit_timer, jiffies + ic->autocommit_jiffies);
1189 }
1191 static void submit_flush_bio(struct dm_integrity_c *ic, struct dm_integrity_io *dio)
1192 {
1193 struct bio *bio;
1194 unsigned long flags;
1196 spin_lock_irqsave(&ic->endio_wait.lock, flags);
1197 bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1198 bio_list_add(&ic->flush_bio_list, bio);
1199 spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
1201 queue_work(ic->commit_wq, &ic->commit_work);
1202 }
1204 static void do_endio(struct dm_integrity_c *ic, struct bio *bio)
1205 {
1206 int r = dm_integrity_failed(ic);
1207 if (unlikely(r) && !bio->bi_status)
1208 bio->bi_status = errno_to_blk_status(r);
1209 bio_endio(bio);
1210 }
1212 static void do_endio_flush(struct dm_integrity_c *ic, struct dm_integrity_io *dio)
1213 {
1214 struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1216 if (unlikely(dio->fua) && likely(!bio->bi_status) && likely(!dm_integrity_failed(ic)))
1217 submit_flush_bio(ic, dio);
1218 else
1219 do_endio(ic, bio);
1220 }
1222 static void dec_in_flight(struct dm_integrity_io *dio)
1223 {
1224 if (atomic_dec_and_test(&dio->in_flight)) {
1225 struct dm_integrity_c *ic = dio->ic;
1226 struct bio *bio;
1228 remove_range(ic, &dio->range);
1230 if (unlikely(dio->write))
1231 schedule_autocommit(ic);
1233 bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1235 if (unlikely(dio->bi_status) && !bio->bi_status)
1236 bio->bi_status = dio->bi_status;
1237 if (likely(!bio->bi_status) && unlikely(bio_sectors(bio) != dio->range.n_sectors)) {
1238 dio->range.logical_sector += dio->range.n_sectors;
1239 bio_advance(bio, dio->range.n_sectors << SECTOR_SHIFT);
1240 INIT_WORK(&dio->work, integrity_bio_wait);
1241 queue_work(ic->wait_wq, &dio->work);
1242 return;
1243 }
1244 do_endio_flush(ic, dio);
1245 }
1246 }
1248 static void integrity_end_io(struct bio *bio)
1249 {
1250 struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
1252 bio->bi_iter = dio->orig_bi_iter;
1253 bio->bi_disk = dio->orig_bi_disk;
1254 bio->bi_partno = dio->orig_bi_partno;
1255 if (dio->orig_bi_integrity) {
1256 bio->bi_integrity = dio->orig_bi_integrity;
1257 bio->bi_opf |= REQ_INTEGRITY;
1258 }
1259 bio->bi_end_io = dio->orig_bi_end_io;
1261 if (dio->completion)
1262 complete(dio->completion);
1264 dec_in_flight(dio);
1265 }
1267 static void integrity_sector_checksum(struct dm_integrity_c *ic, sector_t sector,
1268 const char *data, char *result)
1269 {
1270 __u64 sector_le = cpu_to_le64(sector);
1271 SHASH_DESC_ON_STACK(req, ic->internal_hash);
1272 int r;
1273 unsigned digest_size;
1275 req->tfm = ic->internal_hash;
1276 req->flags = 0;
1278 r = crypto_shash_init(req);
1279 if (unlikely(r < 0)) {
1280 dm_integrity_io_error(ic, "crypto_shash_init", r);
1281 goto failed;
1282 }
1284 r = crypto_shash_update(req, (const __u8 *)§or_le, sizeof sector_le);
1285 if (unlikely(r < 0)) {
1286 dm_integrity_io_error(ic, "crypto_shash_update", r);
1287 goto failed;
1288 }
1290 r = crypto_shash_update(req, data, ic->sectors_per_block << SECTOR_SHIFT);
1291 if (unlikely(r < 0)) {
1292 dm_integrity_io_error(ic, "crypto_shash_update", r);
1293 goto failed;
1294 }
1296 r = crypto_shash_final(req, result);
1297 if (unlikely(r < 0)) {
1298 dm_integrity_io_error(ic, "crypto_shash_final", r);
1299 goto failed;
1300 }
1302 digest_size = crypto_shash_digestsize(ic->internal_hash);
1303 if (unlikely(digest_size < ic->tag_size))
1304 memset(result + digest_size, 0, ic->tag_size - digest_size);
1306 return;
1308 failed:
1309 /* this shouldn't happen anyway, the hash functions have no reason to fail */
1310 get_random_bytes(result, ic->tag_size);
1311 }
1313 static void integrity_metadata(struct work_struct *w)
1314 {
1315 struct dm_integrity_io *dio = container_of(w, struct dm_integrity_io, work);
1316 struct dm_integrity_c *ic = dio->ic;
1318 int r;
1320 if (ic->internal_hash) {
1321 struct bvec_iter iter;
1322 struct bio_vec bv;
1323 unsigned digest_size = crypto_shash_digestsize(ic->internal_hash);
1324 struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1325 char *checksums;
1326 unsigned extra_space = unlikely(digest_size > ic->tag_size) ? digest_size - ic->tag_size : 0;
1327 char checksums_onstack[ic->tag_size + extra_space];
1328 unsigned sectors_to_process = dio->range.n_sectors;
1329 sector_t sector = dio->range.logical_sector;
1331 if (unlikely(ic->mode == 'R'))
1332 goto skip_io;
1334 checksums = kmalloc((PAGE_SIZE >> SECTOR_SHIFT >> ic->sb->log2_sectors_per_block) * ic->tag_size + extra_space,
1335 GFP_NOIO | __GFP_NORETRY | __GFP_NOWARN);
1336 if (!checksums)
1337 checksums = checksums_onstack;
1339 __bio_for_each_segment(bv, bio, iter, dio->orig_bi_iter) {
1340 unsigned pos;
1341 char *mem, *checksums_ptr;
1343 again:
1344 mem = (char *)kmap_atomic(bv.bv_page) + bv.bv_offset;
1345 pos = 0;
1346 checksums_ptr = checksums;
1347 do {
1348 integrity_sector_checksum(ic, sector, mem + pos, checksums_ptr);
1349 checksums_ptr += ic->tag_size;
1350 sectors_to_process -= ic->sectors_per_block;
1351 pos += ic->sectors_per_block << SECTOR_SHIFT;
1352 sector += ic->sectors_per_block;
1353 } while (pos < bv.bv_len && sectors_to_process && checksums != checksums_onstack);
1354 kunmap_atomic(mem);
1356 r = dm_integrity_rw_tag(ic, checksums, &dio->metadata_block, &dio->metadata_offset,
1357 checksums_ptr - checksums, !dio->write ? TAG_CMP : TAG_WRITE);
1358 if (unlikely(r)) {
1359 if (r > 0) {
1360 DMERR("Checksum failed at sector 0x%llx",
1361 (unsigned long long)(sector - ((r + ic->tag_size - 1) / ic->tag_size)));
1362 r = -EILSEQ;
1363 atomic64_inc(&ic->number_of_mismatches);
1364 }
1365 if (likely(checksums != checksums_onstack))
1366 kfree(checksums);
1367 goto error;
1368 }
1370 if (!sectors_to_process)
1371 break;
1373 if (unlikely(pos < bv.bv_len)) {
1374 bv.bv_offset += pos;
1375 bv.bv_len -= pos;
1376 goto again;
1377 }
1378 }
1380 if (likely(checksums != checksums_onstack))
1381 kfree(checksums);
1382 } else {
1383 struct bio_integrity_payload *bip = dio->orig_bi_integrity;
1385 if (bip) {
1386 struct bio_vec biv;
1387 struct bvec_iter iter;
1388 unsigned data_to_process = dio->range.n_sectors;
1389 sector_to_block(ic, data_to_process);
1390 data_to_process *= ic->tag_size;
1392 bip_for_each_vec(biv, bip, iter) {
1393 unsigned char *tag;
1394 unsigned this_len;
1396 BUG_ON(PageHighMem(biv.bv_page));
1397 tag = lowmem_page_address(biv.bv_page) + biv.bv_offset;
1398 this_len = min(biv.bv_len, data_to_process);
1399 r = dm_integrity_rw_tag(ic, tag, &dio->metadata_block, &dio->metadata_offset,
1400 this_len, !dio->write ? TAG_READ : TAG_WRITE);
1401 if (unlikely(r))
1402 goto error;
1403 data_to_process -= this_len;
1404 if (!data_to_process)
1405 break;
1406 }
1407 }
1408 }
1409 skip_io:
1410 dec_in_flight(dio);
1411 return;
1412 error:
1413 dio->bi_status = errno_to_blk_status(r);
1414 dec_in_flight(dio);
1415 }
1417 static int dm_integrity_map(struct dm_target *ti, struct bio *bio)
1418 {
1419 struct dm_integrity_c *ic = ti->private;
1420 struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
1421 struct bio_integrity_payload *bip;
1423 sector_t area, offset;
1425 dio->ic = ic;
1426 dio->bi_status = 0;
1428 if (unlikely(bio->bi_opf & REQ_PREFLUSH)) {
1429 submit_flush_bio(ic, dio);
1430 return DM_MAPIO_SUBMITTED;
1431 }
1433 dio->range.logical_sector = dm_target_offset(ti, bio->bi_iter.bi_sector);
1434 dio->write = bio_op(bio) == REQ_OP_WRITE;
1435 dio->fua = dio->write && bio->bi_opf & REQ_FUA;
1436 if (unlikely(dio->fua)) {
1437 /*
1438 * Don't pass down the FUA flag because we have to flush
1439 * disk cache anyway.
1440 */
1441 bio->bi_opf &= ~REQ_FUA;
1442 }
1443 if (unlikely(dio->range.logical_sector + bio_sectors(bio) > ic->provided_data_sectors)) {
1444 DMERR("Too big sector number: 0x%llx + 0x%x > 0x%llx",
1445 (unsigned long long)dio->range.logical_sector, bio_sectors(bio),
1446 (unsigned long long)ic->provided_data_sectors);
1447 return DM_MAPIO_KILL;
1448 }
1449 if (unlikely((dio->range.logical_sector | bio_sectors(bio)) & (unsigned)(ic->sectors_per_block - 1))) {
1450 DMERR("Bio not aligned on %u sectors: 0x%llx, 0x%x",
1451 ic->sectors_per_block,
1452 (unsigned long long)dio->range.logical_sector, bio_sectors(bio));
1453 return DM_MAPIO_KILL;
1454 }
1456 if (ic->sectors_per_block > 1) {
1457 struct bvec_iter iter;
1458 struct bio_vec bv;
1459 bio_for_each_segment(bv, bio, iter) {
1460 if (unlikely(bv.bv_len & ((ic->sectors_per_block << SECTOR_SHIFT) - 1))) {
1461 DMERR("Bio vector (%u,%u) is not aligned on %u-sector boundary",
1462 bv.bv_offset, bv.bv_len, ic->sectors_per_block);
1463 return DM_MAPIO_KILL;
1464 }
1465 }
1466 }
1468 bip = bio_integrity(bio);
1469 if (!ic->internal_hash) {
1470 if (bip) {
1471 unsigned wanted_tag_size = bio_sectors(bio) >> ic->sb->log2_sectors_per_block;
1472 if (ic->log2_tag_size >= 0)
1473 wanted_tag_size <<= ic->log2_tag_size;
1474 else
1475 wanted_tag_size *= ic->tag_size;
1476 if (unlikely(wanted_tag_size != bip->bip_iter.bi_size)) {
1477 DMERR("Invalid integrity data size %u, expected %u", bip->bip_iter.bi_size, wanted_tag_size);
1478 return DM_MAPIO_KILL;
1479 }
1480 }
1481 } else {
1482 if (unlikely(bip != NULL)) {
1483 DMERR("Unexpected integrity data when using internal hash");
1484 return DM_MAPIO_KILL;
1485 }
1486 }
1488 if (unlikely(ic->mode == 'R') && unlikely(dio->write))
1489 return DM_MAPIO_KILL;
1491 get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
1492 dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset, &dio->metadata_offset);
1493 bio->bi_iter.bi_sector = get_data_sector(ic, area, offset);
1495 dm_integrity_map_continue(dio, true);
1496 return DM_MAPIO_SUBMITTED;
1497 }
1499 static bool __journal_read_write(struct dm_integrity_io *dio, struct bio *bio,
1500 unsigned journal_section, unsigned journal_entry)
1501 {
1502 struct dm_integrity_c *ic = dio->ic;
1503 sector_t logical_sector;
1504 unsigned n_sectors;
1506 logical_sector = dio->range.logical_sector;
1507 n_sectors = dio->range.n_sectors;
1508 do {
1509 struct bio_vec bv = bio_iovec(bio);
1510 char *mem;
1512 if (unlikely(bv.bv_len >> SECTOR_SHIFT > n_sectors))
1513 bv.bv_len = n_sectors << SECTOR_SHIFT;
1514 n_sectors -= bv.bv_len >> SECTOR_SHIFT;
1515 bio_advance_iter(bio, &bio->bi_iter, bv.bv_len);
1516 retry_kmap:
1517 mem = kmap_atomic(bv.bv_page);
1518 if (likely(dio->write))
1519 flush_dcache_page(bv.bv_page);
1521 do {
1522 struct journal_entry *je = access_journal_entry(ic, journal_section, journal_entry);
1524 if (unlikely(!dio->write)) {
1525 struct journal_sector *js;
1526 char *mem_ptr;
1527 unsigned s;
1529 if (unlikely(journal_entry_is_inprogress(je))) {
1530 flush_dcache_page(bv.bv_page);
1531 kunmap_atomic(mem);
1533 __io_wait_event(ic->copy_to_journal_wait, !journal_entry_is_inprogress(je));
1534 goto retry_kmap;
1535 }
1536 smp_rmb();
1537 BUG_ON(journal_entry_get_sector(je) != logical_sector);
1538 js = access_journal_data(ic, journal_section, journal_entry);
1539 mem_ptr = mem + bv.bv_offset;
1540 s = 0;
1541 do {
1542 memcpy(mem_ptr, js, JOURNAL_SECTOR_DATA);
1543 *(commit_id_t *)(mem_ptr + JOURNAL_SECTOR_DATA) = je->last_bytes[s];
1544 js++;
1545 mem_ptr += 1 << SECTOR_SHIFT;
1546 } while (++s < ic->sectors_per_block);
1547 #ifdef INTERNAL_VERIFY
1548 if (ic->internal_hash) {
1549 char checksums_onstack[max(crypto_shash_digestsize(ic->internal_hash), ic->tag_size)];
1551 integrity_sector_checksum(ic, logical_sector, mem + bv.bv_offset, checksums_onstack);
1552 if (unlikely(memcmp(checksums_onstack, journal_entry_tag(ic, je), ic->tag_size))) {
1553 DMERR("Checksum failed when reading from journal, at sector 0x%llx",
1554 (unsigned long long)logical_sector);
1555 }
1556 }
1557 #endif
1558 }
1560 if (!ic->internal_hash) {
1561 struct bio_integrity_payload *bip = bio_integrity(bio);
1562 unsigned tag_todo = ic->tag_size;
1563 char *tag_ptr = journal_entry_tag(ic, je);
1565 if (bip) do {
1566 struct bio_vec biv = bvec_iter_bvec(bip->bip_vec, bip->bip_iter);
1567 unsigned tag_now = min(biv.bv_len, tag_todo);
1568 char *tag_addr;
1569 BUG_ON(PageHighMem(biv.bv_page));
1570 tag_addr = lowmem_page_address(biv.bv_page) + biv.bv_offset;
1571 if (likely(dio->write))
1572 memcpy(tag_ptr, tag_addr, tag_now);
1573 else
1574 memcpy(tag_addr, tag_ptr, tag_now);
1575 bvec_iter_advance(bip->bip_vec, &bip->bip_iter, tag_now);
1576 tag_ptr += tag_now;
1577 tag_todo -= tag_now;
1578 } while (unlikely(tag_todo)); else {
1579 if (likely(dio->write))
1580 memset(tag_ptr, 0, tag_todo);
1581 }
1582 }
1584 if (likely(dio->write)) {
1585 struct journal_sector *js;
1586 unsigned s;
1588 js = access_journal_data(ic, journal_section, journal_entry);
1589 memcpy(js, mem + bv.bv_offset, ic->sectors_per_block << SECTOR_SHIFT);
1591 s = 0;
1592 do {
1593 je->last_bytes[s] = js[s].commit_id;
1594 } while (++s < ic->sectors_per_block);
1596 if (ic->internal_hash) {
1597 unsigned digest_size = crypto_shash_digestsize(ic->internal_hash);
1598 if (unlikely(digest_size > ic->tag_size)) {
1599 char checksums_onstack[digest_size];
1600 integrity_sector_checksum(ic, logical_sector, (char *)js, checksums_onstack);
1601 memcpy(journal_entry_tag(ic, je), checksums_onstack, ic->tag_size);
1602 } else
1603 integrity_sector_checksum(ic, logical_sector, (char *)js, journal_entry_tag(ic, je));
1604 }
1606 journal_entry_set_sector(je, logical_sector);
1607 }
1608 logical_sector += ic->sectors_per_block;
1610 journal_entry++;
1611 if (unlikely(journal_entry == ic->journal_section_entries)) {
1612 journal_entry = 0;
1613 journal_section++;
1614 wraparound_section(ic, &journal_section);
1615 }
1617 bv.bv_offset += ic->sectors_per_block << SECTOR_SHIFT;
1618 } while (bv.bv_len -= ic->sectors_per_block << SECTOR_SHIFT);
1620 if (unlikely(!dio->write))
1621 flush_dcache_page(bv.bv_page);
1622 kunmap_atomic(mem);
1623 } while (n_sectors);
1625 if (likely(dio->write)) {
1626 smp_mb();
1627 if (unlikely(waitqueue_active(&ic->copy_to_journal_wait)))
1628 wake_up(&ic->copy_to_journal_wait);
1629 if (READ_ONCE(ic->free_sectors) <= ic->free_sectors_threshold) {
1630 queue_work(ic->commit_wq, &ic->commit_work);
1631 } else {
1632 schedule_autocommit(ic);
1633 }
1634 } else {
1635 remove_range(ic, &dio->range);
1636 }
1638 if (unlikely(bio->bi_iter.bi_size)) {
1639 sector_t area, offset;
1641 dio->range.logical_sector = logical_sector;
1642 get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
1643 dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset, &dio->metadata_offset);
1644 return true;
1645 }
1647 return false;
1648 }
1650 static void dm_integrity_map_continue(struct dm_integrity_io *dio, bool from_map)
1651 {
1652 struct dm_integrity_c *ic = dio->ic;
1653 struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1654 unsigned journal_section, journal_entry;
1655 unsigned journal_read_pos;
1656 struct completion read_comp;
1657 bool need_sync_io = ic->internal_hash && !dio->write;
1659 if (need_sync_io && from_map) {
1660 INIT_WORK(&dio->work, integrity_bio_wait);
1661 queue_work(ic->metadata_wq, &dio->work);
1662 return;
1663 }
1665 lock_retry:
1666 spin_lock_irq(&ic->endio_wait.lock);
1667 retry:
1668 if (unlikely(dm_integrity_failed(ic))) {
1669 spin_unlock_irq(&ic->endio_wait.lock);
1670 do_endio(ic, bio);
1671 return;
1672 }
1673 dio->range.n_sectors = bio_sectors(bio);
1674 journal_read_pos = NOT_FOUND;
1675 if (likely(ic->mode == 'J')) {
1676 if (dio->write) {
1677 unsigned next_entry, i, pos;
1678 unsigned ws, we, range_sectors;
1680 dio->range.n_sectors = min(dio->range.n_sectors,
1681 ic->free_sectors << ic->sb->log2_sectors_per_block);
1682 if (unlikely(!dio->range.n_sectors)) {
1683 if (from_map)
1684 goto offload_to_thread;
1685 sleep_on_endio_wait(ic);
1686 goto retry;
1687 }
1688 range_sectors = dio->range.n_sectors >> ic->sb->log2_sectors_per_block;
1689 ic->free_sectors -= range_sectors;
1690 journal_section = ic->free_section;
1691 journal_entry = ic->free_section_entry;
1693 next_entry = ic->free_section_entry + range_sectors;
1694 ic->free_section_entry = next_entry % ic->journal_section_entries;
1695 ic->free_section += next_entry / ic->journal_section_entries;
1696 ic->n_uncommitted_sections += next_entry / ic->journal_section_entries;
1697 wraparound_section(ic, &ic->free_section);
1699 pos = journal_section * ic->journal_section_entries + journal_entry;
1700 ws = journal_section;
1701 we = journal_entry;
1702 i = 0;
1703 do {
1704 struct journal_entry *je;
1706 add_journal_node(ic, &ic->journal_tree[pos], dio->range.logical_sector + i);
1707 pos++;
1708 if (unlikely(pos >= ic->journal_entries))
1709 pos = 0;
1711 je = access_journal_entry(ic, ws, we);
1712 BUG_ON(!journal_entry_is_unused(je));
1713 journal_entry_set_inprogress(je);
1714 we++;
1715 if (unlikely(we == ic->journal_section_entries)) {
1716 we = 0;
1717 ws++;
1718 wraparound_section(ic, &ws);
1719 }
1720 } while ((i += ic->sectors_per_block) < dio->range.n_sectors);
1722 spin_unlock_irq(&ic->endio_wait.lock);
1723 goto journal_read_write;
1724 } else {
1725 sector_t next_sector;
1726 journal_read_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
1727 if (likely(journal_read_pos == NOT_FOUND)) {
1728 if (unlikely(dio->range.n_sectors > next_sector - dio->range.logical_sector))
1729 dio->range.n_sectors = next_sector - dio->range.logical_sector;
1730 } else {
1731 unsigned i;
1732 unsigned jp = journal_read_pos + 1;
1733 for (i = ic->sectors_per_block; i < dio->range.n_sectors; i += ic->sectors_per_block, jp++) {
1734 if (!test_journal_node(ic, jp, dio->range.logical_sector + i))
1735 break;
1736 }
1737 dio->range.n_sectors = i;
1738 }
1739 }
1740 }
1741 if (unlikely(!add_new_range(ic, &dio->range, true))) {
1742 /*
1743 * We must not sleep in the request routine because it could
1744 * stall bios on current->bio_list.
1745 * So, we offload the bio to a workqueue if we have to sleep.
1746 */
1747 if (from_map) {
1748 offload_to_thread:
1749 spin_unlock_irq(&ic->endio_wait.lock);
1750 INIT_WORK(&dio->work, integrity_bio_wait);
1751 queue_work(ic->wait_wq, &dio->work);
1752 return;
1753 }
1754 wait_and_add_new_range(ic, &dio->range);
1755 }
1756 spin_unlock_irq(&ic->endio_wait.lock);
1758 if (unlikely(journal_read_pos != NOT_FOUND)) {
1759 journal_section = journal_read_pos / ic->journal_section_entries;
1760 journal_entry = journal_read_pos % ic->journal_section_entries;
1761 goto journal_read_write;
1762 }
1764 dio->in_flight = (atomic_t)ATOMIC_INIT(2);
1766 if (need_sync_io) {
1767 init_completion(&read_comp);
1768 dio->completion = &read_comp;
1769 } else
1770 dio->completion = NULL;
1772 dio->orig_bi_iter = bio->bi_iter;
1774 dio->orig_bi_disk = bio->bi_disk;
1775 dio->orig_bi_partno = bio->bi_partno;
1776 bio_set_dev(bio, ic->dev->bdev);
1778 dio->orig_bi_integrity = bio_integrity(bio);
1779 bio->bi_integrity = NULL;
1780 bio->bi_opf &= ~REQ_INTEGRITY;
1782 dio->orig_bi_end_io = bio->bi_end_io;
1783 bio->bi_end_io = integrity_end_io;
1785 bio->bi_iter.bi_size = dio->range.n_sectors << SECTOR_SHIFT;
1786 generic_make_request(bio);
1788 if (need_sync_io) {
1789 wait_for_completion_io(&read_comp);
1790 if (unlikely(ic->recalc_wq != NULL) &&
1791 ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING) &&
1792 dio->range.logical_sector + dio->range.n_sectors > le64_to_cpu(ic->sb->recalc_sector))
1793 goto skip_check;
1794 if (likely(!bio->bi_status))
1795 integrity_metadata(&dio->work);
1796 else
1797 skip_check:
1798 dec_in_flight(dio);
1800 } else {
1801 INIT_WORK(&dio->work, integrity_metadata);
1802 queue_work(ic->metadata_wq, &dio->work);
1803 }
1805 return;
1807 journal_read_write:
1808 if (unlikely(__journal_read_write(dio, bio, journal_section, journal_entry)))
1809 goto lock_retry;
1811 do_endio_flush(ic, dio);
1812 }
1815 static void integrity_bio_wait(struct work_struct *w)
1816 {
1817 struct dm_integrity_io *dio = container_of(w, struct dm_integrity_io, work);
1819 dm_integrity_map_continue(dio, false);
1820 }
1822 static void pad_uncommitted(struct dm_integrity_c *ic)
1823 {
1824 if (ic->free_section_entry) {
1825 ic->free_sectors -= ic->journal_section_entries - ic->free_section_entry;
1826 ic->free_section_entry = 0;
1827 ic->free_section++;
1828 wraparound_section(ic, &ic->free_section);
1829 ic->n_uncommitted_sections++;
1830 }
1831 WARN_ON(ic->journal_sections * ic->journal_section_entries !=
1832 (ic->n_uncommitted_sections + ic->n_committed_sections) * ic->journal_section_entries + ic->free_sectors);
1833 }
1835 static void integrity_commit(struct work_struct *w)
1836 {
1837 struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, commit_work);
1838 unsigned commit_start, commit_sections;
1839 unsigned i, j, n;
1840 struct bio *flushes;
1842 del_timer(&ic->autocommit_timer);
1844 spin_lock_irq(&ic->endio_wait.lock);
1845 flushes = bio_list_get(&ic->flush_bio_list);
1846 if (unlikely(ic->mode != 'J')) {
1847 spin_unlock_irq(&ic->endio_wait.lock);
1848 dm_integrity_flush_buffers(ic);
1849 goto release_flush_bios;
1850 }
1852 pad_uncommitted(ic);
1853 commit_start = ic->uncommitted_section;
1854 commit_sections = ic->n_uncommitted_sections;
1855 spin_unlock_irq(&ic->endio_wait.lock);
1857 if (!commit_sections)
1858 goto release_flush_bios;
1860 i = commit_start;
1861 for (n = 0; n < commit_sections; n++) {
1862 for (j = 0; j < ic->journal_section_entries; j++) {
1863 struct journal_entry *je;
1864 je = access_journal_entry(ic, i, j);
1865 io_wait_event(ic->copy_to_journal_wait, !journal_entry_is_inprogress(je));
1866 }
1867 for (j = 0; j < ic->journal_section_sectors; j++) {
1868 struct journal_sector *js;
1869 js = access_journal(ic, i, j);
1870 js->commit_id = dm_integrity_commit_id(ic, i, j, ic->commit_seq);
1871 }
1872 i++;
1873 if (unlikely(i >= ic->journal_sections))
1874 ic->commit_seq = next_commit_seq(ic->commit_seq);
1875 wraparound_section(ic, &i);
1876 }
1877 smp_rmb();
1879 write_journal(ic, commit_start, commit_sections);
1881 spin_lock_irq(&ic->endio_wait.lock);
1882 ic->uncommitted_section += commit_sections;
1883 wraparound_section(ic, &ic->uncommitted_section);
1884 ic->n_uncommitted_sections -= commit_sections;
1885 ic->n_committed_sections += commit_sections;
1886 spin_unlock_irq(&ic->endio_wait.lock);
1888 if (READ_ONCE(ic->free_sectors) <= ic->free_sectors_threshold)
1889 queue_work(ic->writer_wq, &ic->writer_work);
1891 release_flush_bios:
1892 while (flushes) {
1893 struct bio *next = flushes->bi_next;
1894 flushes->bi_next = NULL;
1895 do_endio(ic, flushes);
1896 flushes = next;
1897 }
1898 }
1900 static void complete_copy_from_journal(unsigned long error, void *context)
1901 {
1902 struct journal_io *io = context;
1903 struct journal_completion *comp = io->comp;
1904 struct dm_integrity_c *ic = comp->ic;
1905 remove_range(ic, &io->range);
1906 mempool_free(io, &ic->journal_io_mempool);
1907 if (unlikely(error != 0))
1908 dm_integrity_io_error(ic, "copying from journal", -EIO);
1909 complete_journal_op(comp);
1910 }
1912 static void restore_last_bytes(struct dm_integrity_c *ic, struct journal_sector *js,
1913 struct journal_entry *je)
1914 {
1915 unsigned s = 0;
1916 do {
1917 js->commit_id = je->last_bytes[s];
1918 js++;
1919 } while (++s < ic->sectors_per_block);
1920 }
1922 static void do_journal_write(struct dm_integrity_c *ic, unsigned write_start,
1923 unsigned write_sections, bool from_replay)
1924 {
1925 unsigned i, j, n;
1926 struct journal_completion comp;
1927 struct blk_plug plug;
1929 blk_start_plug(&plug);
1931 comp.ic = ic;
1932 comp.in_flight = (atomic_t)ATOMIC_INIT(1);
1933 init_completion(&comp.comp);
1935 i = write_start;
1936 for (n = 0; n < write_sections; n++, i++, wraparound_section(ic, &i)) {
1937 #ifndef INTERNAL_VERIFY
1938 if (unlikely(from_replay))
1939 #endif
1940 rw_section_mac(ic, i, false);
1941 for (j = 0; j < ic->journal_section_entries; j++) {
1942 struct journal_entry *je = access_journal_entry(ic, i, j);
1943 sector_t sec, area, offset;
1944 unsigned k, l, next_loop;
1945 sector_t metadata_block;
1946 unsigned metadata_offset;
1947 struct journal_io *io;
1949 if (journal_entry_is_unused(je))
1950 continue;
1951 BUG_ON(unlikely(journal_entry_is_inprogress(je)) && !from_replay);
1952 sec = journal_entry_get_sector(je);
1953 if (unlikely(from_replay)) {
1954 if (unlikely(sec & (unsigned)(ic->sectors_per_block - 1))) {
1955 dm_integrity_io_error(ic, "invalid sector in journal", -EIO);
1956 sec &= ~(sector_t)(ic->sectors_per_block - 1);
1957 }
1958 }
1959 get_area_and_offset(ic, sec, &area, &offset);
1960 restore_last_bytes(ic, access_journal_data(ic, i, j), je);
1961 for (k = j + 1; k < ic->journal_section_entries; k++) {
1962 struct journal_entry *je2 = access_journal_entry(ic, i, k);
1963 sector_t sec2, area2, offset2;
1964 if (journal_entry_is_unused(je2))
1965 break;
1966 BUG_ON(unlikely(journal_entry_is_inprogress(je2)) && !from_replay);
1967 sec2 = journal_entry_get_sector(je2);
1968 get_area_and_offset(ic, sec2, &area2, &offset2);
1969 if (area2 != area || offset2 != offset + ((k - j) << ic->sb->log2_sectors_per_block))
1970 break;
1971 restore_last_bytes(ic, access_journal_data(ic, i, k), je2);
1972 }
1973 next_loop = k - 1;
1975 io = mempool_alloc(&ic->journal_io_mempool, GFP_NOIO);
1976 io->comp = ∁
1977 io->range.logical_sector = sec;
1978 io->range.n_sectors = (k - j) << ic->sb->log2_sectors_per_block;
1980 spin_lock_irq(&ic->endio_wait.lock);
1981 if (unlikely(!add_new_range(ic, &io->range, true)))
1982 wait_and_add_new_range(ic, &io->range);
1984 if (likely(!from_replay)) {
1985 struct journal_node *section_node = &ic->journal_tree[i * ic->journal_section_entries];
1987 /* don't write if there is newer committed sector */
1988 while (j < k && find_newer_committed_node(ic, §ion_node[j])) {
1989 struct journal_entry *je2 = access_journal_entry(ic, i, j);
1991 journal_entry_set_unused(je2);
1992 remove_journal_node(ic, §ion_node[j]);
1993 j++;
1994 sec += ic->sectors_per_block;
1995 offset += ic->sectors_per_block;
1996 }
1997 while (j < k && find_newer_committed_node(ic, §ion_node[k - 1])) {
1998 struct journal_entry *je2 = access_journal_entry(ic, i, k - 1);
2000 journal_entry_set_unused(je2);
2001 remove_journal_node(ic, §ion_node[k - 1]);
2002 k--;
2003 }
2004 if (j == k) {
2005 remove_range_unlocked(ic, &io->range);
2006 spin_unlock_irq(&ic->endio_wait.lock);
2007 mempool_free(io, &ic->journal_io_mempool);
2008 goto skip_io;
2009 }
2010 for (l = j; l < k; l++) {
2011 remove_journal_node(ic, §ion_node[l]);
2012 }
2013 }
2014 spin_unlock_irq(&ic->endio_wait.lock);
2016 metadata_block = get_metadata_sector_and_offset(ic, area, offset, &metadata_offset);
2017 for (l = j; l < k; l++) {
2018 int r;
2019 struct journal_entry *je2 = access_journal_entry(ic, i, l);
2021 if (
2022 #ifndef INTERNAL_VERIFY
2023 unlikely(from_replay) &&
2024 #endif
2025 ic->internal_hash) {
2026 char test_tag[max(crypto_shash_digestsize(ic->internal_hash), ic->tag_size)];
2028 integrity_sector_checksum(ic, sec + ((l - j) << ic->sb->log2_sectors_per_block),
2029 (char *)access_journal_data(ic, i, l), test_tag);
2030 if (unlikely(memcmp(test_tag, journal_entry_tag(ic, je2), ic->tag_size)))
2031 dm_integrity_io_error(ic, "tag mismatch when replaying journal", -EILSEQ);
2032 }
2034 journal_entry_set_unused(je2);
2035 r = dm_integrity_rw_tag(ic, journal_entry_tag(ic, je2), &metadata_block, &metadata_offset,
2036 ic->tag_size, TAG_WRITE);
2037 if (unlikely(r)) {
2038 dm_integrity_io_error(ic, "reading tags", r);
2039 }
2040 }
2042 atomic_inc(&comp.in_flight);
2043 copy_from_journal(ic, i, j << ic->sb->log2_sectors_per_block,
2044 (k - j) << ic->sb->log2_sectors_per_block,
2045 get_data_sector(ic, area, offset),
2046 complete_copy_from_journal, io);
2047 skip_io:
2048 j = next_loop;
2049 }
2050 }
2052 dm_bufio_write_dirty_buffers_async(ic->bufio);
2054 blk_finish_plug(&plug);
2056 complete_journal_op(&comp);
2057 wait_for_completion_io(&comp.comp);
2059 dm_integrity_flush_buffers(ic);
2060 }
2062 static void integrity_writer(struct work_struct *w)
2063 {
2064 struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, writer_work);
2065 unsigned write_start, write_sections;
2067 unsigned prev_free_sectors;
2069 /* the following test is not needed, but it tests the replay code */
2070 if (READ_ONCE(ic->suspending) && !ic->meta_dev)
2071 return;
2073 spin_lock_irq(&ic->endio_wait.lock);
2074 write_start = ic->committed_section;
2075 write_sections = ic->n_committed_sections;
2076 spin_unlock_irq(&ic->endio_wait.lock);
2078 if (!write_sections)
2079 return;
2081 do_journal_write(ic, write_start, write_sections, false);
2083 spin_lock_irq(&ic->endio_wait.lock);
2085 ic->committed_section += write_sections;
2086 wraparound_section(ic, &ic->committed_section);
2087 ic->n_committed_sections -= write_sections;
2089 prev_free_sectors = ic->free_sectors;
2090 ic->free_sectors += write_sections * ic->journal_section_entries;
2091 if (unlikely(!prev_free_sectors))
2092 wake_up_locked(&ic->endio_wait);
2094 spin_unlock_irq(&ic->endio_wait.lock);
2095 }
2097 static void recalc_write_super(struct dm_integrity_c *ic)
2098 {
2099 int r;
2101 dm_integrity_flush_buffers(ic);
2102 if (dm_integrity_failed(ic))
2103 return;
2105 sb_set_version(ic);
2106 r = sync_rw_sb(ic, REQ_OP_WRITE, 0);
2107 if (unlikely(r))
2108 dm_integrity_io_error(ic, "writing superblock", r);
2109 }
2111 static void integrity_recalc(struct work_struct *w)
2112 {
2113 struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, recalc_work);
2114 struct dm_integrity_range range;
2115 struct dm_io_request io_req;
2116 struct dm_io_region io_loc;
2117 sector_t area, offset;
2118 sector_t metadata_block;
2119 unsigned metadata_offset;
2120 __u8 *t;
2121 unsigned i;
2122 int r;
2123 unsigned super_counter = 0;
2125 spin_lock_irq(&ic->endio_wait.lock);
2127 next_chunk:
2129 if (unlikely(READ_ONCE(ic->suspending)))
2130 goto unlock_ret;
2132 range.logical_sector = le64_to_cpu(ic->sb->recalc_sector);
2133 if (unlikely(range.logical_sector >= ic->provided_data_sectors))
2134 goto unlock_ret;
2136 get_area_and_offset(ic, range.logical_sector, &area, &offset);
2137 range.n_sectors = min((sector_t)RECALC_SECTORS, ic->provided_data_sectors - range.logical_sector);
2138 if (!ic->meta_dev)
2139 range.n_sectors = min(range.n_sectors, (1U << ic->sb->log2_interleave_sectors) - (unsigned)offset);
2141 if (unlikely(!add_new_range(ic, &range, true)))
2142 wait_and_add_new_range(ic, &range);
2144 spin_unlock_irq(&ic->endio_wait.lock);
2146 if (unlikely(++super_counter == RECALC_WRITE_SUPER)) {
2147 recalc_write_super(ic);
2148 super_counter = 0;
2149 }
2151 if (unlikely(dm_integrity_failed(ic)))
2152 goto err;
2154 io_req.bi_op = REQ_OP_READ;
2155 io_req.bi_op_flags = 0;
2156 io_req.mem.type = DM_IO_VMA;
2157 io_req.mem.ptr.addr = ic->recalc_buffer;
2158 io_req.notify.fn = NULL;
2159 io_req.client = ic->io;
2160 io_loc.bdev = ic->dev->bdev;
2161 io_loc.sector = get_data_sector(ic, area, offset);
2162 io_loc.count = range.n_sectors;
2164 r = dm_io(&io_req, 1, &io_loc, NULL);
2165 if (unlikely(r)) {
2166 dm_integrity_io_error(ic, "reading data", r);
2167 goto err;
2168 }
2170 t = ic->recalc_tags;
2171 for (i = 0; i < range.n_sectors; i += ic->sectors_per_block) {
2172 integrity_sector_checksum(ic, range.logical_sector + i, ic->recalc_buffer + (i << SECTOR_SHIFT), t);
2173 t += ic->tag_size;
2174 }
2176 metadata_block = get_metadata_sector_and_offset(ic, area, offset, &metadata_offset);
2178 r = dm_integrity_rw_tag(ic, ic->recalc_tags, &metadata_block, &metadata_offset, t - ic->recalc_tags, TAG_WRITE);
2179 if (unlikely(r)) {
2180 dm_integrity_io_error(ic, "writing tags", r);
2181 goto err;
2182 }
2184 spin_lock_irq(&ic->endio_wait.lock);
2185 remove_range_unlocked(ic, &range);
2186 ic->sb->recalc_sector = cpu_to_le64(range.logical_sector + range.n_sectors);
2187 goto next_chunk;
2189 err:
2190 remove_range(ic, &range);
2191 return;
2193 unlock_ret:
2194 spin_unlock_irq(&ic->endio_wait.lock);
2196 recalc_write_super(ic);
2197 }
2199 static void init_journal(struct dm_integrity_c *ic, unsigned start_section,
2200 unsigned n_sections, unsigned char commit_seq)
2201 {
2202 unsigned i, j, n;
2204 if (!n_sections)
2205 return;
2207 for (n = 0; n < n_sections; n++) {
2208 i = start_section + n;
2209 wraparound_section(ic, &i);
2210 for (j = 0; j < ic->journal_section_sectors; j++) {
2211 struct journal_sector *js = access_journal(ic, i, j);
2212 memset(&js->entries, 0, JOURNAL_SECTOR_DATA);
2213 js->commit_id = dm_integrity_commit_id(ic, i, j, commit_seq);
2214 }
2215 for (j = 0; j < ic->journal_section_entries; j++) {
2216 struct journal_entry *je = access_journal_entry(ic, i, j);
2217 journal_entry_set_unused(je);
2218 }
2219 }
2221 write_journal(ic, start_section, n_sections);
2222 }
2224 static int find_commit_seq(struct dm_integrity_c *ic, unsigned i, unsigned j, commit_id_t id)
2225 {
2226 unsigned char k;
2227 for (k = 0; k < N_COMMIT_IDS; k++) {
2228 if (dm_integrity_commit_id(ic, i, j, k) == id)
2229 return k;
2230 }
2231 dm_integrity_io_error(ic, "journal commit id", -EIO);
2232 return -EIO;
2233 }
2235 static void replay_journal(struct dm_integrity_c *ic)
2236 {
2237 unsigned i, j;
2238 bool used_commit_ids[N_COMMIT_IDS];
2239 unsigned max_commit_id_sections[N_COMMIT_IDS];
2240 unsigned write_start, write_sections;
2241 unsigned continue_section;
2242 bool journal_empty;
2243 unsigned char unused, last_used, want_commit_seq;
2245 if (ic->mode == 'R')
2246 return;
2248 if (ic->journal_uptodate)
2249 return;
2251 last_used = 0;
2252 write_start = 0;
2254 if (!ic->just_formatted) {
2255 DEBUG_print("reading journal\n");
2256 rw_journal(ic, REQ_OP_READ, 0, 0, ic->journal_sections, NULL);
2257 if (ic->journal_io)
2258 DEBUG_bytes(lowmem_page_address(ic->journal_io[0].page), 64, "read journal");
2259 if (ic->journal_io) {
2260 struct journal_completion crypt_comp;
2261 crypt_comp.ic = ic;
2262 init_completion(&crypt_comp.comp);
2263 crypt_comp.in_flight = (atomic_t)ATOMIC_INIT(0);
2264 encrypt_journal(ic, false, 0, ic->journal_sections, &crypt_comp);
2265 wait_for_completion(&crypt_comp.comp);
2266 }
2267 DEBUG_bytes(lowmem_page_address(ic->journal[0].page), 64, "decrypted journal");
2268 }
2270 if (dm_integrity_failed(ic))
2271 goto clear_journal;
2273 journal_empty = true;
2274 memset(used_commit_ids, 0, sizeof used_commit_ids);
2275 memset(max_commit_id_sections, 0, sizeof max_commit_id_sections);
2276 for (i = 0; i < ic->journal_sections; i++) {
2277 for (j = 0; j < ic->journal_section_sectors; j++) {
2278 int k;
2279 struct journal_sector *js = access_journal(ic, i, j);
2280 k = find_commit_seq(ic, i, j, js->commit_id);
2281 if (k < 0)
2282 goto clear_journal;
2283 used_commit_ids[k] = true;
2284 max_commit_id_sections[k] = i;
2285 }
2286 if (journal_empty) {
2287 for (j = 0; j < ic->journal_section_entries; j++) {
2288 struct journal_entry *je = access_journal_entry(ic, i, j);
2289 if (!journal_entry_is_unused(je)) {
2290 journal_empty = false;
2291 break;
2292 }
2293 }
2294 }
2295 }
2297 if (!used_commit_ids[N_COMMIT_IDS - 1]) {
2298 unused = N_COMMIT_IDS - 1;
2299 while (unused && !used_commit_ids[unused - 1])
2300 unused--;
2301 } else {
2302 for (unused = 0; unused < N_COMMIT_IDS; unused++)
2303 if (!used_commit_ids[unused])
2304 break;
2305 if (unused == N_COMMIT_IDS) {
2306 dm_integrity_io_error(ic, "journal commit ids", -EIO);
2307 goto clear_journal;
2308 }
2309 }
2310 DEBUG_print("first unused commit seq %d [%d,%d,%d,%d]\n",
2311 unused, used_commit_ids[0], used_commit_ids[1],
2312 used_commit_ids[2], used_commit_ids[3]);
2314 last_used = prev_commit_seq(unused);
2315 want_commit_seq = prev_commit_seq(last_used);
2317 if (!used_commit_ids[want_commit_seq] && used_commit_ids[prev_commit_seq(want_commit_seq)])
2318 journal_empty = true;
2320 write_start = max_commit_id_sections[last_used] + 1;
2321 if (unlikely(write_start >= ic->journal_sections))
2322 want_commit_seq = next_commit_seq(want_commit_seq);
2323 wraparound_section(ic, &write_start);
2325 i = write_start;
2326 for (write_sections = 0; write_sections < ic->journal_sections; write_sections++) {
2327 for (j = 0; j < ic->journal_section_sectors; j++) {
2328 struct journal_sector *js = access_journal(ic, i, j);
2330 if (js->commit_id != dm_integrity_commit_id(ic, i, j, want_commit_seq)) {
2331 /*
2332 * This could be caused by crash during writing.
2333 * We won't replay the inconsistent part of the
2334 * journal.
2335 */
2336 DEBUG_print("commit id mismatch at position (%u, %u): %d != %d\n",
2337 i, j, find_commit_seq(ic, i, j, js->commit_id), want_commit_seq);
2338 goto brk;
2339 }
2340 }
2341 i++;
2342 if (unlikely(i >= ic->journal_sections))
2343 want_commit_seq = next_commit_seq(want_commit_seq);
2344 wraparound_section(ic, &i);
2345 }
2346 brk:
2348 if (!journal_empty) {
2349 DEBUG_print("replaying %u sections, starting at %u, commit seq %d\n",
2350 write_sections, write_start, want_commit_seq);
2351 do_journal_write(ic, write_start, write_sections, true);
2352 }
2354 if (write_sections == ic->journal_sections && (ic->mode == 'J' || journal_empty)) {
2355 continue_section = write_start;
2356 ic->commit_seq = want_commit_seq;
2357 DEBUG_print("continuing from section %u, commit seq %d\n", write_start, ic->commit_seq);
2358 } else {
2359 unsigned s;
2360 unsigned char erase_seq;
2361 clear_journal:
2362 DEBUG_print("clearing journal\n");
2364 erase_seq = prev_commit_seq(prev_commit_seq(last_used));
2365 s = write_start;
2366 init_journal(ic, s, 1, erase_seq);
2367 s++;
2368 wraparound_section(ic, &s);
2369 if (ic->journal_sections >= 2) {
2370 init_journal(ic, s, ic->journal_sections - 2, erase_seq);
2371 s += ic->journal_sections - 2;
2372 wraparound_section(ic, &s);
2373 init_journal(ic, s, 1, erase_seq);
2374 }
2376 continue_section = 0;
2377 ic->commit_seq = next_commit_seq(erase_seq);
2378 }
2380 ic->committed_section = continue_section;
2381 ic->n_committed_sections = 0;
2383 ic->uncommitted_section = continue_section;
2384 ic->n_uncommitted_sections = 0;
2386 ic->free_section = continue_section;
2387 ic->free_section_entry = 0;
2388 ic->free_sectors = ic->journal_entries;
2390 ic->journal_tree_root = RB_ROOT;
2391 for (i = 0; i < ic->journal_entries; i++)
2392 init_journal_node(&ic->journal_tree[i]);
2393 }
2395 static void dm_integrity_postsuspend(struct dm_target *ti)
2396 {
2397 struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
2399 del_timer_sync(&ic->autocommit_timer);
2401 WRITE_ONCE(ic->suspending, 1);
2403 if (ic->recalc_wq)
2404 drain_workqueue(ic->recalc_wq);
2406 queue_work(ic->commit_wq, &ic->commit_work);
2407 drain_workqueue(ic->commit_wq);
2409 if (ic->mode == 'J') {
2410 if (ic->meta_dev)
2411 queue_work(ic->writer_wq, &ic->writer_work);
2412 drain_workqueue(ic->writer_wq);
2413 dm_integrity_flush_buffers(ic);
2414 }
2416 WRITE_ONCE(ic->suspending, 0);
2418 BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
2420 ic->journal_uptodate = true;
2421 }
2423 static void dm_integrity_resume(struct dm_target *ti)
2424 {
2425 struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
2427 replay_journal(ic);
2429 if (ic->recalc_wq && ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
2430 __u64 recalc_pos = le64_to_cpu(ic->sb->recalc_sector);
2431 if (recalc_pos < ic->provided_data_sectors) {
2432 queue_work(ic->recalc_wq, &ic->recalc_work);
2433 } else if (recalc_pos > ic->provided_data_sectors) {
2434 ic->sb->recalc_sector = cpu_to_le64(ic->provided_data_sectors);
2435 recalc_write_super(ic);
2436 }
2437 }
2438 }
2440 static void dm_integrity_status(struct dm_target *ti, status_type_t type,
2441 unsigned status_flags, char *result, unsigned maxlen)
2442 {
2443 struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
2444 unsigned arg_count;
2445 size_t sz = 0;
2447 switch (type) {
2448 case STATUSTYPE_INFO:
2449 DMEMIT("%llu %llu",
2450 (unsigned long long)atomic64_read(&ic->number_of_mismatches),
2451 (unsigned long long)ic->provided_data_sectors);
2452 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
2453 DMEMIT(" %llu", (unsigned long long)le64_to_cpu(ic->sb->recalc_sector));
2454 else
2455 DMEMIT(" -");
2456 break;
2458 case STATUSTYPE_TABLE: {
2459 __u64 watermark_percentage = (__u64)(ic->journal_entries - ic->free_sectors_threshold) * 100;
2460 watermark_percentage += ic->journal_entries / 2;
2461 do_div(watermark_percentage, ic->journal_entries);
2462 arg_count = 5;
2463 arg_count += !!ic->meta_dev;
2464 arg_count += ic->sectors_per_block != 1;
2465 arg_count += !!(ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING));
2466 arg_count += !!ic->internal_hash_alg.alg_string;
2467 arg_count += !!ic->journal_crypt_alg.alg_string;
2468 arg_count += !!ic->journal_mac_alg.alg_string;
2469 DMEMIT("%s %llu %u %c %u", ic->dev->name, (unsigned long long)ic->start,
2470 ic->tag_size, ic->mode, arg_count);
2471 if (ic->meta_dev)
2472 DMEMIT(" meta_device:%s", ic->meta_dev->name);
2473 if (ic->sectors_per_block != 1)
2474 DMEMIT(" block_size:%u", ic->sectors_per_block << SECTOR_SHIFT);
2475 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
2476 DMEMIT(" recalculate");
2477 DMEMIT(" journal_sectors:%u", ic->initial_sectors - SB_SECTORS);
2478 DMEMIT(" interleave_sectors:%u", 1U << ic->sb->log2_interleave_sectors);
2479 DMEMIT(" buffer_sectors:%u", 1U << ic->log2_buffer_sectors);
2480 DMEMIT(" journal_watermark:%u", (unsigned)watermark_percentage);
2481 DMEMIT(" commit_time:%u", ic->autocommit_msec);
2483 #define EMIT_ALG(a, n) \
2484 do { \
2485 if (ic->a.alg_string) { \
2486 DMEMIT(" %s:%s", n, ic->a.alg_string); \
2487 if (ic->a.key_string) \
2488 DMEMIT(":%s", ic->a.key_string);\
2489 } \
2490 } while (0)
2491 EMIT_ALG(internal_hash_alg, "internal_hash");
2492 EMIT_ALG(journal_crypt_alg, "journal_crypt");
2493 EMIT_ALG(journal_mac_alg, "journal_mac");
2494 break;
2495 }
2496 }
2497 }
2499 static int dm_integrity_iterate_devices(struct dm_target *ti,
2500 iterate_devices_callout_fn fn, void *data)
2501 {
2502 struct dm_integrity_c *ic = ti->private;
2504 if (!ic->meta_dev)
2505 return fn(ti, ic->dev, ic->start + ic->initial_sectors + ic->metadata_run, ti->len, data);
2506 else
2507 return fn(ti, ic->dev, 0, ti->len, data);
2508 }
2510 static void dm_integrity_io_hints(struct dm_target *ti, struct queue_limits *limits)
2511 {
2512 struct dm_integrity_c *ic = ti->private;
2514 if (ic->sectors_per_block > 1) {
2515 limits->logical_block_size = ic->sectors_per_block << SECTOR_SHIFT;
2516 limits->physical_block_size = ic->sectors_per_block << SECTOR_SHIFT;
2517 blk_limits_io_min(limits, ic->sectors_per_block << SECTOR_SHIFT);
2518 }
2519 }
2521 static void calculate_journal_section_size(struct dm_integrity_c *ic)
2522 {
2523 unsigned sector_space = JOURNAL_SECTOR_DATA;
2525 ic->journal_sections = le32_to_cpu(ic->sb->journal_sections);
2526 ic->journal_entry_size = roundup(offsetof(struct journal_entry, last_bytes[ic->sectors_per_block]) + ic->tag_size,
2527 JOURNAL_ENTRY_ROUNDUP);
2529 if (ic->sb->flags & cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC))
2530 sector_space -= JOURNAL_MAC_PER_SECTOR;
2531 ic->journal_entries_per_sector = sector_space / ic->journal_entry_size;
2532 ic->journal_section_entries = ic->journal_entries_per_sector * JOURNAL_BLOCK_SECTORS;
2533 ic->journal_section_sectors = (ic->journal_section_entries << ic->sb->log2_sectors_per_block) + JOURNAL_BLOCK_SECTORS;
2534 ic->journal_entries = ic->journal_section_entries * ic->journal_sections;
2535 }
2537 static int calculate_device_limits(struct dm_integrity_c *ic)
2538 {
2539 __u64 initial_sectors;
2541 calculate_journal_section_size(ic);
2542 initial_sectors = SB_SECTORS + (__u64)ic->journal_section_sectors * ic->journal_sections;
2543 if (initial_sectors + METADATA_PADDING_SECTORS >= ic->meta_device_sectors || initial_sectors > UINT_MAX)
2544 return -EINVAL;
2545 ic->initial_sectors = initial_sectors;
2547 if (!ic->meta_dev) {
2548 sector_t last_sector, last_area, last_offset;
2550 ic->metadata_run = roundup((__u64)ic->tag_size << (ic->sb->log2_interleave_sectors - ic->sb->log2_sectors_per_block),
2551 (__u64)(1 << SECTOR_SHIFT << METADATA_PADDING_SECTORS)) >> SECTOR_SHIFT;
2552 if (!(ic->metadata_run & (ic->metadata_run - 1)))
2553 ic->log2_metadata_run = __ffs(ic->metadata_run);
2554 else
2555 ic->log2_metadata_run = -1;
2557 get_area_and_offset(ic, ic->provided_data_sectors - 1, &last_area, &last_offset);
2558 last_sector = get_data_sector(ic, last_area, last_offset);
2559 if (last_sector < ic->start || last_sector >= ic->meta_device_sectors)
2560 return -EINVAL;
2561 } else {
2562 __u64 meta_size = ic->provided_data_sectors * ic->tag_size;
2563 meta_size = (meta_size + ((1U << (ic->log2_buffer_sectors + SECTOR_SHIFT)) - 1))
2564 >> (ic->log2_buffer_sectors + SECTOR_SHIFT);
2565 meta_size <<= ic->log2_buffer_sectors;
2566 if (ic->initial_sectors + meta_size < ic->initial_sectors ||
2567 ic->initial_sectors + meta_size > ic->meta_device_sectors)
2568 return -EINVAL;
2569 ic->metadata_run = 1;
2570 ic->log2_metadata_run = 0;
2571 }
2573 return 0;
2574 }
2576 static int initialize_superblock(struct dm_integrity_c *ic, unsigned journal_sectors, unsigned interleave_sectors)
2577 {
2578 unsigned journal_sections;
2579 int test_bit;
2581 memset(ic->sb, 0, SB_SECTORS << SECTOR_SHIFT);
2582 memcpy(ic->sb->magic, SB_MAGIC, 8);
2583 ic->sb->integrity_tag_size = cpu_to_le16(ic->tag_size);
2584 ic->sb->log2_sectors_per_block = __ffs(ic->sectors_per_block);
2585 if (ic->journal_mac_alg.alg_string)
2586 ic->sb->flags |= cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC);
2588 calculate_journal_section_size(ic);
2589 journal_sections = journal_sectors / ic->journal_section_sectors;
2590 if (!journal_sections)
2591 journal_sections = 1;
2593 if (!ic->meta_dev) {
2594 ic->sb->journal_sections = cpu_to_le32(journal_sections);
2595 if (!interleave_sectors)
2596 interleave_sectors = DEFAULT_INTERLEAVE_SECTORS;
2597 ic->sb->log2_interleave_sectors = __fls(interleave_sectors);
2598 ic->sb->log2_interleave_sectors = max((__u8)MIN_LOG2_INTERLEAVE_SECTORS, ic->sb->log2_interleave_sectors);
2599 ic->sb->log2_interleave_sectors = min((__u8)MAX_LOG2_INTERLEAVE_SECTORS, ic->sb->log2_interleave_sectors);
2601 ic->provided_data_sectors = 0;
2602 for (test_bit = fls64(ic->meta_device_sectors) - 1; test_bit >= 3; test_bit--) {
2603 __u64 prev_data_sectors = ic->provided_data_sectors;
2605 ic->provided_data_sectors |= (sector_t)1 << test_bit;
2606 if (calculate_device_limits(ic))
2607 ic->provided_data_sectors = prev_data_sectors;
2608 }
2609 if (!ic->provided_data_sectors)
2610 return -EINVAL;
2611 } else {
2612 ic->sb->log2_interleave_sectors = 0;
2613 ic->provided_data_sectors = ic->data_device_sectors;
2614 ic->provided_data_sectors &= ~(sector_t)(ic->sectors_per_block - 1);
2616 try_smaller_buffer:
2617 ic->sb->journal_sections = cpu_to_le32(0);
2618 for (test_bit = fls(journal_sections) - 1; test_bit >= 0; test_bit--) {
2619 __u32 prev_journal_sections = le32_to_cpu(ic->sb->journal_sections);
2620 __u32 test_journal_sections = prev_journal_sections | (1U << test_bit);
2621 if (test_journal_sections > journal_sections)
2622 continue;
2623 ic->sb->journal_sections = cpu_to_le32(test_journal_sections);
2624 if (calculate_device_limits(ic))
2625 ic->sb->journal_sections = cpu_to_le32(prev_journal_sections);
2627 }
2628 if (!le32_to_cpu(ic->sb->journal_sections)) {
2629 if (ic->log2_buffer_sectors > 3) {
2630 ic->log2_buffer_sectors--;
2631 goto try_smaller_buffer;
2632 }
2633 return -EINVAL;
2634 }
2635 }
2637 ic->sb->provided_data_sectors = cpu_to_le64(ic->provided_data_sectors);
2639 sb_set_version(ic);
2641 return 0;
2642 }
2644 static void dm_integrity_set(struct dm_target *ti, struct dm_integrity_c *ic)
2645 {
2646 struct gendisk *disk = dm_disk(dm_table_get_md(ti->table));
2647 struct blk_integrity bi;
2649 memset(&bi, 0, sizeof(bi));
2650 bi.profile = &dm_integrity_profile;
2651 bi.tuple_size = ic->tag_size;
2652 bi.tag_size = bi.tuple_size;
2653 bi.interval_exp = ic->sb->log2_sectors_per_block + SECTOR_SHIFT;
2655 blk_integrity_register(disk, &bi);
2656 blk_queue_max_integrity_segments(disk->queue, UINT_MAX);
2657 }
2659 static void dm_integrity_free_page_list(struct dm_integrity_c *ic, struct page_list *pl)
2660 {
2661 unsigned i;
2663 if (!pl)
2664 return;
2665 for (i = 0; i < ic->journal_pages; i++)
2666 if (pl[i].page)
2667 __free_page(pl[i].page);
2668 kvfree(pl);
2669 }
2671 static struct page_list *dm_integrity_alloc_page_list(struct dm_integrity_c *ic)
2672 {
2673 size_t page_list_desc_size = ic->journal_pages * sizeof(struct page_list);
2674 struct page_list *pl;
2675 unsigned i;
2677 pl = kvmalloc(page_list_desc_size, GFP_KERNEL | __GFP_ZERO);
2678 if (!pl)
2679 return NULL;
2681 for (i = 0; i < ic->journal_pages; i++) {
2682 pl[i].page = alloc_page(GFP_KERNEL);
2683 if (!pl[i].page) {
2684 dm_integrity_free_page_list(ic, pl);
2685 return NULL;
2686 }
2687 if (i)
2688 pl[i - 1].next = &pl[i];
2689 }
2691 return pl;
2692 }
2694 static void dm_integrity_free_journal_scatterlist(struct dm_integrity_c *ic, struct scatterlist **sl)
2695 {
2696 unsigned i;
2697 for (i = 0; i < ic->journal_sections; i++)
2698 kvfree(sl[i]);
2699 kvfree(sl);
2700 }
2702 static struct scatterlist **dm_integrity_alloc_journal_scatterlist(struct dm_integrity_c *ic, struct page_list *pl)
2703 {
2704 struct scatterlist **sl;
2705 unsigned i;
2707 sl = kvmalloc_array(ic->journal_sections,
2708 sizeof(struct scatterlist *),
2709 GFP_KERNEL | __GFP_ZERO);
2710 if (!sl)
2711 return NULL;
2713 for (i = 0; i < ic->journal_sections; i++) {
2714 struct scatterlist *s;
2715 unsigned start_index, start_offset;
2716 unsigned end_index, end_offset;
2717 unsigned n_pages;
2718 unsigned idx;
2720 page_list_location(ic, i, 0, &start_index, &start_offset);
2721 page_list_location(ic, i, ic->journal_section_sectors - 1, &end_index, &end_offset);
2723 n_pages = (end_index - start_index + 1);
2725 s = kvmalloc_array(n_pages, sizeof(struct scatterlist),
2726 GFP_KERNEL);
2727 if (!s) {
2728 dm_integrity_free_journal_scatterlist(ic, sl);
2729 return NULL;
2730 }
2732 sg_init_table(s, n_pages);
2733 for (idx = start_index; idx <= end_index; idx++) {
2734 char *va = lowmem_page_address(pl[idx].page);
2735 unsigned start = 0, end = PAGE_SIZE;
2736 if (idx == start_index)
2737 start = start_offset;
2738 if (idx == end_index)
2739 end = end_offset + (1 << SECTOR_SHIFT);
2740 sg_set_buf(&s[idx - start_index], va + start, end - start);
2741 }
2743 sl[i] = s;
2744 }
2746 return sl;
2747 }
2749 static void free_alg(struct alg_spec *a)
2750 {
2751 kzfree(a->alg_string);
2752 kzfree(a->key);
2753 memset(a, 0, sizeof *a);
2754 }
2756 static int get_alg_and_key(const char *arg, struct alg_spec *a, char **error, char *error_inval)
2757 {
2758 char *k;
2760 free_alg(a);
2762 a->alg_string = kstrdup(strchr(arg, ':') + 1, GFP_KERNEL);
2763 if (!a->alg_string)
2764 goto nomem;
2766 k = strchr(a->alg_string, ':');
2767 if (k) {
2768 *k = 0;
2769 a->key_string = k + 1;
2770 if (strlen(a->key_string) & 1)
2771 goto inval;
2773 a->key_size = strlen(a->key_string) / 2;
2774 a->key = kmalloc(a->key_size, GFP_KERNEL);
2775 if (!a->key)
2776 goto nomem;
2777 if (hex2bin(a->key, a->key_string, a->key_size))
2778 goto inval;
2779 }
2781 return 0;
2782 inval:
2783 *error = error_inval;
2784 return -EINVAL;
2785 nomem:
2786 *error = "Out of memory for an argument";
2787 return -ENOMEM;
2788 }
2790 static int get_mac(struct crypto_shash **hash, struct alg_spec *a, char **error,
2791 char *error_alg, char *error_key)
2792 {
2793 int r;
2795 if (a->alg_string) {
2796 *hash = crypto_alloc_shash(a->alg_string, 0, CRYPTO_ALG_ASYNC);
2797 if (IS_ERR(*hash)) {
2798 *error = error_alg;
2799 r = PTR_ERR(*hash);
2800 *hash = NULL;
2801 return r;
2802 }
2804 if (a->key) {
2805 r = crypto_shash_setkey(*hash, a->key, a->key_size);
2806 if (r) {
2807 *error = error_key;
2808 return r;
2809 }
2810 } else if (crypto_shash_get_flags(*hash) & CRYPTO_TFM_NEED_KEY) {
2811 *error = error_key;
2812 return -ENOKEY;
2813 }
2814 }
2816 return 0;
2817 }
2819 static int create_journal(struct dm_integrity_c *ic, char **error)
2820 {
2821 int r = 0;
2822 unsigned i;
2823 __u64 journal_pages, journal_desc_size, journal_tree_size;
2824 unsigned char *crypt_data = NULL, *crypt_iv = NULL;
2825 struct skcipher_request *req = NULL;
2827 ic->commit_ids[0] = cpu_to_le64(0x1111111111111111ULL);
2828 ic->commit_ids[1] = cpu_to_le64(0x2222222222222222ULL);
2829 ic->commit_ids[2] = cpu_to_le64(0x3333333333333333ULL);
2830 ic->commit_ids[3] = cpu_to_le64(0x4444444444444444ULL);
2832 journal_pages = roundup((__u64)ic->journal_sections * ic->journal_section_sectors,
2833 PAGE_SIZE >> SECTOR_SHIFT) >> (PAGE_SHIFT - SECTOR_SHIFT);
2834 journal_desc_size = journal_pages * sizeof(struct page_list);
2835 if (journal_pages >= totalram_pages - totalhigh_pages || journal_desc_size > ULONG_MAX) {
2836 *error = "Journal doesn't fit into memory";
2837 r = -ENOMEM;
2838 goto bad;
2839 }
2840 ic->journal_pages = journal_pages;
2842 ic->journal = dm_integrity_alloc_page_list(ic);
2843 if (!ic->journal) {
2844 *error = "Could not allocate memory for journal";
2845 r = -ENOMEM;
2846 goto bad;
2847 }
2848 if (ic->journal_crypt_alg.alg_string) {
2849 unsigned ivsize, blocksize;
2850 struct journal_completion comp;
2852 comp.ic = ic;
2853 ic->journal_crypt = crypto_alloc_skcipher(ic->journal_crypt_alg.alg_string, 0, 0);
2854 if (IS_ERR(ic->journal_crypt)) {
2855 *error = "Invalid journal cipher";
2856 r = PTR_ERR(ic->journal_crypt);
2857 ic->journal_crypt = NULL;
2858 goto bad;
2859 }
2860 ivsize = crypto_skcipher_ivsize(ic->journal_crypt);
2861 blocksize = crypto_skcipher_blocksize(ic->journal_crypt);
2863 if (ic->journal_crypt_alg.key) {
2864 r = crypto_skcipher_setkey(ic->journal_crypt, ic->journal_crypt_alg.key,
2865 ic->journal_crypt_alg.key_size);
2866 if (r) {
2867 *error = "Error setting encryption key";
2868 goto bad;
2869 }
2870 }
2871 DEBUG_print("cipher %s, block size %u iv size %u\n",
2872 ic->journal_crypt_alg.alg_string, blocksize, ivsize);
2874 ic->journal_io = dm_integrity_alloc_page_list(ic);
2875 if (!ic->journal_io) {
2876 *error = "Could not allocate memory for journal io";
2877 r = -ENOMEM;
2878 goto bad;
2879 }
2881 if (blocksize == 1) {
2882 struct scatterlist *sg;
2884 req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
2885 if (!req) {
2886 *error = "Could not allocate crypt request";
2887 r = -ENOMEM;
2888 goto bad;
2889 }
2891 crypt_iv = kmalloc(ivsize, GFP_KERNEL);
2892 if (!crypt_iv) {
2893 *error = "Could not allocate iv";
2894 r = -ENOMEM;
2895 goto bad;
2896 }
2898 ic->journal_xor = dm_integrity_alloc_page_list(ic);
2899 if (!ic->journal_xor) {
2900 *error = "Could not allocate memory for journal xor";
2901 r = -ENOMEM;
2902 goto bad;
2903 }
2905 sg = kvmalloc_array(ic->journal_pages + 1,
2906 sizeof(struct scatterlist),
2907 GFP_KERNEL);
2908 if (!sg) {
2909 *error = "Unable to allocate sg list";
2910 r = -ENOMEM;
2911 goto bad;
2912 }
2913 sg_init_table(sg, ic->journal_pages + 1);
2914 for (i = 0; i < ic->journal_pages; i++) {
2915 char *va = lowmem_page_address(ic->journal_xor[i].page);
2916 clear_page(va);
2917 sg_set_buf(&sg[i], va, PAGE_SIZE);
2918 }
2919 sg_set_buf(&sg[i], &ic->commit_ids, sizeof ic->commit_ids);
2920 memset(crypt_iv, 0x00, ivsize);
2922 skcipher_request_set_crypt(req, sg, sg, PAGE_SIZE * ic->journal_pages + sizeof ic->commit_ids, crypt_iv);
2923 init_completion(&comp.comp);
2924 comp.in_flight = (atomic_t)ATOMIC_INIT(1);
2925 if (do_crypt(true, req, &comp))
2926 wait_for_completion(&comp.comp);
2927 kvfree(sg);
2928 r = dm_integrity_failed(ic);
2929 if (r) {
2930 *error = "Unable to encrypt journal";
2931 goto bad;
2932 }
2933 DEBUG_bytes(lowmem_page_address(ic->journal_xor[0].page), 64, "xor data");
2935 crypto_free_skcipher(ic->journal_crypt);
2936 ic->journal_crypt = NULL;
2937 } else {
2938 unsigned crypt_len = roundup(ivsize, blocksize);
2940 req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
2941 if (!req) {
2942 *error = "Could not allocate crypt request";
2943 r = -ENOMEM;
2944 goto bad;
2945 }
2947 crypt_iv = kmalloc(ivsize, GFP_KERNEL);
2948 if (!crypt_iv) {
2949 *error = "Could not allocate iv";
2950 r = -ENOMEM;
2951 goto bad;
2952 }
2954 crypt_data = kmalloc(crypt_len, GFP_KERNEL);
2955 if (!crypt_data) {
2956 *error = "Unable to allocate crypt data";
2957 r = -ENOMEM;
2958 goto bad;
2959 }
2961 ic->journal_scatterlist = dm_integrity_alloc_journal_scatterlist(ic, ic->journal);
2962 if (!ic->journal_scatterlist) {
2963 *error = "Unable to allocate sg list";
2964 r = -ENOMEM;
2965 goto bad;
2966 }
2967 ic->journal_io_scatterlist = dm_integrity_alloc_journal_scatterlist(ic, ic->journal_io);
2968 if (!ic->journal_io_scatterlist) {
2969 *error = "Unable to allocate sg list";
2970 r = -ENOMEM;
2971 goto bad;
2972 }
2973 ic->sk_requests = kvmalloc_array(ic->journal_sections,
2974 sizeof(struct skcipher_request *),
2975 GFP_KERNEL | __GFP_ZERO);
2976 if (!ic->sk_requests) {
2977 *error = "Unable to allocate sk requests";
2978 r = -ENOMEM;
2979 goto bad;
2980 }
2981 for (i = 0; i < ic->journal_sections; i++) {
2982 struct scatterlist sg;
2983 struct skcipher_request *section_req;
2984 __u32 section_le = cpu_to_le32(i);
2986 memset(crypt_iv, 0x00, ivsize);
2987 memset(crypt_data, 0x00, crypt_len);
2988 memcpy(crypt_data, §ion_le, min((size_t)crypt_len, sizeof(section_le)));
2990 sg_init_one(&sg, crypt_data, crypt_len);
2991 skcipher_request_set_crypt(req, &sg, &sg, crypt_len, crypt_iv);
2992 init_completion(&comp.comp);
2993 comp.in_flight = (atomic_t)ATOMIC_INIT(1);
2994 if (do_crypt(true, req, &comp))
2995 wait_for_completion(&comp.comp);
2997 r = dm_integrity_failed(ic);
2998 if (r) {
2999 *error = "Unable to generate iv";
3000 goto bad;
3001 }
3003 section_req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
3004 if (!section_req) {
3005 *error = "Unable to allocate crypt request";
3006 r = -ENOMEM;
3007 goto bad;
3008 }
3009 section_req->iv = kmalloc_array(ivsize, 2,
3010 GFP_KERNEL);
3011 if (!section_req->iv) {
3012 skcipher_request_free(section_req);
3013 *error = "Unable to allocate iv";
3014 r = -ENOMEM;
3015 goto bad;
3016 }
3017 memcpy(section_req->iv + ivsize, crypt_data, ivsize);
3018 section_req->cryptlen = (size_t)ic->journal_section_sectors << SECTOR_SHIFT;
3019 ic->sk_requests[i] = section_req;
3020 DEBUG_bytes(crypt_data, ivsize, "iv(%u)", i);
3021 }
3022 }
3023 }
3025 for (i = 0; i < N_COMMIT_IDS; i++) {
3026 unsigned j;
3027 retest_commit_id:
3028 for (j = 0; j < i; j++) {
3029 if (ic->commit_ids[j] == ic->commit_ids[i]) {
3030 ic->commit_ids[i] = cpu_to_le64(le64_to_cpu(ic->commit_ids[i]) + 1);
3031 goto retest_commit_id;
3032 }
3033 }
3034 DEBUG_print("commit id %u: %016llx\n", i, ic->commit_ids[i]);
3035 }
3037 journal_tree_size = (__u64)ic->journal_entries * sizeof(struct journal_node);
3038 if (journal_tree_size > ULONG_MAX) {
3039 *error = "Journal doesn't fit into memory";
3040 r = -ENOMEM;
3041 goto bad;
3042 }
3043 ic->journal_tree = kvmalloc(journal_tree_size, GFP_KERNEL);
3044 if (!ic->journal_tree) {
3045 *error = "Could not allocate memory for journal tree";
3046 r = -ENOMEM;
3047 }
3048 bad:
3049 kfree(crypt_data);
3050 kfree(crypt_iv);
3051 skcipher_request_free(req);
3053 return r;
3054 }
3056 /*
3057 * Construct a integrity mapping
3058 *
3059 * Arguments:
3060 * device
3061 * offset from the start of the device
3062 * tag size
3063 * D - direct writes, J - journal writes, R - recovery mode
3064 * number of optional arguments
3065 * optional arguments:
3066 * journal_sectors
3067 * interleave_sectors
3068 * buffer_sectors
3069 * journal_watermark
3070 * commit_time
3071 * internal_hash
3072 * journal_crypt
3073 * journal_mac
3074 * block_size
3075 */
3076 static int dm_integrity_ctr(struct dm_target *ti, unsigned argc, char **argv)
3077 {
3078 struct dm_integrity_c *ic;
3079 char dummy;
3080 int r;
3081 unsigned extra_args;
3082 struct dm_arg_set as;
3083 static const struct dm_arg _args[] = {
3084 {0, 9, "Invalid number of feature args"},
3085 };
3086 unsigned journal_sectors, interleave_sectors, buffer_sectors, journal_watermark, sync_msec;
3087 bool recalculate;
3088 bool should_write_sb;
3089 __u64 threshold;
3090 unsigned long long start;
3092 #define DIRECT_ARGUMENTS 4
3094 if (argc <= DIRECT_ARGUMENTS) {
3095 ti->error = "Invalid argument count";
3096 return -EINVAL;
3097 }
3099 ic = kzalloc(sizeof(struct dm_integrity_c), GFP_KERNEL);
3100 if (!ic) {
3101 ti->error = "Cannot allocate integrity context";
3102 return -ENOMEM;
3103 }
3104 ti->private = ic;
3105 ti->per_io_data_size = sizeof(struct dm_integrity_io);
3107 ic->in_progress = RB_ROOT;
3108 INIT_LIST_HEAD(&ic->wait_list);
3109 init_waitqueue_head(&ic->endio_wait);
3110 bio_list_init(&ic->flush_bio_list);
3111 init_waitqueue_head(&ic->copy_to_journal_wait);
3112 init_completion(&ic->crypto_backoff);
3113 atomic64_set(&ic->number_of_mismatches, 0);
3115 r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &ic->dev);
3116 if (r) {
3117 ti->error = "Device lookup failed";
3118 goto bad;
3119 }
3121 if (sscanf(argv[1], "%llu%c", &start, &dummy) != 1 || start != (sector_t)start) {
3122 ti->error = "Invalid starting offset";
3123 r = -EINVAL;
3124 goto bad;
3125 }
3126 ic->start = start;
3128 if (strcmp(argv[2], "-")) {
3129 if (sscanf(argv[2], "%u%c", &ic->tag_size, &dummy) != 1 || !ic->tag_size) {
3130 ti->error = "Invalid tag size";
3131 r = -EINVAL;
3132 goto bad;
3133 }
3134 }
3136 if (!strcmp(argv[3], "J") || !strcmp(argv[3], "D") || !strcmp(argv[3], "R"))
3137 ic->mode = argv[3][0];
3138 else {
3139 ti->error = "Invalid mode (expecting J, D, R)";
3140 r = -EINVAL;
3141 goto bad;
3142 }
3144 journal_sectors = 0;
3145 interleave_sectors = DEFAULT_INTERLEAVE_SECTORS;
3146 buffer_sectors = DEFAULT_BUFFER_SECTORS;
3147 journal_watermark = DEFAULT_JOURNAL_WATERMARK;
3148 sync_msec = DEFAULT_SYNC_MSEC;
3149 recalculate = false;
3150 ic->sectors_per_block = 1;
3152 as.argc = argc - DIRECT_ARGUMENTS;
3153 as.argv = argv + DIRECT_ARGUMENTS;
3154 r = dm_read_arg_group(_args, &as, &extra_args, &ti->error);
3155 if (r)
3156 goto bad;
3158 while (extra_args--) {
3159 const char *opt_string;
3160 unsigned val;
3161 opt_string = dm_shift_arg(&as);
3162 if (!opt_string) {
3163 r = -EINVAL;
3164 ti->error = "Not enough feature arguments";
3165 goto bad;
3166 }
3167 if (sscanf(opt_string, "journal_sectors:%u%c", &val, &dummy) == 1)
3168 journal_sectors = val ? val : 1;
3169 else if (sscanf(opt_string, "interleave_sectors:%u%c", &val, &dummy) == 1)
3170 interleave_sectors = val;
3171 else if (sscanf(opt_string, "buffer_sectors:%u%c", &val, &dummy) == 1)
3172 buffer_sectors = val;
3173 else if (sscanf(opt_string, "journal_watermark:%u%c", &val, &dummy) == 1 && val <= 100)
3174 journal_watermark = val;
3175 else if (sscanf(opt_string, "commit_time:%u%c", &val, &dummy) == 1)
3176 sync_msec = val;
3177 else if (!memcmp(opt_string, "meta_device:", strlen("meta_device:"))) {
3178 if (ic->meta_dev) {
3179 dm_put_device(ti, ic->meta_dev);
3180 ic->meta_dev = NULL;
3181 }
3182 r = dm_get_device(ti, strchr(opt_string, ':') + 1, dm_table_get_mode(ti->table), &ic->meta_dev);
3183 if (r) {
3184 ti->error = "Device lookup failed";
3185 goto bad;
3186 }
3187 } else if (sscanf(opt_string, "block_size:%u%c", &val, &dummy) == 1) {
3188 if (val < 1 << SECTOR_SHIFT ||
3189 val > MAX_SECTORS_PER_BLOCK << SECTOR_SHIFT ||
3190 (val & (val -1))) {
3191 r = -EINVAL;
3192 ti->error = "Invalid block_size argument";
3193 goto bad;
3194 }
3195 ic->sectors_per_block = val >> SECTOR_SHIFT;
3196 } else if (!memcmp(opt_string, "internal_hash:", strlen("internal_hash:"))) {
3197 r = get_alg_and_key(opt_string, &ic->internal_hash_alg, &ti->error,
3198 "Invalid internal_hash argument");
3199 if (r)
3200 goto bad;
3201 } else if (!memcmp(opt_string, "journal_crypt:", strlen("journal_crypt:"))) {
3202 r = get_alg_and_key(opt_string, &ic->journal_crypt_alg, &ti->error,
3203 "Invalid journal_crypt argument");
3204 if (r)
3205 goto bad;
3206 } else if (!memcmp(opt_string, "journal_mac:", strlen("journal_mac:"))) {
3207 r = get_alg_and_key(opt_string, &ic->journal_mac_alg, &ti->error,
3208 "Invalid journal_mac argument");
3209 if (r)
3210 goto bad;
3211 } else if (!strcmp(opt_string, "recalculate")) {
3212 recalculate = true;
3213 } else {
3214 r = -EINVAL;
3215 ti->error = "Invalid argument";
3216 goto bad;
3217 }
3218 }
3220 ic->data_device_sectors = i_size_read(ic->dev->bdev->bd_inode) >> SECTOR_SHIFT;
3221 if (!ic->meta_dev)
3222 ic->meta_device_sectors = ic->data_device_sectors;
3223 else
3224 ic->meta_device_sectors = i_size_read(ic->meta_dev->bdev->bd_inode) >> SECTOR_SHIFT;
3226 if (!journal_sectors) {
3227 journal_sectors = min((sector_t)DEFAULT_MAX_JOURNAL_SECTORS,
3228 ic->data_device_sectors >> DEFAULT_JOURNAL_SIZE_FACTOR);
3229 }
3231 if (!buffer_sectors)
3232 buffer_sectors = 1;
3233 ic->log2_buffer_sectors = min((int)__fls(buffer_sectors), 31 - SECTOR_SHIFT);
3235 r = get_mac(&ic->internal_hash, &ic->internal_hash_alg, &ti->error,
3236 "Invalid internal hash", "Error setting internal hash key");
3237 if (r)
3238 goto bad;
3240 r = get_mac(&ic->journal_mac, &ic->journal_mac_alg, &ti->error,
3241 "Invalid journal mac", "Error setting journal mac key");
3242 if (r)
3243 goto bad;
3245 if (!ic->tag_size) {
3246 if (!ic->internal_hash) {
3247 ti->error = "Unknown tag size";
3248 r = -EINVAL;
3249 goto bad;
3250 }
3251 ic->tag_size = crypto_shash_digestsize(ic->internal_hash);
3252 }
3253 if (ic->tag_size > MAX_TAG_SIZE) {
3254 ti->error = "Too big tag size";
3255 r = -EINVAL;
3256 goto bad;
3257 }
3258 if (!(ic->tag_size & (ic->tag_size - 1)))
3259 ic->log2_tag_size = __ffs(ic->tag_size);
3260 else
3261 ic->log2_tag_size = -1;
3263 ic->autocommit_jiffies = msecs_to_jiffies(sync_msec);
3264 ic->autocommit_msec = sync_msec;
3265 timer_setup(&ic->autocommit_timer, autocommit_fn, 0);
3267 ic->io = dm_io_client_create();
3268 if (IS_ERR(ic->io)) {
3269 r = PTR_ERR(ic->io);
3270 ic->io = NULL;
3271 ti->error = "Cannot allocate dm io";
3272 goto bad;
3273 }
3275 r = mempool_init_slab_pool(&ic->journal_io_mempool, JOURNAL_IO_MEMPOOL, journal_io_cache);
3276 if (r) {
3277 ti->error = "Cannot allocate mempool";
3278 goto bad;
3279 }
3281 ic->metadata_wq = alloc_workqueue("dm-integrity-metadata",
3282 WQ_MEM_RECLAIM, METADATA_WORKQUEUE_MAX_ACTIVE);
3283 if (!ic->metadata_wq) {
3284 ti->error = "Cannot allocate workqueue";
3285 r = -ENOMEM;
3286 goto bad;
3287 }
3289 /*
3290 * If this workqueue were percpu, it would cause bio reordering
3291 * and reduced performance.
3292 */
3293 ic->wait_wq = alloc_workqueue("dm-integrity-wait", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3294 if (!ic->wait_wq) {
3295 ti->error = "Cannot allocate workqueue";
3296 r = -ENOMEM;
3297 goto bad;
3298 }
3300 ic->commit_wq = alloc_workqueue("dm-integrity-commit", WQ_MEM_RECLAIM, 1);
3301 if (!ic->commit_wq) {
3302 ti->error = "Cannot allocate workqueue";
3303 r = -ENOMEM;
3304 goto bad;
3305 }
3306 INIT_WORK(&ic->commit_work, integrity_commit);
3308 if (ic->mode == 'J') {
3309 ic->writer_wq = alloc_workqueue("dm-integrity-writer", WQ_MEM_RECLAIM, 1);
3310 if (!ic->writer_wq) {
3311 ti->error = "Cannot allocate workqueue";
3312 r = -ENOMEM;
3313 goto bad;
3314 }
3315 INIT_WORK(&ic->writer_work, integrity_writer);
3316 }
3318 ic->sb = alloc_pages_exact(SB_SECTORS << SECTOR_SHIFT, GFP_KERNEL);
3319 if (!ic->sb) {
3320 r = -ENOMEM;
3321 ti->error = "Cannot allocate superblock area";
3322 goto bad;
3323 }
3325 r = sync_rw_sb(ic, REQ_OP_READ, 0);
3326 if (r) {
3327 ti->error = "Error reading superblock";
3328 goto bad;
3329 }
3330 should_write_sb = false;
3331 if (memcmp(ic->sb->magic, SB_MAGIC, 8)) {
3332 if (ic->mode != 'R') {
3333 if (memchr_inv(ic->sb, 0, SB_SECTORS << SECTOR_SHIFT)) {
3334 r = -EINVAL;
3335 ti->error = "The device is not initialized";
3336 goto bad;
3337 }
3338 }
3340 r = initialize_superblock(ic, journal_sectors, interleave_sectors);
3341 if (r) {
3342 ti->error = "Could not initialize superblock";
3343 goto bad;
3344 }
3345 if (ic->mode != 'R')
3346 should_write_sb = true;
3347 }
3349 if (!ic->sb->version || ic->sb->version > SB_VERSION_2) {
3350 r = -EINVAL;
3351 ti->error = "Unknown version";
3352 goto bad;
3353 }
3354 if (le16_to_cpu(ic->sb->integrity_tag_size) != ic->tag_size) {
3355 r = -EINVAL;
3356 ti->error = "Tag size doesn't match the information in superblock";
3357 goto bad;
3358 }
3359 if (ic->sb->log2_sectors_per_block != __ffs(ic->sectors_per_block)) {
3360 r = -EINVAL;
3361 ti->error = "Block size doesn't match the information in superblock";
3362 goto bad;
3363 }
3364 if (!le32_to_cpu(ic->sb->journal_sections)) {
3365 r = -EINVAL;
3366 ti->error = "Corrupted superblock, journal_sections is 0";
3367 goto bad;
3368 }
3369 /* make sure that ti->max_io_len doesn't overflow */
3370 if (!ic->meta_dev) {
3371 if (ic->sb->log2_interleave_sectors < MIN_LOG2_INTERLEAVE_SECTORS ||
3372 ic->sb->log2_interleave_sectors > MAX_LOG2_INTERLEAVE_SECTORS) {
3373 r = -EINVAL;
3374 ti->error = "Invalid interleave_sectors in the superblock";
3375 goto bad;
3376 }
3377 } else {
3378 if (ic->sb->log2_interleave_sectors) {
3379 r = -EINVAL;
3380 ti->error = "Invalid interleave_sectors in the superblock";
3381 goto bad;
3382 }
3383 }
3384 ic->provided_data_sectors = le64_to_cpu(ic->sb->provided_data_sectors);
3385 if (ic->provided_data_sectors != le64_to_cpu(ic->sb->provided_data_sectors)) {
3386 /* test for overflow */
3387 r = -EINVAL;
3388 ti->error = "The superblock has 64-bit device size, but the kernel was compiled with 32-bit sectors";
3389 goto bad;
3390 }
3391 if (!!(ic->sb->flags & cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC)) != !!ic->journal_mac_alg.alg_string) {
3392 r = -EINVAL;
3393 ti->error = "Journal mac mismatch";
3394 goto bad;
3395 }
3397 try_smaller_buffer:
3398 r = calculate_device_limits(ic);
3399 if (r) {
3400 if (ic->meta_dev) {
3401 if (ic->log2_buffer_sectors > 3) {
3402 ic->log2_buffer_sectors--;
3403 goto try_smaller_buffer;
3404 }
3405 }
3406 ti->error = "The device is too small";
3407 goto bad;
3408 }
3409 if (!ic->meta_dev)
3410 ic->log2_buffer_sectors = min(ic->log2_buffer_sectors, (__u8)__ffs(ic->metadata_run));
3412 if (ti->len > ic->provided_data_sectors) {
3413 r = -EINVAL;
3414 ti->error = "Not enough provided sectors for requested mapping size";
3415 goto bad;
3416 }
3419 threshold = (__u64)ic->journal_entries * (100 - journal_watermark);
3420 threshold += 50;
3421 do_div(threshold, 100);
3422 ic->free_sectors_threshold = threshold;
3424 DEBUG_print("initialized:\n");
3425 DEBUG_print(" integrity_tag_size %u\n", le16_to_cpu(ic->sb->integrity_tag_size));
3426 DEBUG_print(" journal_entry_size %u\n", ic->journal_entry_size);
3427 DEBUG_print(" journal_entries_per_sector %u\n", ic->journal_entries_per_sector);
3428 DEBUG_print(" journal_section_entries %u\n", ic->journal_section_entries);
3429 DEBUG_print(" journal_section_sectors %u\n", ic->journal_section_sectors);
3430 DEBUG_print(" journal_sections %u\n", (unsigned)le32_to_cpu(ic->sb->journal_sections));
3431 DEBUG_print(" journal_entries %u\n", ic->journal_entries);
3432 DEBUG_print(" log2_interleave_sectors %d\n", ic->sb->log2_interleave_sectors);
3433 DEBUG_print(" device_sectors 0x%llx\n", (unsigned long long)ic->device_sectors);
3434 DEBUG_print(" initial_sectors 0x%x\n", ic->initial_sectors);
3435 DEBUG_print(" metadata_run 0x%x\n", ic->metadata_run);
3436 DEBUG_print(" log2_metadata_run %d\n", ic->log2_metadata_run);
3437 DEBUG_print(" provided_data_sectors 0x%llx (%llu)\n", (unsigned long long)ic->provided_data_sectors,
3438 (unsigned long long)ic->provided_data_sectors);
3439 DEBUG_print(" log2_buffer_sectors %u\n", ic->log2_buffer_sectors);
3441 if (recalculate && !(ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))) {
3442 ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
3443 ic->sb->recalc_sector = cpu_to_le64(0);
3444 }
3446 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
3447 if (!ic->internal_hash) {
3448 r = -EINVAL;
3449 ti->error = "Recalculate is only valid with internal hash";
3450 goto bad;
3451 }
3452 ic->recalc_wq = alloc_workqueue("dm-intergrity-recalc", WQ_MEM_RECLAIM, 1);
3453 if (!ic->recalc_wq ) {
3454 ti->error = "Cannot allocate workqueue";
3455 r = -ENOMEM;
3456 goto bad;
3457 }
3458 INIT_WORK(&ic->recalc_work, integrity_recalc);
3459 ic->recalc_buffer = vmalloc(RECALC_SECTORS << SECTOR_SHIFT);
3460 if (!ic->recalc_buffer) {
3461 ti->error = "Cannot allocate buffer for recalculating";
3462 r = -ENOMEM;
3463 goto bad;
3464 }
3465 ic->recalc_tags = kvmalloc((RECALC_SECTORS >> ic->sb->log2_sectors_per_block) * ic->tag_size, GFP_KERNEL);
3466 if (!ic->recalc_tags) {
3467 ti->error = "Cannot allocate tags for recalculating";
3468 r = -ENOMEM;
3469 goto bad;
3470 }
3471 }
3473 ic->bufio = dm_bufio_client_create(ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev,
3474 1U << (SECTOR_SHIFT + ic->log2_buffer_sectors), 1, 0, NULL, NULL);
3475 if (IS_ERR(ic->bufio)) {
3476 r = PTR_ERR(ic->bufio);
3477 ti->error = "Cannot initialize dm-bufio";
3478 ic->bufio = NULL;
3479 goto bad;
3480 }
3481 dm_bufio_set_sector_offset(ic->bufio, ic->start + ic->initial_sectors);
3483 if (ic->mode != 'R') {
3484 r = create_journal(ic, &ti->error);
3485 if (r)
3486 goto bad;
3487 }
3489 if (should_write_sb) {
3490 int r;
3492 init_journal(ic, 0, ic->journal_sections, 0);
3493 r = dm_integrity_failed(ic);
3494 if (unlikely(r)) {
3495 ti->error = "Error initializing journal";
3496 goto bad;
3497 }
3498 r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
3499 if (r) {
3500 ti->error = "Error initializing superblock";
3501 goto bad;
3502 }
3503 ic->just_formatted = true;
3504 }
3506 if (!ic->meta_dev) {
3507 r = dm_set_target_max_io_len(ti, 1U << ic->sb->log2_interleave_sectors);
3508 if (r)
3509 goto bad;
3510 }
3512 if (!ic->internal_hash)
3513 dm_integrity_set(ti, ic);
3515 ti->num_flush_bios = 1;
3516 ti->flush_supported = true;
3518 return 0;
3519 bad:
3520 dm_integrity_dtr(ti);
3521 return r;
3522 }
3524 static void dm_integrity_dtr(struct dm_target *ti)
3525 {
3526 struct dm_integrity_c *ic = ti->private;
3528 BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
3529 BUG_ON(!list_empty(&ic->wait_list));
3531 if (ic->metadata_wq)
3532 destroy_workqueue(ic->metadata_wq);
3533 if (ic->wait_wq)
3534 destroy_workqueue(ic->wait_wq);
3535 if (ic->commit_wq)
3536 destroy_workqueue(ic->commit_wq);
3537 if (ic->writer_wq)
3538 destroy_workqueue(ic->writer_wq);
3539 if (ic->recalc_wq)
3540 destroy_workqueue(ic->recalc_wq);
3541 if (ic->recalc_buffer)
3542 vfree(ic->recalc_buffer);
3543 if (ic->recalc_tags)
3544 kvfree(ic->recalc_tags);
3545 if (ic->bufio)
3546 dm_bufio_client_destroy(ic->bufio);
3547 mempool_exit(&ic->journal_io_mempool);
3548 if (ic->io)
3549 dm_io_client_destroy(ic->io);
3550 if (ic->dev)
3551 dm_put_device(ti, ic->dev);
3552 if (ic->meta_dev)
3553 dm_put_device(ti, ic->meta_dev);
3554 dm_integrity_free_page_list(ic, ic->journal);
3555 dm_integrity_free_page_list(ic, ic->journal_io);
3556 dm_integrity_free_page_list(ic, ic->journal_xor);
3557 if (ic->journal_scatterlist)
3558 dm_integrity_free_journal_scatterlist(ic, ic->journal_scatterlist);
3559 if (ic->journal_io_scatterlist)
3560 dm_integrity_free_journal_scatterlist(ic, ic->journal_io_scatterlist);
3561 if (ic->sk_requests) {
3562 unsigned i;
3564 for (i = 0; i < ic->journal_sections; i++) {
3565 struct skcipher_request *req = ic->sk_requests[i];
3566 if (req) {
3567 kzfree(req->iv);
3568 skcipher_request_free(req);
3569 }
3570 }
3571 kvfree(ic->sk_requests);
3572 }
3573 kvfree(ic->journal_tree);
3574 if (ic->sb)
3575 free_pages_exact(ic->sb, SB_SECTORS << SECTOR_SHIFT);
3577 if (ic->internal_hash)
3578 crypto_free_shash(ic->internal_hash);
3579 free_alg(&ic->internal_hash_alg);
3581 if (ic->journal_crypt)
3582 crypto_free_skcipher(ic->journal_crypt);
3583 free_alg(&ic->journal_crypt_alg);
3585 if (ic->journal_mac)
3586 crypto_free_shash(ic->journal_mac);
3587 free_alg(&ic->journal_mac_alg);
3589 kfree(ic);
3590 }
3592 static struct target_type integrity_target = {
3593 .name = "integrity",
3594 .version = {1, 2, 0},
3595 .module = THIS_MODULE,
3596 .features = DM_TARGET_SINGLETON | DM_TARGET_INTEGRITY,
3597 .ctr = dm_integrity_ctr,
3598 .dtr = dm_integrity_dtr,
3599 .map = dm_integrity_map,
3600 .postsuspend = dm_integrity_postsuspend,
3601 .resume = dm_integrity_resume,
3602 .status = dm_integrity_status,
3603 .iterate_devices = dm_integrity_iterate_devices,
3604 .io_hints = dm_integrity_io_hints,
3605 };
3607 int __init dm_integrity_init(void)
3608 {
3609 int r;
3611 journal_io_cache = kmem_cache_create("integrity_journal_io",
3612 sizeof(struct journal_io), 0, 0, NULL);
3613 if (!journal_io_cache) {
3614 DMERR("can't allocate journal io cache");
3615 return -ENOMEM;
3616 }
3618 r = dm_register_target(&integrity_target);
3620 if (r < 0)
3621 DMERR("register failed %d", r);
3623 return r;
3624 }
3626 void dm_integrity_exit(void)
3627 {
3628 dm_unregister_target(&integrity_target);
3629 kmem_cache_destroy(journal_io_cache);
3630 }
3632 module_init(dm_integrity_init);
3633 module_exit(dm_integrity_exit);
3635 MODULE_AUTHOR("Milan Broz");
3636 MODULE_AUTHOR("Mikulas Patocka");
3637 MODULE_DESCRIPTION(DM_NAME " target for integrity tags extension");
3638 MODULE_LICENSE("GPL");