dm integrity: fix deadlock with overlapping I/O
[muen/linux.git] / drivers / md / dm-integrity.c
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 <linux/compiler.h>
10 #include <linux/module.h>
11 #include <linux/device-mapper.h>
12 #include <linux/dm-io.h>
13 #include <linux/vmalloc.h>
14 #include <linux/sort.h>
15 #include <linux/rbtree.h>
16 #include <linux/delay.h>
17 #include <linux/random.h>
18 #include <crypto/hash.h>
19 #include <crypto/skcipher.h>
20 #include <linux/async_tx.h>
21 #include <linux/dm-bufio.h>
22
23 #define DM_MSG_PREFIX "integrity"
24
25 #define DEFAULT_INTERLEAVE_SECTORS      32768
26 #define DEFAULT_JOURNAL_SIZE_FACTOR     7
27 #define DEFAULT_BUFFER_SECTORS          128
28 #define DEFAULT_JOURNAL_WATERMARK       50
29 #define DEFAULT_SYNC_MSEC               10000
30 #define DEFAULT_MAX_JOURNAL_SECTORS     131072
31 #define MIN_LOG2_INTERLEAVE_SECTORS     3
32 #define MAX_LOG2_INTERLEAVE_SECTORS     31
33 #define METADATA_WORKQUEUE_MAX_ACTIVE   16
34 #define RECALC_SECTORS                  8192
35 #define RECALC_WRITE_SUPER              16
36
37 /*
38  * Warning - DEBUG_PRINT prints security-sensitive data to the log,
39  * so it should not be enabled in the official kernel
40  */
41 //#define DEBUG_PRINT
42 //#define INTERNAL_VERIFY
43
44 /*
45  * On disk structures
46  */
47
48 #define SB_MAGIC                        "integrt"
49 #define SB_VERSION_1                    1
50 #define SB_VERSION_2                    2
51 #define SB_SECTORS                      8
52 #define MAX_SECTORS_PER_BLOCK           8
53
54 struct superblock {
55         __u8 magic[8];
56         __u8 version;
57         __u8 log2_interleave_sectors;
58         __u16 integrity_tag_size;
59         __u32 journal_sections;
60         __u64 provided_data_sectors;    /* userspace uses this value */
61         __u32 flags;
62         __u8 log2_sectors_per_block;
63         __u8 pad[3];
64         __u64 recalc_sector;
65 };
66
67 #define SB_FLAG_HAVE_JOURNAL_MAC        0x1
68 #define SB_FLAG_RECALCULATING           0x2
69
70 #define JOURNAL_ENTRY_ROUNDUP           8
71
72 typedef __u64 commit_id_t;
73 #define JOURNAL_MAC_PER_SECTOR          8
74
75 struct journal_entry {
76         union {
77                 struct {
78                         __u32 sector_lo;
79                         __u32 sector_hi;
80                 } s;
81                 __u64 sector;
82         } u;
83         commit_id_t last_bytes[0];
84         /* __u8 tag[0]; */
85 };
86
87 #define journal_entry_tag(ic, je)               ((__u8 *)&(je)->last_bytes[(ic)->sectors_per_block])
88
89 #if BITS_PER_LONG == 64
90 #define journal_entry_set_sector(je, x)         do { smp_wmb(); WRITE_ONCE((je)->u.sector, cpu_to_le64(x)); } while (0)
91 #define journal_entry_get_sector(je)            le64_to_cpu((je)->u.sector)
92 #elif defined(CONFIG_LBDAF)
93 #define journal_entry_set_sector(je, x)         do { (je)->u.s.sector_lo = cpu_to_le32(x); smp_wmb(); WRITE_ONCE((je)->u.s.sector_hi, cpu_to_le32((x) >> 32)); } while (0)
94 #define journal_entry_get_sector(je)            le64_to_cpu((je)->u.sector)
95 #else
96 #define journal_entry_set_sector(je, x)         do { (je)->u.s.sector_lo = cpu_to_le32(x); smp_wmb(); WRITE_ONCE((je)->u.s.sector_hi, cpu_to_le32(0)); } while (0)
97 #define journal_entry_get_sector(je)            le32_to_cpu((je)->u.s.sector_lo)
98 #endif
99 #define journal_entry_is_unused(je)             ((je)->u.s.sector_hi == cpu_to_le32(-1))
100 #define journal_entry_set_unused(je)            do { ((je)->u.s.sector_hi = cpu_to_le32(-1)); } while (0)
101 #define journal_entry_is_inprogress(je)         ((je)->u.s.sector_hi == cpu_to_le32(-2))
102 #define journal_entry_set_inprogress(je)        do { ((je)->u.s.sector_hi = cpu_to_le32(-2)); } while (0)
103
104 #define JOURNAL_BLOCK_SECTORS           8
105 #define JOURNAL_SECTOR_DATA             ((1 << SECTOR_SHIFT) - sizeof(commit_id_t))
106 #define JOURNAL_MAC_SIZE                (JOURNAL_MAC_PER_SECTOR * JOURNAL_BLOCK_SECTORS)
107
108 struct journal_sector {
109         __u8 entries[JOURNAL_SECTOR_DATA - JOURNAL_MAC_PER_SECTOR];
110         __u8 mac[JOURNAL_MAC_PER_SECTOR];
111         commit_id_t commit_id;
112 };
113
114 #define MAX_TAG_SIZE                    (JOURNAL_SECTOR_DATA - JOURNAL_MAC_PER_SECTOR - offsetof(struct journal_entry, last_bytes[MAX_SECTORS_PER_BLOCK]))
115
116 #define METADATA_PADDING_SECTORS        8
117
118 #define N_COMMIT_IDS                    4
119
120 static unsigned char prev_commit_seq(unsigned char seq)
121 {
122         return (seq + N_COMMIT_IDS - 1) % N_COMMIT_IDS;
123 }
124
125 static unsigned char next_commit_seq(unsigned char seq)
126 {
127         return (seq + 1) % N_COMMIT_IDS;
128 }
129
130 /*
131  * In-memory structures
132  */
133
134 struct journal_node {
135         struct rb_node node;
136         sector_t sector;
137 };
138
139 struct alg_spec {
140         char *alg_string;
141         char *key_string;
142         __u8 *key;
143         unsigned key_size;
144 };
145
146 struct dm_integrity_c {
147         struct dm_dev *dev;
148         struct dm_dev *meta_dev;
149         unsigned tag_size;
150         __s8 log2_tag_size;
151         sector_t start;
152         mempool_t journal_io_mempool;
153         struct dm_io_client *io;
154         struct dm_bufio_client *bufio;
155         struct workqueue_struct *metadata_wq;
156         struct superblock *sb;
157         unsigned journal_pages;
158         struct page_list *journal;
159         struct page_list *journal_io;
160         struct page_list *journal_xor;
161
162         struct crypto_skcipher *journal_crypt;
163         struct scatterlist **journal_scatterlist;
164         struct scatterlist **journal_io_scatterlist;
165         struct skcipher_request **sk_requests;
166
167         struct crypto_shash *journal_mac;
168
169         struct journal_node *journal_tree;
170         struct rb_root journal_tree_root;
171
172         sector_t provided_data_sectors;
173
174         unsigned short journal_entry_size;
175         unsigned char journal_entries_per_sector;
176         unsigned char journal_section_entries;
177         unsigned short journal_section_sectors;
178         unsigned journal_sections;
179         unsigned journal_entries;
180         sector_t data_device_sectors;
181         sector_t meta_device_sectors;
182         unsigned initial_sectors;
183         unsigned metadata_run;
184         __s8 log2_metadata_run;
185         __u8 log2_buffer_sectors;
186         __u8 sectors_per_block;
187
188         unsigned char mode;
189         int suspending;
190
191         int failed;
192
193         struct crypto_shash *internal_hash;
194
195         /* these variables are locked with endio_wait.lock */
196         struct rb_root in_progress;
197         struct list_head wait_list;
198         wait_queue_head_t endio_wait;
199         struct workqueue_struct *wait_wq;
200
201         unsigned char commit_seq;
202         commit_id_t commit_ids[N_COMMIT_IDS];
203
204         unsigned committed_section;
205         unsigned n_committed_sections;
206
207         unsigned uncommitted_section;
208         unsigned n_uncommitted_sections;
209
210         unsigned free_section;
211         unsigned char free_section_entry;
212         unsigned free_sectors;
213
214         unsigned free_sectors_threshold;
215
216         struct workqueue_struct *commit_wq;
217         struct work_struct commit_work;
218
219         struct workqueue_struct *writer_wq;
220         struct work_struct writer_work;
221
222         struct workqueue_struct *recalc_wq;
223         struct work_struct recalc_work;
224         u8 *recalc_buffer;
225         u8 *recalc_tags;
226
227         struct bio_list flush_bio_list;
228
229         unsigned long autocommit_jiffies;
230         struct timer_list autocommit_timer;
231         unsigned autocommit_msec;
232
233         wait_queue_head_t copy_to_journal_wait;
234
235         struct completion crypto_backoff;
236
237         bool journal_uptodate;
238         bool just_formatted;
239
240         struct alg_spec internal_hash_alg;
241         struct alg_spec journal_crypt_alg;
242         struct alg_spec journal_mac_alg;
243
244         atomic64_t number_of_mismatches;
245 };
246
247 struct dm_integrity_range {
248         sector_t logical_sector;
249         unsigned n_sectors;
250         bool waiting;
251         union {
252                 struct rb_node node;
253                 struct {
254                         struct task_struct *task;
255                         struct list_head wait_entry;
256                 };
257         };
258 };
259
260 struct dm_integrity_io {
261         struct work_struct work;
262
263         struct dm_integrity_c *ic;
264         bool write;
265         bool fua;
266
267         struct dm_integrity_range range;
268
269         sector_t metadata_block;
270         unsigned metadata_offset;
271
272         atomic_t in_flight;
273         blk_status_t bi_status;
274
275         struct completion *completion;
276
277         struct gendisk *orig_bi_disk;
278         u8 orig_bi_partno;
279         bio_end_io_t *orig_bi_end_io;
280         struct bio_integrity_payload *orig_bi_integrity;
281         struct bvec_iter orig_bi_iter;
282 };
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[HASH_MAX_DIGESTSIZE];
563
564                 if (WARN_ON(size > sizeof(digest))) {
565                         dm_integrity_io_error(ic, "digest_size", -EINVAL);
566                         goto err;
567                 }
568                 r = crypto_shash_final(desc, digest);
569                 if (unlikely(r)) {
570                         dm_integrity_io_error(ic, "crypto_shash_final", r);
571                         goto err;
572                 }
573                 memcpy(result, digest, JOURNAL_MAC_SIZE);
574         }
575
576         return;
577 err:
578         memset(result, 0, JOURNAL_MAC_SIZE);
579 }
580
581 static void rw_section_mac(struct dm_integrity_c *ic, unsigned section, bool wr)
582 {
583         __u8 result[JOURNAL_MAC_SIZE];
584         unsigned j;
585
586         if (!ic->journal_mac)
587                 return;
588
589         section_mac(ic, section, result);
590
591         for (j = 0; j < JOURNAL_BLOCK_SECTORS; j++) {
592                 struct journal_sector *js = access_journal(ic, section, j);
593
594                 if (likely(wr))
595                         memcpy(&js->mac, result + (j * JOURNAL_MAC_PER_SECTOR), JOURNAL_MAC_PER_SECTOR);
596                 else {
597                         if (memcmp(&js->mac, result + (j * JOURNAL_MAC_PER_SECTOR), JOURNAL_MAC_PER_SECTOR))
598                                 dm_integrity_io_error(ic, "journal mac", -EILSEQ);
599                 }
600         }
601 }
602
603 static void complete_journal_op(void *context)
604 {
605         struct journal_completion *comp = context;
606         BUG_ON(!atomic_read(&comp->in_flight));
607         if (likely(atomic_dec_and_test(&comp->in_flight)))
608                 complete(&comp->comp);
609 }
610
611 static void xor_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
612                         unsigned n_sections, struct journal_completion *comp)
613 {
614         struct async_submit_ctl submit;
615         size_t n_bytes = (size_t)(n_sections * ic->journal_section_sectors) << SECTOR_SHIFT;
616         unsigned pl_index, pl_offset, section_index;
617         struct page_list *source_pl, *target_pl;
618
619         if (likely(encrypt)) {
620                 source_pl = ic->journal;
621                 target_pl = ic->journal_io;
622         } else {
623                 source_pl = ic->journal_io;
624                 target_pl = ic->journal;
625         }
626
627         page_list_location(ic, section, 0, &pl_index, &pl_offset);
628
629         atomic_add(roundup(pl_offset + n_bytes, PAGE_SIZE) >> PAGE_SHIFT, &comp->in_flight);
630
631         init_async_submit(&submit, ASYNC_TX_XOR_ZERO_DST, NULL, complete_journal_op, comp, NULL);
632
633         section_index = pl_index;
634
635         do {
636                 size_t this_step;
637                 struct page *src_pages[2];
638                 struct page *dst_page;
639
640                 while (unlikely(pl_index == section_index)) {
641                         unsigned dummy;
642                         if (likely(encrypt))
643                                 rw_section_mac(ic, section, true);
644                         section++;
645                         n_sections--;
646                         if (!n_sections)
647                                 break;
648                         page_list_location(ic, section, 0, &section_index, &dummy);
649                 }
650
651                 this_step = min(n_bytes, (size_t)PAGE_SIZE - pl_offset);
652                 dst_page = target_pl[pl_index].page;
653                 src_pages[0] = source_pl[pl_index].page;
654                 src_pages[1] = ic->journal_xor[pl_index].page;
655
656                 async_xor(dst_page, src_pages, pl_offset, 2, this_step, &submit);
657
658                 pl_index++;
659                 pl_offset = 0;
660                 n_bytes -= this_step;
661         } while (n_bytes);
662
663         BUG_ON(n_sections);
664
665         async_tx_issue_pending_all();
666 }
667
668 static void complete_journal_encrypt(struct crypto_async_request *req, int err)
669 {
670         struct journal_completion *comp = req->data;
671         if (unlikely(err)) {
672                 if (likely(err == -EINPROGRESS)) {
673                         complete(&comp->ic->crypto_backoff);
674                         return;
675                 }
676                 dm_integrity_io_error(comp->ic, "asynchronous encrypt", err);
677         }
678         complete_journal_op(comp);
679 }
680
681 static bool do_crypt(bool encrypt, struct skcipher_request *req, struct journal_completion *comp)
682 {
683         int r;
684         skcipher_request_set_callback(req, CRYPTO_TFM_REQ_MAY_BACKLOG,
685                                       complete_journal_encrypt, comp);
686         if (likely(encrypt))
687                 r = crypto_skcipher_encrypt(req);
688         else
689                 r = crypto_skcipher_decrypt(req);
690         if (likely(!r))
691                 return false;
692         if (likely(r == -EINPROGRESS))
693                 return true;
694         if (likely(r == -EBUSY)) {
695                 wait_for_completion(&comp->ic->crypto_backoff);
696                 reinit_completion(&comp->ic->crypto_backoff);
697                 return true;
698         }
699         dm_integrity_io_error(comp->ic, "encrypt", r);
700         return false;
701 }
702
703 static void crypt_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
704                           unsigned n_sections, struct journal_completion *comp)
705 {
706         struct scatterlist **source_sg;
707         struct scatterlist **target_sg;
708
709         atomic_add(2, &comp->in_flight);
710
711         if (likely(encrypt)) {
712                 source_sg = ic->journal_scatterlist;
713                 target_sg = ic->journal_io_scatterlist;
714         } else {
715                 source_sg = ic->journal_io_scatterlist;
716                 target_sg = ic->journal_scatterlist;
717         }
718
719         do {
720                 struct skcipher_request *req;
721                 unsigned ivsize;
722                 char *iv;
723
724                 if (likely(encrypt))
725                         rw_section_mac(ic, section, true);
726
727                 req = ic->sk_requests[section];
728                 ivsize = crypto_skcipher_ivsize(ic->journal_crypt);
729                 iv = req->iv;
730
731                 memcpy(iv, iv + ivsize, ivsize);
732
733                 req->src = source_sg[section];
734                 req->dst = target_sg[section];
735
736                 if (unlikely(do_crypt(encrypt, req, comp)))
737                         atomic_inc(&comp->in_flight);
738
739                 section++;
740                 n_sections--;
741         } while (n_sections);
742
743         atomic_dec(&comp->in_flight);
744         complete_journal_op(comp);
745 }
746
747 static void encrypt_journal(struct dm_integrity_c *ic, bool encrypt, unsigned section,
748                             unsigned n_sections, struct journal_completion *comp)
749 {
750         if (ic->journal_xor)
751                 return xor_journal(ic, encrypt, section, n_sections, comp);
752         else
753                 return crypt_journal(ic, encrypt, section, n_sections, comp);
754 }
755
756 static void complete_journal_io(unsigned long error, void *context)
757 {
758         struct journal_completion *comp = context;
759         if (unlikely(error != 0))
760                 dm_integrity_io_error(comp->ic, "writing journal", -EIO);
761         complete_journal_op(comp);
762 }
763
764 static void rw_journal(struct dm_integrity_c *ic, int op, int op_flags, unsigned section,
765                        unsigned n_sections, struct journal_completion *comp)
766 {
767         struct dm_io_request io_req;
768         struct dm_io_region io_loc;
769         unsigned sector, n_sectors, pl_index, pl_offset;
770         int r;
771
772         if (unlikely(dm_integrity_failed(ic))) {
773                 if (comp)
774                         complete_journal_io(-1UL, comp);
775                 return;
776         }
777
778         sector = section * ic->journal_section_sectors;
779         n_sectors = n_sections * ic->journal_section_sectors;
780
781         pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
782         pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
783
784         io_req.bi_op = op;
785         io_req.bi_op_flags = op_flags;
786         io_req.mem.type = DM_IO_PAGE_LIST;
787         if (ic->journal_io)
788                 io_req.mem.ptr.pl = &ic->journal_io[pl_index];
789         else
790                 io_req.mem.ptr.pl = &ic->journal[pl_index];
791         io_req.mem.offset = pl_offset;
792         if (likely(comp != NULL)) {
793                 io_req.notify.fn = complete_journal_io;
794                 io_req.notify.context = comp;
795         } else {
796                 io_req.notify.fn = NULL;
797         }
798         io_req.client = ic->io;
799         io_loc.bdev = ic->meta_dev ? ic->meta_dev->bdev : ic->dev->bdev;
800         io_loc.sector = ic->start + SB_SECTORS + sector;
801         io_loc.count = n_sectors;
802
803         r = dm_io(&io_req, 1, &io_loc, NULL);
804         if (unlikely(r)) {
805                 dm_integrity_io_error(ic, op == REQ_OP_READ ? "reading journal" : "writing journal", r);
806                 if (comp) {
807                         WARN_ONCE(1, "asynchronous dm_io failed: %d", r);
808                         complete_journal_io(-1UL, comp);
809                 }
810         }
811 }
812
813 static void write_journal(struct dm_integrity_c *ic, unsigned commit_start, unsigned commit_sections)
814 {
815         struct journal_completion io_comp;
816         struct journal_completion crypt_comp_1;
817         struct journal_completion crypt_comp_2;
818         unsigned i;
819
820         io_comp.ic = ic;
821         init_completion(&io_comp.comp);
822
823         if (commit_start + commit_sections <= ic->journal_sections) {
824                 io_comp.in_flight = (atomic_t)ATOMIC_INIT(1);
825                 if (ic->journal_io) {
826                         crypt_comp_1.ic = ic;
827                         init_completion(&crypt_comp_1.comp);
828                         crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
829                         encrypt_journal(ic, true, commit_start, commit_sections, &crypt_comp_1);
830                         wait_for_completion_io(&crypt_comp_1.comp);
831                 } else {
832                         for (i = 0; i < commit_sections; i++)
833                                 rw_section_mac(ic, commit_start + i, true);
834                 }
835                 rw_journal(ic, REQ_OP_WRITE, REQ_FUA | REQ_SYNC, commit_start,
836                            commit_sections, &io_comp);
837         } else {
838                 unsigned to_end;
839                 io_comp.in_flight = (atomic_t)ATOMIC_INIT(2);
840                 to_end = ic->journal_sections - commit_start;
841                 if (ic->journal_io) {
842                         crypt_comp_1.ic = ic;
843                         init_completion(&crypt_comp_1.comp);
844                         crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
845                         encrypt_journal(ic, true, commit_start, to_end, &crypt_comp_1);
846                         if (try_wait_for_completion(&crypt_comp_1.comp)) {
847                                 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
848                                 reinit_completion(&crypt_comp_1.comp);
849                                 crypt_comp_1.in_flight = (atomic_t)ATOMIC_INIT(0);
850                                 encrypt_journal(ic, true, 0, commit_sections - to_end, &crypt_comp_1);
851                                 wait_for_completion_io(&crypt_comp_1.comp);
852                         } else {
853                                 crypt_comp_2.ic = ic;
854                                 init_completion(&crypt_comp_2.comp);
855                                 crypt_comp_2.in_flight = (atomic_t)ATOMIC_INIT(0);
856                                 encrypt_journal(ic, true, 0, commit_sections - to_end, &crypt_comp_2);
857                                 wait_for_completion_io(&crypt_comp_1.comp);
858                                 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
859                                 wait_for_completion_io(&crypt_comp_2.comp);
860                         }
861                 } else {
862                         for (i = 0; i < to_end; i++)
863                                 rw_section_mac(ic, commit_start + i, true);
864                         rw_journal(ic, REQ_OP_WRITE, REQ_FUA, commit_start, to_end, &io_comp);
865                         for (i = 0; i < commit_sections - to_end; i++)
866                                 rw_section_mac(ic, i, true);
867                 }
868                 rw_journal(ic, REQ_OP_WRITE, REQ_FUA, 0, commit_sections - to_end, &io_comp);
869         }
870
871         wait_for_completion_io(&io_comp.comp);
872 }
873
874 static void copy_from_journal(struct dm_integrity_c *ic, unsigned section, unsigned offset,
875                               unsigned n_sectors, sector_t target, io_notify_fn fn, void *data)
876 {
877         struct dm_io_request io_req;
878         struct dm_io_region io_loc;
879         int r;
880         unsigned sector, pl_index, pl_offset;
881
882         BUG_ON((target | n_sectors | offset) & (unsigned)(ic->sectors_per_block - 1));
883
884         if (unlikely(dm_integrity_failed(ic))) {
885                 fn(-1UL, data);
886                 return;
887         }
888
889         sector = section * ic->journal_section_sectors + JOURNAL_BLOCK_SECTORS + offset;
890
891         pl_index = sector >> (PAGE_SHIFT - SECTOR_SHIFT);
892         pl_offset = (sector << SECTOR_SHIFT) & (PAGE_SIZE - 1);
893
894         io_req.bi_op = REQ_OP_WRITE;
895         io_req.bi_op_flags = 0;
896         io_req.mem.type = DM_IO_PAGE_LIST;
897         io_req.mem.ptr.pl = &ic->journal[pl_index];
898         io_req.mem.offset = pl_offset;
899         io_req.notify.fn = fn;
900         io_req.notify.context = data;
901         io_req.client = ic->io;
902         io_loc.bdev = ic->dev->bdev;
903         io_loc.sector = target;
904         io_loc.count = n_sectors;
905
906         r = dm_io(&io_req, 1, &io_loc, NULL);
907         if (unlikely(r)) {
908                 WARN_ONCE(1, "asynchronous dm_io failed: %d", r);
909                 fn(-1UL, data);
910         }
911 }
912
913 static bool ranges_overlap(struct dm_integrity_range *range1, struct dm_integrity_range *range2)
914 {
915         return range1->logical_sector < range2->logical_sector + range2->n_sectors &&
916                range1->logical_sector + range1->n_sectors > range2->logical_sector;
917 }
918
919 static bool add_new_range(struct dm_integrity_c *ic, struct dm_integrity_range *new_range, bool check_waiting)
920 {
921         struct rb_node **n = &ic->in_progress.rb_node;
922         struct rb_node *parent;
923
924         BUG_ON((new_range->logical_sector | new_range->n_sectors) & (unsigned)(ic->sectors_per_block - 1));
925
926         if (likely(check_waiting)) {
927                 struct dm_integrity_range *range;
928                 list_for_each_entry(range, &ic->wait_list, wait_entry) {
929                         if (unlikely(ranges_overlap(range, new_range)))
930                                 return false;
931                 }
932         }
933
934         parent = NULL;
935
936         while (*n) {
937                 struct dm_integrity_range *range = container_of(*n, struct dm_integrity_range, node);
938
939                 parent = *n;
940                 if (new_range->logical_sector + new_range->n_sectors <= range->logical_sector) {
941                         n = &range->node.rb_left;
942                 } else if (new_range->logical_sector >= range->logical_sector + range->n_sectors) {
943                         n = &range->node.rb_right;
944                 } else {
945                         return false;
946                 }
947         }
948
949         rb_link_node(&new_range->node, parent, n);
950         rb_insert_color(&new_range->node, &ic->in_progress);
951
952         return true;
953 }
954
955 static void remove_range_unlocked(struct dm_integrity_c *ic, struct dm_integrity_range *range)
956 {
957         rb_erase(&range->node, &ic->in_progress);
958         while (unlikely(!list_empty(&ic->wait_list))) {
959                 struct dm_integrity_range *last_range =
960                         list_first_entry(&ic->wait_list, struct dm_integrity_range, wait_entry);
961                 struct task_struct *last_range_task;
962                 last_range_task = last_range->task;
963                 list_del(&last_range->wait_entry);
964                 if (!add_new_range(ic, last_range, false)) {
965                         last_range->task = last_range_task;
966                         list_add(&last_range->wait_entry, &ic->wait_list);
967                         break;
968                 }
969                 last_range->waiting = false;
970                 wake_up_process(last_range_task);
971         }
972 }
973
974 static void remove_range(struct dm_integrity_c *ic, struct dm_integrity_range *range)
975 {
976         unsigned long flags;
977
978         spin_lock_irqsave(&ic->endio_wait.lock, flags);
979         remove_range_unlocked(ic, range);
980         spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
981 }
982
983 static void wait_and_add_new_range(struct dm_integrity_c *ic, struct dm_integrity_range *new_range)
984 {
985         new_range->waiting = true;
986         list_add_tail(&new_range->wait_entry, &ic->wait_list);
987         new_range->task = current;
988         do {
989                 __set_current_state(TASK_UNINTERRUPTIBLE);
990                 spin_unlock_irq(&ic->endio_wait.lock);
991                 io_schedule();
992                 spin_lock_irq(&ic->endio_wait.lock);
993         } while (unlikely(new_range->waiting));
994 }
995
996 static void init_journal_node(struct journal_node *node)
997 {
998         RB_CLEAR_NODE(&node->node);
999         node->sector = (sector_t)-1;
1000 }
1001
1002 static void add_journal_node(struct dm_integrity_c *ic, struct journal_node *node, sector_t sector)
1003 {
1004         struct rb_node **link;
1005         struct rb_node *parent;
1006
1007         node->sector = sector;
1008         BUG_ON(!RB_EMPTY_NODE(&node->node));
1009
1010         link = &ic->journal_tree_root.rb_node;
1011         parent = NULL;
1012
1013         while (*link) {
1014                 struct journal_node *j;
1015                 parent = *link;
1016                 j = container_of(parent, struct journal_node, node);
1017                 if (sector < j->sector)
1018                         link = &j->node.rb_left;
1019                 else
1020                         link = &j->node.rb_right;
1021         }
1022
1023         rb_link_node(&node->node, parent, link);
1024         rb_insert_color(&node->node, &ic->journal_tree_root);
1025 }
1026
1027 static void remove_journal_node(struct dm_integrity_c *ic, struct journal_node *node)
1028 {
1029         BUG_ON(RB_EMPTY_NODE(&node->node));
1030         rb_erase(&node->node, &ic->journal_tree_root);
1031         init_journal_node(node);
1032 }
1033
1034 #define NOT_FOUND       (-1U)
1035
1036 static unsigned find_journal_node(struct dm_integrity_c *ic, sector_t sector, sector_t *next_sector)
1037 {
1038         struct rb_node *n = ic->journal_tree_root.rb_node;
1039         unsigned found = NOT_FOUND;
1040         *next_sector = (sector_t)-1;
1041         while (n) {
1042                 struct journal_node *j = container_of(n, struct journal_node, node);
1043                 if (sector == j->sector) {
1044                         found = j - ic->journal_tree;
1045                 }
1046                 if (sector < j->sector) {
1047                         *next_sector = j->sector;
1048                         n = j->node.rb_left;
1049                 } else {
1050                         n = j->node.rb_right;
1051                 }
1052         }
1053
1054         return found;
1055 }
1056
1057 static bool test_journal_node(struct dm_integrity_c *ic, unsigned pos, sector_t sector)
1058 {
1059         struct journal_node *node, *next_node;
1060         struct rb_node *next;
1061
1062         if (unlikely(pos >= ic->journal_entries))
1063                 return false;
1064         node = &ic->journal_tree[pos];
1065         if (unlikely(RB_EMPTY_NODE(&node->node)))
1066                 return false;
1067         if (unlikely(node->sector != sector))
1068                 return false;
1069
1070         next = rb_next(&node->node);
1071         if (unlikely(!next))
1072                 return true;
1073
1074         next_node = container_of(next, struct journal_node, node);
1075         return next_node->sector != sector;
1076 }
1077
1078 static bool find_newer_committed_node(struct dm_integrity_c *ic, struct journal_node *node)
1079 {
1080         struct rb_node *next;
1081         struct journal_node *next_node;
1082         unsigned next_section;
1083
1084         BUG_ON(RB_EMPTY_NODE(&node->node));
1085
1086         next = rb_next(&node->node);
1087         if (unlikely(!next))
1088                 return false;
1089
1090         next_node = container_of(next, struct journal_node, node);
1091
1092         if (next_node->sector != node->sector)
1093                 return false;
1094
1095         next_section = (unsigned)(next_node - ic->journal_tree) / ic->journal_section_entries;
1096         if (next_section >= ic->committed_section &&
1097             next_section < ic->committed_section + ic->n_committed_sections)
1098                 return true;
1099         if (next_section + ic->journal_sections < ic->committed_section + ic->n_committed_sections)
1100                 return true;
1101
1102         return false;
1103 }
1104
1105 #define TAG_READ        0
1106 #define TAG_WRITE       1
1107 #define TAG_CMP         2
1108
1109 static int dm_integrity_rw_tag(struct dm_integrity_c *ic, unsigned char *tag, sector_t *metadata_block,
1110                                unsigned *metadata_offset, unsigned total_size, int op)
1111 {
1112         do {
1113                 unsigned char *data, *dp;
1114                 struct dm_buffer *b;
1115                 unsigned to_copy;
1116                 int r;
1117
1118                 r = dm_integrity_failed(ic);
1119                 if (unlikely(r))
1120                         return r;
1121
1122                 data = dm_bufio_read(ic->bufio, *metadata_block, &b);
1123                 if (IS_ERR(data))
1124                         return PTR_ERR(data);
1125
1126                 to_copy = min((1U << SECTOR_SHIFT << ic->log2_buffer_sectors) - *metadata_offset, total_size);
1127                 dp = data + *metadata_offset;
1128                 if (op == TAG_READ) {
1129                         memcpy(tag, dp, to_copy);
1130                 } else if (op == TAG_WRITE) {
1131                         memcpy(dp, tag, to_copy);
1132                         dm_bufio_mark_partial_buffer_dirty(b, *metadata_offset, *metadata_offset + to_copy);
1133                 } else  {
1134                         /* e.g.: op == TAG_CMP */
1135                         if (unlikely(memcmp(dp, tag, to_copy))) {
1136                                 unsigned i;
1137
1138                                 for (i = 0; i < to_copy; i++) {
1139                                         if (dp[i] != tag[i])
1140                                                 break;
1141                                         total_size--;
1142                                 }
1143                                 dm_bufio_release(b);
1144                                 return total_size;
1145                         }
1146                 }
1147                 dm_bufio_release(b);
1148
1149                 tag += to_copy;
1150                 *metadata_offset += to_copy;
1151                 if (unlikely(*metadata_offset == 1U << SECTOR_SHIFT << ic->log2_buffer_sectors)) {
1152                         (*metadata_block)++;
1153                         *metadata_offset = 0;
1154                 }
1155                 total_size -= to_copy;
1156         } while (unlikely(total_size));
1157
1158         return 0;
1159 }
1160
1161 static void dm_integrity_flush_buffers(struct dm_integrity_c *ic)
1162 {
1163         int r;
1164         r = dm_bufio_write_dirty_buffers(ic->bufio);
1165         if (unlikely(r))
1166                 dm_integrity_io_error(ic, "writing tags", r);
1167 }
1168
1169 static void sleep_on_endio_wait(struct dm_integrity_c *ic)
1170 {
1171         DECLARE_WAITQUEUE(wait, current);
1172         __add_wait_queue(&ic->endio_wait, &wait);
1173         __set_current_state(TASK_UNINTERRUPTIBLE);
1174         spin_unlock_irq(&ic->endio_wait.lock);
1175         io_schedule();
1176         spin_lock_irq(&ic->endio_wait.lock);
1177         __remove_wait_queue(&ic->endio_wait, &wait);
1178 }
1179
1180 static void autocommit_fn(struct timer_list *t)
1181 {
1182         struct dm_integrity_c *ic = from_timer(ic, t, autocommit_timer);
1183
1184         if (likely(!dm_integrity_failed(ic)))
1185                 queue_work(ic->commit_wq, &ic->commit_work);
1186 }
1187
1188 static void schedule_autocommit(struct dm_integrity_c *ic)
1189 {
1190         if (!timer_pending(&ic->autocommit_timer))
1191                 mod_timer(&ic->autocommit_timer, jiffies + ic->autocommit_jiffies);
1192 }
1193
1194 static void submit_flush_bio(struct dm_integrity_c *ic, struct dm_integrity_io *dio)
1195 {
1196         struct bio *bio;
1197         unsigned long flags;
1198
1199         spin_lock_irqsave(&ic->endio_wait.lock, flags);
1200         bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1201         bio_list_add(&ic->flush_bio_list, bio);
1202         spin_unlock_irqrestore(&ic->endio_wait.lock, flags);
1203
1204         queue_work(ic->commit_wq, &ic->commit_work);
1205 }
1206
1207 static void do_endio(struct dm_integrity_c *ic, struct bio *bio)
1208 {
1209         int r = dm_integrity_failed(ic);
1210         if (unlikely(r) && !bio->bi_status)
1211                 bio->bi_status = errno_to_blk_status(r);
1212         bio_endio(bio);
1213 }
1214
1215 static void do_endio_flush(struct dm_integrity_c *ic, struct dm_integrity_io *dio)
1216 {
1217         struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1218
1219         if (unlikely(dio->fua) && likely(!bio->bi_status) && likely(!dm_integrity_failed(ic)))
1220                 submit_flush_bio(ic, dio);
1221         else
1222                 do_endio(ic, bio);
1223 }
1224
1225 static void dec_in_flight(struct dm_integrity_io *dio)
1226 {
1227         if (atomic_dec_and_test(&dio->in_flight)) {
1228                 struct dm_integrity_c *ic = dio->ic;
1229                 struct bio *bio;
1230
1231                 remove_range(ic, &dio->range);
1232
1233                 if (unlikely(dio->write))
1234                         schedule_autocommit(ic);
1235
1236                 bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1237
1238                 if (unlikely(dio->bi_status) && !bio->bi_status)
1239                         bio->bi_status = dio->bi_status;
1240                 if (likely(!bio->bi_status) && unlikely(bio_sectors(bio) != dio->range.n_sectors)) {
1241                         dio->range.logical_sector += dio->range.n_sectors;
1242                         bio_advance(bio, dio->range.n_sectors << SECTOR_SHIFT);
1243                         INIT_WORK(&dio->work, integrity_bio_wait);
1244                         queue_work(ic->wait_wq, &dio->work);
1245                         return;
1246                 }
1247                 do_endio_flush(ic, dio);
1248         }
1249 }
1250
1251 static void integrity_end_io(struct bio *bio)
1252 {
1253         struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
1254
1255         bio->bi_iter = dio->orig_bi_iter;
1256         bio->bi_disk = dio->orig_bi_disk;
1257         bio->bi_partno = dio->orig_bi_partno;
1258         if (dio->orig_bi_integrity) {
1259                 bio->bi_integrity = dio->orig_bi_integrity;
1260                 bio->bi_opf |= REQ_INTEGRITY;
1261         }
1262         bio->bi_end_io = dio->orig_bi_end_io;
1263
1264         if (dio->completion)
1265                 complete(dio->completion);
1266
1267         dec_in_flight(dio);
1268 }
1269
1270 static void integrity_sector_checksum(struct dm_integrity_c *ic, sector_t sector,
1271                                       const char *data, char *result)
1272 {
1273         __u64 sector_le = cpu_to_le64(sector);
1274         SHASH_DESC_ON_STACK(req, ic->internal_hash);
1275         int r;
1276         unsigned digest_size;
1277
1278         req->tfm = ic->internal_hash;
1279         req->flags = 0;
1280
1281         r = crypto_shash_init(req);
1282         if (unlikely(r < 0)) {
1283                 dm_integrity_io_error(ic, "crypto_shash_init", r);
1284                 goto failed;
1285         }
1286
1287         r = crypto_shash_update(req, (const __u8 *)&sector_le, sizeof sector_le);
1288         if (unlikely(r < 0)) {
1289                 dm_integrity_io_error(ic, "crypto_shash_update", r);
1290                 goto failed;
1291         }
1292
1293         r = crypto_shash_update(req, data, ic->sectors_per_block << SECTOR_SHIFT);
1294         if (unlikely(r < 0)) {
1295                 dm_integrity_io_error(ic, "crypto_shash_update", r);
1296                 goto failed;
1297         }
1298
1299         r = crypto_shash_final(req, result);
1300         if (unlikely(r < 0)) {
1301                 dm_integrity_io_error(ic, "crypto_shash_final", r);
1302                 goto failed;
1303         }
1304
1305         digest_size = crypto_shash_digestsize(ic->internal_hash);
1306         if (unlikely(digest_size < ic->tag_size))
1307                 memset(result + digest_size, 0, ic->tag_size - digest_size);
1308
1309         return;
1310
1311 failed:
1312         /* this shouldn't happen anyway, the hash functions have no reason to fail */
1313         get_random_bytes(result, ic->tag_size);
1314 }
1315
1316 static void integrity_metadata(struct work_struct *w)
1317 {
1318         struct dm_integrity_io *dio = container_of(w, struct dm_integrity_io, work);
1319         struct dm_integrity_c *ic = dio->ic;
1320
1321         int r;
1322
1323         if (ic->internal_hash) {
1324                 struct bvec_iter iter;
1325                 struct bio_vec bv;
1326                 unsigned digest_size = crypto_shash_digestsize(ic->internal_hash);
1327                 struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1328                 char *checksums;
1329                 unsigned extra_space = unlikely(digest_size > ic->tag_size) ? digest_size - ic->tag_size : 0;
1330                 char checksums_onstack[HASH_MAX_DIGESTSIZE];
1331                 unsigned sectors_to_process = dio->range.n_sectors;
1332                 sector_t sector = dio->range.logical_sector;
1333
1334                 if (unlikely(ic->mode == 'R'))
1335                         goto skip_io;
1336
1337                 checksums = kmalloc((PAGE_SIZE >> SECTOR_SHIFT >> ic->sb->log2_sectors_per_block) * ic->tag_size + extra_space,
1338                                     GFP_NOIO | __GFP_NORETRY | __GFP_NOWARN);
1339                 if (!checksums) {
1340                         checksums = checksums_onstack;
1341                         if (WARN_ON(extra_space &&
1342                                     digest_size > sizeof(checksums_onstack))) {
1343                                 r = -EINVAL;
1344                                 goto error;
1345                         }
1346                 }
1347
1348                 __bio_for_each_segment(bv, bio, iter, dio->orig_bi_iter) {
1349                         unsigned pos;
1350                         char *mem, *checksums_ptr;
1351
1352 again:
1353                         mem = (char *)kmap_atomic(bv.bv_page) + bv.bv_offset;
1354                         pos = 0;
1355                         checksums_ptr = checksums;
1356                         do {
1357                                 integrity_sector_checksum(ic, sector, mem + pos, checksums_ptr);
1358                                 checksums_ptr += ic->tag_size;
1359                                 sectors_to_process -= ic->sectors_per_block;
1360                                 pos += ic->sectors_per_block << SECTOR_SHIFT;
1361                                 sector += ic->sectors_per_block;
1362                         } while (pos < bv.bv_len && sectors_to_process && checksums != checksums_onstack);
1363                         kunmap_atomic(mem);
1364
1365                         r = dm_integrity_rw_tag(ic, checksums, &dio->metadata_block, &dio->metadata_offset,
1366                                                 checksums_ptr - checksums, !dio->write ? TAG_CMP : TAG_WRITE);
1367                         if (unlikely(r)) {
1368                                 if (r > 0) {
1369                                         DMERR_LIMIT("Checksum failed at sector 0x%llx",
1370                                                     (unsigned long long)(sector - ((r + ic->tag_size - 1) / ic->tag_size)));
1371                                         r = -EILSEQ;
1372                                         atomic64_inc(&ic->number_of_mismatches);
1373                                 }
1374                                 if (likely(checksums != checksums_onstack))
1375                                         kfree(checksums);
1376                                 goto error;
1377                         }
1378
1379                         if (!sectors_to_process)
1380                                 break;
1381
1382                         if (unlikely(pos < bv.bv_len)) {
1383                                 bv.bv_offset += pos;
1384                                 bv.bv_len -= pos;
1385                                 goto again;
1386                         }
1387                 }
1388
1389                 if (likely(checksums != checksums_onstack))
1390                         kfree(checksums);
1391         } else {
1392                 struct bio_integrity_payload *bip = dio->orig_bi_integrity;
1393
1394                 if (bip) {
1395                         struct bio_vec biv;
1396                         struct bvec_iter iter;
1397                         unsigned data_to_process = dio->range.n_sectors;
1398                         sector_to_block(ic, data_to_process);
1399                         data_to_process *= ic->tag_size;
1400
1401                         bip_for_each_vec(biv, bip, iter) {
1402                                 unsigned char *tag;
1403                                 unsigned this_len;
1404
1405                                 BUG_ON(PageHighMem(biv.bv_page));
1406                                 tag = lowmem_page_address(biv.bv_page) + biv.bv_offset;
1407                                 this_len = min(biv.bv_len, data_to_process);
1408                                 r = dm_integrity_rw_tag(ic, tag, &dio->metadata_block, &dio->metadata_offset,
1409                                                         this_len, !dio->write ? TAG_READ : TAG_WRITE);
1410                                 if (unlikely(r))
1411                                         goto error;
1412                                 data_to_process -= this_len;
1413                                 if (!data_to_process)
1414                                         break;
1415                         }
1416                 }
1417         }
1418 skip_io:
1419         dec_in_flight(dio);
1420         return;
1421 error:
1422         dio->bi_status = errno_to_blk_status(r);
1423         dec_in_flight(dio);
1424 }
1425
1426 static int dm_integrity_map(struct dm_target *ti, struct bio *bio)
1427 {
1428         struct dm_integrity_c *ic = ti->private;
1429         struct dm_integrity_io *dio = dm_per_bio_data(bio, sizeof(struct dm_integrity_io));
1430         struct bio_integrity_payload *bip;
1431
1432         sector_t area, offset;
1433
1434         dio->ic = ic;
1435         dio->bi_status = 0;
1436
1437         if (unlikely(bio->bi_opf & REQ_PREFLUSH)) {
1438                 submit_flush_bio(ic, dio);
1439                 return DM_MAPIO_SUBMITTED;
1440         }
1441
1442         dio->range.logical_sector = dm_target_offset(ti, bio->bi_iter.bi_sector);
1443         dio->write = bio_op(bio) == REQ_OP_WRITE;
1444         dio->fua = dio->write && bio->bi_opf & REQ_FUA;
1445         if (unlikely(dio->fua)) {
1446                 /*
1447                  * Don't pass down the FUA flag because we have to flush
1448                  * disk cache anyway.
1449                  */
1450                 bio->bi_opf &= ~REQ_FUA;
1451         }
1452         if (unlikely(dio->range.logical_sector + bio_sectors(bio) > ic->provided_data_sectors)) {
1453                 DMERR("Too big sector number: 0x%llx + 0x%x > 0x%llx",
1454                       (unsigned long long)dio->range.logical_sector, bio_sectors(bio),
1455                       (unsigned long long)ic->provided_data_sectors);
1456                 return DM_MAPIO_KILL;
1457         }
1458         if (unlikely((dio->range.logical_sector | bio_sectors(bio)) & (unsigned)(ic->sectors_per_block - 1))) {
1459                 DMERR("Bio not aligned on %u sectors: 0x%llx, 0x%x",
1460                       ic->sectors_per_block,
1461                       (unsigned long long)dio->range.logical_sector, bio_sectors(bio));
1462                 return DM_MAPIO_KILL;
1463         }
1464
1465         if (ic->sectors_per_block > 1) {
1466                 struct bvec_iter iter;
1467                 struct bio_vec bv;
1468                 bio_for_each_segment(bv, bio, iter) {
1469                         if (unlikely(bv.bv_len & ((ic->sectors_per_block << SECTOR_SHIFT) - 1))) {
1470                                 DMERR("Bio vector (%u,%u) is not aligned on %u-sector boundary",
1471                                         bv.bv_offset, bv.bv_len, ic->sectors_per_block);
1472                                 return DM_MAPIO_KILL;
1473                         }
1474                 }
1475         }
1476
1477         bip = bio_integrity(bio);
1478         if (!ic->internal_hash) {
1479                 if (bip) {
1480                         unsigned wanted_tag_size = bio_sectors(bio) >> ic->sb->log2_sectors_per_block;
1481                         if (ic->log2_tag_size >= 0)
1482                                 wanted_tag_size <<= ic->log2_tag_size;
1483                         else
1484                                 wanted_tag_size *= ic->tag_size;
1485                         if (unlikely(wanted_tag_size != bip->bip_iter.bi_size)) {
1486                                 DMERR("Invalid integrity data size %u, expected %u", bip->bip_iter.bi_size, wanted_tag_size);
1487                                 return DM_MAPIO_KILL;
1488                         }
1489                 }
1490         } else {
1491                 if (unlikely(bip != NULL)) {
1492                         DMERR("Unexpected integrity data when using internal hash");
1493                         return DM_MAPIO_KILL;
1494                 }
1495         }
1496
1497         if (unlikely(ic->mode == 'R') && unlikely(dio->write))
1498                 return DM_MAPIO_KILL;
1499
1500         get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
1501         dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset, &dio->metadata_offset);
1502         bio->bi_iter.bi_sector = get_data_sector(ic, area, offset);
1503
1504         dm_integrity_map_continue(dio, true);
1505         return DM_MAPIO_SUBMITTED;
1506 }
1507
1508 static bool __journal_read_write(struct dm_integrity_io *dio, struct bio *bio,
1509                                  unsigned journal_section, unsigned journal_entry)
1510 {
1511         struct dm_integrity_c *ic = dio->ic;
1512         sector_t logical_sector;
1513         unsigned n_sectors;
1514
1515         logical_sector = dio->range.logical_sector;
1516         n_sectors = dio->range.n_sectors;
1517         do {
1518                 struct bio_vec bv = bio_iovec(bio);
1519                 char *mem;
1520
1521                 if (unlikely(bv.bv_len >> SECTOR_SHIFT > n_sectors))
1522                         bv.bv_len = n_sectors << SECTOR_SHIFT;
1523                 n_sectors -= bv.bv_len >> SECTOR_SHIFT;
1524                 bio_advance_iter(bio, &bio->bi_iter, bv.bv_len);
1525 retry_kmap:
1526                 mem = kmap_atomic(bv.bv_page);
1527                 if (likely(dio->write))
1528                         flush_dcache_page(bv.bv_page);
1529
1530                 do {
1531                         struct journal_entry *je = access_journal_entry(ic, journal_section, journal_entry);
1532
1533                         if (unlikely(!dio->write)) {
1534                                 struct journal_sector *js;
1535                                 char *mem_ptr;
1536                                 unsigned s;
1537
1538                                 if (unlikely(journal_entry_is_inprogress(je))) {
1539                                         flush_dcache_page(bv.bv_page);
1540                                         kunmap_atomic(mem);
1541
1542                                         __io_wait_event(ic->copy_to_journal_wait, !journal_entry_is_inprogress(je));
1543                                         goto retry_kmap;
1544                                 }
1545                                 smp_rmb();
1546                                 BUG_ON(journal_entry_get_sector(je) != logical_sector);
1547                                 js = access_journal_data(ic, journal_section, journal_entry);
1548                                 mem_ptr = mem + bv.bv_offset;
1549                                 s = 0;
1550                                 do {
1551                                         memcpy(mem_ptr, js, JOURNAL_SECTOR_DATA);
1552                                         *(commit_id_t *)(mem_ptr + JOURNAL_SECTOR_DATA) = je->last_bytes[s];
1553                                         js++;
1554                                         mem_ptr += 1 << SECTOR_SHIFT;
1555                                 } while (++s < ic->sectors_per_block);
1556 #ifdef INTERNAL_VERIFY
1557                                 if (ic->internal_hash) {
1558                                         char checksums_onstack[max(HASH_MAX_DIGESTSIZE, MAX_TAG_SIZE)];
1559
1560                                         integrity_sector_checksum(ic, logical_sector, mem + bv.bv_offset, checksums_onstack);
1561                                         if (unlikely(memcmp(checksums_onstack, journal_entry_tag(ic, je), ic->tag_size))) {
1562                                                 DMERR_LIMIT("Checksum failed when reading from journal, at sector 0x%llx",
1563                                                             (unsigned long long)logical_sector);
1564                                         }
1565                                 }
1566 #endif
1567                         }
1568
1569                         if (!ic->internal_hash) {
1570                                 struct bio_integrity_payload *bip = bio_integrity(bio);
1571                                 unsigned tag_todo = ic->tag_size;
1572                                 char *tag_ptr = journal_entry_tag(ic, je);
1573
1574                                 if (bip) do {
1575                                         struct bio_vec biv = bvec_iter_bvec(bip->bip_vec, bip->bip_iter);
1576                                         unsigned tag_now = min(biv.bv_len, tag_todo);
1577                                         char *tag_addr;
1578                                         BUG_ON(PageHighMem(biv.bv_page));
1579                                         tag_addr = lowmem_page_address(biv.bv_page) + biv.bv_offset;
1580                                         if (likely(dio->write))
1581                                                 memcpy(tag_ptr, tag_addr, tag_now);
1582                                         else
1583                                                 memcpy(tag_addr, tag_ptr, tag_now);
1584                                         bvec_iter_advance(bip->bip_vec, &bip->bip_iter, tag_now);
1585                                         tag_ptr += tag_now;
1586                                         tag_todo -= tag_now;
1587                                 } while (unlikely(tag_todo)); else {
1588                                         if (likely(dio->write))
1589                                                 memset(tag_ptr, 0, tag_todo);
1590                                 }
1591                         }
1592
1593                         if (likely(dio->write)) {
1594                                 struct journal_sector *js;
1595                                 unsigned s;
1596
1597                                 js = access_journal_data(ic, journal_section, journal_entry);
1598                                 memcpy(js, mem + bv.bv_offset, ic->sectors_per_block << SECTOR_SHIFT);
1599
1600                                 s = 0;
1601                                 do {
1602                                         je->last_bytes[s] = js[s].commit_id;
1603                                 } while (++s < ic->sectors_per_block);
1604
1605                                 if (ic->internal_hash) {
1606                                         unsigned digest_size = crypto_shash_digestsize(ic->internal_hash);
1607                                         if (unlikely(digest_size > ic->tag_size)) {
1608                                                 char checksums_onstack[HASH_MAX_DIGESTSIZE];
1609                                                 integrity_sector_checksum(ic, logical_sector, (char *)js, checksums_onstack);
1610                                                 memcpy(journal_entry_tag(ic, je), checksums_onstack, ic->tag_size);
1611                                         } else
1612                                                 integrity_sector_checksum(ic, logical_sector, (char *)js, journal_entry_tag(ic, je));
1613                                 }
1614
1615                                 journal_entry_set_sector(je, logical_sector);
1616                         }
1617                         logical_sector += ic->sectors_per_block;
1618
1619                         journal_entry++;
1620                         if (unlikely(journal_entry == ic->journal_section_entries)) {
1621                                 journal_entry = 0;
1622                                 journal_section++;
1623                                 wraparound_section(ic, &journal_section);
1624                         }
1625
1626                         bv.bv_offset += ic->sectors_per_block << SECTOR_SHIFT;
1627                 } while (bv.bv_len -= ic->sectors_per_block << SECTOR_SHIFT);
1628
1629                 if (unlikely(!dio->write))
1630                         flush_dcache_page(bv.bv_page);
1631                 kunmap_atomic(mem);
1632         } while (n_sectors);
1633
1634         if (likely(dio->write)) {
1635                 smp_mb();
1636                 if (unlikely(waitqueue_active(&ic->copy_to_journal_wait)))
1637                         wake_up(&ic->copy_to_journal_wait);
1638                 if (READ_ONCE(ic->free_sectors) <= ic->free_sectors_threshold) {
1639                         queue_work(ic->commit_wq, &ic->commit_work);
1640                 } else {
1641                         schedule_autocommit(ic);
1642                 }
1643         } else {
1644                 remove_range(ic, &dio->range);
1645         }
1646
1647         if (unlikely(bio->bi_iter.bi_size)) {
1648                 sector_t area, offset;
1649
1650                 dio->range.logical_sector = logical_sector;
1651                 get_area_and_offset(ic, dio->range.logical_sector, &area, &offset);
1652                 dio->metadata_block = get_metadata_sector_and_offset(ic, area, offset, &dio->metadata_offset);
1653                 return true;
1654         }
1655
1656         return false;
1657 }
1658
1659 static void dm_integrity_map_continue(struct dm_integrity_io *dio, bool from_map)
1660 {
1661         struct dm_integrity_c *ic = dio->ic;
1662         struct bio *bio = dm_bio_from_per_bio_data(dio, sizeof(struct dm_integrity_io));
1663         unsigned journal_section, journal_entry;
1664         unsigned journal_read_pos;
1665         struct completion read_comp;
1666         bool need_sync_io = ic->internal_hash && !dio->write;
1667
1668         if (need_sync_io && from_map) {
1669                 INIT_WORK(&dio->work, integrity_bio_wait);
1670                 queue_work(ic->metadata_wq, &dio->work);
1671                 return;
1672         }
1673
1674 lock_retry:
1675         spin_lock_irq(&ic->endio_wait.lock);
1676 retry:
1677         if (unlikely(dm_integrity_failed(ic))) {
1678                 spin_unlock_irq(&ic->endio_wait.lock);
1679                 do_endio(ic, bio);
1680                 return;
1681         }
1682         dio->range.n_sectors = bio_sectors(bio);
1683         journal_read_pos = NOT_FOUND;
1684         if (likely(ic->mode == 'J')) {
1685                 if (dio->write) {
1686                         unsigned next_entry, i, pos;
1687                         unsigned ws, we, range_sectors;
1688
1689                         dio->range.n_sectors = min(dio->range.n_sectors,
1690                                                    ic->free_sectors << ic->sb->log2_sectors_per_block);
1691                         if (unlikely(!dio->range.n_sectors)) {
1692                                 if (from_map)
1693                                         goto offload_to_thread;
1694                                 sleep_on_endio_wait(ic);
1695                                 goto retry;
1696                         }
1697                         range_sectors = dio->range.n_sectors >> ic->sb->log2_sectors_per_block;
1698                         ic->free_sectors -= range_sectors;
1699                         journal_section = ic->free_section;
1700                         journal_entry = ic->free_section_entry;
1701
1702                         next_entry = ic->free_section_entry + range_sectors;
1703                         ic->free_section_entry = next_entry % ic->journal_section_entries;
1704                         ic->free_section += next_entry / ic->journal_section_entries;
1705                         ic->n_uncommitted_sections += next_entry / ic->journal_section_entries;
1706                         wraparound_section(ic, &ic->free_section);
1707
1708                         pos = journal_section * ic->journal_section_entries + journal_entry;
1709                         ws = journal_section;
1710                         we = journal_entry;
1711                         i = 0;
1712                         do {
1713                                 struct journal_entry *je;
1714
1715                                 add_journal_node(ic, &ic->journal_tree[pos], dio->range.logical_sector + i);
1716                                 pos++;
1717                                 if (unlikely(pos >= ic->journal_entries))
1718                                         pos = 0;
1719
1720                                 je = access_journal_entry(ic, ws, we);
1721                                 BUG_ON(!journal_entry_is_unused(je));
1722                                 journal_entry_set_inprogress(je);
1723                                 we++;
1724                                 if (unlikely(we == ic->journal_section_entries)) {
1725                                         we = 0;
1726                                         ws++;
1727                                         wraparound_section(ic, &ws);
1728                                 }
1729                         } while ((i += ic->sectors_per_block) < dio->range.n_sectors);
1730
1731                         spin_unlock_irq(&ic->endio_wait.lock);
1732                         goto journal_read_write;
1733                 } else {
1734                         sector_t next_sector;
1735                         journal_read_pos = find_journal_node(ic, dio->range.logical_sector, &next_sector);
1736                         if (likely(journal_read_pos == NOT_FOUND)) {
1737                                 if (unlikely(dio->range.n_sectors > next_sector - dio->range.logical_sector))
1738                                         dio->range.n_sectors = next_sector - dio->range.logical_sector;
1739                         } else {
1740                                 unsigned i;
1741                                 unsigned jp = journal_read_pos + 1;
1742                                 for (i = ic->sectors_per_block; i < dio->range.n_sectors; i += ic->sectors_per_block, jp++) {
1743                                         if (!test_journal_node(ic, jp, dio->range.logical_sector + i))
1744                                                 break;
1745                                 }
1746                                 dio->range.n_sectors = i;
1747                         }
1748                 }
1749         }
1750         if (unlikely(!add_new_range(ic, &dio->range, true))) {
1751                 /*
1752                  * We must not sleep in the request routine because it could
1753                  * stall bios on current->bio_list.
1754                  * So, we offload the bio to a workqueue if we have to sleep.
1755                  */
1756                 if (from_map) {
1757 offload_to_thread:
1758                         spin_unlock_irq(&ic->endio_wait.lock);
1759                         INIT_WORK(&dio->work, integrity_bio_wait);
1760                         queue_work(ic->wait_wq, &dio->work);
1761                         return;
1762                 }
1763                 wait_and_add_new_range(ic, &dio->range);
1764         }
1765         spin_unlock_irq(&ic->endio_wait.lock);
1766
1767         if (unlikely(journal_read_pos != NOT_FOUND)) {
1768                 journal_section = journal_read_pos / ic->journal_section_entries;
1769                 journal_entry = journal_read_pos % ic->journal_section_entries;
1770                 goto journal_read_write;
1771         }
1772
1773         dio->in_flight = (atomic_t)ATOMIC_INIT(2);
1774
1775         if (need_sync_io) {
1776                 init_completion(&read_comp);
1777                 dio->completion = &read_comp;
1778         } else
1779                 dio->completion = NULL;
1780
1781         dio->orig_bi_iter = bio->bi_iter;
1782
1783         dio->orig_bi_disk = bio->bi_disk;
1784         dio->orig_bi_partno = bio->bi_partno;
1785         bio_set_dev(bio, ic->dev->bdev);
1786
1787         dio->orig_bi_integrity = bio_integrity(bio);
1788         bio->bi_integrity = NULL;
1789         bio->bi_opf &= ~REQ_INTEGRITY;
1790
1791         dio->orig_bi_end_io = bio->bi_end_io;
1792         bio->bi_end_io = integrity_end_io;
1793
1794         bio->bi_iter.bi_size = dio->range.n_sectors << SECTOR_SHIFT;
1795         generic_make_request(bio);
1796
1797         if (need_sync_io) {
1798                 wait_for_completion_io(&read_comp);
1799                 if (unlikely(ic->recalc_wq != NULL) &&
1800                     ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING) &&
1801                     dio->range.logical_sector + dio->range.n_sectors > le64_to_cpu(ic->sb->recalc_sector))
1802                         goto skip_check;
1803                 if (likely(!bio->bi_status))
1804                         integrity_metadata(&dio->work);
1805                 else
1806 skip_check:
1807                         dec_in_flight(dio);
1808
1809         } else {
1810                 INIT_WORK(&dio->work, integrity_metadata);
1811                 queue_work(ic->metadata_wq, &dio->work);
1812         }
1813
1814         return;
1815
1816 journal_read_write:
1817         if (unlikely(__journal_read_write(dio, bio, journal_section, journal_entry)))
1818                 goto lock_retry;
1819
1820         do_endio_flush(ic, dio);
1821 }
1822
1823
1824 static void integrity_bio_wait(struct work_struct *w)
1825 {
1826         struct dm_integrity_io *dio = container_of(w, struct dm_integrity_io, work);
1827
1828         dm_integrity_map_continue(dio, false);
1829 }
1830
1831 static void pad_uncommitted(struct dm_integrity_c *ic)
1832 {
1833         if (ic->free_section_entry) {
1834                 ic->free_sectors -= ic->journal_section_entries - ic->free_section_entry;
1835                 ic->free_section_entry = 0;
1836                 ic->free_section++;
1837                 wraparound_section(ic, &ic->free_section);
1838                 ic->n_uncommitted_sections++;
1839         }
1840         WARN_ON(ic->journal_sections * ic->journal_section_entries !=
1841                 (ic->n_uncommitted_sections + ic->n_committed_sections) * ic->journal_section_entries + ic->free_sectors);
1842 }
1843
1844 static void integrity_commit(struct work_struct *w)
1845 {
1846         struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, commit_work);
1847         unsigned commit_start, commit_sections;
1848         unsigned i, j, n;
1849         struct bio *flushes;
1850
1851         del_timer(&ic->autocommit_timer);
1852
1853         spin_lock_irq(&ic->endio_wait.lock);
1854         flushes = bio_list_get(&ic->flush_bio_list);
1855         if (unlikely(ic->mode != 'J')) {
1856                 spin_unlock_irq(&ic->endio_wait.lock);
1857                 dm_integrity_flush_buffers(ic);
1858                 goto release_flush_bios;
1859         }
1860
1861         pad_uncommitted(ic);
1862         commit_start = ic->uncommitted_section;
1863         commit_sections = ic->n_uncommitted_sections;
1864         spin_unlock_irq(&ic->endio_wait.lock);
1865
1866         if (!commit_sections)
1867                 goto release_flush_bios;
1868
1869         i = commit_start;
1870         for (n = 0; n < commit_sections; n++) {
1871                 for (j = 0; j < ic->journal_section_entries; j++) {
1872                         struct journal_entry *je;
1873                         je = access_journal_entry(ic, i, j);
1874                         io_wait_event(ic->copy_to_journal_wait, !journal_entry_is_inprogress(je));
1875                 }
1876                 for (j = 0; j < ic->journal_section_sectors; j++) {
1877                         struct journal_sector *js;
1878                         js = access_journal(ic, i, j);
1879                         js->commit_id = dm_integrity_commit_id(ic, i, j, ic->commit_seq);
1880                 }
1881                 i++;
1882                 if (unlikely(i >= ic->journal_sections))
1883                         ic->commit_seq = next_commit_seq(ic->commit_seq);
1884                 wraparound_section(ic, &i);
1885         }
1886         smp_rmb();
1887
1888         write_journal(ic, commit_start, commit_sections);
1889
1890         spin_lock_irq(&ic->endio_wait.lock);
1891         ic->uncommitted_section += commit_sections;
1892         wraparound_section(ic, &ic->uncommitted_section);
1893         ic->n_uncommitted_sections -= commit_sections;
1894         ic->n_committed_sections += commit_sections;
1895         spin_unlock_irq(&ic->endio_wait.lock);
1896
1897         if (READ_ONCE(ic->free_sectors) <= ic->free_sectors_threshold)
1898                 queue_work(ic->writer_wq, &ic->writer_work);
1899
1900 release_flush_bios:
1901         while (flushes) {
1902                 struct bio *next = flushes->bi_next;
1903                 flushes->bi_next = NULL;
1904                 do_endio(ic, flushes);
1905                 flushes = next;
1906         }
1907 }
1908
1909 static void complete_copy_from_journal(unsigned long error, void *context)
1910 {
1911         struct journal_io *io = context;
1912         struct journal_completion *comp = io->comp;
1913         struct dm_integrity_c *ic = comp->ic;
1914         remove_range(ic, &io->range);
1915         mempool_free(io, &ic->journal_io_mempool);
1916         if (unlikely(error != 0))
1917                 dm_integrity_io_error(ic, "copying from journal", -EIO);
1918         complete_journal_op(comp);
1919 }
1920
1921 static void restore_last_bytes(struct dm_integrity_c *ic, struct journal_sector *js,
1922                                struct journal_entry *je)
1923 {
1924         unsigned s = 0;
1925         do {
1926                 js->commit_id = je->last_bytes[s];
1927                 js++;
1928         } while (++s < ic->sectors_per_block);
1929 }
1930
1931 static void do_journal_write(struct dm_integrity_c *ic, unsigned write_start,
1932                              unsigned write_sections, bool from_replay)
1933 {
1934         unsigned i, j, n;
1935         struct journal_completion comp;
1936         struct blk_plug plug;
1937
1938         blk_start_plug(&plug);
1939
1940         comp.ic = ic;
1941         comp.in_flight = (atomic_t)ATOMIC_INIT(1);
1942         init_completion(&comp.comp);
1943
1944         i = write_start;
1945         for (n = 0; n < write_sections; n++, i++, wraparound_section(ic, &i)) {
1946 #ifndef INTERNAL_VERIFY
1947                 if (unlikely(from_replay))
1948 #endif
1949                         rw_section_mac(ic, i, false);
1950                 for (j = 0; j < ic->journal_section_entries; j++) {
1951                         struct journal_entry *je = access_journal_entry(ic, i, j);
1952                         sector_t sec, area, offset;
1953                         unsigned k, l, next_loop;
1954                         sector_t metadata_block;
1955                         unsigned metadata_offset;
1956                         struct journal_io *io;
1957
1958                         if (journal_entry_is_unused(je))
1959                                 continue;
1960                         BUG_ON(unlikely(journal_entry_is_inprogress(je)) && !from_replay);
1961                         sec = journal_entry_get_sector(je);
1962                         if (unlikely(from_replay)) {
1963                                 if (unlikely(sec & (unsigned)(ic->sectors_per_block - 1))) {
1964                                         dm_integrity_io_error(ic, "invalid sector in journal", -EIO);
1965                                         sec &= ~(sector_t)(ic->sectors_per_block - 1);
1966                                 }
1967                         }
1968                         get_area_and_offset(ic, sec, &area, &offset);
1969                         restore_last_bytes(ic, access_journal_data(ic, i, j), je);
1970                         for (k = j + 1; k < ic->journal_section_entries; k++) {
1971                                 struct journal_entry *je2 = access_journal_entry(ic, i, k);
1972                                 sector_t sec2, area2, offset2;
1973                                 if (journal_entry_is_unused(je2))
1974                                         break;
1975                                 BUG_ON(unlikely(journal_entry_is_inprogress(je2)) && !from_replay);
1976                                 sec2 = journal_entry_get_sector(je2);
1977                                 get_area_and_offset(ic, sec2, &area2, &offset2);
1978                                 if (area2 != area || offset2 != offset + ((k - j) << ic->sb->log2_sectors_per_block))
1979                                         break;
1980                                 restore_last_bytes(ic, access_journal_data(ic, i, k), je2);
1981                         }
1982                         next_loop = k - 1;
1983
1984                         io = mempool_alloc(&ic->journal_io_mempool, GFP_NOIO);
1985                         io->comp = &comp;
1986                         io->range.logical_sector = sec;
1987                         io->range.n_sectors = (k - j) << ic->sb->log2_sectors_per_block;
1988
1989                         spin_lock_irq(&ic->endio_wait.lock);
1990                         if (unlikely(!add_new_range(ic, &io->range, true)))
1991                                 wait_and_add_new_range(ic, &io->range);
1992
1993                         if (likely(!from_replay)) {
1994                                 struct journal_node *section_node = &ic->journal_tree[i * ic->journal_section_entries];
1995
1996                                 /* don't write if there is newer committed sector */
1997                                 while (j < k && find_newer_committed_node(ic, &section_node[j])) {
1998                                         struct journal_entry *je2 = access_journal_entry(ic, i, j);
1999
2000                                         journal_entry_set_unused(je2);
2001                                         remove_journal_node(ic, &section_node[j]);
2002                                         j++;
2003                                         sec += ic->sectors_per_block;
2004                                         offset += ic->sectors_per_block;
2005                                 }
2006                                 while (j < k && find_newer_committed_node(ic, &section_node[k - 1])) {
2007                                         struct journal_entry *je2 = access_journal_entry(ic, i, k - 1);
2008
2009                                         journal_entry_set_unused(je2);
2010                                         remove_journal_node(ic, &section_node[k - 1]);
2011                                         k--;
2012                                 }
2013                                 if (j == k) {
2014                                         remove_range_unlocked(ic, &io->range);
2015                                         spin_unlock_irq(&ic->endio_wait.lock);
2016                                         mempool_free(io, &ic->journal_io_mempool);
2017                                         goto skip_io;
2018                                 }
2019                                 for (l = j; l < k; l++) {
2020                                         remove_journal_node(ic, &section_node[l]);
2021                                 }
2022                         }
2023                         spin_unlock_irq(&ic->endio_wait.lock);
2024
2025                         metadata_block = get_metadata_sector_and_offset(ic, area, offset, &metadata_offset);
2026                         for (l = j; l < k; l++) {
2027                                 int r;
2028                                 struct journal_entry *je2 = access_journal_entry(ic, i, l);
2029
2030                                 if (
2031 #ifndef INTERNAL_VERIFY
2032                                     unlikely(from_replay) &&
2033 #endif
2034                                     ic->internal_hash) {
2035                                         char test_tag[max_t(size_t, HASH_MAX_DIGESTSIZE, MAX_TAG_SIZE)];
2036
2037                                         integrity_sector_checksum(ic, sec + ((l - j) << ic->sb->log2_sectors_per_block),
2038                                                                   (char *)access_journal_data(ic, i, l), test_tag);
2039                                         if (unlikely(memcmp(test_tag, journal_entry_tag(ic, je2), ic->tag_size)))
2040                                                 dm_integrity_io_error(ic, "tag mismatch when replaying journal", -EILSEQ);
2041                                 }
2042
2043                                 journal_entry_set_unused(je2);
2044                                 r = dm_integrity_rw_tag(ic, journal_entry_tag(ic, je2), &metadata_block, &metadata_offset,
2045                                                         ic->tag_size, TAG_WRITE);
2046                                 if (unlikely(r)) {
2047                                         dm_integrity_io_error(ic, "reading tags", r);
2048                                 }
2049                         }
2050
2051                         atomic_inc(&comp.in_flight);
2052                         copy_from_journal(ic, i, j << ic->sb->log2_sectors_per_block,
2053                                           (k - j) << ic->sb->log2_sectors_per_block,
2054                                           get_data_sector(ic, area, offset),
2055                                           complete_copy_from_journal, io);
2056 skip_io:
2057                         j = next_loop;
2058                 }
2059         }
2060
2061         dm_bufio_write_dirty_buffers_async(ic->bufio);
2062
2063         blk_finish_plug(&plug);
2064
2065         complete_journal_op(&comp);
2066         wait_for_completion_io(&comp.comp);
2067
2068         dm_integrity_flush_buffers(ic);
2069 }
2070
2071 static void integrity_writer(struct work_struct *w)
2072 {
2073         struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, writer_work);
2074         unsigned write_start, write_sections;
2075
2076         unsigned prev_free_sectors;
2077
2078         /* the following test is not needed, but it tests the replay code */
2079         if (READ_ONCE(ic->suspending) && !ic->meta_dev)
2080                 return;
2081
2082         spin_lock_irq(&ic->endio_wait.lock);
2083         write_start = ic->committed_section;
2084         write_sections = ic->n_committed_sections;
2085         spin_unlock_irq(&ic->endio_wait.lock);
2086
2087         if (!write_sections)
2088                 return;
2089
2090         do_journal_write(ic, write_start, write_sections, false);
2091
2092         spin_lock_irq(&ic->endio_wait.lock);
2093
2094         ic->committed_section += write_sections;
2095         wraparound_section(ic, &ic->committed_section);
2096         ic->n_committed_sections -= write_sections;
2097
2098         prev_free_sectors = ic->free_sectors;
2099         ic->free_sectors += write_sections * ic->journal_section_entries;
2100         if (unlikely(!prev_free_sectors))
2101                 wake_up_locked(&ic->endio_wait);
2102
2103         spin_unlock_irq(&ic->endio_wait.lock);
2104 }
2105
2106 static void recalc_write_super(struct dm_integrity_c *ic)
2107 {
2108         int r;
2109
2110         dm_integrity_flush_buffers(ic);
2111         if (dm_integrity_failed(ic))
2112                 return;
2113
2114         sb_set_version(ic);
2115         r = sync_rw_sb(ic, REQ_OP_WRITE, 0);
2116         if (unlikely(r))
2117                 dm_integrity_io_error(ic, "writing superblock", r);
2118 }
2119
2120 static void integrity_recalc(struct work_struct *w)
2121 {
2122         struct dm_integrity_c *ic = container_of(w, struct dm_integrity_c, recalc_work);
2123         struct dm_integrity_range range;
2124         struct dm_io_request io_req;
2125         struct dm_io_region io_loc;
2126         sector_t area, offset;
2127         sector_t metadata_block;
2128         unsigned metadata_offset;
2129         __u8 *t;
2130         unsigned i;
2131         int r;
2132         unsigned super_counter = 0;
2133
2134         spin_lock_irq(&ic->endio_wait.lock);
2135
2136 next_chunk:
2137
2138         if (unlikely(READ_ONCE(ic->suspending)))
2139                 goto unlock_ret;
2140
2141         range.logical_sector = le64_to_cpu(ic->sb->recalc_sector);
2142         if (unlikely(range.logical_sector >= ic->provided_data_sectors))
2143                 goto unlock_ret;
2144
2145         get_area_and_offset(ic, range.logical_sector, &area, &offset);
2146         range.n_sectors = min((sector_t)RECALC_SECTORS, ic->provided_data_sectors - range.logical_sector);
2147         if (!ic->meta_dev)
2148                 range.n_sectors = min(range.n_sectors, (1U << ic->sb->log2_interleave_sectors) - (unsigned)offset);
2149
2150         if (unlikely(!add_new_range(ic, &range, true)))
2151                 wait_and_add_new_range(ic, &range);
2152
2153         spin_unlock_irq(&ic->endio_wait.lock);
2154
2155         if (unlikely(++super_counter == RECALC_WRITE_SUPER)) {
2156                 recalc_write_super(ic);
2157                 super_counter = 0;
2158         }
2159
2160         if (unlikely(dm_integrity_failed(ic)))
2161                 goto err;
2162
2163         io_req.bi_op = REQ_OP_READ;
2164         io_req.bi_op_flags = 0;
2165         io_req.mem.type = DM_IO_VMA;
2166         io_req.mem.ptr.addr = ic->recalc_buffer;
2167         io_req.notify.fn = NULL;
2168         io_req.client = ic->io;
2169         io_loc.bdev = ic->dev->bdev;
2170         io_loc.sector = get_data_sector(ic, area, offset);
2171         io_loc.count = range.n_sectors;
2172
2173         r = dm_io(&io_req, 1, &io_loc, NULL);
2174         if (unlikely(r)) {
2175                 dm_integrity_io_error(ic, "reading data", r);
2176                 goto err;
2177         }
2178
2179         t = ic->recalc_tags;
2180         for (i = 0; i < range.n_sectors; i += ic->sectors_per_block) {
2181                 integrity_sector_checksum(ic, range.logical_sector + i, ic->recalc_buffer + (i << SECTOR_SHIFT), t);
2182                 t += ic->tag_size;
2183         }
2184
2185         metadata_block = get_metadata_sector_and_offset(ic, area, offset, &metadata_offset);
2186
2187         r = dm_integrity_rw_tag(ic, ic->recalc_tags, &metadata_block, &metadata_offset, t - ic->recalc_tags, TAG_WRITE);
2188         if (unlikely(r)) {
2189                 dm_integrity_io_error(ic, "writing tags", r);
2190                 goto err;
2191         }
2192
2193         spin_lock_irq(&ic->endio_wait.lock);
2194         remove_range_unlocked(ic, &range);
2195         ic->sb->recalc_sector = cpu_to_le64(range.logical_sector + range.n_sectors);
2196         goto next_chunk;
2197
2198 err:
2199         remove_range(ic, &range);
2200         return;
2201
2202 unlock_ret:
2203         spin_unlock_irq(&ic->endio_wait.lock);
2204
2205         recalc_write_super(ic);
2206 }
2207
2208 static void init_journal(struct dm_integrity_c *ic, unsigned start_section,
2209                          unsigned n_sections, unsigned char commit_seq)
2210 {
2211         unsigned i, j, n;
2212
2213         if (!n_sections)
2214                 return;
2215
2216         for (n = 0; n < n_sections; n++) {
2217                 i = start_section + n;
2218                 wraparound_section(ic, &i);
2219                 for (j = 0; j < ic->journal_section_sectors; j++) {
2220                         struct journal_sector *js = access_journal(ic, i, j);
2221                         memset(&js->entries, 0, JOURNAL_SECTOR_DATA);
2222                         js->commit_id = dm_integrity_commit_id(ic, i, j, commit_seq);
2223                 }
2224                 for (j = 0; j < ic->journal_section_entries; j++) {
2225                         struct journal_entry *je = access_journal_entry(ic, i, j);
2226                         journal_entry_set_unused(je);
2227                 }
2228         }
2229
2230         write_journal(ic, start_section, n_sections);
2231 }
2232
2233 static int find_commit_seq(struct dm_integrity_c *ic, unsigned i, unsigned j, commit_id_t id)
2234 {
2235         unsigned char k;
2236         for (k = 0; k < N_COMMIT_IDS; k++) {
2237                 if (dm_integrity_commit_id(ic, i, j, k) == id)
2238                         return k;
2239         }
2240         dm_integrity_io_error(ic, "journal commit id", -EIO);
2241         return -EIO;
2242 }
2243
2244 static void replay_journal(struct dm_integrity_c *ic)
2245 {
2246         unsigned i, j;
2247         bool used_commit_ids[N_COMMIT_IDS];
2248         unsigned max_commit_id_sections[N_COMMIT_IDS];
2249         unsigned write_start, write_sections;
2250         unsigned continue_section;
2251         bool journal_empty;
2252         unsigned char unused, last_used, want_commit_seq;
2253
2254         if (ic->mode == 'R')
2255                 return;
2256
2257         if (ic->journal_uptodate)
2258                 return;
2259
2260         last_used = 0;
2261         write_start = 0;
2262
2263         if (!ic->just_formatted) {
2264                 DEBUG_print("reading journal\n");
2265                 rw_journal(ic, REQ_OP_READ, 0, 0, ic->journal_sections, NULL);
2266                 if (ic->journal_io)
2267                         DEBUG_bytes(lowmem_page_address(ic->journal_io[0].page), 64, "read journal");
2268                 if (ic->journal_io) {
2269                         struct journal_completion crypt_comp;
2270                         crypt_comp.ic = ic;
2271                         init_completion(&crypt_comp.comp);
2272                         crypt_comp.in_flight = (atomic_t)ATOMIC_INIT(0);
2273                         encrypt_journal(ic, false, 0, ic->journal_sections, &crypt_comp);
2274                         wait_for_completion(&crypt_comp.comp);
2275                 }
2276                 DEBUG_bytes(lowmem_page_address(ic->journal[0].page), 64, "decrypted journal");
2277         }
2278
2279         if (dm_integrity_failed(ic))
2280                 goto clear_journal;
2281
2282         journal_empty = true;
2283         memset(used_commit_ids, 0, sizeof used_commit_ids);
2284         memset(max_commit_id_sections, 0, sizeof max_commit_id_sections);
2285         for (i = 0; i < ic->journal_sections; i++) {
2286                 for (j = 0; j < ic->journal_section_sectors; j++) {
2287                         int k;
2288                         struct journal_sector *js = access_journal(ic, i, j);
2289                         k = find_commit_seq(ic, i, j, js->commit_id);
2290                         if (k < 0)
2291                                 goto clear_journal;
2292                         used_commit_ids[k] = true;
2293                         max_commit_id_sections[k] = i;
2294                 }
2295                 if (journal_empty) {
2296                         for (j = 0; j < ic->journal_section_entries; j++) {
2297                                 struct journal_entry *je = access_journal_entry(ic, i, j);
2298                                 if (!journal_entry_is_unused(je)) {
2299                                         journal_empty = false;
2300                                         break;
2301                                 }
2302                         }
2303                 }
2304         }
2305
2306         if (!used_commit_ids[N_COMMIT_IDS - 1]) {
2307                 unused = N_COMMIT_IDS - 1;
2308                 while (unused && !used_commit_ids[unused - 1])
2309                         unused--;
2310         } else {
2311                 for (unused = 0; unused < N_COMMIT_IDS; unused++)
2312                         if (!used_commit_ids[unused])
2313                                 break;
2314                 if (unused == N_COMMIT_IDS) {
2315                         dm_integrity_io_error(ic, "journal commit ids", -EIO);
2316                         goto clear_journal;
2317                 }
2318         }
2319         DEBUG_print("first unused commit seq %d [%d,%d,%d,%d]\n",
2320                     unused, used_commit_ids[0], used_commit_ids[1],
2321                     used_commit_ids[2], used_commit_ids[3]);
2322
2323         last_used = prev_commit_seq(unused);
2324         want_commit_seq = prev_commit_seq(last_used);
2325
2326         if (!used_commit_ids[want_commit_seq] && used_commit_ids[prev_commit_seq(want_commit_seq)])
2327                 journal_empty = true;
2328
2329         write_start = max_commit_id_sections[last_used] + 1;
2330         if (unlikely(write_start >= ic->journal_sections))
2331                 want_commit_seq = next_commit_seq(want_commit_seq);
2332         wraparound_section(ic, &write_start);
2333
2334         i = write_start;
2335         for (write_sections = 0; write_sections < ic->journal_sections; write_sections++) {
2336                 for (j = 0; j < ic->journal_section_sectors; j++) {
2337                         struct journal_sector *js = access_journal(ic, i, j);
2338
2339                         if (js->commit_id != dm_integrity_commit_id(ic, i, j, want_commit_seq)) {
2340                                 /*
2341                                  * This could be caused by crash during writing.
2342                                  * We won't replay the inconsistent part of the
2343                                  * journal.
2344                                  */
2345                                 DEBUG_print("commit id mismatch at position (%u, %u): %d != %d\n",
2346                                             i, j, find_commit_seq(ic, i, j, js->commit_id), want_commit_seq);
2347                                 goto brk;
2348                         }
2349                 }
2350                 i++;
2351                 if (unlikely(i >= ic->journal_sections))
2352                         want_commit_seq = next_commit_seq(want_commit_seq);
2353                 wraparound_section(ic, &i);
2354         }
2355 brk:
2356
2357         if (!journal_empty) {
2358                 DEBUG_print("replaying %u sections, starting at %u, commit seq %d\n",
2359                             write_sections, write_start, want_commit_seq);
2360                 do_journal_write(ic, write_start, write_sections, true);
2361         }
2362
2363         if (write_sections == ic->journal_sections && (ic->mode == 'J' || journal_empty)) {
2364                 continue_section = write_start;
2365                 ic->commit_seq = want_commit_seq;
2366                 DEBUG_print("continuing from section %u, commit seq %d\n", write_start, ic->commit_seq);
2367         } else {
2368                 unsigned s;
2369                 unsigned char erase_seq;
2370 clear_journal:
2371                 DEBUG_print("clearing journal\n");
2372
2373                 erase_seq = prev_commit_seq(prev_commit_seq(last_used));
2374                 s = write_start;
2375                 init_journal(ic, s, 1, erase_seq);
2376                 s++;
2377                 wraparound_section(ic, &s);
2378                 if (ic->journal_sections >= 2) {
2379                         init_journal(ic, s, ic->journal_sections - 2, erase_seq);
2380                         s += ic->journal_sections - 2;
2381                         wraparound_section(ic, &s);
2382                         init_journal(ic, s, 1, erase_seq);
2383                 }
2384
2385                 continue_section = 0;
2386                 ic->commit_seq = next_commit_seq(erase_seq);
2387         }
2388
2389         ic->committed_section = continue_section;
2390         ic->n_committed_sections = 0;
2391
2392         ic->uncommitted_section = continue_section;
2393         ic->n_uncommitted_sections = 0;
2394
2395         ic->free_section = continue_section;
2396         ic->free_section_entry = 0;
2397         ic->free_sectors = ic->journal_entries;
2398
2399         ic->journal_tree_root = RB_ROOT;
2400         for (i = 0; i < ic->journal_entries; i++)
2401                 init_journal_node(&ic->journal_tree[i]);
2402 }
2403
2404 static void dm_integrity_postsuspend(struct dm_target *ti)
2405 {
2406         struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
2407
2408         del_timer_sync(&ic->autocommit_timer);
2409
2410         WRITE_ONCE(ic->suspending, 1);
2411
2412         if (ic->recalc_wq)
2413                 drain_workqueue(ic->recalc_wq);
2414
2415         queue_work(ic->commit_wq, &ic->commit_work);
2416         drain_workqueue(ic->commit_wq);
2417
2418         if (ic->mode == 'J') {
2419                 if (ic->meta_dev)
2420                         queue_work(ic->writer_wq, &ic->writer_work);
2421                 drain_workqueue(ic->writer_wq);
2422                 dm_integrity_flush_buffers(ic);
2423         }
2424
2425         WRITE_ONCE(ic->suspending, 0);
2426
2427         BUG_ON(!RB_EMPTY_ROOT(&ic->in_progress));
2428
2429         ic->journal_uptodate = true;
2430 }
2431
2432 static void dm_integrity_resume(struct dm_target *ti)
2433 {
2434         struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
2435
2436         replay_journal(ic);
2437
2438         if (ic->recalc_wq && ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING)) {
2439                 __u64 recalc_pos = le64_to_cpu(ic->sb->recalc_sector);
2440                 if (recalc_pos < ic->provided_data_sectors) {
2441                         queue_work(ic->recalc_wq, &ic->recalc_work);
2442                 } else if (recalc_pos > ic->provided_data_sectors) {
2443                         ic->sb->recalc_sector = cpu_to_le64(ic->provided_data_sectors);
2444                         recalc_write_super(ic);
2445                 }
2446         }
2447 }
2448
2449 static void dm_integrity_status(struct dm_target *ti, status_type_t type,
2450                                 unsigned status_flags, char *result, unsigned maxlen)
2451 {
2452         struct dm_integrity_c *ic = (struct dm_integrity_c *)ti->private;
2453         unsigned arg_count;
2454         size_t sz = 0;
2455
2456         switch (type) {
2457         case STATUSTYPE_INFO:
2458                 DMEMIT("%llu %llu",
2459                         (unsigned long long)atomic64_read(&ic->number_of_mismatches),
2460                         (unsigned long long)ic->provided_data_sectors);
2461                 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
2462                         DMEMIT(" %llu", (unsigned long long)le64_to_cpu(ic->sb->recalc_sector));
2463                 else
2464                         DMEMIT(" -");
2465                 break;
2466
2467         case STATUSTYPE_TABLE: {
2468                 __u64 watermark_percentage = (__u64)(ic->journal_entries - ic->free_sectors_threshold) * 100;
2469                 watermark_percentage += ic->journal_entries / 2;
2470                 do_div(watermark_percentage, ic->journal_entries);
2471                 arg_count = 5;
2472                 arg_count += !!ic->meta_dev;
2473                 arg_count += ic->sectors_per_block != 1;
2474                 arg_count += !!(ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING));
2475                 arg_count += !!ic->internal_hash_alg.alg_string;
2476                 arg_count += !!ic->journal_crypt_alg.alg_string;
2477                 arg_count += !!ic->journal_mac_alg.alg_string;
2478                 DMEMIT("%s %llu %u %c %u", ic->dev->name, (unsigned long long)ic->start,
2479                        ic->tag_size, ic->mode, arg_count);
2480                 if (ic->meta_dev)
2481                         DMEMIT(" meta_device:%s", ic->meta_dev->name);
2482                 if (ic->sectors_per_block != 1)
2483                         DMEMIT(" block_size:%u", ic->sectors_per_block << SECTOR_SHIFT);
2484                 if (ic->sb->flags & cpu_to_le32(SB_FLAG_RECALCULATING))
2485                         DMEMIT(" recalculate");
2486                 DMEMIT(" journal_sectors:%u", ic->initial_sectors - SB_SECTORS);
2487                 DMEMIT(" interleave_sectors:%u", 1U << ic->sb->log2_interleave_sectors);
2488                 DMEMIT(" buffer_sectors:%u", 1U << ic->log2_buffer_sectors);
2489                 DMEMIT(" journal_watermark:%u", (unsigned)watermark_percentage);
2490                 DMEMIT(" commit_time:%u", ic->autocommit_msec);
2491
2492 #define EMIT_ALG(a, n)                                                  \
2493                 do {                                                    \
2494                         if (ic->a.alg_string) {                         \
2495                                 DMEMIT(" %s:%s", n, ic->a.alg_string);  \
2496                                 if (ic->a.key_string)                   \
2497                                         DMEMIT(":%s", ic->a.key_string);\
2498                         }                                               \
2499                 } while (0)
2500                 EMIT_ALG(internal_hash_alg, "internal_hash");
2501                 EMIT_ALG(journal_crypt_alg, "journal_crypt");
2502                 EMIT_ALG(journal_mac_alg, "journal_mac");
2503                 break;
2504         }
2505         }
2506 }
2507
2508 static int dm_integrity_iterate_devices(struct dm_target *ti,
2509                                         iterate_devices_callout_fn fn, void *data)
2510 {
2511         struct dm_integrity_c *ic = ti->private;
2512
2513         if (!ic->meta_dev)
2514                 return fn(ti, ic->dev, ic->start + ic->initial_sectors + ic->metadata_run, ti->len, data);
2515         else
2516                 return fn(ti, ic->dev, 0, ti->len, data);
2517 }
2518
2519 static void dm_integrity_io_hints(struct dm_target *ti, struct queue_limits *limits)
2520 {
2521         struct dm_integrity_c *ic = ti->private;
2522
2523         if (ic->sectors_per_block > 1) {
2524                 limits->logical_block_size = ic->sectors_per_block << SECTOR_SHIFT;
2525                 limits->physical_block_size = ic->sectors_per_block << SECTOR_SHIFT;
2526                 blk_limits_io_min(limits, ic->sectors_per_block << SECTOR_SHIFT);
2527         }
2528 }
2529
2530 static void calculate_journal_section_size(struct dm_integrity_c *ic)
2531 {
2532         unsigned sector_space = JOURNAL_SECTOR_DATA;
2533
2534         ic->journal_sections = le32_to_cpu(ic->sb->journal_sections);
2535         ic->journal_entry_size = roundup(offsetof(struct journal_entry, last_bytes[ic->sectors_per_block]) + ic->tag_size,
2536                                          JOURNAL_ENTRY_ROUNDUP);
2537
2538         if (ic->sb->flags & cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC))
2539                 sector_space -= JOURNAL_MAC_PER_SECTOR;
2540         ic->journal_entries_per_sector = sector_space / ic->journal_entry_size;
2541         ic->journal_section_entries = ic->journal_entries_per_sector * JOURNAL_BLOCK_SECTORS;
2542         ic->journal_section_sectors = (ic->journal_section_entries << ic->sb->log2_sectors_per_block) + JOURNAL_BLOCK_SECTORS;
2543         ic->journal_entries = ic->journal_section_entries * ic->journal_sections;
2544 }
2545
2546 static int calculate_device_limits(struct dm_integrity_c *ic)
2547 {
2548         __u64 initial_sectors;
2549
2550         calculate_journal_section_size(ic);
2551         initial_sectors = SB_SECTORS + (__u64)ic->journal_section_sectors * ic->journal_sections;
2552         if (initial_sectors + METADATA_PADDING_SECTORS >= ic->meta_device_sectors || initial_sectors > UINT_MAX)
2553                 return -EINVAL;
2554         ic->initial_sectors = initial_sectors;
2555
2556         if (!ic->meta_dev) {
2557                 sector_t last_sector, last_area, last_offset;
2558
2559                 ic->metadata_run = roundup((__u64)ic->tag_size << (ic->sb->log2_interleave_sectors - ic->sb->log2_sectors_per_block),
2560                                            (__u64)(1 << SECTOR_SHIFT << METADATA_PADDING_SECTORS)) >> SECTOR_SHIFT;
2561                 if (!(ic->metadata_run & (ic->metadata_run - 1)))
2562                         ic->log2_metadata_run = __ffs(ic->metadata_run);
2563                 else
2564                         ic->log2_metadata_run = -1;
2565
2566                 get_area_and_offset(ic, ic->provided_data_sectors - 1, &last_area, &last_offset);
2567                 last_sector = get_data_sector(ic, last_area, last_offset);
2568                 if (last_sector < ic->start || last_sector >= ic->meta_device_sectors)
2569                         return -EINVAL;
2570         } else {
2571                 __u64 meta_size = ic->provided_data_sectors * ic->tag_size;
2572                 meta_size = (meta_size + ((1U << (ic->log2_buffer_sectors + SECTOR_SHIFT)) - 1))
2573                                 >> (ic->log2_buffer_sectors + SECTOR_SHIFT);
2574                 meta_size <<= ic->log2_buffer_sectors;
2575                 if (ic->initial_sectors + meta_size < ic->initial_sectors ||
2576                     ic->initial_sectors + meta_size > ic->meta_device_sectors)
2577                         return -EINVAL;
2578                 ic->metadata_run = 1;
2579                 ic->log2_metadata_run = 0;
2580         }
2581
2582         return 0;
2583 }
2584
2585 static int initialize_superblock(struct dm_integrity_c *ic, unsigned journal_sectors, unsigned interleave_sectors)
2586 {
2587         unsigned journal_sections;
2588         int test_bit;
2589
2590         memset(ic->sb, 0, SB_SECTORS << SECTOR_SHIFT);
2591         memcpy(ic->sb->magic, SB_MAGIC, 8);
2592         ic->sb->integrity_tag_size = cpu_to_le16(ic->tag_size);
2593         ic->sb->log2_sectors_per_block = __ffs(ic->sectors_per_block);
2594         if (ic->journal_mac_alg.alg_string)
2595                 ic->sb->flags |= cpu_to_le32(SB_FLAG_HAVE_JOURNAL_MAC);
2596
2597         calculate_journal_section_size(ic);
2598         journal_sections = journal_sectors / ic->journal_section_sectors;
2599         if (!journal_sections)
2600                 journal_sections = 1;
2601
2602         if (!ic->meta_dev) {
2603                 ic->sb->journal_sections = cpu_to_le32(journal_sections);
2604                 if (!interleave_sectors)
2605                         interleave_sectors = DEFAULT_INTERLEAVE_SECTORS;
2606                 ic->sb->log2_interleave_sectors = __fls(interleave_sectors);
2607                 ic->sb->log2_interleave_sectors = max((__u8)MIN_LOG2_INTERLEAVE_SECTORS, ic->sb->log2_interleave_sectors);
2608                 ic->sb->log2_interleave_sectors = min((__u8)MAX_LOG2_INTERLEAVE_SECTORS, ic->sb->log2_interleave_sectors);
2609
2610                 ic->provided_data_sectors = 0;
2611                 for (test_bit = fls64(ic->meta_device_sectors) - 1; test_bit >= 3; test_bit--) {
2612                         __u64 prev_data_sectors = ic->provided_data_sectors;
2613
2614                         ic->provided_data_sectors |= (sector_t)1 << test_bit;
2615                         if (calculate_device_limits(ic))
2616                                 ic->provided_data_sectors = prev_data_sectors;
2617                 }
2618                 if (!ic->provided_data_sectors)
2619                         return -EINVAL;
2620         } else {
2621                 ic->sb->log2_interleave_sectors = 0;
2622                 ic->provided_data_sectors = ic->data_device_sectors;
2623                 ic->provided_data_sectors &= ~(sector_t)(ic->sectors_per_block - 1);
2624
2625 try_smaller_buffer:
2626                 ic->sb->journal_sections = cpu_to_le32(0);
2627                 for (test_bit = fls(journal_sections) - 1; test_bit >= 0; test_bit--) {
2628                         __u32 prev_journal_sections = le32_to_cpu(ic->sb->journal_sections);
2629                         __u32 test_journal_sections = prev_journal_sections | (1U << test_bit);
2630                         if (test_journal_sections > journal_sections)
2631                                 continue;
2632                         ic->sb->journal_sections = cpu_to_le32(test_journal_sections);
2633                         if (calculate_device_limits(ic))
2634                                 ic->sb->journal_sections = cpu_to_le32(prev_journal_sections);
2635
2636                 }
2637                 if (!le32_to_cpu(ic->sb->journal_sections)) {
2638                         if (ic->log2_buffer_sectors > 3) {
2639                                 ic->log2_buffer_sectors--;
2640                                 goto try_smaller_buffer;
2641                         }
2642                         return -EINVAL;
2643                 }
2644         }
2645
2646         ic->sb->provided_data_sectors = cpu_to_le64(ic->provided_data_sectors);
2647
2648         sb_set_version(ic);
2649
2650         return 0;
2651 }
2652
2653 static void dm_integrity_set(struct dm_target *ti, struct dm_integrity_c *ic)
2654 {
2655         struct gendisk *disk = dm_disk(dm_table_get_md(ti->table));
2656         struct blk_integrity bi;
2657
2658         memset(&bi, 0, sizeof(bi));
2659         bi.profile = &dm_integrity_profile;
2660         bi.tuple_size = ic->tag_size;
2661         bi.tag_size = bi.tuple_size;
2662         bi.interval_exp = ic->sb->log2_sectors_per_block + SECTOR_SHIFT;
2663
2664         blk_integrity_register(disk, &bi);
2665         blk_queue_max_integrity_segments(disk->queue, UINT_MAX);
2666 }
2667
2668 static void dm_integrity_free_page_list(struct dm_integrity_c *ic, struct page_list *pl)
2669 {
2670         unsigned i;
2671
2672         if (!pl)
2673                 return;
2674         for (i = 0; i < ic->journal_pages; i++)
2675                 if (pl[i].page)
2676                         __free_page(pl[i].page);
2677         kvfree(pl);
2678 }
2679
2680 static struct page_list *dm_integrity_alloc_page_list(struct dm_integrity_c *ic)
2681 {
2682         size_t page_list_desc_size = ic->journal_pages * sizeof(struct page_list);
2683         struct page_list *pl;
2684         unsigned i;
2685
2686         pl = kvmalloc(page_list_desc_size, GFP_KERNEL | __GFP_ZERO);
2687         if (!pl)
2688                 return NULL;
2689
2690         for (i = 0; i < ic->journal_pages; i++) {
2691                 pl[i].page = alloc_page(GFP_KERNEL);
2692                 if (!pl[i].page) {
2693                         dm_integrity_free_page_list(ic, pl);
2694                         return NULL;
2695                 }
2696                 if (i)
2697                         pl[i - 1].next = &pl[i];
2698         }
2699
2700         return pl;
2701 }
2702
2703 static void dm_integrity_free_journal_scatterlist(struct dm_integrity_c *ic, struct scatterlist **sl)
2704 {
2705         unsigned i;
2706         for (i = 0; i < ic->journal_sections; i++)
2707                 kvfree(sl[i]);
2708         kvfree(sl);
2709 }
2710
2711 static struct scatterlist **dm_integrity_alloc_journal_scatterlist(struct dm_integrity_c *ic, struct page_list *pl)
2712 {
2713         struct scatterlist **sl;
2714         unsigned i;
2715
2716         sl = kvmalloc_array(ic->journal_sections,
2717                             sizeof(struct scatterlist *),
2718                             GFP_KERNEL | __GFP_ZERO);
2719         if (!sl)
2720                 return NULL;
2721
2722         for (i = 0; i < ic->journal_sections; i++) {
2723                 struct scatterlist *s;
2724                 unsigned start_index, start_offset;
2725                 unsigned end_index, end_offset;
2726                 unsigned n_pages;
2727                 unsigned idx;
2728
2729                 page_list_location(ic, i, 0, &start_index, &start_offset);
2730                 page_list_location(ic, i, ic->journal_section_sectors - 1, &end_index, &end_offset);
2731
2732                 n_pages = (end_index - start_index + 1);
2733
2734                 s = kvmalloc_array(n_pages, sizeof(struct scatterlist),
2735                                    GFP_KERNEL);
2736                 if (!s) {
2737                         dm_integrity_free_journal_scatterlist(ic, sl);
2738                         return NULL;
2739                 }
2740
2741                 sg_init_table(s, n_pages);
2742                 for (idx = start_index; idx <= end_index; idx++) {
2743                         char *va = lowmem_page_address(pl[idx].page);
2744                         unsigned start = 0, end = PAGE_SIZE;
2745                         if (idx == start_index)
2746                                 start = start_offset;
2747                         if (idx == end_index)
2748                                 end = end_offset + (1 << SECTOR_SHIFT);
2749                         sg_set_buf(&s[idx - start_index], va + start, end - start);
2750                 }
2751
2752                 sl[i] = s;
2753         }
2754
2755         return sl;
2756 }
2757
2758 static void free_alg(struct alg_spec *a)
2759 {
2760         kzfree(a->alg_string);
2761         kzfree(a->key);
2762         memset(a, 0, sizeof *a);
2763 }
2764
2765 static int get_alg_and_key(const char *arg, struct alg_spec *a, char **error, char *error_inval)
2766 {
2767         char *k;
2768
2769         free_alg(a);
2770
2771         a->alg_string = kstrdup(strchr(arg, ':') + 1, GFP_KERNEL);
2772         if (!a->alg_string)
2773                 goto nomem;
2774
2775         k = strchr(a->alg_string, ':');
2776         if (k) {
2777                 *k = 0;
2778                 a->key_string = k + 1;
2779                 if (strlen(a->key_string) & 1)
2780                         goto inval;
2781
2782                 a->key_size = strlen(a->key_string) / 2;
2783                 a->key = kmalloc(a->key_size, GFP_KERNEL);
2784                 if (!a->key)
2785                         goto nomem;
2786                 if (hex2bin(a->key, a->key_string, a->key_size))
2787                         goto inval;
2788         }
2789
2790         return 0;
2791 inval:
2792         *error = error_inval;
2793         return -EINVAL;
2794 nomem:
2795         *error = "Out of memory for an argument";
2796         return -ENOMEM;
2797 }
2798
2799 static int get_mac(struct crypto_shash **hash, struct alg_spec *a, char **error,
2800                    char *error_alg, char *error_key)
2801 {
2802         int r;
2803
2804         if (a->alg_string) {
2805                 *hash = crypto_alloc_shash(a->alg_string, 0, 0);
2806                 if (IS_ERR(*hash)) {
2807                         *error = error_alg;
2808                         r = PTR_ERR(*hash);
2809                         *hash = NULL;
2810                         return r;
2811                 }
2812
2813                 if (a->key) {
2814                         r = crypto_shash_setkey(*hash, a->key, a->key_size);
2815                         if (r) {
2816                                 *error = error_key;
2817                                 return r;
2818                         }
2819                 } else if (crypto_shash_get_flags(*hash) & CRYPTO_TFM_NEED_KEY) {
2820                         *error = error_key;
2821                         return -ENOKEY;
2822                 }
2823         }
2824
2825         return 0;
2826 }
2827
2828 static int create_journal(struct dm_integrity_c *ic, char **error)
2829 {
2830         int r = 0;
2831         unsigned i;
2832         __u64 journal_pages, journal_desc_size, journal_tree_size;
2833         unsigned char *crypt_data = NULL, *crypt_iv = NULL;
2834         struct skcipher_request *req = NULL;
2835
2836         ic->commit_ids[0] = cpu_to_le64(0x1111111111111111ULL);
2837         ic->commit_ids[1] = cpu_to_le64(0x2222222222222222ULL);
2838         ic->commit_ids[2] = cpu_to_le64(0x3333333333333333ULL);
2839         ic->commit_ids[3] = cpu_to_le64(0x4444444444444444ULL);
2840
2841         journal_pages = roundup((__u64)ic->journal_sections * ic->journal_section_sectors,
2842                                 PAGE_SIZE >> SECTOR_SHIFT) >> (PAGE_SHIFT - SECTOR_SHIFT);
2843         journal_desc_size = journal_pages * sizeof(struct page_list);
2844         if (journal_pages >= totalram_pages() - totalhigh_pages() || journal_desc_size > ULONG_MAX) {
2845                 *error = "Journal doesn't fit into memory";
2846                 r = -ENOMEM;
2847                 goto bad;
2848         }
2849         ic->journal_pages = journal_pages;
2850
2851         ic->journal = dm_integrity_alloc_page_list(ic);
2852         if (!ic->journal) {
2853                 *error = "Could not allocate memory for journal";
2854                 r = -ENOMEM;
2855                 goto bad;
2856         }
2857         if (ic->journal_crypt_alg.alg_string) {
2858                 unsigned ivsize, blocksize;
2859                 struct journal_completion comp;
2860
2861                 comp.ic = ic;
2862                 ic->journal_crypt = crypto_alloc_skcipher(ic->journal_crypt_alg.alg_string, 0, 0);
2863                 if (IS_ERR(ic->journal_crypt)) {
2864                         *error = "Invalid journal cipher";
2865                         r = PTR_ERR(ic->journal_crypt);
2866                         ic->journal_crypt = NULL;
2867                         goto bad;
2868                 }
2869                 ivsize = crypto_skcipher_ivsize(ic->journal_crypt);
2870                 blocksize = crypto_skcipher_blocksize(ic->journal_crypt);
2871
2872                 if (ic->journal_crypt_alg.key) {
2873                         r = crypto_skcipher_setkey(ic->journal_crypt, ic->journal_crypt_alg.key,
2874                                                    ic->journal_crypt_alg.key_size);
2875                         if (r) {
2876                                 *error = "Error setting encryption key";
2877                                 goto bad;
2878                         }
2879                 }
2880                 DEBUG_print("cipher %s, block size %u iv size %u\n",
2881                             ic->journal_crypt_alg.alg_string, blocksize, ivsize);
2882
2883                 ic->journal_io = dm_integrity_alloc_page_list(ic);
2884                 if (!ic->journal_io) {
2885                         *error = "Could not allocate memory for journal io";
2886                         r = -ENOMEM;
2887                         goto bad;
2888                 }
2889
2890                 if (blocksize == 1) {
2891                         struct scatterlist *sg;
2892
2893                         req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
2894                         if (!req) {
2895                                 *error = "Could not allocate crypt request";
2896                                 r = -ENOMEM;
2897                                 goto bad;
2898                         }
2899
2900                         crypt_iv = kmalloc(ivsize, GFP_KERNEL);
2901                         if (!crypt_iv) {
2902                                 *error = "Could not allocate iv";
2903                                 r = -ENOMEM;
2904                                 goto bad;
2905                         }
2906
2907                         ic->journal_xor = dm_integrity_alloc_page_list(ic);
2908                         if (!ic->journal_xor) {
2909                                 *error = "Could not allocate memory for journal xor";
2910                                 r = -ENOMEM;
2911                                 goto bad;
2912                         }
2913
2914                         sg = kvmalloc_array(ic->journal_pages + 1,
2915                                             sizeof(struct scatterlist),
2916                                             GFP_KERNEL);
2917                         if (!sg) {
2918                                 *error = "Unable to allocate sg list";
2919                                 r = -ENOMEM;
2920                                 goto bad;
2921                         }
2922                         sg_init_table(sg, ic->journal_pages + 1);
2923                         for (i = 0; i < ic->journal_pages; i++) {
2924                                 char *va = lowmem_page_address(ic->journal_xor[i].page);
2925                                 clear_page(va);
2926                                 sg_set_buf(&sg[i], va, PAGE_SIZE);
2927                         }
2928                         sg_set_buf(&sg[i], &ic->commit_ids, sizeof ic->commit_ids);
2929                         memset(crypt_iv, 0x00, ivsize);
2930
2931                         skcipher_request_set_crypt(req, sg, sg, PAGE_SIZE * ic->journal_pages + sizeof ic->commit_ids, crypt_iv);
2932                         init_completion(&comp.comp);
2933                         comp.in_flight = (atomic_t)ATOMIC_INIT(1);
2934                         if (do_crypt(true, req, &comp))
2935                                 wait_for_completion(&comp.comp);
2936                         kvfree(sg);
2937                         r = dm_integrity_failed(ic);
2938                         if (r) {
2939                                 *error = "Unable to encrypt journal";
2940                                 goto bad;
2941                         }
2942                         DEBUG_bytes(lowmem_page_address(ic->journal_xor[0].page), 64, "xor data");
2943
2944                         crypto_free_skcipher(ic->journal_crypt);
2945                         ic->journal_crypt = NULL;
2946                 } else {
2947                         unsigned crypt_len = roundup(ivsize, blocksize);
2948
2949                         req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
2950                         if (!req) {
2951                                 *error = "Could not allocate crypt request";
2952                                 r = -ENOMEM;
2953                                 goto bad;
2954                         }
2955
2956                         crypt_iv = kmalloc(ivsize, GFP_KERNEL);
2957                         if (!crypt_iv) {
2958                                 *error = "Could not allocate iv";
2959                                 r = -ENOMEM;
2960                                 goto bad;
2961                         }
2962
2963                         crypt_data = kmalloc(crypt_len, GFP_KERNEL);
2964                         if (!crypt_data) {
2965                                 *error = "Unable to allocate crypt data";
2966                                 r = -ENOMEM;
2967                                 goto bad;
2968                         }
2969
2970                         ic->journal_scatterlist = dm_integrity_alloc_journal_scatterlist(ic, ic->journal);
2971                         if (!ic->journal_scatterlist) {
2972                                 *error = "Unable to allocate sg list";
2973                                 r = -ENOMEM;
2974                                 goto bad;
2975                         }
2976                         ic->journal_io_scatterlist = dm_integrity_alloc_journal_scatterlist(ic, ic->journal_io);
2977                         if (!ic->journal_io_scatterlist) {
2978                                 *error = "Unable to allocate sg list";
2979                                 r = -ENOMEM;
2980                                 goto bad;
2981                         }
2982                         ic->sk_requests = kvmalloc_array(ic->journal_sections,
2983                                                          sizeof(struct skcipher_request *),
2984                                                          GFP_KERNEL | __GFP_ZERO);
2985                         if (!ic->sk_requests) {
2986                                 *error = "Unable to allocate sk requests";
2987                                 r = -ENOMEM;
2988                                 goto bad;
2989                         }
2990                         for (i = 0; i < ic->journal_sections; i++) {
2991                                 struct scatterlist sg;
2992                                 struct skcipher_request *section_req;
2993                                 __u32 section_le = cpu_to_le32(i);
2994
2995                                 memset(crypt_iv, 0x00, ivsize);
2996                                 memset(crypt_data, 0x00, crypt_len);
2997                                 memcpy(crypt_data, &section_le, min((size_t)crypt_len, sizeof(section_le)));
2998
2999                                 sg_init_one(&sg, crypt_data, crypt_len);
3000                                 skcipher_request_set_crypt(req, &sg, &sg, crypt_len, crypt_iv);
3001                                 init_completion(&comp.comp);
3002                                 comp.in_flight = (atomic_t)ATOMIC_INIT(1);
3003                                 if (do_crypt(true, req, &comp))
3004                                         wait_for_completion(&comp.comp);
3005
3006                                 r = dm_integrity_failed(ic);
3007                                 if (r) {
3008                                         *error = "Unable to generate iv";
3009                                         goto bad;
3010                                 }
3011
3012                                 section_req = skcipher_request_alloc(ic->journal_crypt, GFP_KERNEL);
3013                                 if (!section_req) {
3014                                         *error = "Unable to allocate crypt request";
3015                                         r = -ENOMEM;
3016                                         goto bad;
3017                                 }
3018                                 section_req->iv = kmalloc_array(ivsize, 2,
3019                                                                 GFP_KERNEL);
3020                                 if (!section_req->iv) {
3021                                         skcipher_request_free(section_req);
3022                                         *error = "Unable to allocate iv";
3023                                         r = -ENOMEM;
3024                                         goto bad;
3025                                 }
3026                                 memcpy(section_req->iv + ivsize, crypt_data, ivsize);
3027                                 section_req->cryptlen = (size_t)ic->journal_section_sectors << SECTOR_SHIFT;
3028                                 ic->sk_requests[i] = section_req;
3029                                 DEBUG_bytes(crypt_data, ivsize, "iv(%u)", i);
3030                         }
3031                 }
3032         }
3033
3034         for (i = 0; i < N_COMMIT_IDS; i++) {
3035                 unsigned j;
3036 retest_commit_id:
3037                 for (j = 0; j < i; j++) {
3038                         if (ic->commit_ids[j] == ic->commit_ids[i]) {
3039                                 ic->commit_ids[i] = cpu_to_le64(le64_to_cpu(ic->commit_ids[i]) + 1);
3040                                 goto retest_commit_id;
3041                         }
3042                 }
3043                 DEBUG_print("commit id %u: %016llx\n", i, ic->commit_ids[i]);
3044         }
3045
3046         journal_tree_size = (__u64)ic->journal_entries * sizeof(struct journal_node);
3047         if (journal_tree_size > ULONG_MAX) {
3048                 *error = "Journal doesn't fit into memory";
3049                 r = -ENOMEM;
3050                 goto bad;
3051         }
3052         ic->journal_tree = kvmalloc(journal_tree_size, GFP_KERNEL);
3053         if (!ic->journal_tree) {
3054                 *error = "Could not allocate memory for journal tree";
3055                 r = -ENOMEM;
3056         }
3057 bad:
3058         kfree(crypt_data);
3059         kfree(crypt_iv);
3060         skcipher_request_free(req);
3061
3062         return r;
3063 }
3064
3065 /*
3066  * Construct a integrity mapping
3067  *
3068  * Arguments:
3069  *      device
3070  *      offset from the start of the device
3071  *      tag size
3072  *      D - direct writes, J - journal writes, R - recovery mode
3073  *      number of optional arguments
3074  *      optional arguments:
3075  *              journal_sectors
3076  *              interleave_sectors
3077  *              buffer_sectors
3078  *              journal_watermark
3079  *              commit_time
3080  *              internal_hash
3081  *              journal_crypt
3082  *              journal_mac
3083  *              block_size
3084  */
3085 static int dm_integrity_ctr(struct dm_target *ti, unsigned argc, char **argv)
3086 {
3087         struct dm_integrity_c *ic;
3088         char dummy;
3089         int r;
3090         unsigned extra_args;
3091         struct dm_arg_set as;
3092         static const struct dm_arg _args[] = {
3093                 {0, 9, "Invalid number of feature args"},
3094         };
3095         unsigned journal_sectors, interleave_sectors, buffer_sectors, journal_watermark, sync_msec;
3096         bool recalculate;
3097         bool should_write_sb;
3098         __u64 threshold;
3099         unsigned long long start;
3100
3101 #define DIRECT_ARGUMENTS        4
3102
3103         if (argc <= DIRECT_ARGUMENTS) {
3104                 ti->error = "Invalid argument count";
3105                 return -EINVAL;
3106         }
3107
3108         ic = kzalloc(sizeof(struct dm_integrity_c), GFP_KERNEL);
3109         if (!ic) {
3110                 ti->error = "Cannot allocate integrity context";
3111                 return -ENOMEM;
3112         }
3113         ti->private = ic;
3114         ti->per_io_data_size = sizeof(struct dm_integrity_io);
3115
3116         ic->in_progress = RB_ROOT;
3117         INIT_LIST_HEAD(&ic->wait_list);
3118         init_waitqueue_head(&ic->endio_wait);
3119         bio_list_init(&ic->flush_bio_list);
3120         init_waitqueue_head(&ic->copy_to_journal_wait);
3121         init_completion(&ic->crypto_backoff);
3122         atomic64_set(&ic->number_of_mismatches, 0);
3123
3124         r = dm_get_device(ti, argv[0], dm_table_get_mode(ti->table), &ic->dev);
3125         if (r) {
3126                 ti->error = "Device lookup failed";
3127                 goto bad;
3128         }
3129
3130         if (sscanf(argv[1], "%llu%c", &start, &dummy) != 1 || start != (sector_t)start) {
3131                 ti->error = "Invalid starting offset";
3132                 r = -EINVAL;
3133                 goto bad;
3134         }
3135         ic->start = start;
3136
3137         if (strcmp(argv[2], "-")) {
3138                 if (sscanf(argv[2], "%u%c", &ic->tag_size, &dummy) != 1 || !ic->tag_size) {
3139                         ti->error = "Invalid tag size";
3140                         r = -EINVAL;
3141                         goto bad;
3142                 }
3143         }
3144
3145         if (!strcmp(argv[3], "J") || !strcmp(argv[3], "D") || !strcmp(argv[3], "R"))
3146                 ic->mode = argv[3][0];
3147         else {
3148                 ti->error = "Invalid mode (expecting J, D, R)";
3149                 r = -EINVAL;
3150                 goto bad;
3151         }
3152
3153         journal_sectors = 0;
3154         interleave_sectors = DEFAULT_INTERLEAVE_SECTORS;
3155         buffer_sectors = DEFAULT_BUFFER_SECTORS;
3156         journal_watermark = DEFAULT_JOURNAL_WATERMARK;
3157         sync_msec = DEFAULT_SYNC_MSEC;
3158         recalculate = false;
3159         ic->sectors_per_block = 1;
3160
3161         as.argc = argc - DIRECT_ARGUMENTS;
3162         as.argv = argv + DIRECT_ARGUMENTS;
3163         r = dm_read_arg_group(_args, &as, &extra_args, &ti->error);
3164         if (r)
3165                 goto bad;
3166
3167         while (extra_args--) {
3168                 const char *opt_string;
3169                 unsigned val;
3170                 opt_string = dm_shift_arg(&as);
3171                 if (!opt_string) {
3172                         r = -EINVAL;
3173                         ti->error = "Not enough feature arguments";
3174                         goto bad;
3175                 }
3176                 if (sscanf(opt_string, "journal_sectors:%u%c", &val, &dummy) == 1)
3177                         journal_sectors = val ? val : 1;
3178                 else if (sscanf(opt_string, "interleave_sectors:%u%c", &val, &dummy) == 1)
3179                         interleave_sectors = val;
3180                 else if (sscanf(opt_string, "buffer_sectors:%u%c", &val, &dummy) == 1)
3181                         buffer_sectors = val;
3182                 else if (sscanf(opt_string, "journal_watermark:%u%c", &val, &dummy) == 1 && val <= 100)
3183                         journal_watermark = val;
3184                 else if (sscanf(opt_string, "commit_time:%u%c", &val, &dummy) == 1)
3185                         sync_msec = val;
3186                 else if (!strncmp(opt_string, "meta_device:", strlen("meta_device:"))) {
3187                         if (ic->meta_dev) {
3188                                 dm_put_device(ti, ic->meta_dev);
3189                 &nbs