?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Linux 4.19.133
[linux/fpc-iii.git] / drivers / md / dm-integrity.c
blobd75a4ce7d12ae1689ecf7fb45f061a474f546c6c
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 */
8
9 #include "dm-bio-record.h"
10
11 #include <linux/compiler.h>
12 #include <linux/module.h>
13 #include <linux/device-mapper.h>
14 #include <linux/dm-io.h>
15 #include <linux/vmalloc.h>
16 #include <linux/sort.h>
17 #include <linux/rbtree.h>
18 #include <linux/delay.h>
19 #include <linux/random.h>
20 #include <crypto/hash.h>
21 #include <crypto/skcipher.h>
22 #include <linux/async_tx.h>
23 #include <linux/dm-bufio.h>
24
25 #define DM_MSG_PREFIX "integrity"
26
27 #define DEFAULT_INTERLEAVE_SECTORS 32768
28 #define DEFAULT_JOURNAL_SIZE_FACTOR 7
29 #define DEFAULT_BUFFER_SECTORS 128
30 #define DEFAULT_JOURNAL_WATERMARK 50
31 #define DEFAULT_SYNC_MSEC 10000
32 #define DEFAULT_MAX_JOURNAL_SECTORS 131072
33 #define MIN_LOG2_INTERLEAVE_SECTORS 3
34 #define MAX_LOG2_INTERLEAVE_SECTORS 31
35 #define METADATA_WORKQUEUE_MAX_ACTIVE 16
36 #define RECALC_SECTORS 8192
37 #define RECALC_WRITE_SUPER 16
38
39 /*
40 * Warning - DEBUG_PRINT prints security-sensitive data to the log,
41 * so it should not be enabled in the official kernel
42 */
43 //#define DEBUG_PRINT
44 //#define INTERNAL_VERIFY
45
46 /*
47 * On disk structures
48 */
49
50 #define SB_MAGIC "integrt"
51 #define SB_VERSION_1 1
52 #define SB_VERSION_2 2
53 #define SB_SECTORS 8
54 #define MAX_SECTORS_PER_BLOCK 8
55
56 struct superblock {
57 __u8 magic[8];
58 __u8 version;
59 __u8 log2_interleave_sectors;
60 __u16 integrity_tag_size;
61 __u32 journal_sections;
62 __u64 provided_data_sectors; /* userspace uses this value */
63 __u32 flags;
64 __u8 log2_sectors_per_block;
65 __u8 pad[3];
66 __u64 recalc_sector;
67 };
68
69 #define SB_FLAG_HAVE_JOURNAL_MAC 0x1
70 #define SB_FLAG_RECALCULATING 0x2
71
72 #define JOURNAL_ENTRY_ROUNDUP 8
73
74 typedef __u64 commit_id_t;
75 #define JOURNAL_MAC_PER_SECTOR 8
76
77 struct journal_entry {
78 union {
79 struct {
80 __u32 sector_lo;
81 __u32 sector_hi;
82 } s;
83 __u64 sector;
84 } u;
85 commit_id_t last_bytes[0];
86 /* __u8 tag[0]; */
87 };
88
89 #define journal_entry_tag(ic, je) ((__u8 *)&(je)->last_bytes[(ic)->sectors_per_block])
90
91 #if BITS_PER_LONG == 64
92 #define journal_entry_set_sector(je, x) do { smp_wmb(); WRITE_ONCE((je)->u.sector, cpu_to_le64(x)); } while (0)
93 #define journal_entry_get_sector(je) le64_to_cpu((je)->u.sector)
94 #elif defined(CONFIG_LBDAF)
95 #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)
96 #define journal_entry_get_sector(je) le64_to_cpu((je)->u.sector)
97 #else
98 #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)
99 #define journal_entry_get_sector(je) le32_to_cpu((je)->u.s.sector_lo)
100 #endif
101 #define journal_entry_is_unused(je) ((je)->u.s.sector_hi == cpu_to_le32(-1))
102 #define journal_entry_set_unused(je) do { ((je)->u.s.sector_hi = cpu_to_le32(-1)); } while (0)
103 #define journal_entry_is_inprogress(je) ((je)->u.s.sector_hi == cpu_to_le32(-2))
104 #define journal_entry_set_inprogress(je) do { ((je)->u.s.sector_hi = cpu_to_le32(-2)); } while (0)
105
106 #define JOURNAL_BLOCK_SECTORS 8
107 #define JOURNAL_SECTOR_DATA ((1 << SECTOR_SHIFT) - sizeof(commit_id_t))
108 #define JOURNAL_MAC_SIZE (JOURNAL_MAC_PER_SECTOR * JOURNAL_BLOCK_SECTORS)
109
110 struct journal_sector {
111 __u8 entries[JOURNAL_SECTOR_DATA - JOURNAL_MAC_PER_SECTOR];
112 __u8 mac[JOURNAL_MAC_PER_SECTOR];
113 commit_id_t commit_id;
114 };
115
116 #define MAX_TAG_SIZE (JOURNAL_SECTOR_DATA - JOURNAL_MAC_PER_SECTOR - offsetof(struct journal_entry, last_bytes[MAX_SECTORS_PER_BLOCK]))
117
118 #define METADATA_PADDING_SECTORS 8
119
120 #define N_COMMIT_IDS 4
121
122 static unsigned char prev_commit_seq(unsigned char seq)
123 {
124 return (seq + N_COMMIT_IDS - 1) % N_COMMIT_IDS;
125 }
126
127 static unsigned char next_commit_seq(unsigned char seq)
128 {
129 return (seq + 1) % N_COMMIT_IDS;
130 }
131
132 /*
133 * In-memory structures
134 */
135
136 struct journal_node {
137 struct rb_node node;
138 sector_t sector;
139 };
140
141 struct alg_spec {
142 char *alg_string;
143 char *key_string;
144 __u8 *key;
145 unsigned key_size;
146 };
147
148 struct dm_integrity_c {
149 struct dm_dev *dev;
150 struct dm_dev *meta_dev;
151 unsigned tag_size;
152 __s8 log2_tag_size;
153 sector_t start;
154 mempool_t journal_io_mempool;
155 struct dm_io_client *io;
156 struct dm_bufio_client *bufio;
157 struct workqueue_struct *metadata_wq;
158 struct superblock *sb;
159 unsigned journal_pages;
160 struct page_list *journal;
161 struct page_list *journal_io;
162 struct page_list *journal_xor;
163
164 struct crypto_skcipher *journal_crypt;
165 struct scatterlist **journal_scatterlist;
166 struct scatterlist **journal_io_scatterlist;
167 struct skcipher_request **sk_requests;
168
169 struct crypto_shash *journal_mac;
170
171 struct journal_node *journal_tree;
172 struct rb_root journal_tree_root;
173
174 sector_t provided_data_sectors;
175
176 unsigned short journal_entry_size;
177 unsigned char journal_entries_per_sector;
178 unsigned char journal_section_entries;
179 unsigned short journal_section_sectors;
180 unsigned journal_sections;
181 unsigned journal_entries;
182 sector_t data_device_sectors;
183 sector_t meta_device_sectors;
184 unsigned initial_sectors;
185 unsigned metadata_run;
186 __s8 log2_metadata_run;
187 __u8 log2_buffer_sectors;
188 __u8 sectors_per_block;
189
190 unsigned char mode;
191
192 int failed;
193
194 struct crypto_shash *internal_hash;
195
196 struct dm_target *ti;
197
198 /* these variables are locked with endio_wait.lock */
199 struct rb_root in_progress;
200 struct list_head wait_list;
201 wait_queue_head_t endio_wait;
202 struct workqueue_struct *wait_wq;
203 struct workqueue_struct *offload_wq;
204
205 unsigned char commit_seq;
206 commit_id_t commit_ids[N_COMMIT_IDS];
207
208 unsigned committed_section;
209 unsigned n_committed_sections;
210
211 unsigned uncommitted_section;
212 unsigned n_uncommitted_sections;
213
214 unsigned free_section;
215 unsigned char free_section_entry;
216 unsigned free_sectors;
217
218 unsigned free_sectors_threshold;
219
220 struct workqueue_struct *commit_wq;
221 struct work_struct commit_work;
222
223 struct workqueue_struct *writer_wq;
224 struct work_struct writer_work;
225
226 struct workqueue_struct *recalc_wq;
227 struct work_struct recalc_work;
228 u8 *recalc_buffer;
229 u8 *recalc_tags;
230
231 struct bio_list flush_bio_list;
232
233 unsigned long autocommit_jiffies;
234 struct timer_list autocommit_timer;
235 unsigned autocommit_msec;
236
237 wait_queue_head_t copy_to_journal_wait;
238
239 struct completion crypto_backoff;
240
241 bool journal_uptodate;
242 bool just_formatted;
243
244 struct alg_spec internal_hash_alg;
245 struct alg_spec journal_crypt_alg;
246 struct alg_spec journal_mac_alg;
247
248 atomic64_t number_of_mismatches;
249 };
250
251 struct dm_integrity_range {
252 sector_t logical_sector;
253 unsigned n_sectors;
254 bool waiting;
255 union {
256 struct rb_node node;
257 struct {
258 struct task_struct *task;
259 struct list_head wait_entry;
260 };
261 };
262 };
263
264 struct dm_integrity_io {
265 struct work_struct work;
266
267 struct dm_integrity_c *ic;
268 bool write;
269 bool fua;
270
271 struct dm_integrity_range range;
272
273 sector_t metadata_block;
274 unsigned metadata_offset;
275
276 atomic_t in_flight;
277 blk_status_t bi_status;
278
279 struct completion *completion;
280
281 struct dm_bio_details bio_details;
282 };
283
284 struct journal_completion {
285 struct dm_integrity_c *ic;
286 atomic_t in_flight;
287 struct completion comp;
288 };
289
290 struct journal_io {
291 struct dm_integrity_range range;
292 struct journal_completion *comp;
293 };
294
295 static struct kmem_cache *journal_io_cache;
296
297 #define JOURNAL_IO_MEMPOOL 32
298
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
321
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 };
330
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);
334
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 }
342
343 static int dm_integrity_failed(struct dm_integrity_c *ic)
344 {
345 return READ_ONCE(ic->failed);
346 }
347
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 }
357
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 }
370
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)
376
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;
382
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;
389
390 sector_to_block(ic, offset);
391
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 }
402
403 static sector_t get_data_sector(struct dm_integrity_c *ic, sector_t area, sector_t offset)
404 {
405 sector_t result;
406
407 if (ic->meta_dev)
408 return offset;
409
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;
415
416 result += (sector_t)ic->initial_sectors + offset;
417 result += ic->start;
418
419 return result;
420 }
421
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 }
427
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 }
435
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;
440
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;
450
451 return dm_io(&io_req, 1, &io_loc, NULL);
452 }
453
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;
459
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 }
468
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;
473
474 access_journal_check(ic, section, offset, false, "page_list_location");
475
476 sector = section * ic->journal_section_sectors + offset;
477
478 *pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
479 *pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
480 }
481
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;
487
488 page_list_location(ic, section, offset, &pl_index, &pl_offset);
489
490 if (n_sectors)
491 *n_sectors = (PAGE_SIZE - pl_offset) >> SECTOR_SHIFT;
492
493 va = lowmem_page_address(pl[pl_index].page);
494
495 return (struct journal_sector *)(va + pl_offset);
496 }
497
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 }
502
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;
507
508 access_journal_check(ic, section, n, true, "access_journal_entry");
509
510 rel_sector = n % JOURNAL_BLOCK_SECTORS;
511 offset = n / JOURNAL_BLOCK_SECTORS;
512
513 js = access_journal(ic, section, rel_sector);
514 return (struct journal_entry *)((char *)js + offset * ic->journal_entry_size);
515 }
516
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;
520
521 n += JOURNAL_BLOCK_SECTORS;
522
523 access_journal_check(ic, section, n, false, "access_journal_data");
524
525 return access_journal(ic, section, n);
526 }
527
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;
533
534 desc->tfm = ic->journal_mac;
535 desc->flags = 0;
536
537 r = crypto_shash_init(desc);
538 if (unlikely(r)) {
539 dm_integrity_io_error(ic, "crypto_shash_init", r);
540 goto err;
541 }
542
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 }
551
552 size = crypto_shash_digestsize(ic->journal_mac);
553
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 }
570
571 return;
572 err:
573 memset(result, 0, JOURNAL_MAC_SIZE);
574 }
575
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;
580
581 if (!ic->journal_mac)
582 return;
583
584 section_mac(ic, section, result);
585
586 for (j = 0; j < JOURNAL_BLOCK_SECTORS; j++) {
587 struct journal_sector *js = access_journal(ic, section, j);
588
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 }
597
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 }
605
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;
613
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 }
621
622 page_list_location(ic, section, 0, &pl_index, &pl_offset);
623
624 atomic_add(roundup(pl_offset + n_bytes, PAGE_SIZE) >> PAGE_SHIFT, &comp->in_flight);
625
626 init_async_submit(&submit, ASYNC_TX_XOR_ZERO_DST, NULL, complete_journal_op, comp, NULL);
627
628 section_index = pl_index;
629
630 do {
631 size_t this_step;
632 struct page *src_pages[2];
633 struct page *dst_page;
634
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, &section_index, &dummy);
644 }
645
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;
650
651 async_xor(dst_page, src_pages, pl_offset, 2, this_step, &submit);
652
653 pl_index++;
654 pl_offset = 0;
655 n_bytes -= this_step;
656 } while (n_bytes);
657
658 BUG_ON(n_sections);
659
660 async_tx_issue_pending_all();
661 }
662
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 }
675
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 }
697
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;
703
704 atomic_add(2, &comp->in_flight);
705
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 }
713
714 do {
715 struct skcipher_request *req;
716 unsigned ivsize;
717 char *iv;
718
719 if (likely(encrypt))
720 rw_section_mac(ic, section, true);
721
722 req = ic->sk_requests[section];
723 ivsize = crypto_skcipher_ivsize(ic->journal_crypt);
724 iv = req->iv;
725
726 memcpy(iv, iv + ivsize, ivsize);
727
728 req->src = source_sg[section];
729 req->dst = target_sg[section];
730
731 if (unlikely(do_crypt(encrypt, req, comp)))
732 atomic_inc(&comp->in_flight);
733
734 section++;
735 n_sections--;
736 } while (n_sections);
737
738 atomic_dec(&comp->in_flight);
739 complete_journal_op(comp);
740 }
741
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 }
750
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 }
758
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;
766
767 if (unlikely(dm_integrity_failed(ic))) {
768 if (comp)
769 complete_journal_io(-1UL, comp);
770 return;
771 }
772
773 sector = section * ic->journal_section_sectors;
774 n_sectors = n_sections * ic->journal_section_sectors;
775
776 pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
777 pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
778
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;
797
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 }
807
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;
814
815 io_comp.ic = ic;
816 init_completion(&io_comp.comp);
817
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 }
865
866 wait_for_completion_io(&io_comp.comp);
867 }
868
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;
876
877 BUG_ON((target | n_sectors | offset) & (unsigned)(ic->sectors_per_block - 1));
878
879 if (unlikely(dm_integrity_failed(ic))) {
880 fn(-1UL, data);
881 return;
882 }
883
884 sector = section * ic->journal_section_sectors + JOURNAL_BLOCK_SECTORS + offset;
885
886 pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
887 pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
888
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;
900
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 }
907
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 range1->logical_sector + range1->n_sectors > range2->logical_sector;
912 }
913
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;
918
919 BUG_ON((new_range->logical_sector | new_range->n_sectors) & (unsigned)(ic->sectors_per_block - 1));
920
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 }
928
929 parent = NULL;
930
931 while (*n) {
932 struct dm_integrity_range *range = container_of(*n, struct dm_integrity_range, node);
933
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 }
943
944 rb_link_node(&new_range->node, parent, n);
945 rb_insert_color(&new_range->node, &ic->in_progress);
946
947 return true;
948 }
949
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 last_range_task = last_range->task;
958 list_del(&last_range->wait_entry);
959 if (!add_new_range(ic, last_range, false)) {
960 last_range->task = last_range_task;
961 list_add(&last_range->wait_entry, &ic->wait_list);
962 break;
963 }
964 last_range->waiting = false;
965 wake_up_process(last_range_task);
966 }
967 }
968
969 static void remove_range(struct dm_integrity_c *ic, struct dm_integrity_range *range)
970 {
971 unsigned long flags;
972
973 spin_lock_irqsave(&ic->endio_wait.lock, flags);
974 remove_range_unlocked(ic, range);
975 spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
976 }
977
978 static void wait_and_add_new_range(struct dm_integrity_c *ic, struct dm_integrity_range *new_range)
979 {
980 new_range->waiting = true;
981 list_add_tail(&new_range->wait_entry, &ic->wait_list);
982 new_range->task = current;
983 do {
984 __set_current_state(TASK_UNINTERRUPTIBLE);
985 spin_unlock_irq(&ic->endio_wait.lock);
986 io_schedule();
987 spin_lock_irq(&ic->endio_wait.lock);
988 } while (unlikely(new_range->waiting));
989 }
990
991 static void init_journal_node(struct journal_node *node)
992 {
993 RB_CLEAR_NODE(&node->node);
994 node->sector = (sector_t)-1;
995 }
996
997 static void add_journal_node(struct dm_integrity_c *ic, struct journal_node *node, sector_t sector)
998 {
999 struct rb_node **link;
1000 struct rb_node *parent;
1001
1002 node->sector = sector;
1003 BUG_ON(!RB_EMPTY_NODE(&node->node));
1004
1005 link = &ic->journal_tree_root.rb_node;
1006 parent = NULL;
1007
1008 while (*link) {
1009 struct journal_node *j;
1010 parent = *link;
1011 j = container_of(parent, struct journal_node, node);
1012 if (sector < j->sector)
1013 link = &j->node.rb_left;
1014 else
1015 link = &j->node.rb_right;
1016 }
1017
1018 rb_link_node(&node->node, parent, link);
1019 rb_insert_color(&node->node, &ic->journal_tree_root);
1020 }
1021
1022 static void remove_journal_node(struct dm_integrity_c *ic, struct journal_node *node)
1023 {
1024 BUG_ON(RB_EMPTY_NODE(&node->node));
1025 rb_erase(&node->node, &ic->journal_tree_root);
1026 init_journal_node(node);
1027 }
1028
1029 #define NOT_FOUND (-1U)
1030
1031 static unsigned find_journal_node(struct dm_integrity_c *ic, sector_t sector, sector_t *next_sector)
1032 {
1033 struct rb_node *n = ic->journal_tree_root.rb_node;
1034 unsigned found = NOT_FOUND;
1035 *next_sector = (sector_t)-1;
1036 while (n) {
1037 struct journal_node *j = container_of(n, struct journal_node, node);
1038 if (sector == j->sector) {
1039 found = j - ic->journal_tree;
1040 }
1041 if (sector < j->sector) {
1042 *next_sector = j->sector;
1043 n = j->node.rb_left;
1044 } else {
1045 n = j->node.rb_right;
1046 }
1047 }
1048
1049 return found;
1050 }
1051
1052 static bool test_journal_node(struct dm_integrity_c *ic, unsigned pos, sector_t sector)
1053 {
1054 struct journal_node *node, *next_node;
1055 struct rb_node *next;
1056
1057 if (unlikely(pos >= ic->journal_entries))
1058 return false;
1059 node = &ic->journal_tree[pos];
1060 if (unlikely(RB_EMPTY_NODE(&node->node)))
1061 return false;
1062 if (unlikely(node->sector != sector))
1063 return false;
1064
1065 next = rb_next(&node->node);
1066 if (unlikely(!next))
1067 return true;
1068
1069 next_node = container_of(next, struct journal_node, node);
1070 return next_node->sector != sector;
1071 }
1072
1073 static bool find_newer_committed_node(struct dm_integrity_c *ic, struct journal_node *node)
1074 {
1075 struct rb_node *next;
1076 struct journal_node *next_node;
1077 unsigned next_section;
1078
1079 BUG_ON(RB_EMPTY_NODE(&node->node));
1080
1081 next = rb_next(&node->node);
1082 if (unlikely(!next))
1083 return false;
1084
1085 next_node = container_of(next, struct journal_node, node);
1086
1087 if (next_node->sector != node->sector)
1088 return false;
1089
1090 next_section = (unsigned)(next_node - ic->journal_tree) / ic->journal_section_entries;
1091 if (next_section >= ic->committed_section &&
1092 next_section < ic->committed_section + ic->n_committed_sections)
1093 return true;
1094 if (next_section + ic->journal_sections < ic->committed_section + ic->n_committed_sections)
1095 return true;
1096
1097 return false;
1098 }
1099
1100 #define TAG_READ 0
1101 #define TAG_WRITE 1
1102 #define TAG_CMP 2
1103
1104 static int dm_integrity_rw_tag(struct dm_integrity_c *ic, unsigned char *tag, sector_t *metadata_block,
1105 unsigned *metadata_offset, unsigned total_size, int op)
1106 {
1107 do {
1108 unsigned char *data, *dp;
1109 struct dm_buffer *b;
1110 unsigned to_copy;
1111 int r;
1112
1113 r = dm_integrity_failed(ic);
1114 if (unlikely(r))
1115 return r;
1116
1117 data = dm_bufio_read(ic->bufio, *metadata_block, &b);
1118 if (unlikely(IS_ERR(data)))
1119 return PTR_ERR(data);
1120
1121 to_copy = min((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - *metadata_offset, total_size);
1122 dp = data + *metadata_offset;
1123 if (op == TAG_READ) {
1124 memcpy(tag, dp, to_copy);
1125 } else if (op == TAG_WRITE) {
1126 memcpy(dp, tag, to_copy);
1127 dm_bufio_mark_partial_buffer_dirty(b, *metadata_offset, *metadata_offset + to_copy);
1128 } else {
1129 /* e.g.: op == TAG_CMP */
1130 if (unlikely(memcmp(dp, tag, to_copy))) {
1131 unsigned i;
1132
1133 for (i = 0; i < to_copy; i++) {
1134 if (dp[i] != tag[i])
1135 break;
1136 total_size--;
1137 }
1138 dm_bufio_release(b);
1139 return total_size;
1140 }
1141 }
1142 dm_bufio_release(b);
1143
1144 tag += to_copy;
1145 *metadata_offset += to_copy;
1146 if (unlikely(*metadata_offset == 1U << SECTOR_SHIFT << ic->log2_buffer_sectors)) {
1147 (*metadata_block)++;
1148 *metadata_offset = 0;
1149 }
1150 total_size -= to_copy;
1151 } while (unlikely(total_size));
1152
1153 return 0;
1154 }
1155
1156 static void dm_integrity_flush_buffers(struct dm_integrity_c *ic)
1157 {
1158 int r;
1159 r = dm_bufio_write_dirty_buffers(ic->bufio);
1160 if (unlikely(r))
1161 dm_integrity_io_error(ic, "writing tags", r);
1162 }
1163
1164 static void sleep_on_endio_wait(struct dm_integrity_c *ic)
1165 {
1166 DECLARE_WAITQUEUE(wait, current);
1167 __add_wait_queue(&ic->endio_wait, &wait);
1168 __set_current_state(TASK_UNINTERRUPTIBLE);
1169 spin_unlock_irq(&ic->endio_wait.lock);
1170 io_schedule();
1171 spin_lock_irq(&ic->endio_wait.lock);
1172 __remove_wait_queue(&ic->endio_wait, &wait);
1173 }
1174
1175 static void autocommit_fn(struct timer_list *t)
1176 {
1177 struct dm_integrity_c *ic = from_timer(ic, t, autocommit_timer);
1178
1179 if (likely(!dm_integrity_failed(ic)))
1180 queue_work(ic->commit_wq, &ic->commit_work);
1181 }
1182
1183 static void schedule_autocommit(struct dm_integrity_c *ic)
1184 {
1185 if (!timer_pending(&ic->autocommit_timer))
1186 mod_timer(&ic->autocommit_timer, jiffies + ic->autocommit_jiffies);
1187 }
1188
1189 static void submit_flush_bio(struct dm_integrity_c *ic, struct dm_integrity_io *dio)
1190 {
1191 struct bio *bio;
1192 unsigned long flags;
1193
1194 spin_lock_irqsave(&ic->endio_wait.lock, flags);
1195 bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1196 bio_list_add(&ic->flush_bio_list, bio);
1197 spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
1198
1199 queue_work(ic->commit_wq, &ic->commit_work);
1200 }
1201
1202 static void do_endio(struct dm_integrity_c *ic, struct bio *bio)
1203 {
1204 int r = dm_integrity_failed(ic);
1205 if (unlikely(r) && !bio->bi_status)
1206 bio->bi_status = errno_to_blk_status(r);
1207 bio_endio(bio);
1208 }
1209
1210 static void do_endio_flush(struct dm_integrity_c *ic, struct dm_integrity_io *dio)
1211 {
1212 struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1213
1214 if (unlikely(dio->fua) && likely(!bio->bi_status) && likely(!dm_integrity_failed(ic)))
1215 submit_flush_bio(ic, dio);
1216 else
1217 do_endio(ic, bio);
1218 }
1219
1220 static void dec_in_flight(struct dm_integrity_io *dio)
1221 {
1222 if (atomic_dec_and_test(&dio->in_flight)) {
1223 struct dm_integrity_c *ic = dio->ic;
1224 struct bio *bio;
1225
1226 remove_range(ic, &dio->range);
1227
1228 if (unlikely(dio->write))
1229 schedule_autocommit(ic);
1230
1231 bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1232
1233 if (unlikely(dio->bi_status) && !bio->bi_status)
1234 bio->bi_status = dio->bi_status;
1235 if (likely(!bio->bi_status) && unlikely(bio_sectors(bio) != dio->range.n_sectors)) {
1236 dio->range.logical_sector += dio->range.n_sectors;
1237 bio_advance(bio, dio->range.n_sectors << SECTOR_SHIFT);
1238 INIT_WORK(&dio->work, integrity_bio_wait);
1239 queue_work(ic->offload_wq, &dio->work);
1240 return;
1241 }
1242 do_endio_flush(ic, dio);
1243 }
1244 }
1245
1246 static void integrity_end_io(struct bio *bio)
1247 {
1248 struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
1249
1250 dm_bio_restore(&dio->bio_details, bio);
1251 if (bio->bi_integrity)
1252 bio->bi_opf |= REQ_INTEGRITY;
1253
1254 if (dio->completion)
1255 complete(dio->completion);
1256
1257 dec_in_flight(dio);
1258 }
1259
1260 static void integrity_sector_checksum(struct dm_integrity_c *ic, sector_t sector,
1261 const char *data, char *result)
1262 {
1263 __u64 sector_le = cpu_to_le64(sector);
1264 SHASH_DESC_ON_STACK(req, ic->internal_hash);
1265 int r;
1266 unsigned digest_size;
1267
1268 req->tfm = ic->internal_hash;
1269 req->flags = 0;
1270
1271 r = crypto_shash_init(req);
1272 if (unlikely(r < 0)) {
1273 dm_integrity_io_error(ic, "crypto_shash_init", r);
1274 goto failed;
1275 }
1276
1277 r = crypto_shash_update(req, (const __u8 *)&sector_le, sizeof sector_le);
1278 if (unlikely(r < 0)) {
1279 dm_integrity_io_error(ic, "crypto_shash_update", r);
1280 goto failed;
1281 }
1282
1283 r = crypto_shash_update(req, data, ic->sectors_per_block << SECTOR_SHIFT);
1284 if (unlikely(r < 0)) {
1285 dm_integrity_io_error(ic, "crypto_shash_update", r);
1286 goto failed;
1287 }
1288
1289 r = crypto_shash_final(req, result);
1290 if (unlikely(r < 0)) {
1291 dm_integrity_io_error(ic, "crypto_shash_final", r);
1292 goto failed;
1293 }
1294
1295 digest_size = crypto_shash_digestsize(ic->internal_hash);
1296 if (unlikely(digest_size < ic->tag_size))
1297 memset(result + digest_size, 0, ic->tag_size - digest_size);
1298
1299 return;
1300
1301 failed:
1302 /* this shouldn't happen anyway, the hash functions have no reason to fail */
1303 get_random_bytes(result, ic->tag_size);
1304 }
1305
1306 static void integrity_metadata(struct work_struct *w)
1307 {
1308 struct dm_integrity_io *dio = container_of(w, struct dm_integrity_io, work);
1309 struct dm_integrity_c *ic = dio->ic;
1310
1311 int r;
1312
1313 if (ic->internal_hash) {
1314 struct bvec_iter iter;
1315 struct bio_vec bv;
1316 unsigned digest_size = crypto_shash_digestsize(ic->internal_hash);
1317 struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1318 char *checksums;
1319 unsigned extra_space = unlikely(digest_size > ic->tag_size) ? digest_size - ic->tag_size : 0;
1320 char checksums_onstack[ic->tag_size + extra_space];
1321 unsigned sectors_to_process = dio->range.n_sectors;
1322 sector_t sector = dio->range.logical_sector;
1323
1324 if (unlikely(ic->mode == 'R'))
1325 goto skip_io;
1326
1327 checksums = kmalloc((PAGE_SIZE >> SECTOR_SHIFT >> ic->sb->log2_sectors_per_block) * ic->tag_size + extra_space,
1328 GFP_NOIO | __GFP_NORETRY | __GFP_NOWARN);
1329 if (!checksums)
1330 checksums = checksums_onstack;
1331
1332 __bio_for_each_segment(bv, bio, iter, dio->bio_details.bi_iter) {
1333 unsigned pos;
1334 char *mem, *checksums_ptr;
1335
1336 again:
1337 mem = (char *)kmap_atomic(bv.bv_page) + bv.bv_offset;
1338 pos = 0;
1339 checksums_ptr = checksums;
1340 do {
1341 integrity_sector_checksum(ic, sector, mem + pos, checksums_ptr);
1342 checksums_ptr += ic->tag_size;
1343 sectors_to_process -= ic->sectors_per_block;
1344 pos += ic->sectors_per_block << SECTOR_SHIFT;
1345 sector += ic->sectors_per_block;
1346 } while (pos < bv.bv_len && sectors_to_process && checksums != checksums_onstack);
1347 kunmap_atomic(mem);
1348
1349 r = dm_integrity_rw_tag(ic, checksums, &dio->metadata_block, &dio->metadata_offset,
1350 checksums_ptr - checksums, !dio->write ? TAG_CMP : TAG_WRITE);
1351 if (unlikely(r)) {
1352 if (r > 0) {
1353 DMERR_LIMIT("Checksum failed at sector 0x%llx",
1354 (unsigned long long)(sector - ((r + ic->tag_size - 1) / ic->tag_size)));
1355 r = -EILSEQ;
1356 atomic64_inc(&ic->number_of_mismatches);
1357 }
1358 if (likely(checksums != checksums_onstack))
1359 kfree(checksums);
1360 goto error;
1361 }
1362
1363 if (!sectors_to_process)
1364 break;
1365
1366 if (unlikely(pos < bv.bv_len)) {
1367 bv.bv_offset += pos;
1368 bv.bv_len -= pos;
1369 goto again;
1370 }
1371 }
1372
1373 if (likely(checksums != checksums_onstack))
1374 kfree(checksums);
1375 } else {
1376 struct bio_integrity_payload *bip = dio->bio_details.bi_integrity;
1377
1378 if (bip) {
1379 struct bio_vec biv;
1380 struct bvec_iter iter;
1381 unsigned data_to_process = dio->range.n_sectors;
1382 sector_to_block(ic, data_to_process);
1383 data_to_process *= ic->tag_size;
1384
1385 bip_for_each_vec(biv, bip, iter) {
1386 unsigned char *tag;
1387 unsigned this_len;
1388
1389 BUG_ON(PageHighMem(biv.bv_page));
1390 tag = lowmem_page_address(biv.bv_page) + biv.bv_offset;
1391 this_len = min(biv.bv_len, data_to_process);
1392 r = dm_integrity_rw_tag(ic, tag, &dio->metadata_block, &dio->metadata_offset,
1393 this_len, !dio->write ? TAG_READ : TAG_WRITE);
1394 if (unlikely(r))
1395 goto error;
1396 data_to_process -= this_len;
1397 if (!data_to_process)
1398 break;
1399 }
1400 }
1401 }
1402 skip_io:
1403 dec_in_flight(dio);
1404 return;
1405 error:
1406 dio->bi_status = errno_to_blk_status(r);
1407 dec_in_flight(dio);
1408 }
1409
1410 static int dm_integrity_map(struct dm_target *ti, struct bio *bio)
1411 {
1412 struct dm_integrity_c *ic = ti->private;
1413 struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
1414 struct bio_integrity_payload *bip;
1415
1416 sector_t area, offset;
1417
1418 dio->ic = ic;
1419 dio->bi_status = 0;
1420
1421 if (unlikely(bio->bi_opf & REQ_PREFLUSH)) {
1422 submit_flush_bio(ic, dio);
1423 return DM_MAPIO_SUBMITTED;
1424 }
1425
1426 dio->range.logical_sector = dm_target_offset(ti, bio->bi_iter.bi_sector);
1427 dio->write = bio_op(bio) == REQ_OP_WRITE;
1428 dio->fua = dio->write && bio->bi_opf & REQ_FUA;
1429 if (unlikely(dio->fua)) {
1430 /*
1431 * Don't pass down the FUA flag because we have to flush
1432 * disk cache anyway.
1433 */
1434 bio->bi_opf &= ~REQ_FUA;
1435 }
1436 if (unlikely(dio->range.logical_sector + bio_sectors(bio) > ic->provided_data_sectors)) {
1437 DMERR("Too big sector number: 0x%llx + 0x%x > 0x%llx",
1438 (unsigned long long)dio->range.logical_sector, bio_sectors(bio),
1439 (unsigned long long)ic->provided_data_sectors);
1440 return DM_MAPIO_KILL;
1441 }
1442 if (unlikely((dio->range.logical_sector | bio_sectors(bio)) & (unsigned)(ic->sectors_per_block - 1))) {
1443 DMERR("Bio not aligned on %u sectors: 0x%llx, 0x%x",
1444 ic->sectors_per_block,
1445 (unsigned long long)dio->range.logical_sector, bio_sectors(bio));
1446 return DM_MAPIO_KILL;
1447 }
1448
1449 if (ic->sectors_per_block > 1) {
1450 struct bvec_iter iter;
1451 struct bio_vec bv;
1452 bio_for_each_segment(bv, bio, iter) {
1453 if (unlikely(bv.bv_len & ((ic->sectors_per_block << SECTOR_SHIFT) - 1))) {
1454 DMERR("Bio vector (%u,%u) is not aligned on %u-sector boundary",
1455 bv.bv_offset, bv.bv_len, ic->sectors_per_block);
1456 return DM_MAPIO_KILL;
1457 }
1458 }
1459 }
1460
1461 bip = bio_integrity(bio);
1462 if (!ic->internal_hash) {
1463 if (bip) {
1464 unsigned wanted_tag_size = bio_sectors(bio) >> ic->sb->log2_sectors_per_block;
1465 if (ic->log2_tag_size >= 0)
1466 wanted_tag_size <<= ic->log2_tag_size;
1467 else
1468 wanted_tag_size *= ic->tag_size;
1469 if (unlikely(wanted_tag_size != bip->bip_iter.bi_size)) {
1470 DMERR("Invalid integrity data size %u, expected %u", bip->bip_iter.bi_size, wanted_tag_size);
1471 return DM_MAPIO_KILL;
1472 }
1473 }
1474 } else {
1475 if (unlikely(bip != NULL)) {
1476 DMERR("Unexpected integrity data when using internal hash");
1477 return DM_MAPIO_KILL;
1478 }
1479 }
1480
1481 if (unlikely(ic->mode == 'R') && unlikely(dio->write))
1482 return DM_MAPIO_KILL;
1483
1484 get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
1485 dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset, &dio->metadata_offset);
1486 bio->bi_iter.bi_sector = get_data_sector(ic, area, offset);
1487
1488 dm_integrity_map_continue(dio, true);
1489 return DM_MAPIO_SUBMITTED;
1490 }
1491
1492 static bool __journal_read_write(struct dm_integrity_io *dio, struct bio *bio,
1493 unsigned journal_section, unsigned journal_entry)
1494 {
1495 struct dm_integrity_c *ic = dio->ic;
1496 sector_t logical_sector;
1497 unsigned n_sectors;
1498
1499 logical_sector = dio->range.logical_sector;
1500 n_sectors = dio->range.n_sectors;
1501 do {
1502 struct bio_vec bv = bio_iovec(bio);
1503 char *mem;
1504
1505 if (unlikely(bv.bv_len >> SECTOR_SHIFT > n_sectors))
1506 bv.bv_len = n_sectors << SECTOR_SHIFT;
1507 n_sectors -= bv.bv_len >> SECTOR_SHIFT;
1508 bio_advance_iter(bio, &bio->bi_iter, bv.bv_len);
1509 retry_kmap:
1510 mem = kmap_atomic(bv.bv_page);
1511 if (likely(dio->write))
1512 flush_dcache_page(bv.bv_page);
1513
1514 do {
1515 struct journal_entry *je = access_journal_entry(ic, journal_section, journal_entry);
1516
1517 if (unlikely(!dio->write)) {
1518 struct journal_sector *js;
1519 char *mem_ptr;
1520 unsigned s;
1521
1522 if (unlikely(journal_entry_is_inprogress(je))) {
1523 flush_dcache_page(bv.bv_page);
1524 kunmap_atomic(mem);
1525
1526 __io_wait_event(ic->copy_to_journal_wait, !journal_entry_is_inprogress(je));
1527 goto retry_kmap;
1528 }
1529 smp_rmb();
1530 BUG_ON(journal_entry_get_sector(je) != logical_sector);
1531 js = access_journal_data(ic, journal_section, journal_entry);
1532 mem_ptr = mem + bv.bv_offset;
1533 s = 0;
1534 do {
1535 memcpy(mem_ptr, js, JOURNAL_SECTOR_DATA);
1536 *(commit_id_t *)(mem_ptr + JOURNAL_SECTOR_DATA) = je->last_bytes[s];
1537 js++;
1538 mem_ptr += 1 << SECTOR_SHIFT;
1539 } while (++s < ic->sectors_per_block);
1540 #ifdef INTERNAL_VERIFY
1541 if (ic->internal_hash) {
1542 char checksums_onstack[max(crypto_shash_digestsize(ic->internal_hash), ic->tag_size)];
1543
1544 integrity_sector_checksum(ic, logical_sector, mem + bv.bv_offset, checksums_onstack);
1545 if (unlikely(memcmp(checksums_onstack, journal_entry_tag(ic, je), ic->tag_size))) {
1546 DMERR_LIMIT("Checksum failed when reading from journal, at sector 0x%llx",
1547 (unsigned long long)logical_sector);
1548 }
1549 }
1550 #endif
1551 }
1552
1553 if (!ic->internal_hash) {
1554 struct bio_integrity_payload *bip = bio_integrity(bio);
1555 unsigned tag_todo = ic->tag_size;
1556 char *tag_ptr = journal_entry_tag(ic, je);
1557
1558 if (bip) do {
1559 struct bio_vec biv = bvec_iter_bvec(bip->bip_vec, bip->bip_iter);
1560 unsigned tag_now = min(biv.bv_len, tag_todo);
1561 char *tag_addr;
1562 BUG_ON(PageHighMem(biv.bv_page));
1563 tag_addr = lowmem_page_address(biv.bv_page) + biv.bv_offset;
1564 if (likely(dio->write))
1565 memcpy(tag_ptr, tag_addr, tag_now);
1566 else
1567 memcpy(tag_addr, tag_ptr, tag_now);
1568 bvec_iter_advance(bip->bip_vec, &bip->bip_iter, tag_now);
1569 tag_ptr += tag_now;
1570 tag_todo -= tag_now;
1571 } while (unlikely(tag_todo)); else {
1572 if (likely(dio->write))
1573 memset(tag_ptr, 0, tag_todo);
1574 }
1575 }
1576
1577 if (likely(dio->write)) {
1578 struct journal_sector *js;
1579 unsigned s;
1580
1581 js = access_journal_data(ic, journal_section, journal_entry);
1582 memcpy(js, mem + bv.bv_offset, ic->sectors_per_block << SECTOR_SHIFT);
1583
1584 s = 0;
1585 do {
1586 je->last_bytes[s] = js[s].commit_id;
1587 } while (++s < ic->sectors_per_block);
1588
1589 if (ic->internal_hash) {
1590 unsigned digest_size = crypto_shash_digestsize(ic->internal_hash);
1591 if (unlikely(digest_size > ic->tag_size)) {
1592 char checksums_onstack[digest_size];
1593 integrity_sector_checksum(ic, logical_sector, (char *)js, checksums_onstack);
1594 memcpy(journal_entry_tag(ic, je), checksums_onstack, ic->tag_size);
1595 } else
1596 integrity_sector_checksum(ic, logical_sector, (char *)js, journal_entry_tag(ic, je));
1597 }
1598
1599 journal_entry_set_sector(je, logical_sector);
1600 }
1601 logical_sector += ic->sectors_per_block;
1602
1603 journal_entry++;
1604 if (unlikely(journal_entry == ic->journal_section_entries)) {
1605 journal_entry = 0;
1606 journal_section++;
1607 wraparound_section(ic, &journal_section);
1608 }
1609
1610 bv.bv_offset += ic->sectors_per_block << SECTOR_SHIFT;
1611 } while (bv.bv_len -= ic->sectors_per_block << SECTOR_SHIFT);
1612
1613 if (unlikely(!dio->write))
1614 flush_dcache_page(bv.bv_page);
1615 kunmap_atomic(mem);
1616 } while (n_sectors);
1617
1618 if (likely(dio->write)) {
1619 smp_mb();
1620 if (unlikely(waitqueue_active(&ic->copy_to_journal_wait)))
1621 wake_up(&ic->copy_to_journal_wait);
1622 if (READ_ONCE(ic->free_sectors) <= ic->free_sectors_threshold) {
1623 queue_work(ic->commit_wq, &ic->commit_work);
1624 } else {
1625 schedule_autocommit(ic);
1626 }
1627 } else {
1628 remove_range(ic, &dio->range);
1629 }
1630
1631 if (unlikely(bio->bi_iter.bi_size)) {
1632 sector_t area, offset;
1633
1634 dio->range.logical_sector = logical_sector;
1635 get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
1636 dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset, &dio->metadata_offset);
1637 return true;
1638 }
1639
1640 return false;
1641 }
1642
1643 static void dm_integrity_map_continue(struct dm_integrity_io *dio, bool from_map)
1644 {
1645 struct dm_integrity_c *ic = dio->ic;
1646 struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1647 unsigned journal_section, journal_entry;
1648 unsigned journal_read_pos;
1649 struct completion read_comp;
1650 bool need_sync_io = ic->internal_hash && !dio->write;
1651
1652 if (need_sync_io && from_map) {
1653 INIT_WORK(&dio->work, integrity_bio_wait);
1654 queue_work(ic->offload_wq, &dio->work);
1655 return;
1656 }
1657
1658 lock_retry:
1659 spin_lock_irq(&ic->endio_wait.lock);
1660 retry:
1661 if (unlikely(dm_integrity_failed(ic))) {
1662 spin_unlock_irq(&ic->endio_wait.lock);
1663 do_endio(ic, bio);
1664 return;
1665 }
1666 dio->range.n_sectors = bio_sectors(bio);
1667 journal_read_pos = NOT_FOUND;
1668 if (likely(ic->mode == 'J')) {
1669 if (dio->write) {
1670 unsigned next_entry, i, pos;
1671 unsigned ws, we, range_sectors;
1672
1673 dio->range.n_sectors = min(dio->range.n_sectors,
1674 ic->free_sectors << ic->sb->log2_sectors_per_block);
1675 if (unlikely(!dio->range.n_sectors)) {
1676 if (from_map)
1677 goto offload_to_thread;
1678 sleep_on_endio_wait(ic);
1679 goto retry;
1680 }
1681 range_sectors = dio->range.n_sectors >> ic->sb->log2_sectors_per_block;
1682 ic->free_sectors -= range_sectors;
1683 journal_section = ic->free_section;
1684 journal_entry = ic->free_section_entry;
1685
1686 next_entry = ic->free_section_entry + range_sectors;
1687 ic->free_section_entry = next_entry % ic->journal_section_entries;
1688 ic->free_section += next_entry / ic->journal_section_entries;
1689 ic->n_uncommitted_sections += next_entry / ic->journal_section_entries;
1690 wraparound_section(ic, &ic->free_section);
1691
1692 pos = journal_section * ic->journal_section_entries + journal_entry;
1693 ws = journal_section;
1694 we = journal_entry;
1695 i = 0;
1696 do {
1697 struct journal_entry *je;
1698
1699 add_journal_node(ic, &ic->journal_tree[pos], dio->range.logical_sector + i);
1700 pos++;
1701 if (unlikely(pos >= ic->journal_entries))
1702 pos = 0;
1703
1704 je = access_journal_entry(ic, ws, we);
1705 BUG_ON(!journal_entry_is_unused(je));
1706 journal_entry_set_inprogress(je);
1707 we++;
1708 if (unlikely(we == ic->journal_section_entries)) {
1709 we = 0;
1710 ws++;
1711 wraparound_section(ic, &ws);
1712 }
1713 } while ((i += ic->sectors_per_block) < dio->range.n_sectors);
1714
1715 spin_unlock_irq(&ic->endio_wait.lock);
1716 goto journal_read_write;
1717 } else {
1718 sector_t next_sector;
1719 journal_read_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
1720 if (likely(journal_read_pos == NOT_FOUND)) {
1721 if (unlikely(dio->range.n_sectors > next_sector - dio->range.logical_sector))
1722 dio->range.n_sectors = next_sector - dio->range.logical_sector;
1723 } else {
1724 unsigned i;
1725 unsigned jp = journal_read_pos + 1;
1726 for (i = ic->sectors_per_block; i < dio->range.n_sectors; i += ic->sectors_per_block, jp++) {
1727 if (!test_journal_node(ic, jp, dio->range.logical_sector + i))
1728 break;
1729 }
1730 dio->range.n_sectors = i;
1731 }
1732 }
1733 }
1734 if (unlikely(!add_new_range(ic, &dio->range, true))) {
1735 /*
1736 * We must not sleep in the request routine because it could
1737 * stall bios on current->bio_list.
1738 * So, we offload the bio to a workqueue if we have to sleep.
1739 */
1740 if (from_map) {
1741 offload_to_thread:
1742 spin_unlock_irq(&ic->endio_wait.lock);
1743 INIT_WORK(&dio->work, integrity_bio_wait);
1744 queue_work(ic->wait_wq, &dio->work);
1745 return;
1746 }
1747 if (journal_read_pos != NOT_FOUND)
1748 dio->range.n_sectors = ic->sectors_per_block;
1749 wait_and_add_new_range(ic, &dio->range);
1750 /*
1751 * wait_and_add_new_range drops the spinlock, so the journal
1752 * may have been changed arbitrarily. We need to recheck.
1753 * To simplify the code, we restrict I/O size to just one block.
1754 */
1755 if (journal_read_pos != NOT_FOUND) {
1756 sector_t next_sector;
1757 unsigned new_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
1758 if (unlikely(new_pos != journal_read_pos)) {
1759 remove_range_unlocked(ic, &dio->range);
1760 goto retry;
1761 }
1762 }
1763 }
1764 spin_unlock_irq(&ic->endio_wait.lock);
1765
1766 if (unlikely(journal_read_pos != NOT_FOUND)) {
1767 journal_section = journal_read_pos / ic->journal_section_entries;
1768 journal_entry = journal_read_pos % ic->journal_section_entries;
1769 goto journal_read_write;
1770 }
1771
1772 dio->in_flight = (atomic_t)ATOMIC_INIT(2);
1773
1774 if (need_sync_io) {
1775 init_completion(&read_comp);
1776 dio->completion = &read_comp;
1777 } else
1778 dio->completion = NULL;
1779
1780 dm_bio_record(&dio->bio_details, bio);
1781 bio_set_dev(bio, ic->dev->bdev);
1782 bio->bi_integrity = NULL;
1783 bio->bi_opf &= ~REQ_INTEGRITY;
1784 bio->bi_end_io = integrity_end_io;
1785 bio->bi_iter.bi_size = dio->range.n_sectors << SECTOR_SHIFT;
1786
1787 generic_make_request(bio);
1788
1789 if (need_sync_io) {
1790 wait_for_completion_io(&read_comp);
1791 if (unlikely(ic->recalc_wq != NULL) &&
1792 ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING) &&
1793 dio->range.logical_sector + dio->range.n_sectors > le64_to_cpu(ic->sb->recalc_sector))
1794 goto skip_check;
1795 if (likely(!bio->bi_status))
1796 integrity_metadata(&dio->work);
1797 else
1798 skip_check:
1799 dec_in_flight(dio);
1800
1801 } else {
1802 INIT_WORK(&dio->work, integrity_metadata);
1803 queue_work(ic->metadata_wq, &dio->work);
1804 }
1805
1806 return;
1807
1808 journal_read_write:
1809 if (unlikely(__journal_read_write(dio, bio, journal_section, journal_entry)))
1810 goto lock_retry;
1811
1812 do_endio_flush(ic, dio);
1813 }
1814
1815
1816 static void integrity_bio_wait(struct work_struct *w)
1817 {
1818 struct dm_integrity_io *dio = container_of(w, struct dm_integrity_io, work);
1819
1820 dm_integrity_map_continue(dio, false);
1821 }
1822
1823 static void pad_uncommitted(struct dm_integrity_c *ic)
1824 {
1825 if (ic->free_section_entry) {
1826 ic->free_sectors -= ic->journal_section_entries - ic->free_section_entry;
1827 ic->free_section_entry = 0;
1828 ic->free_section++;
1829 wraparound_section(ic, &ic->free_section);
1830 ic->n_uncommitted_sections++;
1831 }
1832 WARN_ON(ic->journal_sections * ic->journal_section_entries !=
1833 (ic->n_uncommitted_sections + ic->n_committed_sections) * ic->journal_section_entries + ic->free_sectors);
1834 }
1835
1836 static void integrity_commit(struct work_struct *w)
1837 {
1838 struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, commit_work);
1839 unsigned commit_start, commit_sections;
1840 unsigned i, j, n;
1841 struct bio *flushes;
1842
1843 del_timer(&ic->autocommit_timer);
1844
1845 spin_lock_irq(&ic->endio_wait.lock);
1846 flushes = bio_list_get(&ic->flush_bio_list);
1847 if (unlikely(ic->mode != 'J')) {
1848 spin_unlock_irq(&ic->endio_wait.lock);
1849 dm_integrity_flush_buffers(ic);
1850 goto release_flush_bios;
1851 }
1852
1853 pad_uncommitted(ic);
1854 commit_start = ic->uncommitted_section;
1855 commit_sections = ic->n_uncommitted_sections;
1856 spin_unlock_irq(&ic->endio_wait.lock);
1857
1858 if (!commit_sections)
1859 goto release_flush_bios;
1860
1861 i = commit_start;
1862 for (n = 0; n < commit_sections; n++) {
1863 for (j = 0; j < ic->journal_section_entries; j++) {
1864 struct journal_entry *je;
1865 je = access_journal_entry(ic, i, j);
1866 io_wait_event(ic->copy_to_journal_wait, !journal_entry_is_inprogress(je));
1867 }
1868 for (j = 0; j < ic->journal_section_sectors; j++) {
1869 struct journal_sector *js;
1870 js = access_journal(ic, i, j);
1871 js->commit_id = dm_integrity_commit_id(ic, i, j, ic->commit_seq);
1872 }
1873 i++;
1874 if (unlikely(i >= ic->journal_sections))
1875 ic->commit_seq = next_commit_seq(ic->commit_seq);
1876 wraparound_section(ic, &i);
1877 }
1878 smp_rmb();
1879
1880 write_journal(ic, commit_start, commit_sections);
1881
1882 spin_lock_irq(&ic->endio_wait.lock);
1883 ic->uncommitted_section += commit_sections;
1884 wraparound_section(ic, &ic->uncommitted_section);
1885 ic->n_uncommitted_sections -= commit_sections;
1886 ic->n_committed_sections += commit_sections;
1887 spin_unlock_irq(&ic->endio_wait.lock);
1888
1889 if (READ_ONCE(ic->free_sectors) <= ic->free_sectors_threshold)
1890 queue_work(ic->writer_wq, &ic->writer_work);
1891
1892 release_flush_bios:
1893 while (flushes) {
1894 struct bio *next = flushes->bi_next;
1895 flushes->bi_next = NULL;
1896 do_endio(ic, flushes);
1897 flushes = next;
1898 }
1899 }
1900
1901 static void complete_copy_from_journal(unsigned long error, void *context)
1902 {
1903 struct journal_io *io = context;
1904 struct journal_completion *comp = io->comp;
1905 struct dm_integrity_c *ic = comp->ic;
1906 remove_range(ic, &io->range);
1907 mempool_free(io, &ic->journal_io_mempool);
1908 if (unlikely(error != 0))
1909 dm_integrity_io_error(ic, "copying from journal", -EIO);
1910 complete_journal_op(comp);
1911 }
1912
1913 static void restore_last_bytes(struct dm_integrity_c *ic, struct journal_sector *js,
1914 struct journal_entry *je)
1915 {
1916 unsigned s = 0;
1917 do {
1918 js->commit_id = je->last_bytes[s];
1919 js++;
1920 } while (++s < ic->sectors_per_block);
1921 }
1922
1923 static void do_journal_write(struct dm_integrity_c *ic, unsigned write_start,
1924 unsigned write_sections, bool from_replay)
1925 {
1926 unsigned i, j, n;
1927 struct journal_completion comp;
1928 struct blk_plug plug;
1929
1930 blk_start_plug(&plug);
1931
1932 comp.ic = ic;
1933 comp.in_flight = (atomic_t)ATOMIC_INIT(1);
1934 init_completion(&comp.comp);
1935
1936 i = write_start;
1937 for (n = 0; n < write_sections; n++, i++, wraparound_section(ic, &i)) {
1938 #ifndef INTERNAL_VERIFY
1939 if (unlikely(from_replay))
1940 #endif
1941 rw_section_mac(ic, i, false);
1942 for (j = 0; j < ic->journal_section_entries; j++) {
1943 struct journal_entry *je = access_journal_entry(ic, i, j);
1944 sector_t sec, area, offset;
1945 unsigned k, l, next_loop;
1946 sector_t metadata_block;
1947 unsigned metadata_offset;
1948 struct journal_io *io;
1949
1950 if (journal_entry_is_unused(je))
1951 continue;
1952 BUG_ON(unlikely(journal_entry_is_inprogress(je)) && !from_replay);
1953 sec = journal_entry_get_sector(je);
1954 if (unlikely(from_replay)) {
1955 if (unlikely(sec & (unsigned)(ic->sectors_per_block - 1))) {
1956 dm_integrity_io_error(ic, "invalid sector in journal", -EIO);
1957 sec &= ~(sector_t)(ic->sectors_per_block - 1);
1958 }
1959 }
1960 get_area_and_offset(ic, sec, &area, &offset);
1961 restore_last_bytes(ic, access_journal_data(ic, i, j), je);
1962 for (k = j + 1; k < ic->journal_section_entries; k++) {
1963 struct journal_entry *je2 = access_journal_entry(ic, i, k);
1964 sector_t sec2, area2, offset2;
1965 if (journal_entry_is_unused(je2))
1966 break;
1967 BUG_ON(unlikely(journal_entry_is_inprogress(je2)) && !from_replay);
1968 sec2 = journal_entry_get_sector(je2);
1969 get_area_and_offset(ic, sec2, &area2, &offset2);
1970 if (area2 != area || offset2 != offset + ((k - j) << ic->sb->log2_sectors_per_block))
1971 break;
1972 restore_last_bytes(ic, access_journal_data(ic, i, k), je2);
1973 }
1974 next_loop = k - 1;
1975
1976 io = mempool_alloc(&ic->journal_io_mempool, GFP_NOIO);
1977 io->comp = &comp;
1978 io->range.logical_sector = sec;
1979 io->range.n_sectors = (k - j) << ic->sb->log2_sectors_per_block;
1980
1981 spin_lock_irq(&ic->endio_wait.lock);
1982 if (unlikely(!add_new_range(ic, &io->range, true)))
1983 wait_and_add_new_range(ic, &io->range);
1984
1985 if (likely(!from_replay)) {
1986 struct journal_node *section_node = &ic->journal_tree[i * ic->journal_section_entries];
1987
1988 /* don't write if there is newer committed sector */
1989 while (j < k && find_newer_committed_node(ic, &section_node[j])) {
1990 struct journal_entry *je2 = access_journal_entry(ic, i, j);
1991
1992 journal_entry_set_unused(je2);
1993 remove_journal_node(ic, &section_node[j]);
1994 j++;
1995 sec += ic->sectors_per_block;
1996 offset += ic->sectors_per_block;
1997 }
1998 while (j < k && find_newer_committed_node(ic, &section_node[k - 1])) {
1999 struct journal_entry *je2 = access_journal_entry(ic, i, k - 1);
2000
2001 journal_entry_set_unused(je2);
2002 remove_journal_node(ic, &section_node[k - 1]);
2003 k--;
2004 }
2005 if (j == k) {
2006 remove_range_unlocked(ic, &io->range);
2007 spin_unlock_irq(&ic->endio_wait.lock);
2008 mempool_free(io, &ic->journal_io_mempool);
2009 goto skip_io;
2010 }
2011 for (l = j; l < k; l++) {
2012 remove_journal_node(ic, &section_node[l]);
2013 }
2014 }
2015 spin_unlock_irq(&ic->endio_wait.lock);
2016
2017 metadata_block = get_metadata_sector_and_offset(ic, area, offset, &metadata_offset);
2018 for (l = j; l < k; l++) {
2019 int r;
2020 struct journal_entry *je2 = access_journal_entry(ic, i, l);
2021
2022 if (
2023 #ifndef INTERNAL_VERIFY
2024 unlikely(from_replay) &&
2025 #endif
2026 ic->internal_hash) {
2027 char test_tag[max(crypto_shash_digestsize(ic->internal_hash), ic->tag_size)];
2028
2029 integrity_sector_checksum(ic, sec + ((l - j) << ic->sb->log2_sectors_per_block),
2030 (char *)access_journal_data(ic, i, l), test_tag);
2031 if (unlikely(memcmp(test_tag, journal_entry_tag(ic, je2), ic->tag_size)))
2032 dm_integrity_io_error(ic, "tag mismatch when replaying journal", -EILSEQ);
2033 }
2034
2035 journal_entry_set_unused(je2);
2036 r = dm_integrity_rw_tag(ic, journal_entry_tag(ic, je2), &metadata_block, &metadata_offset,
2037 ic->tag_size, TAG_WRITE);
2038 if (unlikely(r)) {
2039 dm_integrity_io_error(ic, "reading tags", r);
2040 }
2041 }
2042
2043 atomic_inc(&comp.in_flight);
2044 copy_from_journal(ic, i, j << ic->sb->log2_sectors_per_block,
2045 (k - j) << ic->sb->log2_sectors_per_block,
2046 get_data_sector(ic, area, offset),
2047 complete_copy_from_journal, io);
2048 skip_io:
2049 j = next_loop;
2050 }
2051 }
2052
2053 dm_bufio_write_dirty_buffers_async(ic->bufio);
2054
2055 blk_finish_plug(&plug);
2056
2057 complete_journal_op(&comp);
2058 wait_for_completion_io(&comp.comp);
2059
2060 dm_integrity_flush_buffers(ic);
2061 }
2062
2063 static void integrity_writer(struct work_struct *w)
2064 {
2065 struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, writer_work);
2066 unsigned write_start, write_sections;
2067
2068 unsigned prev_free_sectors;
2069
2070 /* the following test is not needed, but it tests the replay code */
2071 if (unlikely(dm_suspended(ic->ti)) && !ic->meta_dev)
2072 return;
2073
2074 spin_lock_irq(&ic->endio_wait.lock);
2075 write_start = ic->committed_section;
2076 write_sections = ic->n_committed_sections;
2077 spin_unlock_irq(&ic->endio_wait.lock);
2078
2079 if (!write_sections)
2080 return;
2081
2082 do_journal_write(ic, write_start, write_sections, false);
2083
2084 spin_lock_irq(&ic->endio_wait.lock);
2085
2086 ic->committed_section += write_sections;
2087 wraparound_section(ic, &ic->committed_section);
2088 ic->n_committed_sections -= write_sections;
2089
2090 prev_free_sectors = ic->free_sectors;
2091 ic->free_sectors += write_sections * ic->journal_section_entries;
2092 if (unlikely(!prev_free_sectors))
2093 wake_up_locked(&ic->endio_wait);
2094
2095 spin_unlock_irq(&ic->endio_wait.lock);
2096 }
2097
2098 static void recalc_write_super(struct dm_integrity_c *ic)
2099 {
2100 int r;
2101
2102 dm_integrity_flush_buffers(ic);
2103 if (dm_integrity_failed(ic))
2104 return;
2105
2106 sb_set_version(ic);
2107 r = sync_rw_sb(ic, REQ_OP_WRITE, 0);
2108 if (unlikely(r))
2109 dm_integrity_io_error(ic, "writing superblock", r);
2110 }
2111
2112 static void integrity_recalc(struct work_struct *w)
2113 {
2114 struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, recalc_work);
2115 struct dm_integrity_range range;
2116 struct dm_io_request io_req;
2117 struct dm_io_region io_loc;
2118 sector_t area, offset;
2119 sector_t metadata_block;
2120 unsigned metadata_offset;
2121 __u8 *t;
2122 unsigned i;
2123 int r;
2124 unsigned super_counter = 0;
2125
2126 spin_lock_irq(&ic->endio_wait.lock);
2127
2128 next_chunk:
2129
2130 if (unlikely(dm_suspended(ic->ti)))
2131 goto unlock_ret;
2132
2133 range.logical_sector = le64_to_cpu(ic->sb->recalc_sector);
2134 if (unlikely(range.logical_sector >= ic->provided_data_sectors))
2135 goto unlock_ret;
2136
2137 get_area_and_offset(ic, range.logical_sector, &area, &offset);
2138 range.n_sectors = min((sector_t)RECALC_SECTORS, ic->provided_data_sectors - range.logical_sector);
2139 if (!ic->meta_dev)
2140 range.n_sectors = min(range.n_sectors, (1U << ic->sb->log2_interleave_sectors) - (unsigned)offset);
2141
2142 if (unlikely(!add_new_range(ic, &range, true)))
2143 wait_and_add_new_range(ic, &range);
2144
2145 spin_unlock_irq(&ic->endio_wait.lock);
2146
2147 if (unlikely(++super_counter == RECALC_WRITE_SUPER)) {
2148 recalc_write_super(ic);
2149 super_counter = 0;
2150 }
2151
2152 if (unlikely(dm_integrity_failed(ic)))
2153 goto err;
2154
2155 io_req.bi_op = REQ_OP_READ;
2156 io_req.bi_op_flags = 0;
2157 io_req.mem.type = DM_IO_VMA;
2158 io_req.mem.ptr.addr = ic->recalc_buffer;
2159 io_req.notify.fn = NULL;
2160 io_req.client = ic->io;
2161 io_loc.bdev = ic->dev->bdev;
2162 io_loc.sector = get_data_sector(ic, area, offset);
2163 io_loc.count = range.n_sectors;
2164
2165 r = dm_io(&io_req, 1, &io_loc, NULL);
2166 if (unlikely(r)) {
2167 dm_integrity_io_error(ic, "reading data", r);
2168 goto err;
2169 }
2170
2171 t = ic->recalc_tags;
2172 for (i = 0; i < range.n_sectors; i += ic->sectors_per_block) {
2173 integrity_sector_checksum(ic, range.logical_sector + i, ic->recalc_buffer + (i << SECTOR_SHIFT), t);
2174 t += ic->tag_size;
2175 }
2176
2177 metadata_block = get_metadata_sector_and_offset(ic, area, offset, &metadata_offset);
2178
2179 r = dm_integrity_rw_tag(ic, ic->recalc_tags, &metadata_block, &metadata_offset, t - ic->recalc_tags, TAG_WRITE);
2180 if (unlikely(r)) {
2181 dm_integrity_io_error(ic, "writing tags", r);
2182 goto err;
2183 }
2184
2185 spin_lock_irq(&ic->endio_wait.lock);
2186 remove_range_unlocked(ic, &range);
2187 ic->sb->recalc_sector = cpu_to_le64(range.logical_sector + range.n_sectors);
2188 goto next_chunk;
2189
2190 err:
2191 remove_range(ic, &range);
2192 return;
2193
2194 unlock_ret:
2195 spin_unlock_irq(&ic->endio_wait.lock);
2196
2197 recalc_write_super(ic);
2198 }
2199
2200 static void init_journal(struct dm_integrity_c *ic, unsigned start_section,
2201 unsigned n_sections, unsigned char commit_seq)
2202 {
2203 unsigned i, j, n;
2204
2205 if (!n_sections)
2206 return;
2207
2208 for (n = 0; n < n_sections; n++) {
2209 i = start_section + n;
2210 wraparound_section(ic, &i);
2211 for (j = 0; j < ic->journal_section_sectors; j++) {
2212 struct journal_sector *js = access_journal(ic, i, j);
2213 memset(&js->entries, 0, JOURNAL_SECTOR_DATA);
2214 js->commit_id = dm_integrity_commit_id(ic, i, j, commit_seq);
2215 }
2216 for (j = 0; j < ic->journal_section_entries; j++) {
2217 struct journal_entry *je = access_journal_entry(ic, i, j);
2218 journal_entry_set_unused(je);
2219 }
2220 }
2221
2222 write_journal(ic, start_section, n_sections);
2223 }
2224
2225 static int find_commit_seq(struct dm_integrity_c *ic, unsigned i, unsigned j, commit_id_t id)
2226 {
2227 unsigned char k;
2228 for (k = 0; k < N_COMMIT_IDS; k++) {
2229 if (dm_integrity_commit_id(ic, i, j, k) == id)
2230 return k;
2231 }
2232 dm_integrity_io_error(ic, "journal commit id", -EIO);
2233 return -EIO;
2234 }
2235
2236 static void replay_journal(struct dm_integrity_c *ic)
2237 {
2238 unsigned i, j;
2239 bool used_commit_ids[N_COMMIT_IDS];
2240 unsigned max_commit_id_sections[N_COMMIT_IDS];
2241 unsigned write_start, write_sections;
2242 unsigned continue_section;
2243 bool journal_empty;
2244 unsigned char unused, last_used, want_commit_seq;
2245
2246 if (ic->mode == 'R')
2247 return;
2248
2249 if (ic->journal_uptodate)
2250 return;
2251
2252 last_used = 0;
2253 write_start = 0;
2254
2255 if (!ic->just_formatted) {
2256 DEBUG_print("reading journal\n");
2257 rw_journal(ic, REQ_OP_READ, 0, 0, ic->journal_sections, NULL);
2258 if (ic->journal_io)
2259 DEBUG_bytes(lowmem_page_address(ic->journal_io[0].page), 64, "read journal");
2260 if (ic->journal_io) {
2261 struct journal_completion crypt_comp;
2262 crypt_comp.ic = ic;
2263 init_completion(&crypt_comp.comp);
2264 crypt_comp.in_flight = (atomic_t)ATOMIC_INIT(0);
2265 encrypt_journal(ic, false, 0, ic->journal_sections, &crypt_comp);
2266 wait_for_completion(&crypt_comp.comp);
2267 }
2268 DEBUG_bytes(lowmem_page_address(ic->journal[0].page), 64, "decrypted journal");
2269 }
2270
2271 if (dm_integrity_failed(ic))
2272 goto clear_journal;
2273
2274 journal_empty = true;
2275 memset(used_commit_ids, 0, sizeof used_commit_ids);
2276 memset(max_commit_id_sections, 0, sizeof max_commit_id_sections);
2277 for (i = 0; i < ic->journal_sections; i++) {
2278 for (j = 0; j < ic->journal_section_sectors; j++) {
2279 int k;
2280 struct journal_sector *js = access_journal(ic, i, j);
2281 k = find_commit_seq(ic, i, j, js->commit_id);
2282 if (k < 0)
2283 goto clear_journal;
2284 used_commit_ids[k] = true;
2285 max_commit_id_sections[k] = i;
2286 }
2287 if (journal_empty) {
2288 for (j = 0; j < ic->journal_section_entries; j++) {
2289 struct journal_entry *je = access_journal_entry(ic, i, j);
2290 if (!journal_entry_is_unused(je)) {
2291 journal_empty = false;
2292 break;
2293 }
2294 }
2295 }
2296 }
2297
2298 if (!used_commit_ids[N_COMMIT_IDS - 1]) {
2299 unused = N_COMMIT_IDS - 1;
2300 while (unused && !used_commit_ids[unused - 1])
2301 unused--;
2302 } else {
2303 for (unused = 0; unused < N_COMMIT_IDS; unused++)
2304 if (!used_commit_ids[unused])
2305 break;
2306 if (unused == N_COMMIT_IDS) {
2307 dm_integrity_io_error(ic, "journal commit ids", -EIO);
2308 goto clear_journal;
2309 }
2310 }
2311 DEBUG_print("first unused commit seq %d [%d,%d,%d,%d]\n",
2312 unused, used_commit_ids[0], used_commit_ids[1],
2313 used_commit_ids[2], used_commit_ids[3]);
2314
2315 last_used = prev_commit_seq(unused);
2316 want_commit_seq = prev_commit_seq(last_used);
2317
2318 if (!used_commit_ids[want_commit_seq] && used_commit_ids[prev_commit_seq(want_commit_seq)])
2319 journal_empty = true;
2320
2321 write_start = max_commit_id_sections[last_used] + 1;
2322 if (unlikely(write_start >= ic->journal_sections))
2323 want_commit_seq = next_commit_seq(want_commit_seq);
2324 wraparound_section(ic, &write_start);
2325
2326 i = write_start;
2327 for (write_sections = 0; write_sections < ic->journal_sections; write_sections++) {
2328 for (j = 0; j < ic->journal_section_sectors; j++) {
2329 struct journal_sector *js = access_journal(ic, i, j);
2330
2331 if (js->commit_id != dm_integrity_commit_id(ic, i, j, want_commit_seq)) {
2332 /*
2333 * This could be caused by crash during writing.
2334 * We won't replay the inconsistent part of the
2335 * journal.
2336 */
2337 DEBUG_print("commit id mismatch at position (%u, %u): %d != %d\n",
2338 i, j, find_commit_seq(ic, i, j, js->commit_id), want_commit_seq);
2339 goto brk;
2340 }
2341 }
2342 i++;
2343 if (unlikely(i >= ic->journal_sections))
2344 want_commit_seq = next_commit_seq(want_commit_seq);
2345 wraparound_section(ic, &i);
2346 }
2347 brk:
2348
2349 if (!journal_empty) {
2350 DEBUG_print("replaying %u sections, starting at %u, commit seq %d\n",
2351 write_sections, write_start, want_commit_seq);
2352 do_journal_write(ic, write_start, write_sections, true);
2353 }
2354
2355 if (write_sections == ic->journal_sections && (ic->mode == 'J' || journal_empty)) {
2356 continue_section = write_start;
2357 ic->commit_seq = want_commit_seq;
2358 DEBUG_print("continuing from section %u, commit seq %d\n", write_start, ic->commit_seq);
2359 } else {
2360 unsigned s;
2361 unsigned char erase_seq;
2362 clear_journal:
2363 DEBUG_print("clearing journal\n");
2364
2365 erase_seq = prev_commit_seq(prev_commit_seq(last_used));
2366 s = write_start;
2367 init_journal(ic, s, 1, erase_seq);
2368 s++;
2369 wraparound_section(ic, &s);
2370 if (ic->journal_sections >= 2) {
2371 init_journal(ic, s, ic->journal_sections - 2, erase_seq);
2372 s += ic->journal_sections - 2;
2373 wraparound_section(ic, &s);
2374 init_journal(ic, s, 1, erase_seq);
2375 }
2376
2377 continue_section = 0;
2378 ic->commit_seq = next_commit_seq(erase_seq);
2379 }
2380
2381 ic->committed_section = continue_section;
2382 ic->n_committed_sections = 0;
2383
2384 ic->uncommitted_section = continue_section;
2385 ic->n_uncommitted_sections = 0;
2386
2387 ic->free_section = continue_section;
2388 ic->free_section_entry = 0;
2389 ic->free_sectors = ic->journal_entries;
2390
2391 ic->journal_tree_root = RB_ROOT;
2392 for (i = 0; i < ic->journal_entries; i++)
2393 init_journal_node(&ic->journal_tree[i]);
2394 }
2395
2396 static void dm_integrity_postsuspend(struct dm_target *ti)
2397 {
2398 struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
2399
2400 del_timer_sync(&ic->autocommit_timer);
2401
2402 if (ic->recalc_wq)
2403 drain_workqueue(ic->recalc_wq);
2404
2405 queue_work(ic->commit_wq, &ic->commit_work);
2406 drain_workqueue(ic->commit_wq);
2407
2408 if (ic->mode == 'J') {
2409 if (ic->meta_dev)
2410 queue_work(ic->writer_wq, &ic->writer_work);
2411 drain_workqueue(ic->writer_wq);
2412 dm_integrity_flush_buffers(ic);
2413 }
2414
2415 BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
2416
2417 ic->journal_uptodate = true;
2418 }
2419
2420 static void dm_integrity_resume(struct dm_target *ti)
2421 {
2422 struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
2423
2424 replay_journal(ic);
2425
2426 if (ic->recalc_wq && ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
2427 __u64 recalc_pos = le64_to_cpu(ic->sb->recalc_sector);
2428 if (recalc_pos < ic->provided_data_sectors) {
2429 queue_work(ic->recalc_wq, &ic->recalc_work);
2430 } else if (recalc_pos > ic->provided_data_sectors) {
2431 ic->sb->recalc_sector = cpu_to_le64(ic->provided_data_sectors);
2432 recalc_write_super(ic);
2433 }
2434 }
2435 }
2436
2437 static void dm_integrity_status(struct dm_target *ti, status_type_t type,
2438 unsigned status_flags, char *result, unsigned maxlen)
2439 {
2440 struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
2441 unsigned arg_count;
2442 size_t sz = 0;
2443
2444 switch (type) {
2445 case STATUSTYPE_INFO:
2446 DMEMIT("%llu %llu",
2447 (unsigned long long)atomic64_read(&ic->number_of_mismatches),
2448 (unsigned long long)ic->provided_data_sectors);
2449 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
2450 DMEMIT(" %llu", (unsigned long long)le64_to_cpu(ic->sb->recalc_sector));
2451 else
2452 DMEMIT(" -");
2453 break;
2454
2455 case STATUSTYPE_TABLE: {
2456 __u64 watermark_percentage = (__u64)(ic->journal_entries - ic->free_sectors_threshold) * 100;
2457 watermark_percentage += ic->journal_entries / 2;
2458 do_div(watermark_percentage, ic->journal_entries);
2459 arg_count = 5;
2460 arg_count += !!ic->meta_dev;
2461 arg_count += ic->sectors_per_block != 1;
2462 arg_count += !!(ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING));
2463 arg_count += !!ic->internal_hash_alg.alg_string;
2464 arg_count += !!ic->journal_crypt_alg.alg_string;
2465 arg_count += !!ic->journal_mac_alg.alg_string;
2466 DMEMIT("%s %llu %u %c %u", ic->dev->name, (unsigned long long)ic->start,
2467 ic->tag_size, ic->mode, arg_count);
2468 if (ic->meta_dev)
2469 DMEMIT(" meta_device:%s", ic->meta_dev->name);
2470 if (ic->sectors_per_block != 1)
2471 DMEMIT(" block_size:%u", ic->sectors_per_block << SECTOR_SHIFT);
2472 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
2473 DMEMIT(" recalculate");
2474 DMEMIT(" journal_sectors:%u", ic->initial_sectors - SB_SECTORS);
2475 DMEMIT(" interleave_sectors:%u", 1U << ic->sb->log2_interleave_sectors);
2476 DMEMIT(" buffer_sectors:%u", 1U << ic->log2_buffer_sectors);
2477 DMEMIT(" journal_watermark:%u", (unsigned)watermark_percentage);
2478 DMEMIT(" commit_time:%u", ic->autocommit_msec);
2479
2480 #define EMIT_ALG(a, n) \
2481 do { \
2482 if (ic->a.alg_string) { \
2483 DMEMIT(" %s:%s", n, ic->a.alg_string); \
2484 if (ic->a.key_string) \
2485 DMEMIT(":%s", ic->a.key_string);\
2486 } \
2487 } while (0)
2488 EMIT_ALG(internal_hash_alg, "internal_hash");
2489 EMIT_ALG(journal_crypt_alg, "journal_crypt");
2490 EMIT_ALG(journal_mac_alg, "journal_mac");
2491 break;
2492 }
2493 }
2494 }
2495
2496 static int dm_integrity_iterate_devices(struct dm_target *ti,
2497 iterate_devices_callout_fn fn, void *data)
2498 {
2499 struct dm_integrity_c *ic = ti->private;
2500
2501 if (!ic->meta_dev)
2502 return fn(ti, ic->dev, ic->start + ic->initial_sectors + ic->metadata_run, ti->len, data);
2503 else
2504 return fn(ti, ic->dev, 0, ti->len, data);
2505 }
2506
2507 static void dm_integrity_io_hints(struct dm_target *ti, struct queue_limits *limits)
2508 {
2509 struct dm_integrity_c *ic = ti->private;
2510
2511 if (ic->sectors_per_block > 1) {
2512 limits->logical_block_size = ic->sectors_per_block << SECTOR_SHIFT;
2513 limits->physical_block_size = ic->sectors_per_block << SECTOR_SHIFT;
2514 blk_limits_io_min(limits, ic->sectors_per_block << SECTOR_SHIFT);
2515 }
2516 }
2517
2518 static void calculate_journal_section_size(struct dm_integrity_c *ic)
2519 {
2520 unsigned sector_space = JOURNAL_SECTOR_DATA;
2521
2522 ic->journal_sections = le32_to_cpu(ic->sb->journal_sections);
2523 ic->journal_entry_size = roundup(offsetof(struct journal_entry, last_bytes[ic->sectors_per_block]) + ic->tag_size,
2524 JOURNAL_ENTRY_ROUNDUP);
2525
2526 if (ic->sb->flags & cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC))
2527 sector_space -= JOURNAL_MAC_PER_SECTOR;
2528 ic->journal_entries_per_sector = sector_space / ic->journal_entry_size;
2529 ic->journal_section_entries = ic->journal_entries_per_sector * JOURNAL_BLOCK_SECTORS;
2530 ic->journal_section_sectors = (ic->journal_section_entries << ic->sb->log2_sectors_per_block) + JOURNAL_BLOCK_SECTORS;
2531 ic->journal_entries = ic->journal_section_entries * ic->journal_sections;
2532 }
2533
2534 static int calculate_device_limits(struct dm_integrity_c *ic)
2535 {
2536 __u64 initial_sectors;
2537
2538 calculate_journal_section_size(ic);
2539 initial_sectors = SB_SECTORS + (__u64)ic->journal_section_sectors * ic->journal_sections;
2540 if (initial_sectors + METADATA_PADDING_SECTORS >= ic->meta_device_sectors || initial_sectors > UINT_MAX)
2541 return -EINVAL;
2542 ic->initial_sectors = initial_sectors;
2543
2544 if (!ic->meta_dev) {
2545 sector_t last_sector, last_area, last_offset;
2546
2547 ic->metadata_run = roundup((__u64)ic->tag_size << (ic->sb->log2_interleave_sectors - ic->sb->log2_sectors_per_block),
2548 (__u64)(1 << SECTOR_SHIFT << METADATA_PADDING_SECTORS)) >> SECTOR_SHIFT;
2549 if (!(ic->metadata_run & (ic->metadata_run - 1)))
2550 ic->log2_metadata_run = __ffs(ic->metadata_run);
2551 else
2552 ic->log2_metadata_run = -1;
2553
2554 get_area_and_offset(ic, ic->provided_data_sectors - 1, &last_area, &last_offset);
2555 last_sector = get_data_sector(ic, last_area, last_offset);
2556 if (last_sector < ic->start || last_sector >= ic->meta_device_sectors)
2557 return -EINVAL;
2558 } else {
2559 __u64 meta_size = (ic->provided_data_sectors >> ic->sb->log2_sectors_per_block) * ic->tag_size;
2560 meta_size = (meta_size + ((1U << (ic->log2_buffer_sectors + SECTOR_SHIFT)) - 1))
2561 >> (ic->log2_buffer_sectors + SECTOR_SHIFT);
2562 meta_size <<= ic->log2_buffer_sectors;
2563 if (ic->initial_sectors + meta_size < ic->initial_sectors ||
2564 ic->initial_sectors + meta_size > ic->meta_device_sectors)
2565 return -EINVAL;
2566 ic->metadata_run = 1;
2567 ic->log2_metadata_run = 0;
2568 }
2569
2570 return 0;
2571 }
2572
2573 static int initialize_superblock(struct dm_integrity_c *ic, unsigned journal_sectors, unsigned interleave_sectors)
2574 {
2575 unsigned journal_sections;
2576 int test_bit;
2577
2578 memset(ic->sb, 0, SB_SECTORS << SECTOR_SHIFT);
2579 memcpy(ic->sb->magic, SB_MAGIC, 8);
2580 ic->sb->integrity_tag_size = cpu_to_le16(ic->tag_size);
2581 ic->sb->log2_sectors_per_block = __ffs(ic->sectors_per_block);
2582 if (ic->journal_mac_alg.alg_string)
2583 ic->sb->flags |= cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC);
2584
2585 calculate_journal_section_size(ic);
2586 journal_sections = journal_sectors / ic->journal_section_sectors;
2587 if (!journal_sections)
2588 journal_sections = 1;
2589
2590 if (!ic->meta_dev) {
2591 ic->sb->journal_sections = cpu_to_le32(journal_sections);
2592 if (!interleave_sectors)
2593 interleave_sectors = DEFAULT_INTERLEAVE_SECTORS;
2594 ic->sb->log2_interleave_sectors = __fls(interleave_sectors);
2595 ic->sb->log2_interleave_sectors = max((__u8)MIN_LOG2_INTERLEAVE_SECTORS, ic->sb->log2_interleave_sectors);
2596 ic->sb->log2_interleave_sectors = min((__u8)MAX_LOG2_INTERLEAVE_SECTORS, ic->sb->log2_interleave_sectors);
2597
2598 ic->provided_data_sectors = 0;
2599 for (test_bit = fls64(ic->meta_device_sectors) - 1; test_bit >= 3; test_bit--) {
2600 __u64 prev_data_sectors = ic->provided_data_sectors;
2601
2602 ic->provided_data_sectors |= (sector_t)1 << test_bit;
2603 if (calculate_device_limits(ic))
2604 ic->provided_data_sectors = prev_data_sectors;
2605 }
2606 if (!ic->provided_data_sectors)
2607 return -EINVAL;
2608 } else {
2609 ic->sb->log2_interleave_sectors = 0;
2610 ic->provided_data_sectors = ic->data_device_sectors;
2611 ic->provided_data_sectors &= ~(sector_t)(ic->sectors_per_block - 1);
2612
2613 try_smaller_buffer:
2614 ic->sb->journal_sections = cpu_to_le32(0);
2615 for (test_bit = fls(journal_sections) - 1; test_bit >= 0; test_bit--) {
2616 __u32 prev_journal_sections = le32_to_cpu(ic->sb->journal_sections);
2617 __u32 test_journal_sections = prev_journal_sections | (1U << test_bit);
2618 if (test_journal_sections > journal_sections)
2619 continue;
2620 ic->sb->journal_sections = cpu_to_le32(test_journal_sections);
2621 if (calculate_device_limits(ic))
2622 ic->sb->journal_sections = cpu_to_le32(prev_journal_sections);
2623
2624 }
2625 if (!le32_to_cpu(ic->sb->journal_sections)) {
2626 if (ic->log2_buffer_sectors > 3) {
2627 ic->log2_buffer_sectors--;
2628 goto try_smaller_buffer;
2629 }
2630 return -EINVAL;
2631 }
2632 }
2633
2634 ic->sb->provided_data_sectors = cpu_to_le64(ic->provided_data_sectors);
2635
2636 sb_set_version(ic);
2637
2638 return 0;
2639 }
2640
2641 static void dm_integrity_set(struct dm_target *ti, struct dm_integrity_c *ic)
2642 {
2643 struct gendisk *disk = dm_disk(dm_table_get_md(ti->table));
2644 struct blk_integrity bi;
2645
2646 memset(&bi, 0, sizeof(bi));
2647 bi.profile = &dm_integrity_profile;
2648 bi.tuple_size = ic->tag_size;
2649 bi.tag_size = bi.tuple_size;
2650 bi.interval_exp = ic->sb->log2_sectors_per_block + SECTOR_SHIFT;
2651
2652 blk_integrity_register(disk, &bi);
2653 blk_queue_max_integrity_segments(disk->queue, UINT_MAX);
2654 }
2655
2656 static void dm_integrity_free_page_list(struct dm_integrity_c *ic, struct page_list *pl)
2657 {
2658 unsigned i;
2659
2660 if (!pl)
2661 return;
2662 for (i = 0; i < ic->journal_pages; i++)
2663 if (pl[i].page)
2664 __free_page(pl[i].page);
2665 kvfree(pl);
2666 }
2667
2668 static struct page_list *dm_integrity_alloc_page_list(struct dm_integrity_c *ic)
2669 {
2670 size_t page_list_desc_size = ic->journal_pages * sizeof(struct page_list);
2671 struct page_list *pl;
2672 unsigned i;
2673
2674 pl = kvmalloc(page_list_desc_size, GFP_KERNEL | __GFP_ZERO);
2675 if (!pl)
2676 return NULL;
2677
2678 for (i = 0; i < ic->journal_pages; i++) {
2679 pl[i].page = alloc_page(GFP_KERNEL);
2680 if (!pl[i].page) {
2681 dm_integrity_free_page_list(ic, pl);
2682 return NULL;
2683 }
2684 if (i)
2685 pl[i - 1].next = &pl[i];
2686 }
2687
2688 return pl;
2689 }
2690
2691 static void dm_integrity_free_journal_scatterlist(struct dm_integrity_c *ic, struct scatterlist **sl)
2692 {
2693 unsigned i;
2694 for (i = 0; i < ic->journal_sections; i++)
2695 kvfree(sl[i]);
2696 kvfree(sl);
2697 }
2698
2699 static struct scatterlist **dm_integrity_alloc_journal_scatterlist(struct dm_integrity_c *ic, struct page_list *pl)
2700 {
2701 struct scatterlist **sl;
2702 unsigned i;
2703
2704 sl = kvmalloc_array(ic->journal_sections,
2705 sizeof(struct scatterlist *),
2706 GFP_KERNEL | __GFP_ZERO);
2707 if (!sl)
2708 return NULL;
2709
2710 for (i = 0; i < ic->journal_sections; i++) {
2711 struct scatterlist *s;
2712 unsigned start_index, start_offset;
2713 unsigned end_index, end_offset;
2714 unsigned n_pages;
2715 unsigned idx;
2716
2717 page_list_location(ic, i, 0, &start_index, &start_offset);
2718 page_list_location(ic, i, ic->journal_section_sectors - 1, &end_index, &end_offset);
2719
2720 n_pages = (end_index - start_index + 1);
2721
2722 s = kvmalloc_array(n_pages, sizeof(struct scatterlist),
2723 GFP_KERNEL);
2724 if (!s) {
2725 dm_integrity_free_journal_scatterlist(ic, sl);
2726 return NULL;
2727 }
2728
2729 sg_init_table(s, n_pages);
2730 for (idx = start_index; idx <= end_index; idx++) {
2731 char *va = lowmem_page_address(pl[idx].page);
2732 unsigned start = 0, end = PAGE_SIZE;
2733 if (idx == start_index)
2734 start = start_offset;
2735 if (idx == end_index)
2736 end = end_offset + (1 << SECTOR_SHIFT);
2737 sg_set_buf(&s[idx - start_index], va + start, end - start);
2738 }
2739
2740 sl[i] = s;
2741 }
2742
2743 return sl;
2744 }
2745
2746 static void free_alg(struct alg_spec *a)
2747 {
2748 kzfree(a->alg_string);
2749 kzfree(a->key);
2750 memset(a, 0, sizeof *a);
2751 }
2752
2753 static int get_alg_and_key(const char *arg, struct alg_spec *a, char **error, char *error_inval)
2754 {
2755 char *k;
2756
2757 free_alg(a);
2758
2759 a->alg_string = kstrdup(strchr(arg, ':') + 1, GFP_KERNEL);
2760 if (!a->alg_string)
2761 goto nomem;
2762
2763 k = strchr(a->alg_string, ':');
2764 if (k) {
2765 *k = 0;
2766 a->key_string = k + 1;
2767 if (strlen(a->key_string) & 1)
2768 goto inval;
2769
2770 a->key_size = strlen(a->key_string) / 2;
2771 a->key = kmalloc(a->key_size, GFP_KERNEL);
2772 if (!a->key)
2773 goto nomem;
2774 if (hex2bin(a->key, a->key_string, a->key_size))
2775 goto inval;
2776 }
2777
2778 return 0;
2779 inval:
2780 *error = error_inval;
2781 return -EINVAL;
2782 nomem:
2783 *error = "Out of memory for an argument";
2784 return -ENOMEM;
2785 }
2786
2787 static int get_mac(struct crypto_shash **hash, struct alg_spec *a, char **error,
2788 char *error_alg, char *error_key)
2789 {
2790 int r;
2791
2792 if (a->alg_string) {
2793 *hash = crypto_alloc_shash(a->alg_string, 0, CRYPTO_ALG_ASYNC);
2794 if (IS_ERR(*hash)) {
2795 *error = error_alg;
2796 r = PTR_ERR(*hash);
2797 *hash = NULL;
2798 return r;
2799 }
2800
2801 if (a->key) {
2802 r = crypto_shash_setkey(*hash, a->key, a->key_size);
2803 if (r) {
2804 *error = error_key;
2805 return r;
2806 }
2807 } else if (crypto_shash_get_flags(*hash) & CRYPTO_TFM_NEED_KEY) {
2808 *error = error_key;
2809 return -ENOKEY;
2810 }
2811 }
2812
2813 return 0;
2814 }
2815
2816 static int create_journal(struct dm_integrity_c *ic, char **error)
2817 {
2818 int r = 0;
2819 unsigned i;
2820 __u64 journal_pages, journal_desc_size, journal_tree_size;
2821 unsigned char *crypt_data = NULL, *crypt_iv = NULL;
2822 struct skcipher_request *req = NULL;
2823
2824 ic->commit_ids[0] = cpu_to_le64(0x1111111111111111ULL);
2825 ic->commit_ids[1] = cpu_to_le64(0x2222222222222222ULL);
2826 ic->commit_ids[2] = cpu_to_le64(0x3333333333333333ULL);
2827 ic->commit_ids[3] = cpu_to_le64(0x4444444444444444ULL);
2828
2829 journal_pages = roundup((__u64)ic->journal_sections * ic->journal_section_sectors,
2830 PAGE_SIZE >> SECTOR_SHIFT) >> (PAGE_SHIFT - SECTOR_SHIFT);
2831 journal_desc_size = journal_pages * sizeof(struct page_list);
2832 if (journal_pages >= totalram_pages - totalhigh_pages || journal_desc_size > ULONG_MAX) {
2833 *error = "Journal doesn't fit into memory";
2834 r = -ENOMEM;
2835 goto bad;
2836 }
2837 ic->journal_pages = journal_pages;
2838
2839 ic->journal = dm_integrity_alloc_page_list(ic);
2840 if (!ic->journal) {
2841 *error = "Could not allocate memory for journal";
2842 r = -ENOMEM;
2843 goto bad;
2844 }
2845 if (ic->journal_crypt_alg.alg_string) {
2846 unsigned ivsize, blocksize;
2847 struct journal_completion comp;
2848
2849 comp.ic = ic;
2850 ic->journal_crypt = crypto_alloc_skcipher(ic->journal_crypt_alg.alg_string, 0, 0);
2851 if (IS_ERR(ic->journal_crypt)) {
2852 *error = "Invalid journal cipher";
2853 r = PTR_ERR(ic->journal_crypt);
2854 ic->journal_crypt = NULL;
2855 goto bad;
2856 }
2857 ivsize = crypto_skcipher_ivsize(ic->journal_crypt);
2858 blocksize = crypto_skcipher_blocksize(ic->journal_crypt);
2859
2860 if (ic->journal_crypt_alg.key) {
2861 r = crypto_skcipher_setkey(ic->journal_crypt, ic->journal_crypt_alg.key,
2862 ic->journal_crypt_alg.key_size);
2863 if (r) {
2864 *error = "Error setting encryption key";
2865 goto bad;
2866 }
2867 }
2868 DEBUG_print("cipher %s, block size %u iv size %u\n",
2869 ic->journal_crypt_alg.alg_string, blocksize, ivsize);
2870
2871 ic->journal_io = dm_integrity_alloc_page_list(ic);
2872 if (!ic->journal_io) {
2873 *error = "Could not allocate memory for journal io";
2874 r = -ENOMEM;
2875 goto bad;
2876 }
2877
2878 if (blocksize == 1) {
2879 struct scatterlist *sg;
2880
2881 req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
2882 if (!req) {
2883 *error = "Could not allocate crypt request";
2884 r = -ENOMEM;
2885 goto bad;
2886 }
2887
2888 crypt_iv = kmalloc(ivsize, GFP_KERNEL);
2889 if (!crypt_iv) {
2890 *error = "Could not allocate iv";
2891 r = -ENOMEM;
2892 goto bad;
2893 }
2894
2895 ic->journal_xor = dm_integrity_alloc_page_list(ic);
2896 if (!ic->journal_xor) {
2897 *error = "Could not allocate memory for journal xor";
2898 r = -ENOMEM;
2899 goto bad;
2900 }
2901
2902 sg = kvmalloc_array(ic->journal_pages + 1,
2903 sizeof(struct scatterlist),
2904 GFP_KERNEL);
2905 if (!sg) {
2906 *error = "Unable to allocate sg list";
2907 r = -ENOMEM;
2908 goto bad;
2909 }
2910 sg_init_table(sg, ic->journal_pages + 1);
2911 for (i = 0; i < ic->journal_pages; i++) {
2912 char *va = lowmem_page_address(ic->journal_xor[i].page);
2913 clear_page(va);
2914 sg_set_buf(&sg[i], va, PAGE_SIZE);
2915 }
2916 sg_set_buf(&sg[i], &ic->commit_ids, sizeof ic->commit_ids);
2917 memset(crypt_iv, 0x00, ivsize);
2918
2919 skcipher_request_set_crypt(req, sg, sg, PAGE_SIZE * ic->journal_pages + sizeof ic->commit_ids, crypt_iv);
2920 init_completion(&comp.comp);
2921 comp.in_flight = (atomic_t)ATOMIC_INIT(1);
2922 if (do_crypt(true, req, &comp))
2923 wait_for_completion(&comp.comp);
2924 kvfree(sg);
2925 r = dm_integrity_failed(ic);
2926 if (r) {
2927 *error = "Unable to encrypt journal";
2928 goto bad;
2929 }
2930 DEBUG_bytes(lowmem_page_address(ic->journal_xor[0].page), 64, "xor data");
2931
2932 crypto_free_skcipher(ic->journal_crypt);
2933 ic->journal_crypt = NULL;
2934 } else {
2935 unsigned crypt_len = roundup(ivsize, blocksize);
2936
2937 req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
2938 if (!req) {
2939 *error = "Could not allocate crypt request";
2940 r = -ENOMEM;
2941 goto bad;
2942 }
2943
2944 crypt_iv = kmalloc(ivsize, GFP_KERNEL);
2945 if (!crypt_iv) {
2946 *error = "Could not allocate iv";
2947 r = -ENOMEM;
2948 goto bad;
2949 }
2950
2951 crypt_data = kmalloc(crypt_len, GFP_KERNEL);
2952 if (!crypt_data) {
2953 *error = "Unable to allocate crypt data";
2954 r = -ENOMEM;
2955 goto bad;
2956 }
2957
2958 ic->journal_scatterlist = dm_integrity_alloc_journal_scatterlist(ic, ic->journal);
2959 if (!ic->journal_scatterlist) {
2960 *error = "Unable to allocate sg list";
2961 r = -ENOMEM;
2962 goto bad;
2963 }
2964 ic->journal_io_scatterlist = dm_integrity_alloc_journal_scatterlist(ic, ic->journal_io);
2965 if (!ic->journal_io_scatterlist) {
2966 *error = "Unable to allocate sg list";
2967 r = -ENOMEM;
2968 goto bad;
2969 }
2970 ic->sk_requests = kvmalloc_array(ic->journal_sections,
2971 sizeof(struct skcipher_request *),
2972 GFP_KERNEL | __GFP_ZERO);
2973 if (!ic->sk_requests) {
2974 *error = "Unable to allocate sk requests";
2975 r = -ENOMEM;
2976 goto bad;
2977 }
2978 for (i = 0; i < ic->journal_sections; i++) {
2979 struct scatterlist sg;
2980 struct skcipher_request *section_req;
2981 __u32 section_le = cpu_to_le32(i);
2982
2983 memset(crypt_iv, 0x00, ivsize);
2984 memset(crypt_data, 0x00, crypt_len);
2985 memcpy(crypt_data, &section_le, min((size_t)crypt_len, sizeof(section_le)));
2986
2987 sg_init_one(&sg, crypt_data, crypt_len);
2988 skcipher_request_set_crypt(req, &sg, &sg, crypt_len, crypt_iv);
2989 init_completion(&comp.comp);
2990 comp.in_flight = (atomic_t)ATOMIC_INIT(1);
2991 if (do_crypt(true, req, &comp))
2992 wait_for_completion(&comp.comp);
2993
2994 r = dm_integrity_failed(ic);
2995 if (r) {
2996 *error = "Unable to generate iv";
2997 goto bad;
2998 }
2999
3000 section_req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
3001 if (!section_req) {
3002 *error = "Unable to allocate crypt request";
3003 r = -ENOMEM;
3004 goto bad;
3005 }
3006 section_req->iv = kmalloc_array(ivsize, 2,
3007 GFP_KERNEL);
3008 if (!section_req->iv) {
3009 skcipher_request_free(section_req);
3010 *error = "Unable to allocate iv";
3011 r = -ENOMEM;
3012 goto bad;
3013 }
3014 memcpy(section_req->iv + ivsize, crypt_data, ivsize);
3015 section_req->cryptlen = (size_t)ic->journal_section_sectors << SECTOR_SHIFT;
3016 ic->sk_requests[i] = section_req;
3017 DEBUG_bytes(crypt_data, ivsize, "iv(%u)", i);
3018 }
3019 }
3020 }
3021
3022 for (i = 0; i < N_COMMIT_IDS; i++) {
3023 unsigned j;
3024 retest_commit_id:
3025 for (j = 0; j < i; j++) {
3026 if (ic->commit_ids[j] == ic->commit_ids[i]) {
3027 ic->commit_ids[i] = cpu_to_le64(le64_to_cpu(ic->commit_ids[i]) + 1);
3028 goto retest_commit_id;
3029 }
3030 }
3031 DEBUG_print("commit id %u: %016llx\n", i, ic->commit_ids[i]);
3032 }
3033
3034 journal_tree_size = (__u64)ic->journal_entries * sizeof(struct journal_node);
3035 if (journal_tree_size > ULONG_MAX) {
3036 *error = "Journal doesn't fit into memory";
3037 r = -ENOMEM;
3038 goto bad;
3039 }
3040 ic->journal_tree = kvmalloc(journal_tree_size, GFP_KERNEL);
3041 if (!ic->journal_tree) {
3042 *error = "Could not allocate memory for journal tree";
3043 r = -ENOMEM;
3044 }
3045 bad:
3046 kfree(crypt_data);
3047 kfree(crypt_iv);
3048 skcipher_request_free(req);
3049
3050 return r;
3051 }
3052
3053 /*
3054 * Construct a integrity mapping
3055 *
3056 * Arguments:
3057 * device
3058 * offset from the start of the device
3059 * tag size
3060 * D - direct writes, J - journal writes, R - recovery mode
3061 * number of optional arguments
3062 * optional arguments:
3063 * journal_sectors
3064 * interleave_sectors
3065 * buffer_sectors
3066 * journal_watermark
3067 * commit_time
3068 * internal_hash
3069 * journal_crypt
3070 * journal_mac
3071 * block_size
3072 */
3073 static int dm_integrity_ctr(struct dm_target *ti, unsigned argc, char **argv)
3074 {
3075 struct dm_integrity_c *ic;
3076 char dummy;
3077 int r;
3078 unsigned extra_args;
3079 struct dm_arg_set as;
3080 static const struct dm_arg _args[] = {
3081 {0, 9, "Invalid number of feature args"},
3082 };
3083 unsigned journal_sectors, interleave_sectors, buffer_sectors, journal_watermark, sync_msec;
3084 bool recalculate;
3085 bool should_write_sb;
3086 __u64 threshold;
3087 unsigned long long start;
3088
3089 #define DIRECT_ARGUMENTS 4
3090
3091 if (argc <= DIRECT_ARGUMENTS) {
3092 ti->error = "Invalid argument count";
3093 return -EINVAL;
3094 }
3095
3096 ic = kzalloc(sizeof(struct dm_integrity_c), GFP_KERNEL);
3097 if (!ic) {
3098 ti->error = "Cannot allocate integrity context";
3099 return -ENOMEM;
3100 }
3101 ti->private = ic;
3102 ti->per_io_data_size = sizeof(struct dm_integrity_io);
3103 ic->ti = ti;
3104
3105 ic->in_progress = RB_ROOT;
3106 INIT_LIST_HEAD(&ic->wait_list);
3107 init_waitqueue_head(&ic->endio_wait);
3108 bio_list_init(&ic->flush_bio_list);
3109 init_waitqueue_head(&ic->copy_to_journal_wait);
3110 init_completion(&ic->crypto_backoff);
3111 atomic64_set(&ic->number_of_mismatches, 0);
3112
3113 r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &ic->dev);
3114 if (r) {
3115 ti->error = "Device lookup failed";
3116 goto bad;
3117 }
3118
3119 if (sscanf(argv[1], "%llu%c", &start, &dummy) != 1 || start != (sector_t)start) {
3120 ti->error = "Invalid starting offset";
3121 r = -EINVAL;
3122 goto bad;
3123 }
3124 ic->start = start;
3125
3126 if (strcmp(argv[2], "-")) {
3127 if (sscanf(argv[2], "%u%c", &ic->tag_size, &dummy) != 1 || !ic->tag_size) {
3128 ti->error = "Invalid tag size";
3129 r = -EINVAL;
3130 goto bad;
3131 }
3132 }
3133
3134 if (!strcmp(argv[3], "J") || !strcmp(argv[3], "D") || !strcmp(argv[3], "R"))
3135 ic->mode = argv[3][0];
3136 else {
3137 ti->error = "Invalid mode (expecting J, D, R)";
3138 r = -EINVAL;
3139 goto bad;
3140 }
3141
3142 journal_sectors = 0;
3143 interleave_sectors = DEFAULT_INTERLEAVE_SECTORS;
3144 buffer_sectors = DEFAULT_BUFFER_SECTORS;
3145 journal_watermark = DEFAULT_JOURNAL_WATERMARK;
3146 sync_msec = DEFAULT_SYNC_MSEC;
3147 recalculate = false;
3148 ic->sectors_per_block = 1;
3149
3150 as.argc = argc - DIRECT_ARGUMENTS;
3151 as.argv = argv + DIRECT_ARGUMENTS;
3152 r = dm_read_arg_group(_args, &as, &extra_args, &ti->error);
3153 if (r)
3154 goto bad;
3155
3156 while (extra_args--) {
3157 const char *opt_string;
3158 unsigned val;
3159 opt_string = dm_shift_arg(&as);
3160 if (!opt_string) {
3161 r = -EINVAL;
3162 ti->error = "Not enough feature arguments";
3163 goto bad;
3164 }
3165 if (sscanf(opt_string, "journal_sectors:%u%c", &val, &dummy) == 1)
3166 journal_sectors = val ? val : 1;
3167 else if (sscanf(opt_string, "interleave_sectors:%u%c", &val, &dummy) == 1)
3168 interleave_sectors = val;
3169 else if (sscanf(opt_string, "buffer_sectors:%u%c", &val, &dummy) == 1)
3170 buffer_sectors = val;
3171 else if (sscanf(opt_string, "journal_watermark:%u%c", &val, &dummy) == 1 && val <= 100)
3172 journal_watermark = val;
3173 else if (sscanf(opt_string, "commit_time:%u%c", &val, &dummy) == 1)
3174 sync_msec = val;
3175 else if (!strncmp(opt_string, "meta_device:", strlen("meta_device:"))) {
3176 if (ic->meta_dev) {
3177 dm_put_device(ti, ic->meta_dev);
3178 ic->meta_dev = NULL;
3179 }
3180 r = dm_get_device(ti, strchr(opt_string, ':') + 1, dm_table_get_mode(ti->table), &ic->meta_dev);
3181 if (r) {
3182 ti->error = "Device lookup failed";
3183 goto bad;
3184 }
3185 } else if (sscanf(opt_string, "block_size:%u%c", &val, &dummy) == 1) {
3186 if (val < 1 << SECTOR_SHIFT ||
3187 val > MAX_SECTORS_PER_BLOCK << SECTOR_SHIFT ||
3188 (val & (val -1))) {
3189 r = -EINVAL;
3190 ti->error = "Invalid block_size argument";
3191 goto bad;
3192 }
3193 ic->sectors_per_block = val >> SECTOR_SHIFT;
3194 } else if (!strncmp(opt_string, "internal_hash:", strlen("internal_hash:"))) {
3195 r = get_alg_and_key(opt_string, &ic->internal_hash_alg, &ti->error,
3196 "Invalid internal_hash argument");
3197 if (r)
3198 goto bad;
3199 } else if (!strncmp(opt_string, "journal_crypt:", strlen("journal_crypt:"))) {
3200 r = get_alg_and_key(opt_string, &ic->journal_crypt_alg, &ti->error,
3201 "Invalid journal_crypt argument");
3202 if (r)
3203 goto bad;
3204 } else if (!strncmp(opt_string, "journal_mac:", strlen("journal_mac:"))) {
3205 r = get_alg_and_key(opt_string, &ic->journal_mac_alg, &ti->error,
3206 "Invalid journal_mac argument");
3207 if (r)
3208 goto bad;
3209 } else if (!strcmp(opt_string, "recalculate")) {
3210 recalculate = true;
3211 } else {
3212 r = -EINVAL;
3213 ti->error = "Invalid argument";
3214 goto bad;
3215 }
3216 }
3217
3218 ic->data_device_sectors = i_size_read(ic->dev->bdev->bd_inode) >> SECTOR_SHIFT;
3219 if (!ic->meta_dev)
3220 ic->meta_device_sectors = ic->data_device_sectors;
3221 else
3222 ic->meta_device_sectors = i_size_read(ic->meta_dev->bdev->bd_inode) >> SECTOR_SHIFT;
3223
3224 if (!journal_sectors) {
3225 journal_sectors = min((sector_t)DEFAULT_MAX_JOURNAL_SECTORS,
3226 ic->data_device_sectors >> DEFAULT_JOURNAL_SIZE_FACTOR);
3227 }
3228
3229 if (!buffer_sectors)
3230 buffer_sectors = 1;
3231 ic->log2_buffer_sectors = min((int)__fls(buffer_sectors), 31 - SECTOR_SHIFT);
3232
3233 r = get_mac(&ic->internal_hash, &ic->internal_hash_alg, &ti->error,
3234 "Invalid internal hash", "Error setting internal hash key");
3235 if (r)
3236 goto bad;
3237
3238 r = get_mac(&ic->journal_mac, &ic->journal_mac_alg, &ti->error,
3239 "Invalid journal mac", "Error setting journal mac key");
3240 if (r)
3241 goto bad;
3242
3243 if (!ic->tag_size) {
3244 if (!ic->internal_hash) {
3245 ti->error = "Unknown tag size";
3246 r = -EINVAL;
3247 goto bad;
3248 }
3249 ic->tag_size = crypto_shash_digestsize(ic->internal_hash);
3250 }
3251 if (ic->tag_size > MAX_TAG_SIZE) {
3252 ti->error = "Too big tag size";
3253 r = -EINVAL;
3254 goto bad;
3255 }
3256 if (!(ic->tag_size & (ic->tag_size - 1)))
3257 ic->log2_tag_size = __ffs(ic->tag_size);
3258 else
3259 ic->log2_tag_size = -1;
3260
3261 ic->autocommit_jiffies = msecs_to_jiffies(sync_msec);
3262 ic->autocommit_msec = sync_msec;
3263 timer_setup(&ic->autocommit_timer, autocommit_fn, 0);
3264
3265 ic->io = dm_io_client_create();
3266 if (IS_ERR(ic->io)) {
3267 r = PTR_ERR(ic->io);
3268 ic->io = NULL;
3269 ti->error = "Cannot allocate dm io";
3270 goto bad;
3271 }
3272
3273 r = mempool_init_slab_pool(&ic->journal_io_mempool, JOURNAL_IO_MEMPOOL, journal_io_cache);
3274 if (r) {
3275 ti->error = "Cannot allocate mempool";
3276 goto bad;
3277 }
3278
3279 ic->metadata_wq = alloc_workqueue("dm-integrity-metadata",
3280 WQ_MEM_RECLAIM, METADATA_WORKQUEUE_MAX_ACTIVE);
3281 if (!ic->metadata_wq) {
3282 ti->error = "Cannot allocate workqueue";
3283 r = -ENOMEM;
3284 goto bad;
3285 }
3286
3287 /*
3288 * If this workqueue were percpu, it would cause bio reordering
3289 * and reduced performance.
3290 */
3291 ic->wait_wq = alloc_workqueue("dm-integrity-wait", WQ_MEM_RECLAIM | WQ_UNBOUND, 1);
3292 if (!ic->wait_wq) {
3293 ti->error = "Cannot allocate workqueue";
3294 r = -ENOMEM;
3295 goto bad;
3296 }
3297
3298 ic->offload_wq = alloc_workqueue("dm-integrity-offload", WQ_MEM_RECLAIM,
3299 METADATA_WORKQUEUE_MAX_ACTIVE);
3300 if (!ic->offload_wq) {
3301 ti->error = "Cannot allocate workqueue";
3302 r = -ENOMEM;
3303 goto bad;
3304 }
3305
3306 ic->commit_wq = alloc_workqueue("dm-integrity-commit", WQ_MEM_RECLAIM, 1);
3307 if (!ic->commit_wq) {
3308 ti->error = "Cannot allocate workqueue";
3309 r = -ENOMEM;
3310 goto bad;
3311 }
3312 INIT_WORK(&ic->commit_work, integrity_commit);
3313
3314 if (ic->mode == 'J') {
3315 ic->writer_wq = alloc_workqueue("dm-integrity-writer", WQ_MEM_RECLAIM, 1);
3316 if (!ic->writer_wq) {
3317 ti->error = "Cannot allocate workqueue";
3318 r = -ENOMEM;
3319 goto bad;
3320 }
3321 INIT_WORK(&ic->writer_work, integrity_writer);
3322 }
3323
3324 ic->sb = alloc_pages_exact(SB_SECTORS << SECTOR_SHIFT, GFP_KERNEL);
3325 if (!ic->sb) {
3326 r = -ENOMEM;
3327 ti->error = "Cannot allocate superblock area";
3328 goto bad;
3329 }
3330
3331 r = sync_rw_sb(ic, REQ_OP_READ, 0);
3332 if (r) {
3333 ti->error = "Error reading superblock";
3334 goto bad;
3335 }
3336 should_write_sb = false;
3337 if (memcmp(ic->sb->magic, SB_MAGIC, 8)) {
3338 if (ic->mode != 'R') {
3339 if (memchr_inv(ic->sb, 0, SB_SECTORS << SECTOR_SHIFT)) {
3340 r = -EINVAL;
3341 ti->error = "The device is not initialized";
3342 goto bad;
3343 }
3344 }
3345
3346 r = initialize_superblock(ic, journal_sectors, interleave_sectors);
3347 if (r) {
3348 ti->error = "Could not initialize superblock";
3349 goto bad;
3350 }
3351 if (ic->mode != 'R')
3352 should_write_sb = true;
3353 }
3354
3355 if (!ic->sb->version || ic->sb->version > SB_VERSION_2) {
3356 r = -EINVAL;
3357 ti->error = "Unknown version";
3358 goto bad;
3359 }
3360 if (le16_to_cpu(ic->sb->integrity_tag_size) != ic->tag_size) {
3361 r = -EINVAL;
3362 ti->error = "Tag size doesn't match the information in superblock";
3363 goto bad;
3364 }
3365 if (ic->sb->log2_sectors_per_block != __ffs(ic->sectors_per_block)) {
3366 r = -EINVAL;
3367 ti->error = "Block size doesn't match the information in superblock";
3368 goto bad;
3369 }
3370 if (!le32_to_cpu(ic->sb->journal_sections)) {
3371 r = -EINVAL;
3372 ti->error = "Corrupted superblock, journal_sections is 0";
3373 goto bad;
3374 }
3375 /* make sure that ti->max_io_len doesn't overflow */
3376 if (!ic->meta_dev) {
3377 if (ic->sb->log2_interleave_sectors < MIN_LOG2_INTERLEAVE_SECTORS ||
3378 ic->sb->log2_interleave_sectors > MAX_LOG2_INTERLEAVE_SECTORS) {
3379 r = -EINVAL;
3380 ti->error = "Invalid interleave_sectors in the superblock";
3381 goto bad;
3382 }
3383 } else {
3384 if (ic->sb->log2_interleave_sectors) {
3385 r = -EINVAL;
3386 ti->error = "Invalid interleave_sectors in the superblock";
3387 goto bad;
3388 }
3389 }
3390 ic->provided_data_sectors = le64_to_cpu(ic->sb->provided_data_sectors);
3391 if (ic->provided_data_sectors != le64_to_cpu(ic->sb->provided_data_sectors)) {
3392 /* test for overflow */
3393 r = -EINVAL;
3394 ti->error = "The superblock has 64-bit device size, but the kernel was compiled with 32-bit sectors";
3395 goto bad;
3396 }
3397 if (!!(ic->sb->flags & cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC)) != !!ic->journal_mac_alg.alg_string) {
3398 r = -EINVAL;
3399 ti->error = "Journal mac mismatch";
3400 goto bad;
3401 }
3402
3403 try_smaller_buffer:
3404 r = calculate_device_limits(ic);
3405 if (r) {
3406 if (ic->meta_dev) {
3407 if (ic->log2_buffer_sectors > 3) {
3408 ic->log2_buffer_sectors--;
3409 goto try_smaller_buffer;
3410 }
3411 }
3412 ti->error = "The device is too small";
3413 goto bad;
3414 }
3415 if (!ic->meta_dev)
3416 ic->log2_buffer_sectors = min(ic->log2_buffer_sectors, (__u8)__ffs(ic->metadata_run));
3417
3418 if (ti->len > ic->provided_data_sectors) {
3419 r = -EINVAL;
3420 ti->error = "Not enough provided sectors for requested mapping size";
3421 goto bad;
3422 }
3423
3424
3425 threshold = (__u64)ic->journal_entries * (100 - journal_watermark);
3426 threshold += 50;
3427 do_div(threshold, 100);
3428 ic->free_sectors_threshold = threshold;
3429
3430 DEBUG_print("initialized:\n");
3431 DEBUG_print(" integrity_tag_size %u\n", le16_to_cpu(ic->sb->integrity_tag_size));
3432 DEBUG_print(" journal_entry_size %u\n", ic->journal_entry_size);
3433 DEBUG_print(" journal_entries_per_sector %u\n", ic->journal_entries_per_sector);
3434 DEBUG_print(" journal_section_entries %u\n", ic->journal_section_entries);
3435 DEBUG_print(" journal_section_sectors %u\n", ic->journal_section_sectors);
3436 DEBUG_print(" journal_sections %u\n", (unsigned)le32_to_cpu(ic->sb->journal_sections));
3437 DEBUG_print(" journal_entries %u\n", ic->journal_entries);
3438 DEBUG_print(" log2_interleave_sectors %d\n", ic->sb->log2_interleave_sectors);
3439 DEBUG_print(" data_device_sectors 0x%llx\n", (unsigned long long)ic->data_device_sectors);
3440 DEBUG_print(" initial_sectors 0x%x\n", ic->initial_sectors);
3441 DEBUG_print(" metadata_run 0x%x\n", ic->metadata_run);
3442 DEBUG_print(" log2_metadata_run %d\n", ic->log2_metadata_run);
3443 DEBUG_print(" provided_data_sectors 0x%llx (%llu)\n", (unsigned long long)ic->provided_data_sectors,
3444 (unsigned long long)ic->provided_data_sectors);
3445 DEBUG_print(" log2_buffer_sectors %u\n", ic->log2_buffer_sectors);
3446
3447 if (recalculate && !(ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))) {
3448 ic->sb->flags |= cpu_to_le32(SB_FLAG_RECALCULATING);
3449 ic->sb->recalc_sector = cpu_to_le64(0);
3450 }
3451
3452 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
3453 if (!ic->internal_hash) {
3454 r = -EINVAL;
3455 ti->error = "Recalculate is only valid with internal hash";
3456 goto bad;
3457 }
3458 ic->recalc_wq = alloc_workqueue("dm-intergrity-recalc", WQ_MEM_RECLAIM, 1);
3459 if (!ic->recalc_wq ) {
3460 ti->error = "Cannot allocate workqueue";
3461 r = -ENOMEM;
3462 goto bad;
3463 }
3464 INIT_WORK(&ic->recalc_work, integrity_recalc);
3465 ic->recalc_buffer = vmalloc(RECALC_SECTORS << SECTOR_SHIFT);
3466 if (!ic->recalc_buffer) {
3467 ti->error = "Cannot allocate buffer for recalculating";
3468 r = -ENOMEM;
3469 goto bad;
3470 }
3471 ic->recalc_tags = kvmalloc_array(RECALC_SECTORS >> ic->sb->log2_sectors_per_block,
3472 ic->tag_size, GFP_KERNEL);
3473 if (!ic->recalc_tags) {
3474 ti->error = "Cannot allocate tags for recalculating";
3475 r = -ENOMEM;
3476 goto bad;
3477 }
3478 }
3479
3480 ic->bufio = dm_bufio_client_create(ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev,
3481 1U << (SECTOR_SHIFT + ic->log2_buffer_sectors), 1, 0, NULL, NULL);
3482 if (IS_ERR(ic->bufio)) {
3483 r = PTR_ERR(ic->bufio);
3484 ti->error = "Cannot initialize dm-bufio";
3485 ic->bufio = NULL;
3486 goto bad;
3487 }
3488 dm_bufio_set_sector_offset(ic->bufio, ic->start + ic->initial_sectors);
3489
3490 if (ic->mode != 'R') {
3491 r = create_journal(ic, &ti->error);
3492 if (r)
3493 goto bad;
3494 }
3495
3496 if (should_write_sb) {
3497 int r;
3498
3499 init_journal(ic, 0, ic->journal_sections, 0);
3500 r = dm_integrity_failed(ic);
3501 if (unlikely(r)) {
3502 ti->error = "Error initializing journal";
3503 goto bad;
3504 }
3505 r = sync_rw_sb(ic, REQ_OP_WRITE, REQ_FUA);
3506 if (r) {
3507 ti->error = "Error initializing superblock";
3508 goto bad;
3509 }
3510 ic->just_formatted = true;
3511 }
3512
3513 if (!ic->meta_dev) {
3514 r = dm_set_target_max_io_len(ti, 1U << ic->sb->log2_interleave_sectors);
3515 if (r)
3516 goto bad;
3517 }
3518
3519 if (!ic->internal_hash)
3520 dm_integrity_set(ti, ic);
3521
3522 ti->num_flush_bios = 1;
3523 ti->flush_supported = true;
3524
3525 return 0;
3526 bad:
3527 dm_integrity_dtr(ti);
3528 return r;
3529 }
3530
3531 static void dm_integrity_dtr(struct dm_target *ti)
3532 {
3533 struct dm_integrity_c *ic = ti->private;
3534
3535 BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
3536 BUG_ON(!list_empty(&ic->wait_list));
3537
3538 if (ic->metadata_wq)
3539 destroy_workqueue(ic->metadata_wq);
3540 if (ic->wait_wq)
3541 destroy_workqueue(ic->wait_wq);
3542 if (ic->offload_wq)
3543 destroy_workqueue(ic->offload_wq);
3544 if (ic->commit_wq)
3545 destroy_workqueue(ic->commit_wq);
3546 if (ic->writer_wq)
3547 destroy_workqueue(ic->writer_wq);
3548 if (ic->recalc_wq)
3549 destroy_workqueue(ic->recalc_wq);
3550 if (ic->recalc_buffer)
3551 vfree(ic->recalc_buffer);
3552 if (ic->recalc_tags)
3553 kvfree(ic->recalc_tags);
3554 if (ic->bufio)
3555 dm_bufio_client_destroy(ic->bufio);
3556 mempool_exit(&ic->journal_io_mempool);
3557 if (ic->io)
3558 dm_io_client_destroy(ic->io);
3559 if (ic->dev)
3560 dm_put_device(ti, ic->dev);
3561 if (ic->meta_dev)
3562 dm_put_device(ti, ic->meta_dev);
3563 dm_integrity_free_page_list(ic, ic->journal);
3564 dm_integrity_free_page_list(ic, ic->journal_io);
3565 dm_integrity_free_page_list(ic, ic->journal_xor);
3566 if (ic->journal_scatterlist)
3567 dm_integrity_free_journal_scatterlist(ic, ic->journal_scatterlist);
3568 if (ic->journal_io_scatterlist)
3569 dm_integrity_free_journal_scatterlist(ic, ic->journal_io_scatterlist);
3570 if (ic->sk_requests) {
3571 unsigned i;
3572
3573 for (i = 0; i < ic->journal_sections; i++) {
3574 struct skcipher_request *req = ic->sk_requests[i];
3575 if (req) {
3576 kzfree(req->iv);
3577 skcipher_request_free(req);
3578 }
3579 }
3580 kvfree(ic->sk_requests);
3581 }
3582 kvfree(ic->journal_tree);
3583 if (ic->sb)
3584 free_pages_exact(ic->sb, SB_SECTORS << SECTOR_SHIFT);
3585
3586 if (ic->internal_hash)
3587 crypto_free_shash(ic->internal_hash);
3588 free_alg(&ic->internal_hash_alg);
3589
3590 if (ic->journal_crypt)
3591 crypto_free_skcipher(ic->journal_crypt);
3592 free_alg(&ic->journal_crypt_alg);
3593
3594 if (ic->journal_mac)
3595 crypto_free_shash(ic->journal_mac);
3596 free_alg(&ic->journal_mac_alg);
3597
3598 kfree(ic);
3599 }
3600
3601 static struct target_type integrity_target = {
3602 .name = "integrity",
3603 .version = {1, 2, 0},
3604 .module = THIS_MODULE,
3605 .features = DM_TARGET_SINGLETON | DM_TARGET_INTEGRITY,
3606 .ctr = dm_integrity_ctr,
3607 .dtr = dm_integrity_dtr,
3608 .map = dm_integrity_map,
3609 .postsuspend = dm_integrity_postsuspend,
3610 .resume = dm_integrity_resume,
3611 .status = dm_integrity_status,
3612 .iterate_devices = dm_integrity_iterate_devices,
3613 .io_hints = dm_integrity_io_hints,
3614 };
3615
3616 int __init dm_integrity_init(void)
3617 {
3618 int r;
3619
3620 journal_io_cache = kmem_cache_create("integrity_journal_io",
3621 sizeof(struct journal_io), 0, 0, NULL);
3622 if (!journal_io_cache) {
3623 DMERR("can't allocate journal io cache");
3624 return -ENOMEM;
3625 }
3626
3627 r = dm_register_target(&integrity_target);
3628
3629 if (r < 0)
3630 DMERR("register failed %d", r);
3631
3632 return r;
3633 }
3634
3635 void dm_integrity_exit(void)
3636 {
3637 dm_unregister_target(&integrity_target);
3638 kmem_cache_destroy(journal_io_cache);
3639 }
3640
3641 module_init(dm_integrity_init);
3642 module_exit(dm_integrity_exit);
3643
3644 MODULE_AUTHOR("Milan Broz");
3645 MODULE_AUTHOR("Mikulas Patocka");
3646 MODULE_DESCRIPTION(DM_NAME " target for integrity tags extension");
3647 MODULE_LICENSE("GPL");
Linux with added FPC-III support